From 79647922dc149b4464325303f30c93a115f17e3a Mon Sep 17 00:00:00 2001 From: lishengzxc <306009337@qq.com> Date: Tue, 28 Jul 2026 10:16:09 +0800 Subject: [PATCH] chore: update bailian-docs-llm-wiki (2026-07-28) --- skills/bailian-docs-llm-wiki/SKILL.md | 27 +- skills/bailian-docs-llm-wiki/llms.txt | 1196 +++-- .../models/families.jsonl | 10 +- .../groups/kimi-models-market-place.json | 203 +- .../groups/qwen-audio-realtime-flash.json | 8 +- .../groups/qwen-audio-realtime-plus.json | 8 +- .../models/groups/qwen-embedding.json | 183 +- .../models/groups/qwen-image-3.0-pro.json | 78 + .../models/groups/qwen3.7-flash.json | 406 ++ .../bailian-docs-llm-wiki/models/index.json | 67 +- skills/bailian-docs-llm-wiki/models/index.md | 14 +- .../bailian-docs-llm-wiki/models/models.jsonl | 28 +- .../api-bailian-2023-12-29-createindex.md | 4 +- ...bailian-2023-12-29-listindexfiledetails.md | 75 +- .../api-bailian-2023-12-29-addchunk.md | 233 + .../api-bailian-2023-12-29-overview.md | 6 + .../api-bailian-2023-12-29-ram.md | 764 +++- .../multimodal-error-code.md | 4 +- .../multimodal-interaction-protocol.md | 42 +- .../multimodal-timbre-list.md | 140 +- .../multimodal-products/product-billing.md | 2 +- ...e-ram-users-for-ccai-dialogue-analysis.md} | 0 ...i-bailianvoicebot-2025-01-01-listvoices.md | 349 ++ .../web-page-translation-jssdk.md | 748 +++- ...ightapp-2024-08-01-getvideoanalysistask.md | 29 +- .../api-aimiaobi-2023-08-01-overview.md | 36 +- .../application-after-sales-service-scope.md | 57 + .../application-related-agreements.md | 2 +- .../billing-for-knowledge-base.md | 92 +- .../rag-knowledge-base-specifications.md | 13 +- .../knowledge-base/rag-knowledge-base.md | 42 +- .../knowledge-base/rag-optimization.md | 100 +- .../application-introduction.md | 7 +- .../new-single-agent-application.md | 16 +- .../llm-application/rich-code-application.md | 10 +- .../managed-agents/managed-agents-context.md | 2 - .../tripo-3d-generation-api-reference.md | 23 +- .../fun-asr-realtime-python-sdk.md | 10 +- ...recorded-speech-recognition-android-sdk.md | 12 +- ...sr-recorded-speech-recognition-http-api.md | 922 +--- ...asr-recorded-speech-recognition-ios-sdk.md | 10 +- ...sr-recorded-speech-recognition-java-sdk.md | 28 +- ...-recorded-speech-recognition-python-sdk.md | 28 +- ...me-speech-recognition-for-fun-asr-flash.md | 561 +++ ...speech-recognition-for-fun-asr-realtime.md | 328 ++ ...r-real-time-speech-recognition-java-sdk.md | 20 +- ...real-time-speech-recognition-python-sdk.md | 30 +- ...recorded-speech-recognition-android-sdk.md | 15 +- ...mer-recorded-speech-recognition-ios-sdk.md | 10 +- ...er-recorded-speech-recognition-java-sdk.md | 28 +- ...-recorded-speech-recognition-python-sdk.md | 28 +- ...recorded-speech-recognition-restful-api.md | 20 +- .../qwen-asr-api-reference.md | 86 +- .../qwen-asr-realtime-interaction-process.md | 8 +- .../cosyvoice-android-sdk.md | 15 +- .../cosyvoice-client-events.md | 49 +- .../cosyvoice-ios-sdk.md | 14 +- .../cosyvoice-java-sdk.md | 59 +- .../cosyvoice-python-sdk.md | 77 +- .../cosyvoice-websocket-api.md | 2 +- .../cosyvoice-tts-http-api.md | 20 +- .../cosyvoice-tts-java-sdk.md | 66 +- .../cosyvoice-tts-python-sdk.md | 20 +- .../qwen-tts-api.md | 140 +- .../voice-clone-design-http-api.md | 18 +- .../sound-reengraving/voice-clone-java-sdk.md | 109 +- .../voice-clone-python-sdk.md | 18 +- ...> fun-audiochat-realtime-websocket-api.md} | 14 +- .../creative-poster-generation-api.md | 1 - ...anx-background-generation-api-reference.md | 12 +- ...ge-generation-and-editing-api-reference.md | 459 ++ .../omni-realtime-api/client-events.md | 6 +- .../omni-realtime-java-sdk.md | 6 +- .../omni-realtime-python-sdk.md | 18 +- .../qwen-omni-voice-cloning.md | 102 +- .../omni-realtime-api/server-events.md | 4 +- .../preparations/error-code.md | 35 +- ...batch-interfaces-compatible-with-openai.md | 32 +- ...compatibility-with-openai-responses-api.md | 4 +- .../toolkits-and-frameworks/completions.md | 14 +- .../openai-compatible-batch-chat.md | 6 +- .../openai-compatible-conversations.md | 6 +- .../text-embedding-batch-api.md | 32 +- .../multimodal-embedding-api-reference.md | 24 +- ...happyhorse-image-to-video-api-reference.md | 31 +- ...yhorse-reference-to-video-api-reference.md | 25 +- .../happyhorse-text-to-video-api-reference.md | 25 +- .../happyhorse-video-edit-api-reference.md | 27 +- .../kling-video-generation-api-reference.md | 23 +- .../pixverse-image-to-video-api-reference.md | 25 +- ...ixverse-keyframe-to-video-api-reference.md | 19 +- .../pixverse-lipsync-api-reference.md | 12 +- .../pixverse-motioncontrol-api-reference.md | 9 +- ...xverse-reference-to-video-api-reference.md | 14 +- .../pixverse-text-to-video-api-reference.md | 43 +- .../pixverse-upscale-api-reference.md | 9 +- .../vidu-image-to-video-api-reference.md | 12 +- .../vidu-keyframe-to-video-api-reference.md | 12 +- .../vidu-reference-to-video-api-reference.md | 24 +- .../vidu-text-to-video-api-reference.md | 32 +- .../image-to-video-general-api-reference.md | 92 +- .../legacy-image-to-video-api-reference.md | 95 +- ...o-by-first-and-last-frame-api-reference.md | 55 +- ...cy-wan-reference-to-video-api-reference.md | 65 +- .../legacy-wan-text-to-video-api-reference.md | 76 +- .../legacy-wanx-vace-api-reference.md | 41 +- .../text-to-video-api-reference.md | 2 +- .../wan-api-reference/wan-animate-mix-api.md | 10 +- .../wan-api-reference/wan-animate-move-api.md | 4 +- .../wan-video-editing-api-reference.md | 49 +- .../fine-tuning-api-guide.md | 10 +- .../model-training-on-console.md | 20 +- .../model-training-overview.md | 540 ++- .../fine-tuning/rl-training-overview.md | 448 ++ .../wan-image-generation-finetune-guide.md | 138 + .../wan-video-generation-finetune-guide.md | 148 +- .../get-started-with-models/models.md | 288 +- .../get-started-with-models/rate-limit.md | 1210 +---- .../get-started-with-models/regions.md | 70 +- .../what-is-model-studio.md | 2 +- .../model-data-overview/model-log-backflow.md | 330 ++ .../model-deployment-introduction.md | 290 +- .../model-deployment-1/model-import.md | 465 +- .../ptu-long-input-and-cache.md | 30 +- .../evaluation-metrics.md | 12 +- .../model-experience/image-model.md | 22 +- .../model-experience/s2s-model.md | 116 +- .../model-experience/text-generation-model.md | 30 +- .../model-experience/tts-model.md | 10 +- .../video-generate-edit-model.md | 14 - .../model-experience/vision-model.md | 91 +- .../tpm-reservation.md | 25 +- .../model-monitoring/model-telemetry.md | 54 +- .../realtime-api-aoq-sdk-desc.md | 17 + .../aoq-audio-features.md | 1179 +++++ .../aoq-connection-management.md | 171 + .../aoq-custom-audio-capture.md | 253 ++ .../aoq-custom-audio-playback.md | 235 + .../aoq-custom-video-input.md | 647 +++ .../aoq-media-stream-control.md | 115 + .../aoq-video-features.md | 1186 +++++ .../best-practice-aoq-omni-realtime.md | 520 +++ .../best-practice-webrtc-multimodal-dialog.md | 512 +++ .../best-practice-webrtc-omni-realtime.md | 388 ++ .../realtime-api-overview.md | 180 + .../realtime-connect-model.md | 291 ++ .../realtime-sdk-download.md | 60 + .../realtime-token-authentication.md | 358 ++ .../release-notes/model-release-notes.md | 26 +- .../release-notes/newly-released-models.md | 648 +-- .../content-security.md | 35 + .../permission-management-overview.md | 45 +- .../support/after-sales-service-scope.md | 57 + .../faq-about-alibaba-cloud-model-studio.md | 4 + .../model-list-3d-generation/tripo-h3-1.md | 207 + .../model-list-3d-generation/tripo-p1-0.md | 153 + .../gte-rerank-v2.md | 109 + .../multimodal-embedding-v1.md | 115 + .../qwen2-5-vl-embedding.md | 115 + .../qwen3-7-text-embedding.md | 109 + .../qwen3-rerank.md | 141 + .../qwen3-vl-embedding.md | 115 + .../qwen3-vl-rerank.md | 115 + .../text-embedding-async-v1.md | 93 + .../text-embedding-async-v2.md | 93 + .../text-embedding-v1.md | 115 + .../text-embedding-v2.md | 115 + .../text-embedding-v3.md | 147 + .../text-embedding-v4.md | 147 + .../tongyi-embedding-vision-flash.md | 271 ++ .../tongyi-embedding-vision-plus.md | 271 ++ .../aitryon-parsing-v1.md | 93 + .../aitryon-plus.md | 93 + .../aitryon-refiner.md | 179 + .../model-list-image-generation/aitryon.md | 93 + .../facechain-facedetect.md | 79 + .../image-erase-completion.md | 79 + .../image-out-painting.md | 79 + .../kling-v3-image-generation.md | 111 + .../kling-v3-omni-image-generation.md | 117 + .../model-facechain-generation.md | 93 + .../model-image-instance-segmentation.md | 79 + .../model-qwen-image-edit.md | 133 + .../qwen-image-2-0-pro.md | 537 +++ .../qwen-image-2-0.md | 269 ++ .../qwen-image-3-0-pro.md | 103 + .../qwen-image-edit-max.md | 269 ++ .../qwen-image-edit-plus.md | 403 ++ .../qwen-image-max.md | 253 ++ .../qwen-image-plus.md | 273 ++ .../model-list-image-generation/qwen-image.md | 137 + .../qwen-mt-image.md | 105 + .../shoemodel-v1.md | 79 + .../vidu-image-reference2image.md | 117 + .../viduq2-fast-reference2image.md | 105 + .../viduq2-pro-reference2image.md | 117 + .../viduq3-fast-reference2image.md | 117 + .../virtualmodel-v2.md | 79 + .../wan2-1-t2i-plus.md | 109 + .../wan2-1-t2i-turbo.md | 109 + .../wan2-2-t2i-flash.md | 133 + .../wan2-2-t2i-plus.md | 133 + .../model-list-image-generation/wan2-5-i2i.md | 269 ++ .../model-list-image-generation/wan2-5-t2i.md | 269 ++ .../wan2-6-image.md | 189 + .../model-list-image-generation/wan2-6-t2i.md | 189 + .../wan2-7-image-pro.md | 133 + .../wan2-7-image.md | 161 + .../wanx-background-generation-v2.md | 93 + .../wanx-poster-generation-v1.md | 79 + .../wanx-sketch-to-image-lite.md | 93 + .../wanx-style-repaint-v1.md | 93 + .../model-list-image-generation/wanx-v1.md | 93 + .../wanx-virtualmodel.md | 79 + .../wanx-x-painting.md | 79 + .../wanx2-0-t2i-turbo.md | 105 + .../wanx2-1-imageedit.md | 93 + .../wanx2-1-t2i-plus.md | 105 + .../wanx2-1-t2i-turbo.md | 105 + .../wordart-semantic.md | 93 + .../wordart-texture.md | 93 + .../z-image-turbo.md | 145 + .../qwen-omni-turbo-realtime.md | 609 +++ .../model-list-omni/qwen-omni-turbo.md | 821 ++++ .../model-list-omni/qwen2-5-omni-7b.md | 185 + .../qwen3-5-omni-flash-realtime.md | 357 ++ .../model-list-omni/qwen3-5-omni-flash.md | 357 ++ .../qwen3-5-omni-plus-realtime.md | 357 ++ .../model-list-omni/qwen3-5-omni-plus.md | 393 ++ .../qwen3-omni-flash-realtime.md | 561 +++ .../model-list-omni/qwen3-omni-flash.md | 747 ++++ .../fun-asr-flash-8k-realtime.md | 213 + .../fun-asr-flash.md | 269 ++ .../fun-asr-mtl.md | 237 + .../fun-asr-realtime.md | 449 ++ .../model-list-speech-recognition/fun-asr.md | 403 ++ .../gummy-chat-v1.md | 105 + .../gummy-realtime-v1.md | 105 + ...odel-qwen3-livetranslate-flash-realtime.md | 357 ++ .../model-qwen3-livetranslate-flash.md | 289 ++ .../paraformer-8k-v1.md | 93 + .../paraformer-8k-v2.md | 93 + .../paraformer-mtl-v1.md | 93 + .../paraformer-realtime-8k-v1.md | 93 + .../paraformer-realtime-8k-v2.md | 93 + .../paraformer-realtime-v1.md | 93 + .../paraformer-realtime-v2.md | 93 + .../paraformer-v1.md | 93 + .../paraformer-v2.md | 93 + .../model-list-speech-recognition/qwen3-6.md | 357 ++ .../qwen3-asr-flash-filetrans.md | 269 ++ .../qwen3-asr-flash-realtime.md | 403 ++ .../qwen3-asr-flash-us.md | 193 + .../qwen3-asr-flash.md | 403 ++ .../qwen3-omni-30b-a3b-captioner.md | 153 + .../speech-biasing.md | 91 + .../cosyvoice-clone-v1.md | 93 + .../cosyvoice-v1.md | 93 + .../cosyvoice-v2.md | 93 + .../cosyvoice-v3-5-flash.md | 105 + .../cosyvoice-v3-5-plus.md | 105 + .../cosyvoice-v3-flash.md | 133 + .../cosyvoice-v3-plus.md | 133 + .../fun-music-v1.md | 105 + .../model-qwen-tts-realtime.md | 351 ++ .../model-qwen-tts.md | 467 ++ .../music-generation-preview.md | 213 + .../qwen-voice-design.md | 133 + .../qwen-voice-enrollment.md | 133 + .../qwen3-tts-flash-realtime.md | 403 ++ .../qwen3-tts-flash.md | 403 ++ .../qwen3-tts-instruct-flash-realtime.md | 269 ++ .../qwen3-tts-instruct-flash.md | 269 ++ .../qwen3-tts-vc-realtime.md | 403 ++ .../qwen3-tts-vc.md | 269 ++ .../qwen3-tts-vd-realtime.md | 403 ++ .../qwen3-tts-vd.md | 269 ++ .../model-list-speech-synthesis/sambert.md | 267 ++ .../speech-02-hd.md | 115 + .../speech-02-turbo.md | 115 + .../speech-2-8-hd.md | 115 + .../speech-2-8-turbo.md | 115 + .../voice-enrollment.md | 103 + .../deepseek-ocr.md | 115 + .../deepseek-r1-distill-qwen-1-5b.md | 95 + .../deepseek-r1-distill-qwen-14b.md | 115 + .../deepseek-r1-distill-qwen-32b.md | 115 + .../deepseek-r1-distill-qwen-7b.md | 115 + .../model-list-text-generation/deepseek-r1.md | 509 +++ .../deepseek-v3-1-terminus.md | 257 ++ .../deepseek-v3-1.md | 121 + .../deepseek-v3-2-exp.md | 115 + .../deepseek-v3-2-think.md | 129 + .../deepseek-v3-3.md | 341 ++ .../model-list-text-generation/deepseek-v3.md | 371 ++ .../deepseek-v4-flash-us.md | 107 + .../deepseek-v4-flash.md | 297 ++ .../deepseek-v4-pro-us.md | 107 + .../deepseek-v4-pro.md | 427 ++ .../model-list-text-generation/farui-plus.md | 115 + .../model-list-text-generation/glm-4-5-air.md | 137 + .../model-list-text-generation/glm-4-6.md | 161 + .../model-list-text-generation/glm-4-7.md | 149 + .../model-list-text-generation/glm-4.md | 137 + .../model-list-text-generation/glm-5-1.md | 647 +++ .../glm-5-2-fast.md | 349 ++ .../model-list-text-generation/glm-5-2-us.md | 119 + .../model-list-text-generation/glm-5-2.md | 399 ++ .../model-list-text-generation/glm-9.md | 295 ++ .../model-list-text-generation/kimi-k2-5.md | 379 ++ .../model-list-text-generation/kimi-k2-6.md | 267 ++ .../kimi-k2-7-code-highspeed.md | 133 + .../kimi-k2-7-code.md | 423 ++ .../kimi-k2-thinking.md | 121 + .../model-list-text-generation/kimi-k3.md | 133 + .../mimo-v2-5-pro.md | 133 + .../minimax-m2-5.md | 255 ++ .../minimax-m2-7.md | 133 + .../model-list-text-generation/minimax-m2.md | 239 + .../model-list-text-generation/minimax-m3.md | 125 + .../model-qwen-deep-research.md | 233 + .../model-qwen3-max.md | 2039 +++++++++ .../moonshot-kimi-k2-instruct.md | 121 + .../qwen-coder-plus.md | 115 + .../qwen-coder-turbo.md | 115 + .../qwen-doc-turbo.md | 133 + .../qwen-flash-character.md | 289 ++ .../qwen-flash-us.md | 261 ++ .../model-list-text-generation/qwen-flash.md | 1051 +++++ .../model-list-text-generation/qwen-long.md | 375 ++ .../qwen-math-plus.md | 467 ++ .../qwen-math-turbo.md | 115 + .../model-list-text-generation/qwen-max.md | 189 + .../qwen-mt-flash.md | 229 + .../qwen-mt-lite-us.md | 101 + .../qwen-mt-lite.md | 229 + .../qwen-mt-plus.md | 229 + .../qwen-mt-turbo.md | 153 + .../qwen-plus-character-ja.md | 125 + .../qwen-plus-character.md | 297 ++ .../qwen-plus-us.md | 341 ++ .../model-list-text-generation/qwen-plus.md | 3975 +++++++++++++++++ .../model-list-text-generation/qwen-turbo.md | 255 ++ .../model-list-text-generation/qwen3-14b.md | 233 + .../qwen3-235b-a22b-instruct-2507.md | 229 + .../qwen3-235b-a22b-thinking-2507.md | 241 + .../qwen3-235b-a22b.md | 289 ++ .../qwen3-30b-a3b-instruct-2507.md | 229 + .../qwen3-30b-a3b-thinking-2507.md | 229 + .../qwen3-30b-a3b.md | 277 ++ .../model-list-text-generation/qwen3-32b.md | 283 ++ .../qwen3-5-122b-a10b.md | 307 ++ .../model-list-text-generation/qwen3-5-27b.md | 307 ++ .../qwen3-5-35b-a3b.md | 307 ++ .../qwen3-5-397b-a17b.md | 307 ++ .../qwen3-5-flash.md | 1009 +++++ .../qwen3-5-plus.md | 1271 ++++++ .../model-list-text-generation/qwen3-6-27b.md | 161 + .../qwen3-6-35b-a3b.md | 241 + .../qwen3-6-flash-us.md | 155 + .../qwen3-6-flash.md | 949 ++++ .../model-list-text-generation/qwen3-6-max.md | 517 +++ .../qwen3-6-plus.md | 949 ++++ .../qwen3-7-max-us.md | 127 + .../model-list-text-generation/qwen3-7-max.md | 1293 ++++++ .../qwen3-7-plus-us.md | 171 + .../qwen3-7-plus.md | 1189 +++++ .../model-list-text-generation/qwen3-8b.md | 283 ++ .../qwen3-coder-30b-a3b-instruct.md | 477 ++ .../qwen3-coder-480b-a35b-instruct.md | 477 ++ .../qwen3-coder-flash.md | 1245 ++++++ .../qwen3-coder-next.md | 317 ++ .../qwen3-coder-plus.md | 1723 +++++++ .../qwen3-next-80b-a3b-instruct.md | 229 + .../qwen3-next-80b-a3b-thinking.md | 229 + .../model-list-text-generation/qwq-plus.md | 165 + .../step-3-7-flash.md | 133 + .../tongyi-intent-detect-v3.md | 115 + .../tongyi-xiaomi-analysis-flash.md | 115 + .../tongyi-xiaomi-analysis-pro.md | 115 + .../acoustic-portraits-videoretalk.md | 93 + .../animate-anyone-detect-gen2.md | 93 + .../animate-anyone-gen2.md | 93 + .../animate-anyone-template-gen2.md | 93 + .../emo-detect-v1.md | 93 + .../model-list-video-generation/emo-v1.md | 99 + .../emoji-detect-v1.md | 93 + .../model-list-video-generation/emoji-v1.md | 93 + .../happyhorse-1-0-i2v.md | 213 + .../happyhorse-1-0-r2v.md | 213 + .../happyhorse-1-0-t2v.md | 213 + .../happyhorse-1-0-video-edit.md | 247 + .../happyhorse-1-1-i2v.md | 251 ++ .../happyhorse-1-1-r2v.md | 247 + .../happyhorse-1-1-t2v.md | 247 + .../kling-v3-omni-video-generation.md | 135 + .../kling-v3-video-generation.md | 123 + .../liveportrait-detect.md | 93 + .../pixverse-c1-it2v.md | 147 + .../pixverse-c1-kf2v.md | 147 + .../pixverse-c1-r2v.md | 147 + .../pixverse-c1-t2v.md | 147 + .../pixverse-lipsync.md | 105 + .../pixverse-motioncontrol.md | 117 + .../pixverse-upscale.md | 105 + .../pixverse-v5-6-it2v.md | 147 + .../pixverse-v5-6-kf2v.md | 147 + .../pixverse-v5-6-r2v.md | 147 + .../pixverse-v5-6-t2v.md | 147 + .../pixverse-v6-it2v.md | 147 + .../pixverse-v6-kf2v.md | 147 + .../pixverse-v6-r2v.md | 147 + .../pixverse-v6-t2v.md | 147 + .../smart-portrait-liveport.md | 93 + .../video-style-transform.md | 99 + .../viduq2-pro-fast-img2video.md | 111 + .../viduq2-pro-img2video.md | 117 + .../viduq2-pro-reference2video.md | 117 + .../viduq2-pro-start-end2video.md | 117 + .../viduq2-reference2video.md | 117 + .../viduq2-text2video.md | 117 + .../viduq2-turbo-img2video.md | 117 + .../viduq2-turbo-start-end2video.md | 117 + .../viduq3-ad-reference2video.md | 111 + .../viduq3-drama-reference2video.md | 105 + .../viduq3-mix-reference2video.md | 111 + .../viduq3-pro-fast-img2video.md | 111 + .../viduq3-pro-img2video.md | 117 + .../viduq3-pro-start-end2video.md | 117 + .../viduq3-pro-text2video.md | 117 + .../viduq3-reference2video.md | 117 + .../viduq3-turbo-img2video.md | 117 + .../viduq3-turbo-reference2video.md | 117 + .../viduq3-turbo-start-end2video.md | 117 + .../viduq3-turbo-text2video.md | 117 + .../wan2-1-i2v-plus.md | 109 + .../wan2-1-i2v-turbo.md | 95 + .../wan2-1-kf2v-plus.md | 109 + .../wan2-1-t2v-plus.md | 109 + .../wan2-1-t2v-turbo.md | 115 + .../wan2-1-vace-plus.md | 109 + .../wan2-2-animate-mix.md | 145 + .../wan2-2-animate-move.md | 145 + .../wan2-2-i2v-flash.md | 151 + .../wan2-2-i2v-plus.md | 145 + .../wan2-2-kf2v-flash.md | 157 + .../wan2-2-s2v-detect.md | 93 + .../model-list-video-generation/wan2-2-s2v.md | 99 + .../wan2-2-t2v-plus.md | 145 + .../model-list-video-generation/wan2-5-i2v.md | 317 ++ .../model-list-video-generation/wan2-5-t2v.md | 317 ++ .../wan2-6-i2v-flash.md | 169 + .../wan2-6-i2v-us.md | 101 + .../model-list-video-generation/wan2-6-i2v.md | 225 + .../wan2-6-r2v-flash.md | 169 + .../model-list-video-generation/wan2-6-r2v.md | 213 + .../wan2-6-t2v-us.md | 101 + .../model-list-video-generation/wan2-6-t2v.md | 225 + .../model-list-video-generation/wan2-7-i2v.md | 327 ++ .../model-list-video-generation/wan2-7-r2v.md | 327 ++ .../model-list-video-generation/wan2-7-t2v.md | 473 ++ .../wan2-7-videoedit.md | 179 + .../wanx2-1-i2v-plus.md | 105 + .../wanx2-1-i2v-turbo.md | 93 + .../wanx2-1-kf2v-plus.md | 105 + .../wanx2-1-t2v-plus.md | 105 + .../wanx2-1-t2v-turbo.md | 111 + .../wanx2-1-vace-plus.md | 105 + .../gui-plus.md | 229 + .../qvq-max.md | 153 + .../qvq-plus.md | 115 + .../qwen-vl-max.md | 183 + .../qwen-vl-plus.md | 183 + .../qwen3-5-ocr.md | 115 + .../qwen3-vl-235b-a22b-instruct.md | 229 + .../qwen3-vl-235b-a22b-thinking.md | 241 + .../qwen3-vl-30b-a3b-instruct.md | 229 + .../qwen3-vl-30b-a3b-thinking.md | 229 + .../qwen3-vl-32b-instruct.md | 229 + .../qwen3-vl-32b-thinking.md | 241 + .../qwen3-vl-8b-instruct.md | 235 + .../qwen3-vl-8b-thinking.md | 235 + .../qwen3-vl-flash-us.md | 483 ++ .../qwen3-vl-flash.md | 1609 +++++++ .../qwen3-vl-plus.md | 1489 ++++++ .../qwenvl-ocr.md | 951 ++++ .../support/related-agreements.md | 2 +- .../test-1/bill-query-and-cost-management.md | 78 +- .../model-user-guide/test-1/model-pricing.md | 1704 ++----- .../model-training-and-deployment-billing.md | 276 +- .../model-user-guide/test-1/new-free-quota.md | 15 +- .../savings-plan-and-resource-package.md | 22 +- .../coding-plan-guide/add-vision-skill.md | 63 +- .../coding-plan-guide/coding-plan.md | 34 +- .../add-vision-skill.md | 329 ++ .../token-plan-harness-tool.md | 79 + .../token-plan-multimodal-gen.md | 78 +- .../token-plan-tool.md | 253 -- .../web-search-mcp.md} | 196 +- .../token-plan-guide/token-plan-faq.md | 292 -- .../token-plan-guide/token-plan-overview.md | 308 +- .../token-plan-personal-faq.md | 186 + .../token-plan-personal-overview.md | 257 ++ .../token-plan-personal-quick-start.md | 81 + .../token-plan-team-faq.md | 211 + .../token-plan-team-management.md | 183 + .../token-plan-team-overview.md | 314 ++ .../token-plan-team-quickstart.md} | 29 +- .../token-plan-guide/token-plan-team.md | 272 -- .../model-training-best-practices.md | 8 +- .../deepseek-api-by-vanchin.md | 27 +- .../deepseek-api.md | 49 +- .../glm-zhipu.md | 12 +- .../glm.md | 88 +- .../kimi-api-by-moonshot-ai.md | 186 +- .../kimi-api.md | 58 +- .../mimo.md | 9 +- .../minimax-api-by-minimax.md | 30 +- .../minimax-api.md | 80 +- .../siliconflow-deepseek-api.md | 22 +- .../stepfun.md | 27 +- .../chatbox.md | 34 +- .../cherry-studio.md | 36 +- .../claude-code.md | 112 +- .../cline.md | 38 +- .../codex.md | 219 +- .../cursor.md | 28 +- .../dify.md | 21 +- .../first-call-to-image-and-video-api.md | 4 +- .../hermes-agent.md | 71 +- .../kilo-cli.md | 122 +- .../lingma-agent.md | 14 +- .../more-tools.md | 36 +- .../openclaw.md | 195 +- .../opencode.md | 138 +- .../qoder-agent.md | 20 +- .../qwen-code.md | 157 +- .../qwenpaw.md | 28 +- .../wiki/api/3d-generation.md | 168 +- .../wiki/api/application-call.md | 19 +- .../application-component-api-reference.md | 185 +- .../wiki/api/file-management-api.md | 15 +- .../wiki/api/frameworks.md | 19 +- .../wiki/api/image-generation.md | 128 +- .../wiki/api/knowledge.md | 23 +- .../wiki/api/long-term-memory-new.md | 21 +- .../wiki/api/managed-agents-api.md | 15 +- .../wiki/api/model-production.md | 45 +- .../wiki/api/more-about-models.md | 29 +- .../wiki/api/more-models.md | 15 +- skills/bailian-docs-llm-wiki/wiki/api/more.md | 19 +- .../wiki/api/omni-realtime-api.md | 199 +- .../wiki/api/preparations.md | 95 +- .../wiki/api/qwen-api-reference.md | 15 +- .../wiki/api/toolkits-and-frameworks.md | 117 +- .../wiki/api/vector-and-sort.md | 116 +- .../wiki/api/video-generation-api.md | 145 +- .../wiki/comparisons/agent-extensibility.md | 70 + .../comparisons/agent-extension-options.md | 76 + .../comparisons/app-eval-vs-monitoring.md | 79 + .../application-calling-approaches.md | 91 + .../application-calling-comparison.md | 103 +- .../application-eval-vs-model-eval.md | 74 + .../application-evaluation-vs-monitoring.md | 87 +- .../application-lifecycle-management.md | 74 + .../comparisons/content-generation-apis.md | 58 + .../wiki/comparisons/evaluation-comparison.md | 102 +- .../wiki/comparisons/evaluation-vs.md | 67 + ...ine-tuning-vs-compression-vs-deployment.md | 148 +- .../comparisons/fine-tuning-vs-compression.md | 70 +- .../wiki/comparisons/frameworks-comparison.md | 47 + .../wiki/comparisons/frameworks-vs.md | 46 + .../comparisons/generation-api-comparison.md | 73 + .../wiki/comparisons/generation-api-vs.md | 78 + .../image-video-3d-generation-comparison.md | 132 +- .../comparisons/image-vs-video-generation.md | 75 + .../image-vs-video-vs-3d-generation.md | 82 +- .../comparisons/knowledge-base-vs-memory.md | 101 + .../knowledge-memory-data-connection.md | 72 + .../knowledge-vs-memory-comparison.md | 56 + .../managed-agents-api-vs-guide.md | 40 + .../managed-agents-api-vs-guides.md | 54 + .../comparisons/managed-agents-comparison.md | 43 + .../managed-agents-guide-vs-api.md | 89 +- .../wiki/comparisons/managed-agents-vs-api.md | 74 + .../comparisons/managed-vs-custom-agents.md | 73 + .../wiki/comparisons/memory-comparison.md | 110 + .../comparisons/memory-features-comparison.md | 64 + .../wiki/comparisons/memory-solutions-vs.md | 56 + .../wiki/comparisons/memory-solutions.md | 61 + .../comparisons/memory-vs-knowledge-base.md | 78 + .../comparisons/model-deployment-inference.md | 61 + .../comparisons/model-eval-vs-app-eval.md | 78 + .../model-monitoring-vs-app-monitoring.md | 67 + .../model-optimization-approaches.md | 119 + .../model-optimization-comparison.md | 100 +- .../model-optimization-lifecycle.md | 60 + .../comparisons/model-optimization-methods.md | 62 + .../wiki/comparisons/model-optimization-vs.md | 59 + .../wiki/comparisons/model-optimization.md | 57 + .../model-vs-application-evaluation.md | 63 + .../model-vs-application-monitoring.md | 64 + .../wiki/comparisons/monitoring-comparison.md | 74 +- .../wiki/comparisons/monitoring-vs.md | 52 + .../multimodal-generation-api-comparison.md | 122 +- .../comparisons/multimodal-generation-api.md | 65 + .../comparisons/multimodal-generation-apis.md | 71 + .../multimodal-generation-comparison.md | 92 +- .../omni-realtime-vs-realtime-user-guide.md | 53 + .../wiki/comparisons/platform-extension.md | 52 + .../realtime-api-vs-omni-realtime.md | 60 + .../comparisons/realtime-apis-comparison.md | 57 + .../wiki/concepts/agent-application.md | 97 +- .../wiki/concepts/api-key.md | 98 +- .../wiki/concepts/async-invocation.md | 72 +- .../wiki/concepts/billing.md | 99 +- .../wiki/concepts/dashscope-sdk.md | 137 +- .../wiki/concepts/embedding.md | 80 +- .../wiki/concepts/evaluation.md | 96 +- .../wiki/concepts/event-stream.md | 85 + .../wiki/concepts/function-calling.md | 76 +- .../wiki/concepts/knowledge-base.md | 103 +- .../wiki/concepts/long-term-memory.md | 112 + .../wiki/concepts/migration.md | 93 + .../wiki/concepts/model-lifecycle.md | 93 + .../wiki/concepts/multimodal.md | 66 +- .../concepts/openai-compatible-interface.md | 65 +- .../wiki/concepts/prompt-engineering.md | 85 +- .../wiki/concepts/rag.md | 79 +- .../wiki/concepts/rate-limit.md | 48 + .../wiki/concepts/rate-limiting.md | 152 +- .../wiki/concepts/region.md | 101 +- .../wiki/concepts/streaming.md | 60 +- .../wiki/concepts/token-billing.md | 76 +- .../wiki/concepts/token.md | 79 +- .../wiki/concepts/workflow.md | 114 +- .../wiki/concepts/workspace.md | 76 +- .../wiki/guides/application-evaluation.md | 31 +- .../wiki/guides/application-monitoring.md | 23 +- .../application-permission-management.md | 59 +- .../application-publishing-and-sharing.md | 33 +- .../wiki/guides/application-support.md | 109 +- .../wiki/guides/application-use-cases.md | 19 +- .../guides/bailian-application-calling.md | 19 +- .../wiki/guides/data-connection-overview.md | 19 +- .../wiki/guides/fine-tuning.md | 117 +- .../wiki/guides/get-started-with-models.md | 97 +- .../wiki/guides/knowledge-base.md | 192 +- .../wiki/guides/llm-application.md | 145 +- .../wiki/guides/managed-agents.md | 66 +- .../wiki/guides/memory-library-overview.md | 19 +- .../wiki/guides/model-compression.md | 89 +- .../wiki/guides/model-context-protocol.md | 29 +- .../wiki/guides/model-data-overview.md | 185 +- .../wiki/guides/model-deployment-1.md | 112 +- .../guides/model-evaluation-introduction.md | 127 +- .../wiki/guides/model-experience.md | 110 +- .../wiki/guides/model-high-speed-inference.md | 96 +- .../wiki/guides/model-monitoring.md | 102 +- .../wiki/guides/plug-in.md | 17 +- .../wiki/guides/prompt.md | 23 +- .../wiki/guides/realtime-api-user-guide.md | 185 + .../wiki/guides/release-notes.md | 70 +- .../wiki/guides/security-and-compliance.md | 223 +- .../wiki/guides/skill.md | 13 + .../wiki/guides/start-using.md | 29 +- .../wiki/guides/support.md | 106 - .../wiki/guides/test-1.md | 99 +- .../wiki/guides/token-plan-guide.md | 150 +- .../wiki/guides/use-cases.md | 110 +- .../use-chat-client-or-development-tool.md | 93 +- skills/bailian-docs-llm-wiki/wiki/index.md | 39 +- 672 files changed, 108569 insertions(+), 11473 deletions(-) create mode 100644 skills/bailian-docs-llm-wiki/models/groups/qwen-image-3.0-pro.json create mode 100644 skills/bailian-docs-llm-wiki/models/groups/qwen3.7-flash.json create mode 100644 skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-addchunk.md rename skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/{lingque-ccai-dialogue-analytics-ram-subaccount-usage-and-authorization-procedures.md => use-and-authorize-ram-users-for-ccai-dialogue-analysis.md} (100%) create mode 100644 skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-lingque-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listvoices.md create mode 100644 skills/bailian-docs-llm-wiki/raw/application-user-guide/application-support/application-after-sales-service-scope.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-realtime.md rename skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/{qwen-audio-realtime-websocket-api.md => fun-audiochat-realtime-websocket-api.md} (67%) create mode 100644 skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-generation-and-editing-api-reference.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/rl-training-overview.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/model-data-overview/model-log-backflow.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-desc.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-audio-features.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-connection-management.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-capture.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-playback.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-video-input.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-media-stream-control.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-video-features.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-aoq-omni-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-multimodal-dialog.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-omni-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-overview.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-connect-model.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-sdk-download.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-token-authentication.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/after-sales-service-scope.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-h3-1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-p1-0.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/gte-rerank-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/multimodal-embedding-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen2-5-vl-embedding.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-7-text-embedding.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-rerank.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-embedding.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-rerank.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v3.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v4.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-parsing-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-refiner.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/facechain-facedetect.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-erase-completion.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-out-painting.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-image-generation.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-omni-image-generation.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-facechain-generation.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-image-instance-segmentation.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-qwen-image-edit.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0-pro.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-3-0-pro.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-max.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-max.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-mt-image.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/shoemodel-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/vidu-image-reference2image.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-fast-reference2image.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-pro-reference2image.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq3-fast-reference2image.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/virtualmodel-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-i2i.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-t2i.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-image.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-t2i.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image-pro.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-background-generation-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-poster-generation-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-sketch-to-image-lite.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-style-repaint-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-virtualmodel.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-x-painting.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-0-t2i-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-imageedit.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-semantic.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-texture.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/z-image-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen2-5-omni-7b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash-8k-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-mtl.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-chat-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-realtime-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-mtl-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-6.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-filetrans.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-omni-30b-a3b-captioner.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/speech-biasing.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-clone-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/fun-music-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/music-generation-preview.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-design.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-enrollment.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/sambert.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-hd.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-hd.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/voice-enrollment.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-ocr.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-1-5b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-14b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-32b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-7b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1-terminus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-exp.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-think.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-3.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/farui-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-5-air.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-6.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-7.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-fast.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-9.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-5.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-6.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code-highspeed.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-thinking.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k3.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/mimo-v2-5-pro.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-5.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-7.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m3.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen-deep-research.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen3-max.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/moonshot-kimi-k2-instruct.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-coder-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-coder-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-doc-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-flash-character.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-flash-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-long.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-math-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-math-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-max.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-lite-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-lite.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-plus-character-ja.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-plus-character.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-plus-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-14b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-235b-a22b-instruct-2507.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-235b-a22b-thinking-2507.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-235b-a22b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-30b-a3b-instruct-2507.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-30b-a3b-thinking-2507.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-30b-a3b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-32b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-122b-a10b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-27b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-35b-a3b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-397b-a17b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-27b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-35b-a3b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-flash-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-max.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-7-max-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-7-max.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-7-plus-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-7-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-8b.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-30b-a3b-instruct.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-480b-a35b-instruct.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-next.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-next-80b-a3b-instruct.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-next-80b-a3b-thinking.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwq-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/step-3-7-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/tongyi-intent-detect-v3.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/tongyi-xiaomi-analysis-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/tongyi-xiaomi-analysis-pro.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/acoustic-portraits-videoretalk.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/animate-anyone-detect-gen2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/animate-anyone-gen2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/animate-anyone-template-gen2.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/emo-detect-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/emo-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/emoji-detect-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/emoji-v1.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-0-i2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-0-r2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-0-t2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-0-video-edit.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-1-i2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-1-r2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-1-t2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/kling-v3-omni-video-generation.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/kling-v3-video-generation.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/liveportrait-detect.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-c1-it2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-c1-kf2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-c1-r2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-c1-t2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-lipsync.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-motioncontrol.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-upscale.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v5-6-it2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v5-6-kf2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v5-6-r2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v5-6-t2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v6-it2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v6-kf2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v6-r2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v6-t2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/smart-portrait-liveport.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/video-style-transform.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-pro-fast-img2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-pro-img2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-pro-reference2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-pro-start-end2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-reference2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-text2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-turbo-img2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-turbo-start-end2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-ad-reference2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-drama-reference2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-mix-reference2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-pro-fast-img2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-pro-img2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-pro-start-end2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-pro-text2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-reference2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-turbo-img2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-turbo-reference2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-turbo-start-end2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-turbo-text2video.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-i2v-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-i2v-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-kf2v-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-t2v-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-t2v-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-vace-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-animate-mix.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-animate-move.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-i2v-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-i2v-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-kf2v-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-s2v-detect.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-s2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-t2v-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-5-i2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-5-t2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-i2v-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-i2v-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-i2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-r2v-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-r2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-t2v-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-t2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-7-i2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-7-r2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-7-t2v.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-7-videoedit.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-i2v-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-i2v-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-kf2v-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-t2v-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-t2v-turbo.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-vace-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/gui-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qvq-max.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qvq-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen-vl-max.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen-vl-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-5-ocr.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-235b-a22b-instruct.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-235b-a22b-thinking.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-30b-a3b-instruct.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-30b-a3b-thinking.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-32b-instruct.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-32b-thinking.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-8b-instruct.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-8b-thinking.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-flash-us.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-flash.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-plus.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwenvl-ocr.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/add-vision-skill.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-harness-tool.md delete mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-tool.md rename skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/{coding-plan-guide/web-search-for-coding-plan.md => token-plan-best-practice/web-search-mcp.md} (51%) delete mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-faq.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-faq.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-overview.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-quick-start.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-faq.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-management.md create mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-overview.md rename skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/{token-plan-quickstart.md => token-plan-team-edition/token-plan-team-quickstart.md} (64%) delete mode 100644 skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/agent-extensibility.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/agent-extension-options.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/app-eval-vs-monitoring.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/application-calling-approaches.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/application-eval-vs-model-eval.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/application-lifecycle-management.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/content-generation-apis.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/evaluation-vs.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/frameworks-comparison.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/frameworks-vs.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/generation-api-comparison.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/generation-api-vs.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/image-vs-video-generation.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-base-vs-memory.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-memory-data-connection.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-vs-memory-comparison.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-api-vs-guide.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-api-vs-guides.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-comparison.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-vs-api.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/managed-vs-custom-agents.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/memory-comparison.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/memory-features-comparison.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/memory-solutions-vs.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/memory-solutions.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/memory-vs-knowledge-base.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-deployment-inference.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-eval-vs-app-eval.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-monitoring-vs-app-monitoring.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-approaches.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-lifecycle.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-methods.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-vs.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-vs-application-evaluation.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/model-vs-application-monitoring.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/monitoring-vs.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-api.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-apis.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/omni-realtime-vs-realtime-user-guide.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/platform-extension.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/realtime-api-vs-omni-realtime.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/comparisons/realtime-apis-comparison.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/concepts/event-stream.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/concepts/long-term-memory.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/concepts/migration.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/concepts/model-lifecycle.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/concepts/rate-limit.md create mode 100644 skills/bailian-docs-llm-wiki/wiki/guides/realtime-api-user-guide.md delete mode 100644 skills/bailian-docs-llm-wiki/wiki/guides/support.md diff --git a/skills/bailian-docs-llm-wiki/SKILL.md b/skills/bailian-docs-llm-wiki/SKILL.md index 7ec690da..3dbef468 100644 --- a/skills/bailian-docs-llm-wiki/SKILL.md +++ b/skills/bailian-docs-llm-wiki/SKILL.md @@ -149,21 +149,20 @@ description: >- | 模型家族总览 / 按能力分桶浏览 | `models/index.md` | | 主题页 / API 文档（按功能领域查找） | `wiki/index.md`（完整索引入口） | | OpenAI 兼容接口 | `wiki/concepts/openai-compatible-interface.md` | -| API Key 鉴权 | `wiki/concepts/api-key.md` | -| 函数调用（Function Calling） | `wiki/concepts/function-calling.md` | +| 异步调用 | `wiki/concepts/async-invocation.md` | +| 流式输出 | `wiki/concepts/streaming.md` | | 检索增强生成（RAG） | `wiki/concepts/rag.md` | -| 异步调用与任务轮询 | `wiki/concepts/async-invocation.md` | -| Token 与计费 | `wiki/concepts/token-and-billing.md` | -| 流式输出 | `wiki/concepts/streaming-output.md` | -| 业务空间（Workspace） | `wiki/concepts/workspace.md` | -| 模型调优与部署 | `wiki/concepts/fine-tuning-and-deployment.md` | -| MCP 与工具扩展 | `wiki/concepts/mcp-and-tools.md` | -| 模型微调、压缩与部署对比 | `wiki/comparisons/fine-tuning-vs-compression-vs-deployment.md` | -| 模型评估与模型监控对比 | `wiki/comparisons/model-evaluation-vs-monitoring.md` | -| 图像、视频与 3D 生成对比 | `wiki/comparisons/image-vs-video-vs-3d-generation.md` | -| 应用评估与应用监控对比 | `wiki/comparisons/app-evaluation-vs-monitoring.md` | -| 知识库与记忆库对比 | `wiki/comparisons/knowledge-base-vs-memory-library.md` | -| 托管智能体：指南与 API 对比 | `wiki/comparisons/managed-agents-guide-vs-api.md` | +| API Key 鉴权 | `wiki/concepts/api-key.md` | +| 业务空间 | `wiki/concepts/workspace.md` | +| Token 计费 | `wiki/concepts/token-billing.md` | +| 多模态 | `wiki/concepts/multimodal.md` | +| 限流 | `wiki/concepts/rate-limit.md` | +| 地域与接入点 | `wiki/concepts/region.md` | +| 图像、视频与3D生成对比 | `wiki/comparisons/multimodal-generation-comparison.md` | +| 模型微调、压缩与高速推理对比 | `wiki/comparisons/model-optimization-comparison.md` | +| 知识库与记忆体系对比 | `wiki/comparisons/knowledge-vs-memory-comparison.md` | +| 模型评测与应用评测对比 | `wiki/comparisons/model-vs-application-evaluation.md` | +| 模型监控与应用监控对比 | `wiki/comparisons/model-vs-application-monitoring.md` | > 实际文件名以 `wiki/index.md` 为准；上表若有出入应回到索引页查找。 diff --git a/skills/bailian-docs-llm-wiki/llms.txt b/skills/bailian-docs-llm-wiki/llms.txt index f711c7c8..2374a770 100644 --- a/skills/bailian-docs-llm-wiki/llms.txt +++ b/skills/bailian-docs-llm-wiki/llms.txt @@ -4,101 +4,109 @@ ## 模型使用指南 -- **模型体验** - - [视觉理解](raw/model-user-guide/model-experience/vision-model.md) - - [图片生成与编辑](raw/model-user-guide/model-experience/image-model.md) - - [视频生成与编辑](raw/model-user-guide/model-experience/video-generate-edit-model.md) - - [文本生成](raw/model-user-guide/model-experience/text-generation-model.md) - - [Tripo 3D模型生成](raw/model-user-guide/model-experience/tripo-3d-generation-guide.md) - - [语音合成](raw/model-user-guide/model-experience/tts-model.md) - - [音乐生成](raw/model-user-guide/model-experience/fun-music.md) - - [语音识别](raw/model-user-guide/model-experience/asr-model.md) - - [语音转语音](raw/model-user-guide/model-experience/s2s-model.md) - - [向量与重排序](raw/model-user-guide/model-experience/embedding-rerank-model.md) - - [全模态](raw/model-user-guide/model-experience/omni.md) - **开始使用** - [什么是阿里云百炼](raw/model-user-guide/get-started-with-models/what-is-model-studio.md) - [首次调用千问API](raw/model-user-guide/get-started-with-models/first-api-call-to-qwen.md) - [选择模型](raw/model-user-guide/get-started-with-models/models.md) - [Base URL总览](raw/model-user-guide/get-started-with-models/base-url.md) - - [选择地域、服务部署范围和接入域名](raw/model-user-guide/get-started-with-models/regions.md) + - [地域及接入域名](raw/model-user-guide/get-started-with-models/regions.md) - [限流](raw/model-user-guide/get-started-with-models/rate-limit.md) -- **产品计费** - - [新人免费额度](raw/model-user-guide/test-1/new-free-quota.md) - - [模型训练与部署计费](raw/model-user-guide/test-1/model-training-and-deployment-billing.md) - - [节省计划与资源包](raw/model-user-guide/test-1/savings-plan-and-resource-package.md) - - [账单查询与成本管理](raw/model-user-guide/test-1/bill-query-and-cost-management.md) - - [模型调用价格](raw/model-user-guide/test-1/model-pricing.md) - **接入客户端/开发工具** - [Hermes Agent](raw/model-user-guide/use-chat-client-or-development-tool/hermes-agent.md) - [OpenClaw](raw/model-user-guide/use-chat-client-or-development-tool/openclaw.md) - [Claude Code](raw/model-user-guide/use-chat-client-or-development-tool/claude-code.md) - [OpenCode](raw/model-user-guide/use-chat-client-or-development-tool/opencode.md) - - [Qwen Code](raw/model-user-guide/use-chat-client-or-development-tool/qwen-code.md) - - [Cursor](raw/model-user-guide/use-chat-client-or-development-tool/cursor.md) - [Codex](raw/model-user-guide/use-chat-client-or-development-tool/codex.md) + - [Cursor](raw/model-user-guide/use-chat-client-or-development-tool/cursor.md) - [QwenPaw](raw/model-user-guide/use-chat-client-or-development-tool/qwenpaw.md) - - [Cline](raw/model-user-guide/use-chat-client-or-development-tool/cline.md) - [Cherry Studio](raw/model-user-guide/use-chat-client-or-development-tool/cherry-studio.md) + - [Qwen Code](raw/model-user-guide/use-chat-client-or-development-tool/qwen-code.md) - [Chatbox](raw/model-user-guide/use-chat-client-or-development-tool/chatbox.md) + - [Cline](raw/model-user-guide/use-chat-client-or-development-tool/cline.md) + - [Kilo CLI](raw/model-user-guide/use-chat-client-or-development-tool/kilo-cli.md) - [Qoder](raw/model-user-guide/use-chat-client-or-development-tool/qoder-agent.md) - [Qoder CN（原 Lingma）](raw/model-user-guide/use-chat-client-or-development-tool/lingma-agent.md) - - [Kilo CLI](raw/model-user-guide/use-chat-client-or-development-tool/kilo-cli.md) - [使用Postman或cURL调用图像/视频生成API](raw/model-user-guide/use-chat-client-or-development-tool/first-call-to-image-and-video-api.md) - [Dify](raw/model-user-guide/use-chat-client-or-development-tool/dify.md) - [更多工具](raw/model-user-guide/use-chat-client-or-development-tool/more-tools.md) -- **Token Plan（团队版）** - - **最佳实践** - - [工具调用](raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-tool.md) - - [接入多模态生成模型](raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-multimodal-gen.md) +- **产品计费** + - [新人免费额度](raw/model-user-guide/test-1/new-free-quota.md) + - [模型训练与部署计费](raw/model-user-guide/test-1/model-training-and-deployment-billing.md) + - [节省计划与资源包](raw/model-user-guide/test-1/savings-plan-and-resource-package.md) + - [账单查询与成本管理](raw/model-user-guide/test-1/bill-query-and-cost-management.md) + - [模型调用价格](raw/model-user-guide/test-1/model-pricing.md) +- **Token Plan** + - **个人版** + - [概述](raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-overview.md) + - [快速开始](raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-quick-start.md) + - [常见问题](raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-faq.md) - **Coding Plan** - [Coding Plan概述](raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan.md) - - [联网搜索](raw/model-user-guide/token-plan-guide/coding-plan-guide/web-search-for-coding-plan.md) - - [添加视觉理解能力](raw/model-user-guide/token-plan-guide/coding-plan-guide/add-vision-skill.md) - [常见问题](raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan-faq.md) - - [快速开始](raw/model-user-guide/token-plan-guide/token-plan-quickstart.md) - - [Token Plan（团队版）概述](raw/model-user-guide/token-plan-guide/token-plan-overview.md) - - [团队管理](raw/model-user-guide/token-plan-guide/token-plan-team.md) - - [常见问题](raw/model-user-guide/token-plan-guide/token-plan-faq.md) + - **团队版** + - [概述](raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-overview.md) + - [快速开始](raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-quickstart.md) + - [团队管理](raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-management.md) + - [常见问题](raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-faq.md) + - **最佳实践** + - [接入 Harness 工具](raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-harness-tool.md) + - [接入多模态生成模型](raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-multimodal-gen.md) + - [联网搜索](raw/model-user-guide/token-plan-guide/token-plan-best-practice/web-search-mcp.md) + - [添加视觉理解能力](raw/model-user-guide/token-plan-guide/token-plan-best-practice/add-vision-skill.md) + - [Token Plan 概述](raw/model-user-guide/token-plan-guide/token-plan-overview.md) - **模型推理** - [TPM 预留](raw/model-user-guide/model-high-speed-inference/tpm-reservation.md) - [快速模式](raw/model-user-guide/model-high-speed-inference/fast-mode.md) +- **模型体验** + - [视觉理解](raw/model-user-guide/model-experience/vision-model.md) + - [文本生成](raw/model-user-guide/model-experience/text-generation-model.md) + - [图片生成与编辑](raw/model-user-guide/model-experience/image-model.md) + - [Tripo 3D模型生成](raw/model-user-guide/model-experience/tripo-3d-generation-guide.md) + - [语音合成](raw/model-user-guide/model-experience/tts-model.md) + - [视频生成与编辑](raw/model-user-guide/model-experience/video-generate-edit-model.md) + - [音乐生成](raw/model-user-guide/model-experience/fun-music.md) + - [语音识别](raw/model-user-guide/model-experience/asr-model.md) + - [语音转语音](raw/model-user-guide/model-experience/s2s-model.md) + - [向量与重排序](raw/model-user-guide/model-experience/embedding-rerank-model.md) + - [全模态](raw/model-user-guide/model-experience/omni.md) - **模型部署** + - [模型导入](raw/model-user-guide/model-deployment-1/model-import.md) - [模型部署](raw/model-user-guide/model-deployment-1/model-deployment-introduction.md) - [预置吞吐长输入与缓存](raw/model-user-guide/model-deployment-1/ptu-long-input-and-cache.md) - - [模型导入](raw/model-user-guide/model-deployment-1/model-import.md) - [使用 API或命令行进行模型部署](raw/model-user-guide/model-deployment-1/model-deployment-quick-start.md) +- **模型评测** + - [模型评测](raw/model-user-guide/model-evaluation-introduction/model-evaluation-overview.md) + - [评测维度](raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md) - **模型调优** - **千问模型调优** - - [模型调优简介](raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md) + - [使用 API 或命令行进行模型调优](raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/fine-tuning-api-guide.md) - [在控制台进行模型调优](raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-on-console.md) - [0 代码强化大模型安全合规能力](raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/enhance-the-security-compliance-of-large-models.md) - - [使用 API 或命令行进行模型调优](raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/fine-tuning-api-guide.md) + - [模型调优简介](raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md) - **语音合成模型调优** - [CosyVoice模型调优](raw/model-user-guide/fine-tuning/fine-tune-speech-synthesis-model/fine-tune-speech-synthesis-model-by-api.md) - [微调图像生成模型](raw/model-user-guide/fine-tuning/wan-image-generation-finetune-guide.md) - [微调视频生成模型](raw/model-user-guide/fine-tuning/wan-video-generation-finetune-guide.md) -- **模型评测** - - [模型评测](raw/model-user-guide/model-evaluation-introduction/model-evaluation-overview.md) - - [评测维度](raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md) -- **模型压缩** - - [模型压缩](raw/model-user-guide/model-compression/model-compression-introduction.md) + - [强化学习训练概述](raw/model-user-guide/fine-tuning/rl-training-overview.md) - **用量统计与性能监控** - [模型用量](raw/model-user-guide/model-monitoring/model-usage-statistics.md) - [模型监控](raw/model-user-guide/model-monitoring/model-telemetry.md) +- **模型压缩** + - [模型压缩](raw/model-user-guide/model-compression/model-compression-introduction.md) - **模型数据** - [训练集与评测集](raw/model-user-guide/model-data-overview/training-set-and-evaluation-set.md) + - [日志回流](raw/model-user-guide/model-data-overview/model-log-backflow.md) - [数据清洗或增强](raw/model-user-guide/model-data-overview/data-processing.md) - **安全合规** - - **安全存储** - - [配置终端节点并发起连接](raw/model-user-guide/security-and-compliance/secure-storage/configure-an-endpoint-and-initiate-a-connection.md) - - [配置可用区IP](raw/model-user-guide/security-and-compliance/secure-storage/configure-zone-ip.md) - - [配置MSE云原生网关](raw/model-user-guide/security-and-compliance/secure-storage/configure-mse.md) - - [配置私有网络中的资源](raw/model-user-guide/security-and-compliance/secure-storage/configure-resources-in-private-network.md) - **传输安全** - [获取RSA的公钥](raw/model-user-guide/security-and-compliance/transmission-security/model-interface-aes-encryption.md) - [以加密的方式接入模型推理功能](raw/model-user-guide/security-and-compliance/transmission-security/encrypted-access-to-model-inference.md) - [通过终端节点私网访问阿里云百炼模型或应用 API](raw/model-user-guide/security-and-compliance/transmission-security/access-model-studio-through-privatelink.md) + - **安全存储** + - [配置终端节点并发起连接](raw/model-user-guide/security-and-compliance/secure-storage/configure-an-endpoint-and-initiate-a-connection.md) + - [配置可用区IP](raw/model-user-guide/security-and-compliance/secure-storage/configure-zone-ip.md) + - [配置私有网络中的资源](raw/model-user-guide/security-and-compliance/secure-storage/configure-resources-in-private-network.md) + - [配置MSE云原生网关](raw/model-user-guide/security-and-compliance/secure-storage/configure-mse.md) - [权限管理](raw/model-user-guide/security-and-compliance/permission-management-overview.md) - [输⼊输出AI安全护栏](raw/model-user-guide/security-and-compliance/content-security.md) - [模型备案信息公示](raw/model-user-guide/security-and-compliance/model-filing-information-publicity.md) @@ -106,33 +114,376 @@ - [合规资质与隐私说明](raw/model-user-guide/security-and-compliance/privacy-notice.md) - **实践教程** - **三方模型调用教程** - - [DeepSeek-硅基流动](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/siliconflow-deepseek-api.md) - [DeepSeek-阿里云](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api.md) - [DeepSeek](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api-by-vanchin.md) - - [Kimi](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api.md) + - [DeepSeek-硅基流动](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/siliconflow-deepseek-api.md) - [Kimi-月之暗面](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api-by-moonshot-ai.md) + - [Kimi](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api.md) - [MiniMax](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api.md) - [GLM-智谱](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm-zhipu.md) - [GLM](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm.md) - - [MiniMax](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api-by-minimax.md) - [Vidu视频生成Prompt指南](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/vidu-video-generation-prompt-guide.md) + - [MiniMax](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api-by-minimax.md) - [MiMo-小米](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/mimo.md) - [Stepfun-阶跃星辰](raw/model-user-guide/use-cases/third-party-model-integration-tutorial/stepfun.md) - - [HappyHorse 打造一站式影视创作平台](raw/model-user-guide/use-cases/infinite-canvas.md) - [高效搭建 AI 智能体与工作流应用](raw/model-user-guide/use-cases/build-ai-applications-based-on-alibaba-cloud-model-studio.md) + - [HappyHorse 打造一站式影视创作平台](raw/model-user-guide/use-cases/infinite-canvas.md) - [深度研究：生成你的独家洞察报告](raw/model-user-guide/use-cases/deep-research.md) - - [AI 解题 + 批改：推动课程教学智变](raw/model-user-guide/use-cases/ai-homework-helper.md) - - [文生文Prompt指南](raw/model-user-guide/use-cases/prompt-engineering-guide.md) - [文生图Prompt指南](raw/model-user-guide/use-cases/text-to-image-prompt.md) + - [文生文Prompt指南](raw/model-user-guide/use-cases/prompt-engineering-guide.md) + - [AI 解题 + 批改：推动课程教学智变](raw/model-user-guide/use-cases/ai-homework-helper.md) - [基于LlamaIndex构建RAG应用](raw/model-user-guide/use-cases/build-rag-applications-based-on-llamaindex.md) - [文生视频/图生视频Prompt指南](raw/model-user-guide/use-cases/text-to-video-prompt.md) - [自定义模型调优、部署与评测](raw/model-user-guide/use-cases/model-training-best-practices.md) - - [借助大模型将文档转换为视频](raw/model-user-guide/use-cases/use-llm-to-convert-document-to-video.md) - [限流应对最佳实践 ](raw/model-user-guide/use-cases/rate-limiting-best-practices.md) + - [借助大模型将文档转换为视频](raw/model-user-guide/use-cases/use-llm-to-convert-document-to-video.md) - [显式缓存最佳实践](raw/model-user-guide/use-cases/explicit-cache-guide.md) - **服务支持** + - **模型列表** + - **文本生成** + - [qwen3.7-max](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-7-max.md) + - [qwen3.7-max-us](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-7-max-us.md) + - [qwen3.6-max-preview](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-max.md) + - [qwen3-max](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen3-max.md) + - [qwen-max](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-max.md) + - [qwen3.7-plus-us](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-7-plus-us.md) + - [qwen3.7-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-7-plus.md) + - [qwen3.6-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-plus.md) + - [qwen3.5-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-plus.md) + - [qwen-plus-us](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-plus-us.md) + - [qwen-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-plus.md) + - [qwen3.6-flash](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-flash.md) + - [qwen3.6-flash-us](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-flash-us.md) + - [qwen3.5-flash](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-flash.md) + - [qwen-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-turbo.md) + - [qwen-flash-us](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-flash-us.md) + - [qwen-flash](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-flash.md) + - [qwen-plus-character](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-plus-character.md) + - [qwq-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwq-plus.md) + - [qwen-plus-character-ja](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-plus-character-ja.md) + - [qwen-flash-character](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-flash-character.md) + - [qwen-long](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-long.md) + - [qwen-math-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-math-turbo.md) + - [qwen-math-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-math-plus.md) + - [qwen3-coder-flash](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-flash.md) + - [qwen3-coder-next](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-next.md) + - [qwen-coder-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-coder-plus.md) + - [qwen3-coder-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-plus.md) + - [qwen-coder-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-coder-turbo.md) + - [qwen-mt-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-plus.md) + - [qwen-mt-flash](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-flash.md) + - [qwen-mt-lite-us](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-lite-us.md) + - [qwen-mt-lite](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-lite.md) + - [qwen-mt-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-mt-turbo.md) + - [qwen-doc-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen-doc-turbo.md) + - [qwen-deep-research](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen-deep-research.md) + - [farui-plus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/farui-plus.md) + - [tongyi-intent-detect-v3](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/tongyi-intent-detect-v3.md) + - [tongyi-xiaomi-analysis-flash](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/tongyi-xiaomi-analysis-flash.md) + - [tongyi-xiaomi-analysis-pro](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/tongyi-xiaomi-analysis-pro.md) + - [qwen3-coder-30b-a3b-instruct](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-30b-a3b-instruct.md) + - [qwen3.6-27b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-27b.md) + - [qwen3-coder-480b-a35b-instruct](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-coder-480b-a35b-instruct.md) + - [qwen3.6-35b-a3b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-6-35b-a3b.md) + - [qwen3.5-27b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-27b.md) + - [qwen3.5-122b-a10b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-122b-a10b.md) + - [qwen3.5-35b-a3b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-35b-a3b.md) + - [qwen3.5-397b-a17b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-5-397b-a17b.md) + - [qwen3-235b-a22b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-235b-a22b.md) + - [qwen3-235b-a22b-thinking-2507](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-235b-a22b-thinking-2507.md) + - [qwen3-235b-a22b-instruct-2507](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-235b-a22b-instruct-2507.md) + - [qwen3-30b-a3b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-30b-a3b.md) + - [qwen3-30b-a3b-instruct-2507](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-30b-a3b-instruct-2507.md) + - [qwen3-30b-a3b-thinking-2507](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-30b-a3b-thinking-2507.md) + - [qwen3-32b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-32b.md) + - [qwen3-14b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-14b.md) + - [qwen3-8b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-8b.md) + - [qwen3-next-80b-a3b-instruct](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-next-80b-a3b-instruct.md) + - [qwen3-next-80b-a3b-thinking](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/qwen3-next-80b-a3b-thinking.md) + - [deepseek-v4-pro](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro.md) + - [deepseek-v4-pro-us](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro-us.md) + - [deepseek-v4-flash](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash.md) + - [deepseek-v4-flash-us](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash-us.md) + - [deepseek-v3.2](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-3.md) + - [deepseek-v3.2-exp](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-exp.md) + - [vanchin/deepseek-v3.2-think](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-think.md) + - [deepseek-v3.1](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1.md) + - [deepseek-v3](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3.md) + - [vanchin/deepseek-v3.1-terminus](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1-terminus.md) + - [deepseek-r1](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1.md) + - [deepseek-r1-distill-qwen-1.5b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-1-5b.md) + - [deepseek-r1-distill-qwen-14b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-14b.md) + - [deepseek-r1-distill-qwen-7b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-7b.md) + - [deepseek-r1-distill-qwen-32b](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-32b.md) + - [vanchin/deepseek-ocr](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-ocr.md) + - [kimi/kimi-k3](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k3.md) + - [kimi-k2.7-code](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code.md) + - [kimi/kimi-k2.7-code-highspeed](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code-highspeed.md) + - [kimi-k2.5](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-5.md) + - [kimi-k2.6](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-6.md) + - [Moonshot-Kimi-K2-Instruct](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/moonshot-kimi-k2-instruct.md) + - [kimi-k2-thinking](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-thinking.md) + - [glm-5.2-fast-preview](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-fast.md) + - [glm-5.2-us](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-us.md) + - [glm-5.2](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2.md) + - [glm-5](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-9.md) + - [glm-4.7](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-7.md) + - [glm-5.1](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-1.md) + - [glm-4.6](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-6.md) + - [glm-4.5](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4.md) + - [glm-4.5-air](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-5-air.md) + - [MiniMax/MiniMax-M3](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m3.md) + - [MiniMax/MiniMax-M2.7](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-7.md) + - [MiniMax-M2.1](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2.md) + - [MiniMax-M2.5](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-5.md) + - [xiaomi/mimo-v2.5-pro](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/mimo-v2-5-pro.md) + - [stepfun/step-3.7-flash](raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/step-3-7-flash.md) + - **视觉理解** + - [qwen3-vl-flash](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-flash.md) + - [qwen3-vl-flash-us](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-flash-us.md) + - [qwen3-vl-plus](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-plus.md) + - [qwen-vl-max](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen-vl-max.md) + - [qwen-vl-ocr](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwenvl-ocr.md) + - [qwen-vl-plus](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen-vl-plus.md) + - [qwen3.5-ocr](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-5-ocr.md) + - [gui-plus](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/gui-plus.md) + - [qvq-max](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qvq-max.md) + - [qvq-plus](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qvq-plus.md) + - [qwen3-vl-235b-a22b-instruct](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-235b-a22b-instruct.md) + - [qwen3-vl-235b-a22b-thinking](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-235b-a22b-thinking.md) + - [qwen3-vl-30b-a3b-instruct](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-30b-a3b-instruct.md) + - [qwen3-vl-30b-a3b-thinking](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-30b-a3b-thinking.md) + - [qwen3-vl-32b-instruct](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-32b-instruct.md) + - [qwen3-vl-32b-thinking](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-32b-thinking.md) + - [qwen3-vl-8b-instruct](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-8b-instruct.md) + - [qwen3-vl-8b-thinking](raw/model-user-guide/support/model-studio-model-list/model-list-visual-understanding/qwen3-vl-8b-thinking.md) + - **全模态** + - [qwen3.5-omni-plus](raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus.md) + - [qwen3.5-omni-flash](raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash.md) + - [qwen3-omni-flash](raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash.md) + - [qwen-omni-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo.md) + - [qwen3.5-omni-plus-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus-realtime.md) + - [qwen3.5-omni-flash-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash-realtime.md) + - [qwen3-omni-flash-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash-realtime.md) + - [qwen-omni-turbo-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo-realtime.md) + - [qwen2.5-omni-7b](raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen2-5-omni-7b.md) + - **图像生成** + - [qwen-image-2.0-pro](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0-pro.md) + - [qwen-image-3.0-pro](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-3-0-pro.md) + - [qwen-image-2.0](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0.md) + - [qwen-image-edit-max](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-max.md) + - [qwen-image-edit-plus](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-plus.md) + - [qwen-image-edit](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-qwen-image-edit.md) + - [qwen-image-max](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-max.md) + - [qwen-image-plus](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-plus.md) + - [qwen-image](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image.md) + - [wan2.7-image](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image.md) + - [qwen-mt-image](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-mt-image.md) + - [wan2.7-image-pro](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image-pro.md) + - [wan2.6-image](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-image.md) + - [wan2.6-t2i](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-t2i.md) + - [wan2.5-t2i-preview](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-t2i.md) + - [wan2.2-t2i-flash](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-flash.md) + - [wan2.5-i2i-preview](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-i2i.md) + - [wan2.2-t2i-plus](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-plus.md) + - [wan2.1-t2i-plus](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-plus.md) + - [wanx2.1-imageedit](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-imageedit.md) + - [wanx2.1-t2i-plus](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-plus.md) + - [wan2.1-t2i-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-turbo.md) + - [wanx2.1-t2i-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-turbo.md) + - [wanx-background-generation-v2](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-background-generation-v2.md) + - [wanx-sketch-to-image-lite](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-sketch-to-image-lite.md) + - [wanx2.0-t2i-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-0-t2i-turbo.md) + - [wanx-poster-generation-v1](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-poster-generation-v1.md) + - [wanx-style-repaint-v1](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-style-repaint-v1.md) + - [wanx-virtualmodel](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-virtualmodel.md) + - [wanx-x-painting](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-x-painting.md) + - [wanx-v1](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-v1.md) + - [wordart-semantic](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-semantic.md) + - [wordart-texture](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-texture.md) + - [z-image-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/z-image-turbo.md) + - [aitryon-parsing-v1](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-parsing-v1.md) + - [shoemodel-v1](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/shoemodel-v1.md) + - [virtualmodel-v2](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/virtualmodel-v2.md) + - [aitryon-plus](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-plus.md) + - [aitryon](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon.md) + - [aitryon-refiner](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-refiner.md) + - [facechain-facedetect](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/facechain-facedetect.md) + - [facechain-generation](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-facechain-generation.md) + - [image-erase-completion](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-erase-completion.md) + - [image-instance-segmentation](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-image-instance-segmentation.md) + - [image-out-painting](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-out-painting.md) + - [kling/kling-v3-image-generation](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-image-generation.md) + - [kling/kling-v3-omni-image-generation](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-omni-image-generation.md) + - [vidu/vidu-image_reference2image](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/vidu-image-reference2image.md) + - [vidu/viduq2-fast_reference2image](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-fast-reference2image.md) + - [vidu/viduq2-pro_reference2image](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-pro-reference2image.md) + - [vidu/viduq3-fast_reference2image](raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq3-fast-reference2image.md) + - **视频生成** + - [happyhorse-1.1-t2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-1-t2v.md) + - [happyhorse-1.0-t2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-0-t2v.md) + - [happyhorse-1.1-i2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-1-i2v.md) + - [happyhorse-1.0-i2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-0-i2v.md) + - [happyhorse-1.1-r2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-1-r2v.md) + - [happyhorse-1.0-r2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-0-r2v.md) + - [wan2.6-t2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-t2v.md) + - [wan2.7-t2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-7-t2v.md) + - [happyhorse-1.0-video-edit](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/happyhorse-1-0-video-edit.md) + - [wan2.6-t2v-us](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-t2v-us.md) + - [wan2.5-t2v-preview](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-5-t2v.md) + - [wan2.2-t2v-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-t2v-plus.md) + - [wanx2.1-t2v-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-t2v-plus.md) + - [wan2.1-t2v-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-t2v-plus.md) + - [wan2.1-t2v-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-t2v-turbo.md) + - [wanx2.1-t2v-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-t2v-turbo.md) + - [wan2.7-i2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-7-i2v.md) + - [wan2.6-i2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-i2v.md) + - [wan2.6-i2v-us](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-i2v-us.md) + - [wan2.6-i2v-flash](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-i2v-flash.md) + - [wan2.5-i2v-preview](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-5-i2v.md) + - [wan2.2-i2v-flash](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-i2v-flash.md) + - [wanx2.1-i2v-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-i2v-plus.md) + - [wan2.2-i2v-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-i2v-plus.md) + - [wanx2.1-i2v-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-i2v-turbo.md) + - [wan2.1-i2v-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-i2v-plus.md) + - [wan2.1-i2v-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-i2v-turbo.md) + - [wan2.7-r2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-7-r2v.md) + - [wan2.6-r2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-r2v.md) + - [wan2.6-r2v-flash](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-6-r2v-flash.md) + - [wan2.2-s2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-s2v.md) + - [wan2.2-s2v-detect](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-s2v-detect.md) + - [wan2.2-kf2v-flash](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-kf2v-flash.md) + - [wanx2.1-kf2v-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-kf2v-plus.md) + - [wan2.1-kf2v-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-kf2v-plus.md) + - [wan2.2-animate-mix](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-animate-mix.md) + - [wan2.2-animate-move](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-2-animate-move.md) + - [wan2.7-videoedit](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-7-videoedit.md) + - [wanx2.1-vace-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wanx2-1-vace-plus.md) + - [wan2.1-vace-plus](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/wan2-1-vace-plus.md) + - [animate-anyone-detect-gen2](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/animate-anyone-detect-gen2.md) + - [animate-anyone-gen2](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/animate-anyone-gen2.md) + - [animate-anyone-template-gen2](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/animate-anyone-template-gen2.md) + - [emo-v1](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/emo-v1.md) + - [emo-detect-v1](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/emo-detect-v1.md) + - [emoji-v1](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/emoji-v1.md) + - [emoji-detect-v1](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/emoji-detect-v1.md) + - [liveportrait](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/smart-portrait-liveport.md) + - [liveportrait-detect](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/liveportrait-detect.md) + - [video-style-transform](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/video-style-transform.md) + - [videoretalk](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/acoustic-portraits-videoretalk.md) + - [kling/kling-v3-omni-video-generation](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/kling-v3-omni-video-generation.md) + - [kling/kling-v3-video-generation](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/kling-v3-video-generation.md) + - [pixverse/pixverse-motioncontrol](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-motioncontrol.md) + - [pixverse/pixverse-lipsync](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-lipsync.md) + - [pixverse/pixverse-upscale](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-upscale.md) + - [pixverse/pixverse-v6-it2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v6-it2v.md) + - [pixverse/pixverse-v6-kf2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v6-kf2v.md) + - [pixverse/pixverse-v6-r2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v6-r2v.md) + - [pixverse/pixverse-v6-t2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v6-t2v.md) + - [pixverse/pixverse-v5.6-it2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v5-6-it2v.md) + - [pixverse/pixverse-v5.6-kf2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v5-6-kf2v.md) + - [pixverse/pixverse-v5.6-t2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v5-6-t2v.md) + - [pixverse/pixverse-v5.6-r2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-v5-6-r2v.md) + - [pixverse/pixverse-c1-it2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-c1-it2v.md) + - [pixverse/pixverse-c1-kf2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-c1-kf2v.md) + - [pixverse/pixverse-c1-r2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-c1-r2v.md) + - [vidu/viduq3-ad_reference2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-ad-reference2video.md) + - [vidu/viduq3-drama_reference2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-drama-reference2video.md) + - [pixverse/pixverse-c1-t2v](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/pixverse-c1-t2v.md) + - [vidu/viduq3-pro-fast_img2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-pro-fast-img2video.md) + - [vidu/viduq2-pro-fast_img2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-pro-fast-img2video.md) + - [vidu/viduq2-pro_start-end2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-pro-start-end2video.md) + - [vidu/viduq2-pro_img2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-pro-img2video.md) + - [vidu/viduq2-pro_reference2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-pro-reference2video.md) + - [vidu/viduq2-turbo_img2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-turbo-img2video.md) + - [vidu/viduq2-turbo_start-end2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-turbo-start-end2video.md) + - [vidu/viduq2_reference2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-reference2video.md) + - [vidu/viduq2_text2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq2-text2video.md) + - [vidu/viduq3-mix_reference2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-mix-reference2video.md) + - [vidu/viduq3-pro_img2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-pro-img2video.md) + - [vidu/viduq3-pro_start-end2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-pro-start-end2video.md) + - [vidu/viduq3-pro_text2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-pro-text2video.md) + - [vidu/viduq3-turbo_reference2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-turbo-reference2video.md) + - [vidu/viduq3-turbo_img2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-turbo-img2video.md) + - [vidu/viduq3-turbo_start-end2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-turbo-start-end2video.md) + - [vidu/viduq3-turbo_text2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-turbo-text2video.md) + - [vidu/viduq3_reference2video](raw/model-user-guide/support/model-studio-model-list/model-list-video-generation/viduq3-reference2video.md) + - **3D生成** + - [Tripo/Tripo-P1.0](raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-p1-0.md) + - [Tripo/Tripo-H3.1](raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-h3-1.md) + - **语音识别** + - [qwen3.5-livetranslate-flash-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-6.md) + - [qwen3-livetranslate-flash](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash.md) + - [qwen3-livetranslate-flash-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash-realtime.md) + - [qwen3-asr-flash-us](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-us.md) + - [qwen3-asr-flash-filetrans](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-filetrans.md) + - [qwen3-asr-flash](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash.md) + - [qwen3-asr-flash-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-realtime.md) + - [fun-asr](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr.md) + - [fun-asr-flash-8k-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash-8k-realtime.md) + - [fun-asr-mtl](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-mtl.md) + - [fun-asr-flash-2026-06-15](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash.md) + - [fun-asr-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-realtime.md) + - [gummy-chat-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-chat-v1.md) + - [gummy-realtime-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-realtime-v1.md) + - [paraformer-8k-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v1.md) + - [paraformer-8k-v2](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v2.md) + - [paraformer-mtl-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-mtl-v1.md) + - [paraformer-realtime-8k-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v1.md) + - [paraformer-realtime-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v1.md) + - [paraformer-realtime-v2](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v2.md) + - [paraformer-realtime-8k-v2](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v2.md) + - [paraformer-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v1.md) + - [paraformer-v2](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v2.md) + - [qwen3-omni-30b-a3b-captioner](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-omni-30b-a3b-captioner.md) + - [speech-biasing](raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/speech-biasing.md) + - **语音合成** + - [qwen3-tts-flash](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash.md) + - [qwen3-tts-instruct-flash](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash.md) + - [qwen3-tts-vc-2026-01-22](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc.md) + - [qwen-tts](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts.md) + - [qwen3-tts-vd-2026-01-26](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd.md) + - [qwen3-tts-instruct-flash-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash-realtime.md) + - [qwen3-tts-flash-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash-realtime.md) + - [qwen3-tts-vc-realtime-2026-01-15](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc-realtime.md) + - [qwen3-tts-vd-realtime-2026-01-15](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd-realtime.md) + - [qwen-tts-realtime](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts-realtime.md) + - [qwen-voice-design](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-design.md) + - [qwen-voice-enrollment](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-enrollment.md) + - [cosyvoice-v3.5-plus](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-plus.md) + - [cosyvoice-v3.5-flash](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-flash.md) + - [cosyvoice-v3-flash](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-flash.md) + - [cosyvoice-v3-plus](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-plus.md) + - [cosyvoice-v2](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v2.md) + - [cosyvoice-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v1.md) + - [cosyvoice-clone-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-clone-v1.md) + - [fun-music-v1](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/fun-music-v1.md) + - [Sambert语音合成](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/sambert.md) + - [voice-enrollment](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/voice-enrollment.md) + - [fun-music-preview](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/music-generation-preview.md) + - [MiniMax/speech-2.8-hd](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-hd.md) + - [MiniMax/speech-2.8-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-turbo.md) + - [MiniMax/speech-02-hd](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-hd.md) + - [MiniMax/speech-02-turbo](raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-turbo.md) + - **向量与重排序** + - [qwen3.7-text-embedding](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-7-text-embedding.md) + - [qwen3-vl-embedding](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-embedding.md) + - [qwen2.5-vl-embedding](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen2-5-vl-embedding.md) + - [qwen3-vl-rerank](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-rerank.md) + - [gte-rerank-v2](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/gte-rerank-v2.md) + - [qwen3-rerank](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-rerank.md) + - [text-embedding-v4](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v4.md) + - [text-embedding-v3](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v3.md) + - [text-embedding-v2](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v2.md) + - [text-embedding-async-v2](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v2.md) + - [text-embedding-v1](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v1.md) + - [text-embedding-async-v1](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v1.md) + - [multimodal-embedding-v1](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/multimodal-embedding-v1.md) + - [tongyi-embedding-vision-flash](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-flash.md) + - [tongyi-embedding-vision-plus](raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-plus.md) - [常见问题](raw/model-user-guide/support/faq-about-alibaba-cloud-model-studio.md) - [相关协议](raw/model-user-guide/support/related-agreements.md) + - [阿里云百炼平台售后服务范围说明](raw/model-user-guide/support/after-sales-service-scope.md) - **产品动态** - [模型平台功能更新](raw/model-user-guide/release-notes/model-release-notes.md) - [模型上下架与更新](raw/model-user-guide/release-notes/newly-released-models.md) @@ -142,10 +493,20 @@ - **开始使用** - [0代码构建私有知识问答应用](raw/application-user-guide/start-using/build-knowledge-base-qa-assistant-without-coding.md) - [应用功能动态](raw/application-user-guide/start-using/application-release-notes.md) +- **记忆库** + - [记忆库](raw/application-user-guide/memory-library-overview/memory-library.md) + - [为 OpenClaw 配置长期记忆插件](raw/application-user-guide/memory-library-overview/modelstudio-memory-for-openclaw.md) + - [长期记忆 API](raw/application-user-guide/memory-library-overview/long-term-memory-2-0.md) +- **Prompt** + - [Prompt模板概述](raw/application-user-guide/prompt/prompt-template.md) + - [自定义Prompt模板](raw/application-user-guide/prompt/prompt-custom-template.md) + - [Prompt自动优化](raw/application-user-guide/prompt/optimize-prompt.md) + - [使用Prompt样例库优化模型输出](raw/application-user-guide/prompt/prompt-sample-optimization.md) + - [基于大模型输入输出样例的Prompt自动优化](raw/application-user-guide/prompt/prompt-feedback-optimization.md) - **应用开发** - [应用类型介绍](raw/application-user-guide/llm-application/application-introduction.md) + - [新版智能体应用](raw/application-user-guide/llm-application/new-single-agent-application.md) - [智能体应用](raw/application-user-guide/llm-application/single-agent-application.md) - - [新版智能体应用（Agent 2.0）](raw/application-user-guide/llm-application/new-single-agent-application.md) - [高代码应用](raw/application-user-guide/llm-application/rich-code-application.md) - [文件问答](raw/application-user-guide/llm-application/file-q-a.md) - [工作流应用](raw/application-user-guide/llm-application/workflow-application.md) @@ -153,59 +514,51 @@ - [概述](raw/application-user-guide/managed-agents/managed-agents-introduction.md) - [快速开始](raw/application-user-guide/managed-agents/managed-agents-quick-start.md) - [构建 Agent](raw/application-user-guide/managed-agents/managed-agents-agent.md) - - [配置 Agent 环境](raw/application-user-guide/managed-agents/managed-agents-environment.md) - [委派任务给 Agent](raw/application-user-guide/managed-agents/managed-agents-session.md) + - [配置 Agent 环境](raw/application-user-guide/managed-agents/managed-agents-environment.md) - [Agent 上下文管理](raw/application-user-guide/managed-agents/managed-agents-context.md) -- **Prompt** - - [Prompt模板概述](raw/application-user-guide/prompt/prompt-template.md) - - [Prompt自动优化](raw/application-user-guide/prompt/optimize-prompt.md) - - [使用Prompt样例库优化模型输出](raw/application-user-guide/prompt/prompt-sample-optimization.md) - - [基于大模型输入输出样例的Prompt自动优化](raw/application-user-guide/prompt/prompt-feedback-optimization.md) - - [自定义Prompt模板](raw/application-user-guide/prompt/prompt-custom-template.md) -- **记忆库** - - [记忆库](raw/application-user-guide/memory-library-overview/memory-library.md) - - [为 OpenClaw 配置长期记忆插件](raw/application-user-guide/memory-library-overview/modelstudio-memory-for-openclaw.md) - - [长期记忆 API](raw/application-user-guide/memory-library-overview/long-term-memory-2-0.md) - **知识库（RAG）** - - [知识库](raw/application-user-guide/knowledge-base/rag-knowledge-base.md) - [RAG效果优化](raw/application-user-guide/knowledge-base/rag-optimization.md) + - [知识库](raw/application-user-guide/knowledge-base/rag-knowledge-base.md) - [知识库日志与监控](raw/application-user-guide/knowledge-base/rag-knowledge-base-log-monitoring.md) - [知识库配额与限制](raw/application-user-guide/knowledge-base/rag-knowledge-base-specifications.md) + - [知识问答](raw/application-user-guide/knowledge-base/rag-knowledge-qa.md) - [知识检索](raw/application-user-guide/knowledge-base/rag-knowledge-retrieval.md) - - [知识库计费说明](raw/application-user-guide/knowledge-base/billing-for-knowledge-base.md) - [知识库API指南](raw/application-user-guide/knowledge-base/rag-knowledge-base-api-guide.md) - - [知识问答](raw/application-user-guide/knowledge-base/rag-knowledge-qa.md) + - [知识库计费说明](raw/application-user-guide/knowledge-base/billing-for-knowledge-base.md) - **数据连接** - [数据连接](raw/application-user-guide/data-connection-overview/data-connection.md) - **Skill** - [Skill](raw/application-user-guide/skill/introduction-to-skill.md) -- **插件** - - [插件概述](raw/application-user-guide/plug-in/plug-in-overview.md) - - [官方和第三方插件](raw/application-user-guide/plug-in/plugins.md) - - [自定义插件](raw/application-user-guide/plug-in/custom-plug-ins.md) - **MCP** + - [官方 MCP 服务](raw/application-user-guide/model-context-protocol/official-and-third-party-mcp.md) - [模型上下文协议（MCP）](raw/application-user-guide/model-context-protocol/mcp-introduction.md) - [自定义 MCP 服务](raw/application-user-guide/model-context-protocol/custom-mcp.md) - - [官方 MCP 服务](raw/application-user-guide/model-context-protocol/official-and-third-party-mcp.md) - [外部调用](raw/application-user-guide/model-context-protocol/mcp-external-calls.md) - [MCP 常见问题](raw/application-user-guide/model-context-protocol/mcp-faq.md) +- **插件** + - [插件概述](raw/application-user-guide/plug-in/plug-in-overview.md) + - [官方和第三方插件](raw/application-user-guide/plug-in/plugins.md) + - [自定义插件](raw/application-user-guide/plug-in/custom-plug-ins.md) - **应用发布与分享** - - [使用智能体或工作流作为组件](raw/application-user-guide/application-publishing-and-sharing/use-agent-or-workflow-as-component.md) - [分享智能体应用](raw/application-user-guide/application-publishing-and-sharing/share-an-application.md) + - [使用智能体或工作流作为组件](raw/application-user-guide/application-publishing-and-sharing/use-agent-or-workflow-as-component.md) - [UI设计器](raw/application-user-guide/application-publishing-and-sharing/ui-designer.md) -- **应用调用** - - [应用的自定义参数传递](raw/application-user-guide/bailian-application-calling/pass-through-of-application-parameters.md) - - [调用工作流应用](raw/application-user-guide/bailian-application-calling/invoke-workflow-application.md) - - [调用智能体应用](raw/application-user-guide/bailian-application-calling/call-single-agent-application.md) - **应用评测** - **新版应用评测** - [新版评测集](raw/application-user-guide/application-evaluation/new-version-of-application-evaluation/new-version-of-evaluation-set.md) - [评测任务](raw/application-user-guide/application-evaluation/new-version-of-application-evaluation/evaluation-task.md) - [标签管理](raw/application-user-guide/application-evaluation/new-version-of-application-evaluation/label-management.md) - [评估器](raw/application-user-guide/application-evaluation/new-version-of-application-evaluation/grader.md) + - [自动评测](raw/application-user-guide/application-evaluation/application-auto-evaluation.md) - [手动评测](raw/application-user-guide/application-evaluation/evaluate-manual-application.md) - [评测集](raw/application-user-guide/application-evaluation/application-evaluation-dataset.md) - - [自动评测](raw/application-user-guide/application-evaluation/application-auto-evaluation.md) +- **应用调用** + - [应用的自定义参数传递](raw/application-user-guide/bailian-application-calling/pass-through-of-application-parameters.md) + - [调用智能体应用](raw/application-user-guide/bailian-application-calling/call-single-agent-application.md) + - [调用工作流应用](raw/application-user-guide/bailian-application-calling/invoke-workflow-application.md) +- **应用观测** + - [应用观测](raw/application-user-guide/application-monitoring/application-observation.md) - **应用广场** - **官方应用-通义拍照解题辅导** - **API参考** @@ -228,8 +581,8 @@ - **官方应用-多模态交互开发套件** - **使用指南** - [应用创建](raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-app-creation.md) - - [应用配置](raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-app-configuration.md) - [应用体验与发布](raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-app-experience-and-publishing.md) + - [应用配置](raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-app-configuration.md) - [百炼应用推荐模板](raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/agent-template.md) - [指令列表](raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/instruction-list.md) - [对话日志接入](raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-chatlog.md) @@ -237,30 +590,33 @@ - [音色列表](raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-timbre-list.md) - [三方Agent接入](raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-integration-a2a.md) - **SDK安装** - - [服务端Java SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/multimodal-sdk-java.md) - [服务端Python SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/multimodal-sdk-python.md) + - [服务端Java SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/multimodal-sdk-java.md) - [服务端 Go SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/server-go-sdk.md) - - [移动端iOS SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/multimodal-sdk-ios.md) - [移动端Android SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/multimodal-sdk-android.md) + - [移动端iOS SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/multimodal-sdk-ios.md) - [移动端Android Lite SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/multimodal-sdk-android-lite.md) - [移动端iOS Lite SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/multimodal-sdk-ios-lite.md) - [Linux C++ SDK](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/multimodal-sdk-linux.md) - [RTOS C SDK（License模式）](raw/application-user-guide/application-gallery/multimodal-products/multimodal-sdk/mmi-rtos-sdk.md) - **API参考** - [实时多模态交互协议（WebSocket）](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-interaction-protocol.md) - - [HTTP协议](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-http-protocol.md) - [调用官方Agent](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/official-agent.md) + - [HTTP协议](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-http-protocol.md) - [调用三方语音模型](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/third-party-voice-integration.md) - [调用插件](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/call-plugins.md) - [管理热词](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/management-hot-words.md) - - [多模态对话结果 extra_info 说明](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/extra-info-description.md) - [多模态交互套件-错误码](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-error-code.md) + - [多模态对话结果 extra_info 说明](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/extra-info-description.md) - [长期记忆开放接口](raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/long-term-memory-api.md) - **最佳实践** - **接入百炼及三方Agent** - [百炼及三方Agent直连调用](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/bailian-and-tripartite-agent/agent-direct-call.md) - [接入百炼智能体应用](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/bailian-and-tripartite-agent/multimodal-call-app.md) - [接入百炼工作流应用](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/bailian-and-tripartite-agent/multimodal-call-workflow.md) + - **接入拍照问答Agent** + - [通过HTTP协议接入拍照问答Agent](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/vqa-agent/vqa-agent-through-the-http-protocol.md) + - [通过WebSocket协议接入拍照问答Agent和语音合成](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/vqa-agent/vqa-agent-via-websocket-protocol.md) - **接入图像生成Agent** - [通过HTTP协议接入图像生成Agent](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/generateimgagent/image-agent.md) - [语音请求直通图像生成Agent(websocket)](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/generateimgagent/audio-to-generateimgagent.md) @@ -268,15 +624,12 @@ - [录音纪要Agent使用教程](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/fast-integrate-tingwu-meeting-agent/recording-summary-agent-tutorial.md) - [快速集成智能纪要Agent](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/fast-integrate-tingwu-meeting-agent/fast-integrate-offline-tingwu-meeting-agent.md) - [实时转写能力集成](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/fast-integrate-tingwu-meeting-agent/realtime-tingwu-meeting-agent-integration.md) - - **接入拍照问答Agent** - - [通过HTTP协议接入拍照问答Agent](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/vqa-agent/vqa-agent-through-the-http-protocol.md) - - [通过WebSocket协议接入拍照问答Agent和语音合成](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/vqa-agent/vqa-agent-via-websocket-protocol.md) - [接入多模态备忘录Agent](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/multimodal-memo-agent.md) - - [接入音乐电台Agent](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/music-agent.md) - [接入视频通话Agent](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/live-api-integration.md) + - [接入音乐电台Agent](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/music-agent.md) - [动作情绪控制实践](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/action-emotion-control-practice.md) - - [自定义指令实践](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/custom-directive.md) - [自定义对话角色实践](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/custom-role.md) + - [自定义指令实践](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/custom-directive.md) - [声音复刻及声音设计实践](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/voice-cloning-and-voice-design.md) - [音频采集和播放说明](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/audio-capture-and-playback-instructions.md) - [基于RTOS SDK (License模式) 实现聊天能力](raw/application-user-guide/application-gallery/multimodal-products/multimodal-best-practices/chat-capability-based-on-rtos-sdk.md) @@ -284,13 +637,6 @@ - [产品计费](raw/application-user-guide/application-gallery/multimodal-products/product-billing.md) - [多模态交互开发套件常见问题](raw/application-user-guide/application-gallery/multimodal-products/multimodal-products-faq.md) - **官方应用-伶鹊CCAI-对话分析AIO** - - **使用指南** - - [如何开通伶鹊CCAI-对话分析AIO](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/product-activation.md) - - [如何对应用进行编辑、删除等管理,如何进行API调用、如何查看调用量](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/application-management.md) - - [如何进行基于对话分析Agent方式创建应用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/create-an-application-based-on-dialogue-analysis-agent.md) - - [如何基于自定义方式创建应用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/create-an-application-based-on-a-custom-method.md) - - [热词组配置管理与使用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/hot-phrase-management.md) - - [知识库的使用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/using-the-knowledge-base.md) - **API参考** - **API目录** - **热词管理** @@ -301,54 +647,61 @@ - [GetVocab - 获取热词](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-dir-hot-word-management/api-contactcenterai-2024-06-03-getvocab.md) - **不推荐或白名单开放** - [AnalyzeAudioSync - 语音文件实时分析](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-dir-not-recommended-or-whitelisted-open/api-contactcenterai-2024-06-03-analyzeaudiosync.md) - - [RunCompletionMessage - 使用原生Prompt调用通义晓蜜CCAI-对话分析AIO应用](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-runcompletionmessage.md) - [RunCompletion - 通过模版ID调用通义晓蜜CCAI-对话分析AIO应用](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-runcompletion.md) + - [RunCompletionMessage - 使用原生Prompt调用通义晓蜜CCAI-对话分析AIO应用](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-runcompletionmessage.md) - [AnalyzeConversation - 通过任务类型调用通义晓蜜CCAI-对话分析AIO应用](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-analyzeconversation.md) - [CreateTask - 通过上传离线任务数据进行通义晓蜜CCAI-对话分析](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-createtask.md) - [GetTaskResult - 通过任务ID获取离线任务分析结果](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-gettaskresult.md) - - [AnalyzeImage - 图片内容分析](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-analyzeimage.md) - [GeneralAnalyzeImage - 通用图片分析](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-generalanalyzeimage.md) + - [AnalyzeImage - 图片内容分析](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-dir/api-contactcenterai-2024-06-03-analyzeimage.md) - [API概览](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-overview.md) - [服务接入点](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-endpoint.md) - [版本说明](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/api-reference-2/api-contactcenterai-2024-06-03-changeset.md) + - **使用指南** + - [如何开通伶鹊CCAI-对话分析AIO](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/product-activation.md) + - [如何对应用进行编辑、删除等管理,如何进行API调用、如何查看调用量](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/application-management.md) + - [如何进行基于对话分析Agent方式创建应用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/create-an-application-based-on-dialogue-analysis-agent.md) + - [如何基于自定义方式创建应用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/create-an-application-based-on-a-custom-method.md) + - [知识库的使用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/using-the-knowledge-base.md) + - [热词组配置管理与使用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/lingque-ccai-aio-user-guide/hot-phrase-management.md) - **最佳实践** - [客服服务质检最佳实践](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/best-practices/customer-service-quality-inspection-best-practices.md) - [字段信息抽取最佳实践](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/best-practices/best-practices-for-automatic-work-order-generation.md) - [摘要生成（含摘要/标题/关键词）最佳实践](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/best-practices/summary-best-practices.md) - **接口调用示例** - - [通过原生Prompt调用伶鹊CCAI-对话分析AIO应用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/use-native-prompt-to-call-tongyi-xiaomi-ccai-aio.md) - [通过模板ID调用伶鹊CCAI-对话分析AIO应用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/use-template-id-to-call-tongyi-xiaomi-ccai-aio.md) + - [通过原生Prompt调用伶鹊CCAI-对话分析AIO应用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/use-native-prompt-to-call-tongyi-xiaomi-ccai-aio.md) - [通过任务类型调用伶鹊CCAI-对话分析AIO应用](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/call-tongyi-xiaomi-ccai-dialogue-analysis-aio-application-through-task-type.md) - [通过上传离线任务数据进行伶鹊CCAI-对话分析](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/tongyi-xiaomi-ccai-dialogue-analysis-by-uploading-offline-task-data.md) - [ROA风格请求体&签名机制](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/roa-style-request-body-signature-mechanism.md) - [通过伶鹊CCAI-对话分析AIO应用进行图片分析](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/picture-analysis-through-tongyi-xiaomi-ccai-dialogue-analysis-aio-application.md) - - [伶鹊CCAI-对话分析RAM子账号使用方式和授权操作](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/lingque-ccai-dialogue-analytics-ram-subaccount-usage-and-authorization-procedures.md) - - [通义晓蜜CCAI更新公告](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/tongyi-xiaomi-ccai-update-announcement.md) + - [伶鹊CCAI-对话分析RAM子账号使用方式和授权操作](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/use-and-authorize-ram-users-for-ccai-dialogue-analysis.md) - [产品概述](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/product-overview-1.md) + - [通义晓蜜CCAI更新公告](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/tongyi-xiaomi-ccai-update-announcement.md) - [伶鹊CCAI-对话分析AIO产品计费](raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/billing-description-magpie-ccai-dialogue-analysis-aio.md) - [CCAI如何进行集成；查看技术集成方案](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-dialogue-analysis-aio/technology-integration-scheme.md) - **官方应用-全妙轻应用系列** - - **计费说明（全妙轻应用）** - - [电商零售推广文案写作计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/e-commerce-retail-promotion-copywriting-billing.md) - - [电商文案智能可控生成计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/e-commerce-copy-intelligent-controllable-generation-billing.md) - - [影视传媒视频理解计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/film-and-television-media-video-understanding-billing.md) - - [影视互娱剧本创作计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/film-and-television-mutual-entertainment-script-creation-billing.md) - - [泛企业VOC挖掘计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/pan-enterprise-voc-mining-billing.md) - - [车机网络热点信息互动问答计费文档](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/car-machine-network-hot-information-interactive-question-and-answer-billing.md) - - [泛企业线索挖掘计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/pan-enterprise-lead-mining-billing.md) - - [网络内容安全审核计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/network-content-security-audit-billing.md) - - [作文批改计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/composition-correction-billing.md) - - [视频智能拆条计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/video-smart-strip-billing.md) - **使用指南** - [电商文案智能可控生成](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/intelligent-and-controllable-generation-of-e-commerce-copywriting.md) - [传媒/零售文章风格与格式学习](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/media-retail-article-style-and-format-learning.md) - - [影视传媒视频理解](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/media-video-understanding.md) - [影视互娱剧本创作](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/film-and-television-script-creation.md) - - [车机网络热点信息互动问答](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/car-machine-content-platform-news-hot-list-interaction.md) - [泛企业VOC挖掘](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/pan-enterprise-voc-mining.md) + - [车机网络热点信息互动问答](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/car-machine-content-platform-news-hot-list-interaction.md) + - [影视传媒视频理解](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/media-video-understanding.md) - [泛企业线索挖掘](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/pan-enterprise-clue-mining.md) - - [作文批改助手](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/composition-correction-assistant.md) - [网络内容安全审核](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/network-content-security-audit.md) + - [作文批改助手](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-guidelines-for-use/composition-correction-assistant.md) + - **计费说明（全妙轻应用）** + - [电商文案智能可控生成计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/e-commerce-copy-intelligent-controllable-generation-billing.md) + - [电商零售推广文案写作计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/e-commerce-retail-promotion-copywriting-billing.md) + - [影视互娱剧本创作计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/film-and-television-mutual-entertainment-script-creation-billing.md) + - [影视传媒视频理解计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/film-and-television-media-video-understanding-billing.md) + - [泛企业线索挖掘计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/pan-enterprise-lead-mining-billing.md) + - [泛企业VOC挖掘计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/pan-enterprise-voc-mining-billing.md) + - [网络内容安全审核计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/network-content-security-audit-billing.md) + - [车机网络热点信息互动问答计费文档](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/car-machine-network-hot-information-interactive-question-and-answer-billing.md) + - [作文批改计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/composition-correction-billing.md) + - [视频智能拆条计费](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-billing-document/video-smart-strip-billing.md) - **开发文档** - **最佳实践** - [应用视频理解和一键成片的最佳实践](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/light-application-best-practices/best-practices-for-applying-video-understanding-and-one-click-film.md) @@ -358,57 +711,57 @@ - **数据结构** - [ModelUsage](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-struct-dir/api-quanmiaolightapp-2024-08-01-struct-modelusage.md) - **API目录** - - **电商零售推广文案写作** - - [RunMarketingInformationWriting - 电商零售推广文案写作](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-e-commerce-retail-promotion-copy-writing/api-quanmiaolightapp-2024-08-01-runmarketinginformationwriting.md) - - [RunMarketingInformationExtract - 电商零售内容实体抽取](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-e-commerce-retail-promotion-copy-writing/api-quanmiaolightapp-2024-08-01-runmarketinginformationextract.md) - - **传媒/零售文章风格与格式学习** - - [RunStyleWriting - 传媒/零售文章风格与格式学习](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-media-retail-article-style-and-format-learning/api-quanmiaolightapp-2024-08-01-runstylewriting.md) - **影视互娱剧本创作** - [RunScriptRefine - 影视互娱剧本创作-剧本整理](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-mutual-entertainment-script-creation/api-quanmiaolightapp-2024-08-01-runscriptrefine.md) - - [RunScriptPlanning - 影视互娱剧本创作-剧本策划](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-mutual-entertainment-script-creation/api-quanmiaolightapp-2024-08-01-runscriptplanning.md) - [RunScriptChat - 影视互娱剧本创作-交互式创作](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-mutual-entertainment-script-creation/api-quanmiaolightapp-2024-08-01-runscriptchat.md) + - [RunScriptPlanning - 影视互娱剧本创作-剧本策划](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-mutual-entertainment-script-creation/api-quanmiaolightapp-2024-08-01-runscriptplanning.md) - [RunScriptContinue - 影视互娱剧本创作-剧本续写](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-mutual-entertainment-script-creation/api-quanmiaolightapp-2024-08-01-runscriptcontinue.md) + - **传媒/零售文章风格与格式学习** + - [RunStyleWriting - 传媒/零售文章风格与格式学习](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-media-retail-article-style-and-format-learning/api-quanmiaolightapp-2024-08-01-runstylewriting.md) + - **电商零售推广文案写作** + - [RunMarketingInformationWriting - 电商零售推广文案写作](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-e-commerce-retail-promotion-copy-writing/api-quanmiaolightapp-2024-08-01-runmarketinginformationwriting.md) + - [RunMarketingInformationExtract - 电商零售内容实体抽取](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-e-commerce-retail-promotion-copy-writing/api-quanmiaolightapp-2024-08-01-runmarketinginformationextract.md) - **影视传媒视频理解** - - [SubmitVideoAnalysisTask - 视频理解-提交异步任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-submitvideoanalysistask.md) - [GetVideoAnalysisTask - 视频理解-获取异步任务状态和结果](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-getvideoanalysistask.md) + - [SubmitVideoAnalysisTask - 视频理解-提交异步任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-submitvideoanalysistask.md) - [UpdateVideoAnalysisConfig - 视频理解-更新配置](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-updatevideoanalysisconfig.md) - [GetVideoAnalysisConfig - 视频理解-获取配置](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-getvideoanalysisconfig.md) - - [RunVideoAnalysis - 视频理解-在线任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-runvideoanalysis.md) - [UpdateVideoAnalysisTask - 视频理解-修改异步任务状态](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-updatevideoanalysistask.md) + - [RunVideoAnalysis - 视频理解-在线任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-runvideoanalysis.md) - [UpdateVideoAnalysisTasks - 视频理解-批量取消任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-updatevideoanalysistasks.md) - **影视传媒智能拆条** - - [GetVideoDetectShotTask - 智能拆条-获取异步任务状态和结果](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-intelligent-strip/api-quanmiaolightapp-2024-08-01-getvideodetectshottask.md) - [SubmitVideoDetectShotTask - 智能拆条-提交异步任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-intelligent-strip/api-quanmiaolightapp-2024-08-01-submitvideodetectshottask.md) + - [GetVideoDetectShotTask - 智能拆条-获取异步任务状态和结果](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-intelligent-strip/api-quanmiaolightapp-2024-08-01-getvideodetectshottask.md) - [UpdateVideoDetectShotTask - 智能拆条-修改异步任务状态](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-intelligent-strip/api-quanmiaolightapp-2024-08-01-updatevideodetectshottask.md) - - [UpdateVideoDetectShotConfig - 智能拆条-更新配置](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-intelligent-strip/api-quanmiaolightapp-2024-08-01-updatevideodetectshotconfig.md) - [GetVideoDetectShotConfig - 智能拆条-获取配置](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-intelligent-strip/api-quanmiaolightapp-2024-08-01-getvideodetectshotconfig.md) + - [UpdateVideoDetectShotConfig - 智能拆条-更新配置](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-intelligent-strip/api-quanmiaolightapp-2024-08-01-updatevideodetectshotconfig.md) - [RunVideoDetectShot - 智能拆条-在线任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-intelligent-strip/api-quanmiaolightapp-2024-08-01-runvideodetectshot.md) + - **泛企业线索挖掘** + - [GenerateOutputFormat - 获取输出格式示例](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-pan-enterprise-clue-mining/api-quanmiaolightapp-2024-08-01-generateoutputformat.md) + - [RunTagMiningAnalysis - 标签挖掘分析](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-pan-enterprise-clue-mining/api-quanmiaolightapp-2024-08-01-runtagmininganalysis.md) - **车机网络热点信息互动问答** - [RunHotTopicChat - 播报单（热榜）问答](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-car-machine-network-hot-information-interactive-question-and-answer/api-quanmiaolightapp-2024-08-01-runhottopicchat.md) - [RunHotTopicSummary - 播报单热点自定义摘要生成](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-car-machine-network-hot-information-interactive-question-and-answer/api-quanmiaolightapp-2024-08-01-runhottopicsummary.md) - **泛企业VOC挖掘** - [RunEnterpriseVocAnalysis - 在线企业VOC分析](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-pan-enterprise-voc-mining/api-quanmiaolightapp-2024-08-01-runenterprisevocanalysis.md) - - **泛企业线索挖掘** - - [GenerateOutputFormat - 获取输出格式示例](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-pan-enterprise-clue-mining/api-quanmiaolightapp-2024-08-01-generateoutputformat.md) - - [RunTagMiningAnalysis - 标签挖掘分析](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-pan-enterprise-clue-mining/api-quanmiaolightapp-2024-08-01-runtagmininganalysis.md) - **网络内容安全审核** - [RunNetworkContentAudit - 网络内容安全审核](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-network-content-security-audit/api-quanmiaolightapp-2024-08-01-runnetworkcontentaudit.md) - **作文批改** - [RunEssayCorrection - 作文批改](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-composition-correction/api-quanmiaolightapp-2024-08-01-runessaycorrection.md) - [RunOcrParse - 图片OCR解析](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-composition-correction/api-quanmiaolightapp-2024-08-01-runocrparse.md) - - [SubmitEssayCorrectionTask - 提交作文批改任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-composition-correction/api-quanmiaolightapp-2024-08-01-submitessaycorrectiontask.md) - [GetEssayCorrectionTask - 获取作文批改任务结果](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-composition-correction/api-quanmiaolightapp-2024-08-01-getessaycorrectiontask.md) + - [SubmitEssayCorrectionTask - 提交作文批改任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-composition-correction/api-quanmiaolightapp-2024-08-01-submitessaycorrectiontask.md) - **其他** - [GenerateBroadcastNews - 播报单（热榜）热点推荐](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-generatebroadcastnews.md) - - [GetTagMiningAnalysisTask - 获取标签挖掘分析任务结果](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-gettagmininganalysistask.md) - - [ListHotTopicSummaries - 查询完整播报单（热榜）](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-listhottopicsummaries.md) - [SubmitTagMiningAnalysisTask - 提交标签挖掘分析任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-submittagmininganalysistask.md) + - [ListHotTopicSummaries - 查询完整播报单（热榜）](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-listhottopicsummaries.md) + - [GetTagMiningAnalysisTask - 获取标签挖掘分析任务结果](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-gettagmininganalysistask.md) - [HotNewsRecommend - 新闻热点推荐](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-hotnewsrecommend.md) - [GetFileContent - 获取文件内容](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-getfilecontent.md) - - [BatchQueryTaskStatus - 批量查询异步任务状态](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-batchquerytaskstatus.md) - - [CancelAsyncTask - 根据任务ID取消异步任务的执行](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-cancelasynctask.md) - [BatchCancelTasks - 批量取消异步任务](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-batchcanceltasks.md) + - [BatchQueryTaskStatus - 批量查询异步任务状态](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-batchquerytaskstatus.md) - [ExportAnalysisTagDetailByTaskId - 根据任务ID导出分析明细](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-exportanalysistagdetailbytaskid.md) + - [CancelAsyncTask - 根据任务ID取消异步任务的执行](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-cancelasynctask.md) - [GetEnterpriseVocAnalysisTask - 获取企业VOC分析任务结果](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-getenterprisevocanalysistask.md) - [GetTaskExecutionStatistics - 查询任务执行情况统计](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-gettaskexecutionstatistics.md) - [ListAnalysisTagDetailByTaskId - 获取挖掘结果明细列表](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-other/api-quanmiaolightapp-2024-08-01-listanalysistagdetailbytaskid.md) @@ -419,33 +772,30 @@ - [版本说明](raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-changeset.md) - [全妙轻应用更新公告](raw/application-user-guide/application-gallery/quanmiao-light-application-series/light-application-update-announcement.md) - [常见问题](raw/application-user-guide/application-gallery/quanmiao-light-application-series/quanmiao-lightapp-faq.md) - - **官方应用-伶鹊CCAI-客服对话Agent** - - **API参考** - - **API目录** - - [SseChat - 问答接口](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-chatbot-agent/api-reference-5/api-bailianchatbot-2024-11-05-dir/api-bailianchatbot-2024-11-05-ssechat.md) - - [API概览](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-chatbot-agent/api-reference-5/api-bailianchatbot-2024-11-05-overview.md) - - [授权信息](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-chatbot-agent/api-reference-5/api-bailianchatbot-2024-11-05-ram.md) - - [产品概述](raw/application-user-guide/application-gallery/official-application-voicepica-ccai-beebot-agent/product-overview-voicepica-beebot-agent.md) - - [计费说明（客服对话Agent）](raw/application-user-guide/application-gallery/official-application-voicepica-ccai-beebot-agent/billing-description-beebot-agent.md) - - [使用指南](raw/application-user-guide/application-gallery/official-application-voicepica-ccai-beebot-agent/guidelines-for-use.md) - **官方应用-伶鹊CCAI-语音对话机器人** - **API参考** - **API目录** - **MQ消息订阅配置** - [UpdateSubscription - 更新订阅信息](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-mq-message-subscription-configuration/api-bailianvoicebot-2025-01-01-updatesubscription.md) - - [GetSubscription - 获取消息订阅配置信息](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-mq-message-subscription-configuration/api-bailianvoicebot-2025-01-01-getsubscription.md) - [DisableSubscription - 关闭消息订阅](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-mq-message-subscription-configuration/api-bailianvoicebot-2025-01-01-disablesubscription.md) + - [GetSubscription - 获取消息订阅配置信息](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-mq-message-subscription-configuration/api-bailianvoicebot-2025-01-01-getsubscription.md) - **变量管理** - [DeleteVariable - 删除变量](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-variable-management/api-bailianvoicebot-2025-01-01-deletevariable.md) - - [ListVariable - 获取变量列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-variable-management/api-bailianvoicebot-2025-01-01-listvariable.md) - [UpdateVariable - 更新变量](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-variable-management/api-bailianvoicebot-2025-01-01-updatevariable.md) + - [ListVariable - 获取变量列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-variable-management/api-bailianvoicebot-2025-01-01-listvariable.md) - [CreateVariable - 创建变量](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-variable-management/api-bailianvoicebot-2025-01-01-createvariable.md) - **三方语音配置** - - [UpdateVoiceAccessProfile - 更新三方语音配置](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-three-way-voice-configuration/api-bailianvoicebot-2025-01-01-updatevoiceaccessprofile.md) - [ListVoiceEngines - 获取三方语音引擎列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-three-way-voice-configuration/api-bailianvoicebot-2025-01-01-listvoiceengines.md) + - [UpdateVoiceAccessProfile - 更新三方语音配置](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-three-way-voice-configuration/api-bailianvoicebot-2025-01-01-updatevoiceaccessprofile.md) - [ListVoiceAccessProfile - 获取三方语音配置列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-three-way-voice-configuration/api-bailianvoicebot-2025-01-01-listvoiceaccessprofile.md) - [DeleteVoiceAccessProfile - 删除三方语音配置](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-three-way-voice-configuration/api-bailianvoicebot-2025-01-01-deletevoiceaccessprofile.md) - [CreateVoiceAccessProfile - 创建三方语音配置](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-three-way-voice-configuration/api-bailianvoicebot-2025-01-01-createvoiceaccessprofile.md) + - **克隆音管理** + - [ListCloneVoiceModels - 获取克隆音模型列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-listclonevoicemodels.md) + - [ListCloneVoice - 获取克隆音列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-listclonevoice.md) + - [DeleteCloneVoice - 删除克隆音](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-deleteclonevoice.md) + - [UpdateCloneVoice - 更新克隆音](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-updateclonevoice.md) + - [CreateCloneVoice - 创建克隆音](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-createclonevoice.md) - **热词管理** - [UpdateVocabulary - 更新热词](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-hot-word-management/api-bailianvoicebot-2025-01-01-updatevocabulary.md) - [ListVocabulary - 获取热词列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-hot-word-management/api-bailianvoicebot-2025-01-01-listvocabulary.md) @@ -454,92 +804,95 @@ - [ExportVocabulary - 导出热词](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-hot-word-management/api-bailianvoicebot-2025-01-01-exportvocabulary.md) - [DeleteVocabulary - 删除热词](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-hot-word-management/api-bailianvoicebot-2025-01-01-deletevocabulary.md) - [CreateVocabulary - 创建热词](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-hot-word-management/api-bailianvoicebot-2025-01-01-createvocabulary.md) - - **克隆音管理** - - [ListCloneVoiceModels - 获取克隆音模型列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-listclonevoicemodels.md) - - [DeleteCloneVoice - 删除克隆音](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-deleteclonevoice.md) - - [ListCloneVoice - 获取克隆音列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-listclonevoice.md) - - [CreateCloneVoice - 创建克隆音](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-createclonevoice.md) - - [UpdateCloneVoice - 更新克隆音](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-clone-tone-management/api-bailianvoicebot-2025-01-01-updateclonevoice.md) - **应用管理** - [DeleteApplication - 删除语音机器人应用](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-deleteapplication.md) - [ListNluModels - 获取对话大模型列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listnlumodels.md) - - [ListBackgroundMusics - 获取背景音列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listbackgroundmusics.md) - [PreviewVoice - TTS合成试听](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-previewvoice.md) - - [ListVoices - 获取音色列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listvoices.md) - - [CreateApplicationVersion - 创建语音机器人应用版本](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-createapplicationversion.md) + - [ListBackgroundMusics - 获取背景音列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listbackgroundmusics.md) + - [ListVoices - 获取音色列表](raw/application-user-guide/application-gallery/official-application-lingque-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listvoices.md) - [ListApplications - 查询语音机器人应用列表](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listapplications.md) - - [CreateApplication - 创建语音机器人应用](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-createapplication.md) + - [CreateApplicationVersion - 创建语音机器人应用版本](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-createapplicationversion.md) - [UpdateApplication - 修改语音机器人应用](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-updateapplication.md) + - [CreateApplication - 创建语音机器人应用](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-createapplication.md) - [GetApplication - 获取语音机器人应用](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-getapplication.md) - - [PublishApplicationVersion - 发布语音机器人](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-publishapplicationversion.md) - [UpdateApplicationVersion - 修改语音机器人应用版本](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-updateapplicationversion.md) + - [PublishApplicationVersion - 发布语音机器人](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-publishapplicationversion.md) - [BridgeWebCall - 软电话测试通话](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-bridgewebcall.md) - [GetDataChannelCredential - 获取数据通道凭证](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-getdatachannelcredential.md) - [GenerateFileUploadParams - 获取文件上传参数](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-generatefileuploadparams.md) - [API概览](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-overview.md) - [授权信息](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-ram.md) - [版本说明](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-changeset.md) - - [伶鹊CCAI-语音对话机器人产品计费说明](raw/application-user-guide/application-gallery/official-application-lingque-ccai-voice-dialogue-robot/billing-information-lingque-ccai-voice-dialogue-robot.md) - - [语音对话机器人产品概述](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/product-0verview.md) - [语音对话机器人操作指南](raw/application-user-guide/application-gallery/official-application-lingque-ccai-voice-dialogue-robot/operation-guide.md) + - [语音对话机器人产品概述](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-voice-dialogue-robot/product-0verview.md) + - [伶鹊CCAI-语音对话机器人产品计费说明](raw/application-user-guide/application-gallery/official-application-lingque-ccai-voice-dialogue-robot/billing-information-lingque-ccai-voice-dialogue-robot.md) + - **官方应用-伶鹊CCAI-客服对话Agent** + - **API参考** + - **API目录** + - [SseChat - 问答接口](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-chatbot-agent/api-reference-5/api-bailianchatbot-2024-11-05-dir/api-bailianchatbot-2024-11-05-ssechat.md) + - [API概览](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-chatbot-agent/api-reference-5/api-bailianchatbot-2024-11-05-overview.md) + - [授权信息](raw/application-user-guide/application-gallery/official-application-tongyi-xiaomi-ccai-chatbot-agent/api-reference-5/api-bailianchatbot-2024-11-05-ram.md) + - [产品概述](raw/application-user-guide/application-gallery/official-application-voicepica-ccai-beebot-agent/product-overview-voicepica-beebot-agent.md) + - [使用指南](raw/application-user-guide/application-gallery/official-application-voicepica-ccai-beebot-agent/guidelines-for-use.md) + - [计费说明（客服对话Agent）](raw/application-user-guide/application-gallery/official-application-voicepica-ccai-beebot-agent/billing-description-beebot-agent.md) - **通义点金** - **API参考** - **API目录** - **平台能力-文档库** - - [GetAppConfig - 获取配置信息](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getappconfig.md) - [UpdateDocumentChunk - 更新文档块内容](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-updatedocumentchunk.md) + - [GetAppConfig - 获取配置信息](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getappconfig.md) - [CreateLibrary - 创建文档库](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-createlibrary.md) - - [GetLibraryList - 获取文档库列表](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getlibrarylist.md) - [GetLibrary - 获取文档库详情](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getlibrary.md) - [UploadDocument - 上传文档](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-uploaddocument.md) + - [GetLibraryList - 获取文档库列表](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getlibrarylist.md) - [GetDocumentUrl - 获取文档的下载链接](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getdocumenturl.md) - - [GetFilterDocumentList - 按元信息过滤查询文档列表](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getfilterdocumentlist.md) - - [GetDocumentList - 获取文档列表](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getdocumentlist.md) - [PreviewDocument - 预览文档](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-previewdocument.md) + - [GetDocumentList - 获取文档列表](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getdocumentlist.md) + - [GetFilterDocumentList - 按元信息过滤查询文档列表](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getfilterdocumentlist.md) - [DeleteDocument - 删除文档](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-deletedocument.md) - [UpdateDocument - 更新文档](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-updatedocument.md) - [CreatePredefinedDocument - 创建预定义文档](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-createpredefineddocument.md) - - [GetDocumentChunkList - 获取文档块列表](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getdocumentchunklist.md) - [RecallDocument - 文档召回](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-recalldocument.md) - - [ReIndex - 重建索引](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-reindex.md) - [GetParseResult - 获取文档解析结果](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getparseresult.md) - - [DeleteLibrary - 删除文档库](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-deletelibrary.md) + - [GetDocumentChunkList - 获取文档块列表](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-getdocumentchunklist.md) + - [ReIndex - 重建索引](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-reindex.md) - [UpdateLibrary - 更新文档库](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-updatelibrary.md) + - [DeleteLibrary - 删除文档库](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-deletelibrary.md) - [GetHistoryListByBizType - 根据业务类型获取对话历史记录](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-gethistorylistbybiztype.md) - - [InvokePlugin - 调用插件](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-invokeplugin.md) - [RunLibraryChatGeneration - 文档库会话生成](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-runlibrarychatgeneration.md) + - [InvokePlugin - 调用插件](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-document-library/api-dianjin-2024-06-28-invokeplugin.md) - **平台能力-应用** - [EndToEndRealTimeDialog - 语音实时对话](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-endtoendrealtimedialog.md) - [RunDialogAnalysis - 会话分析结果生成](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-rundialoganalysis.md) - - [RunAgent - 运行智能体](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-runagent.md) - [CreateDialog - 创建外呼会话](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-createdialog.md) + - [RunAgent - 运行智能体](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-runagent.md) - [RealTimeDialog - 实时会话](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-realtimedialog.md) - [GetDialogDetail - 获取会话详情](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-getdialogdetail.md) - [RealtimeDialogAssist - 实时会话辅助](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-realtimedialogassist.md) + - [GetDialogAnalysisResult - 获取会话分析结果](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-getdialoganalysisresult.md) - [GetDialogLog - 获取对话日志](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-getdialoglog.md) - [CreateDialogAnalysisTask - 创建会话分析任务](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-createdialoganalysistask.md) - - [GetDialogAnalysisResult - 获取会话分析结果](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-getdialoganalysisresult.md) - [RebuildTask - 重建任务](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-rebuildtask.md) - - [GetTaskStatus - 获取任务状态](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-gettaskstatus.md) - [EvictTask - 取消任务](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-evicttask.md) - [CreateDocsSummaryTask - 创建多文档总结任务](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-createdocssummarytask.md) + - [GetTaskStatus - 获取任务状态](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-gettaskstatus.md) - [CreateAnnualDocSummaryTask - 创建按年份总结文档任务](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-createannualdocsummarytask.md) - [CreatePdfTranslateTask - 创建pdf文档翻译任务](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-createpdftranslatetask.md) - [CreateFinReportSummaryTask - 创建财报总结任务](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-createfinreportsummarytask.md) - [GetSummaryTaskResult - 获取财报总结任务结果](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-getsummarytaskresult.md) - - [GetTaskResult - 获取结果](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-gettaskresult.md) - [CreateQualityCheckTask - 创建质检任务](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-createqualitychecktask.md) + - [GetTaskResult - 获取结果](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-gettaskresult.md) - [GetQualityCheckTaskResult - 获取质检结果](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-getqualitychecktaskresult.md) - - [GenDocQaResult - 根据文档解析问答QA](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-gendocqaresult.md) - [RecognizeIntention - 意图识别](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-recognizeintention.md) + - [GenDocQaResult - 根据文档解析问答QA](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-gendocqaresult.md) - [UpdateQaLibrary - 更新QA问答库](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-updateqalibrary.md) - [SubmitChatQuestion - 提交问题列表](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-submitchatquestion.md) - - [GetChatQuestionResp - 获取问答结果](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-getchatquestionresp.md) - [RunChatResultGeneration - 对话结果生成](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-runchatresultgeneration.md) + - [GetChatQuestionResp - 获取问答结果](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-platform-capabilities-application/api-dianjin-2024-06-28-getchatquestionresp.md) - **其他** - [DashscopeAsyncTaskFinishEvent - Dashscope异步任务完成回调事件](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-dir/api-dianjin-2024-06-28-dir-other/api-dianjin-2024-06-28-dashscopeasynctaskfinishevent.md) - - [服务接入点](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-endpoint.md) - [API概览](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-overview.md) + - [服务接入点](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-endpoint.md) - [授权信息](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-ram.md) - [版本说明](raw/application-user-guide/application-gallery/tongyi-dianjin/api-reference-3/api-dianjin-2024-06-28-changeset.md) - [产品简介](raw/application-user-guide/application-gallery/tongyi-dianjin/tongyi-dianjin-overview.md) @@ -550,42 +903,18 @@ - [信息抽取](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-api-reference/docmining-api-directory/document-information-extraction.md) - [文档内容审核](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-api-reference/docmining-api-directory/document-content-audit.md) - [打标分类](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-api-reference/docmining-api-directory/document-tagging.md) - - [文档删除](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-api-reference/docmining-api-directory/document-delete.md) - [摘要生成](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-api-reference/docmining-api-directory/document-summary-generation.md) + - [文档删除](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-api-reference/docmining-api-directory/document-delete.md) - [API概览](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-api-reference/docmining-api-overview.md) - [服务接入点](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-api-reference/docmining-service-access-point.md) - [错误码](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-api-reference/docmining-error-code.md) - [通义数据挖掘产品介绍](raw/application-user-guide/application-gallery/tongyi-docmining/docmining-product-introduction.md) - - **官方应用-通义多模态翻译** - - **API参考** - - **API目录** - - **文本翻译** - - [BatchTranslate - 批量文本翻译](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-batchtranslate.md) - - [TextTranslate - 文本翻译接口](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-texttranslate.md) - - [SubmitLongTextTranslateTask - 提交长文本翻译任务](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-submitlongtexttranslatetask.md) - - [SubmitHtmlTranslateTask - 提交html翻译任务](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-submithtmltranslatetask.md) - - [GetHtmlTranslateTask - 获取html翻译任务结果](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-gethtmltranslatetask.md) - - [TermEdit - 术语库编辑](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-termedit.md) - - [TermQuery - 术语库查询](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-termquery.md) - - [GetLongTextTranslateTask - 获取长文本翻译任务结果](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-getlongtexttranslatetask.md) - - **图片翻译** - - [GetImageTranslateTask - 获取图片翻译任务结果](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-image-translation/api-anytrans-2025-07-07-getimagetranslatetask.md) - - [SubmitImageTranslateTask - 提交图片翻译任务](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-image-translation/api-anytrans-2025-07-07-submitimagetranslatetask.md) - - **文档翻译** - - [SubmitDocTranslateTask - 文档翻译任务提交](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-document-translation/api-anytrans-2025-07-07-submitdoctranslatetask.md) - - [GetDocTranslateTask - 文档翻译结果获取](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-document-translation/api-anytrans-2025-07-07-getdoctranslatetask.md) - - [服务接入点](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-endpoint.md) - - [API概览](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-overview.md) - - [授权信息](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-ram.md) - - [版本说明](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-changeset.md) - - [通义多模态翻译](raw/application-user-guide/application-gallery/official-application-tongyi-translate/official-application-tongyi-translate-overview.md) - - [网页翻译JSSDK](raw/application-user-guide/application-gallery/official-application-tongyi-translate/web-page-translation-jssdk.md) - **官方应用-通义深度搜索** - **API参考** - **API目录** - - [生成对话](raw/application-user-guide/application-gallery/tongyi-deepsearch/deepsearch-api-reference/deepsearch-api-list/deepsearch-chat-generate.md) - - [对话文件管理](raw/application-user-guide/application-gallery/tongyi-deepsearch/deepsearch-api-reference/deepsearch-api-list/deepsearch-session-file-management.md) - [上传文件](raw/application-user-guide/application-gallery/tongyi-deepsearch/deepsearch-api-reference/deepsearch-api-list/deepsearch-file-upload.md) + - [对话文件管理](raw/application-user-guide/application-gallery/tongyi-deepsearch/deepsearch-api-reference/deepsearch-api-list/deepsearch-session-file-management.md) + - [生成对话](raw/application-user-guide/application-gallery/tongyi-deepsearch/deepsearch-api-reference/deepsearch-api-list/deepsearch-chat-generate.md) - [生成报告导出](raw/application-user-guide/application-gallery/tongyi-deepsearch/deepsearch-api-reference/deepsearch-api-list/deepsearch-report-export.md) - [对接自有知识库](raw/application-user-guide/application-gallery/tongyi-deepsearch/deepsearch-api-reference/deepsearch-api-list/docking-self-built-database.md) - [API概览](raw/application-user-guide/application-gallery/tongyi-deepsearch/deepsearch-api-reference/deepsearch-api-overview.md) @@ -597,51 +926,75 @@ - **API参考** - [API概览](raw/application-user-guide/application-gallery/web-search-agent/web-search-agent-api/web-search-agent-api-overview.md) - [服务接入点](raw/application-user-guide/application-gallery/web-search-agent/web-search-agent-api/service-access-point.md) - - [生成对话](raw/application-user-guide/application-gallery/web-search-agent/web-search-agent-api/web-search-agent-api-chat.md) - [多模态文件操作](raw/application-user-guide/application-gallery/web-search-agent/web-search-agent-api/web-search-agent-api-chat-multimodal-file.md) + - [生成对话](raw/application-user-guide/application-gallery/web-search-agent/web-search-agent-api/web-search-agent-api-chat.md) - [千问联网检索Agent产品简介](raw/application-user-guide/application-gallery/web-search-agent/web-search-agent-guide.md) - [错误码-千问联网检索Agent](raw/application-user-guide/application-gallery/web-search-agent/web-search-agent-error-code.md) + - **官方应用-通义多模态翻译** + - **API参考** + - **API目录** + - **图片翻译** + - [GetImageTranslateTask - 获取图片翻译任务结果](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-image-translation/api-anytrans-2025-07-07-getimagetranslatetask.md) + - [SubmitImageTranslateTask - 提交图片翻译任务](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-image-translation/api-anytrans-2025-07-07-submitimagetranslatetask.md) + - **文档翻译** + - [SubmitDocTranslateTask - 文档翻译任务提交](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-document-translation/api-anytrans-2025-07-07-submitdoctranslatetask.md) + - [GetDocTranslateTask - 文档翻译结果获取](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-document-translation/api-anytrans-2025-07-07-getdoctranslatetask.md) + - **文本翻译** + - [TextTranslate - 文本翻译接口](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-texttranslate.md) + - [BatchTranslate - 批量文本翻译](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-batchtranslate.md) + - [SubmitLongTextTranslateTask - 提交长文本翻译任务](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-submitlongtexttranslatetask.md) + - [TermEdit - 术语库编辑](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-termedit.md) + - [SubmitHtmlTranslateTask - 提交html翻译任务](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-submithtmltranslatetask.md) + - [GetHtmlTranslateTask - 获取html翻译任务结果](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-gethtmltranslatetask.md) + - [GetLongTextTranslateTask - 获取长文本翻译任务结果](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-getlongtexttranslatetask.md) + - [TermQuery - 术语库查询](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-dir/api-anytrans-2025-07-07-dir-text-translation/api-anytrans-2025-07-07-termquery.md) + - [服务接入点](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-endpoint.md) + - [API概览](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-overview.md) + - [授权信息](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-ram.md) + - [版本说明](raw/application-user-guide/application-gallery/official-application-tongyi-translate/tongyi-translate-api-reference/api-anytrans-2025-07-07-changeset.md) + - [通义多模态翻译](raw/application-user-guide/application-gallery/official-application-tongyi-translate/official-application-tongyi-translate-overview.md) + - [网页翻译JSSDK](raw/application-user-guide/application-gallery/official-application-tongyi-translate/web-page-translation-jssdk.md) - **通义 UI Agent** - [通义 UI Agent](raw/application-user-guide/application-gallery/ui-agent/ui-agent-api.md) - **官方应用-全妙解决方案类产品** - **妙笔、妙策和审校** + - **文本写作指导** + - **传媒类文体写作指导** + - [快速写一篇传媒稿（prompt一步式撰写）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/media-style-writing-best-practices/quick-media-writing-prompt.md) + - [没有思路，要谋篇布局](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/media-style-writing-best-practices/use-amb-to-help-writing.md) + - [用已有文章，生成标题摘要等](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/media-style-writing-best-practices/generate-titles-summaries-media-text.md) + - **政务公文写作指导** + - [快速写一篇政务稿（prompt一步式撰写）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/government-document-writing-best-practices/quick-gov-writing-prompt.md) + - [分步式撰写政务稿（精准控制章节内容）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/government-document-writing-best-practices/step-by-step-gov-writing.md) + - [用已有文章，生成标题摘要等](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/government-document-writing-best-practices/generate-titles-summaries-gov-text.md) + - [常见FAQ](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/faq-for-using-quanmiao-series-products.md) - **使用指南** - **AI妙笔** - **功能界面** - [妙笔-分布生成创作文章](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/function-interface/step-by-step-generation.md) + - [智能配图](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/function-interface/smart-image-generation.md) - [直接生成](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/function-interface/direct-generation.md) - [搜索素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/function-interface/search-materials.md) - - [智能配图](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/function-interface/smart-image-generation.md) - - [AI妙笔产品概述](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/product-overview-for-amb.md) - [妙笔首页概览](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/amb-homepage-overview.md) + - [AI妙笔产品概述](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/product-overview-for-amb.md) - [AI工具箱](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/ai-toolbox.md) - [素材库](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/material-library.md) - [系统配置](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/system-configuration.md) - [文章风格和格式学习](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/amb/style-imitation.md) - [AI妙策](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/ai-miaoce.md) - - [智能审校](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/article-review.md) - [深度写作](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/deep-writing.md) - - **文本写作指导** - - **传媒类文体写作指导** - - [快速写一篇传媒稿（prompt一步式撰写）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/media-style-writing-best-practices/quick-media-writing-prompt.md) - - [没有思路，要谋篇布局](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/media-style-writing-best-practices/use-amb-to-help-writing.md) - - [用已有文章，生成标题摘要等](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/media-style-writing-best-practices/generate-titles-summaries-media-text.md) - - **政务公文写作指导** - - [快速写一篇政务稿（prompt一步式撰写）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/government-document-writing-best-practices/quick-gov-writing-prompt.md) - - [分步式撰写政务稿（精准控制章节内容）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/government-document-writing-best-practices/step-by-step-gov-writing.md) - - [用已有文章，生成标题摘要等](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/government-document-writing-best-practices/generate-titles-summaries-gov-text.md) - - [常见FAQ](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/document-writing-best-practices/faq-for-using-quanmiao-series-products.md) + - [智能审校](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/usage-guide/article-review.md) - **更新公告** - **功能更新** - [2025年2月26日更新-妙笔](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/update-announcement/miaobi-and-miaoce-function-update/february-26-2025-update-miaobi.md) - [2025年1月24日更新-全妙解决方案类产品](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/update-announcement/miaobi-and-miaoce-function-update/2025-1-24-function-update-announcement-quanmiao-saas.md) - [2024年3月11更新-AI全妙系列 V2.2.2](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/update-announcement/miaobi-and-miaoce-function-update/update-2024-03-11-ai-quanmiao-v2-2.md) - [2024年3月1更新-AI全妙系列 V2.2.1](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/update-announcement/miaobi-and-miaoce-function-update/update-2024-03-01-ai-quanmiao-v2-2.md) - - [2023年12月19更新-AI妙笔V2.1](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/update-announcement/miaobi-and-miaoce-function-update/update-2023-12-19-amb-v2.md) - [2024年2月28更新-AI全妙系列 V2.2](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/update-announcement/miaobi-and-miaoce-function-update/update-2024-02-28-ai-quanmiao-v2.md) + - [2023年12月19更新-AI妙笔V2.1](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/update-announcement/miaobi-and-miaoce-function-update/update-2023-12-19-amb-v2.md) - [计费说明（政务公文配套工具）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/government-document-tool-billing.md) - - [计费说明（妙笔）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/miaobi-billing.md) - [计费说明（PPT生成）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/ppt-generation-billing.md) + - [计费说明（妙笔）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/miaobi-billing.md) - [计费说明（妙策-自定义数据源）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/billing-document-miaoce-custom-data-source.md) - [计费说明（视频混剪）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaobi-miaoce-shenjiao/billing-description-video-mixing.md) - **妙搜和妙读** @@ -649,14 +1002,6 @@ - [妙搜](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaosou-and-miaodu/miaodou-and-miaodu-guidelines-for-use/ai-miaosou.md) - [计费说明（妙搜和妙读）](raw/application-user-guide/application-gallery/quanmiao-solution-products/miaosou-and-miaodu/miaosou-miaodu-api-billing.md) - **开发文档** - - **最佳实践** - - [妙笔API最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-miaobi-api.md) - - [智能审校最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-smart-audit.md) - - [妙策API最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-miaoce-api.md) - - [妙搜API最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-miaosou-api.md) - - [妙读最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/miaodu-best-practices.md) - - [视频混剪最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-video-mixing-and-cutting.md) - - [PPT生成最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/ppt-generation-best-practices.md) - **API参考** - **数据结构** - [GenerateTraceability](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-struct-dir/api-aimiaobi-2023-08-01-struct-generatetraceability.md) @@ -673,53 +1018,53 @@ - [ListDialogues - 生成历史列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-universal-interface/api-aimiaobi-2023-08-01-listdialogues.md) - [ListVersions - 获取版本信息](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-universal-interface/api-aimiaobi-2023-08-01-listversions.md) - [GetProperties - 获取配置信息](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-universal-interface/api-aimiaobi-2023-08-01-getproperties.md) + - **通用接口-文件上传下载** + - [GenerateFileUrlByKey - 生成文件URL](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-file-upload-and-download/api-aimiaobi-2023-08-01-generatefileurlbykey.md) + - [GenerateUploadConfig - 生成上传配置](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-file-upload-and-download/api-aimiaobi-2023-08-01-generateuploadconfig.md) - **通用接口-异步任务管理** - [SubmitAsyncTask - 提交异步任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-asynchronous-task-management/api-aimiaobi-2023-08-01-submitasynctask.md) - [CancelAsyncTask - 取消异步任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-asynchronous-task-management/api-aimiaobi-2023-08-01-cancelasynctask.md) - - [ListAsyncTasks - 获取异步任务列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-asynchronous-task-management/api-aimiaobi-2023-08-01-listasynctasks.md) - [QueryAsyncTask - 查询异步任务明细](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-asynchronous-task-management/api-aimiaobi-2023-08-01-queryasynctask.md) - - **通用接口-文件上传下载** - - [GenerateFileUrlByKey - 生成文件URL](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-file-upload-and-download/api-aimiaobi-2023-08-01-generatefileurlbykey.md) - - [GenerateUploadConfig - 生成上传配置](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-file-upload-and-download/api-aimiaobi-2023-08-01-generateuploadconfig.md) + - [ListAsyncTasks - 获取异步任务列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-asynchronous-task-management/api-aimiaobi-2023-08-01-listasynctasks.md) - **妙笔-创作文章** - [RunAiHelperWriting - AI帮写](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runaihelperwriting.md) - [RunWritingV2 - 智能写作](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runwritingv2.md) - - [RunTranslateGeneration - 中英翻译](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runtranslategeneration.md) - - [RunWriting - 直接写作](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runwriting.md) - [RunStepByStepWriting - 分步骤写作](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runstepbystepwriting.md) + - [RunWriting - 直接写作](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runwriting.md) + - [RunTranslateGeneration - 中英翻译](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runtranslategeneration.md) - [RunTextPolishing - 润色](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runtextpolishing.md) - - [RunKeywordsExtractionGeneration - 关键词抽取](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runkeywordsextractiongeneration.md) - [RunContinueContent - 内容续写](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runcontinuecontent.md) + - [RunKeywordsExtractionGeneration - 关键词抽取](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runkeywordsextractiongeneration.md) - [RunWriteToneGeneration - 文风改写](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runwritetonegeneration.md) - [RunTitleGeneration - 标题生成](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runtitlegeneration.md) - - [RunExpandContent - 内容扩写](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runexpandcontent.md) - [RunSummaryGenerate - 摘要生成](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runsummarygenerate.md) - - [SearchNews - 信息检索](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-searchnews.md) - [RunAbbreviationContent - 内容缩写](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runabbreviationcontent.md) + - [RunExpandContent - 内容扩写](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runexpandcontent.md) + - [SearchNews - 信息检索](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-searchnews.md) - [RunQuickWriting - 快速写作](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-runquickwriting.md) - - [GenerateImageTask - 生成智能配图任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-generateimagetask.md) - [ListBuildConfigs - 获取系统自定义预设](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-listbuildconfigs.md) - - [FetchImageTask - 获取图片任务执行结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-fetchimagetask.md) + - [GenerateImageTask - 生成智能配图任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-generateimagetask.md) - [FeedbackDialogue - 反馈对话](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-feedbackdialogue.md) + - [FetchImageTask - 获取图片任务执行结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-creative-articles/api-aimiaobi-2023-08-01-fetchimagetask.md) - **通用接口-通用配置** - [CreateGeneralConfig - 通用配置-创建](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-common-configuration/api-aimiaobi-2023-08-01-creategeneralconfig.md) - [ListGeneralConfigs - 通用配置-列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-common-configuration/api-aimiaobi-2023-08-01-listgeneralconfigs.md) + - [UpdateGeneralConfig - 通用配置-修改](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-common-configuration/api-aimiaobi-2023-08-01-updategeneralconfig.md) - [GetGeneralConfig - 通用配置-详情](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-common-configuration/api-aimiaobi-2023-08-01-getgeneralconfig.md) - [DeleteGeneralConfig - 通用配置-删除](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-common-configuration/api-aimiaobi-2023-08-01-deletegeneralconfig.md) - - [UpdateGeneralConfig - 通用配置-修改](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-common-interface-common-configuration/api-aimiaobi-2023-08-01-updategeneralconfig.md) + - **妙笔-视频审校** + - [SubmitVideoAudit - 提交视频审校任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-review/api-aimiaobi-2023-08-01-submitvideoaudit.md) + - [QueryVideoAuditResult - 查询视频审校结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-review/api-aimiaobi-2023-08-01-queryvideoauditresult.md) - **妙笔-文体仿写** - [ListStyleLearningResult - 获取文体学习分析结果列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-style-imitation-writing/api-aimiaobi-2023-08-01-liststylelearningresult.md) - - [RunStyleFeatureAnalysis - 内容特点分析](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-style-imitation-writing/api-aimiaobi-2023-08-01-runstylefeatureanalysis.md) - - [GetStyleLearningResult - 获取文体学习分析结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-style-imitation-writing/api-aimiaobi-2023-08-01-getstylelearningresult.md) - [DeleteStyleLearningResult - 删除自定义文体](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-style-imitation-writing/api-aimiaobi-2023-08-01-deletestylelearningresult.md) + - [RunStyleFeatureAnalysis - 内容特点分析](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-style-imitation-writing/api-aimiaobi-2023-08-01-runstylefeatureanalysis.md) - [SaveStyleLearningResult - 保存文体学习分析结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-style-imitation-writing/api-aimiaobi-2023-08-01-savestylelearningresult.md) + - [GetStyleLearningResult - 获取文体学习分析结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-style-imitation-writing/api-aimiaobi-2023-08-01-getstylelearningresult.md) - [ListWritingStyles - 获取写作文体列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-style-imitation-writing/api-aimiaobi-2023-08-01-listwritingstyles.md) - - **妙笔-视频审校** - - [SubmitVideoAudit - 提交视频审校任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-review/api-aimiaobi-2023-08-01-submitvideoaudit.md) - - [QueryVideoAuditResult - 查询视频审校结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-review/api-aimiaobi-2023-08-01-queryvideoauditresult.md) - **妙笔-文章审校-规则库管理** - - [SubmitAuditNote - 提交自定义规则库](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-proofreading-rule-library-management/api-aimiaobi-2023-08-01-submitauditnote.md) - [ConfirmAndPostProcessAuditNote - 确认提交规则库用于审核](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-proofreading-rule-library-management/api-aimiaobi-2023-08-01-confirmandpostprocessauditnote.md) + - [SubmitAuditNote - 提交自定义规则库](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-proofreading-rule-library-management/api-aimiaobi-2023-08-01-submitauditnote.md) - [DownloadAuditNote - 下载规则库](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-proofreading-rule-library-management/api-aimiaobi-2023-08-01-downloadauditnote.md) - [DeleteAuditNote - 删除规则库](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-proofreading-rule-library-management/api-aimiaobi-2023-08-01-deleteauditnote.md) - [GetAuditNotePostProcessingStatus - 获取规则库后处理进度](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-proofreading-rule-library-management/api-aimiaobi-2023-08-01-getauditnotepostprocessingstatus.md) @@ -730,8 +1075,8 @@ - [AddAuditTerms - 添加自定义词库记录](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-thesaurus-management/api-aimiaobi-2023-08-01-addauditterms.md) - [EditAuditTerms - 编辑自定义词库记录](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-thesaurus-management/api-aimiaobi-2023-08-01-editauditterms.md) - [DeleteAuditTerms - 删除指定词库记录](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-thesaurus-management/api-aimiaobi-2023-08-01-deleteauditterms.md) - - [SubmitImportTermsTask - 提交导入词库任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-thesaurus-management/api-aimiaobi-2023-08-01-submitimporttermstask.md) - [FetchImportTermsTask - 获取导入词库任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-thesaurus-management/api-aimiaobi-2023-08-01-fetchimporttermstask.md) + - [SubmitImportTermsTask - 提交导入词库任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-thesaurus-management/api-aimiaobi-2023-08-01-submitimporttermstask.md) - [SubmitExportTermsTask - 提交导出词库任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-thesaurus-management/api-aimiaobi-2023-08-01-submitexporttermstask.md) - [FetchExportTermsTask - 获取导出词库任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-thesaurus-management/api-aimiaobi-2023-08-01-fetchexporttermstask.md) - **妙笔-文章审校-事实性审核** @@ -739,26 +1084,10 @@ - [GetFactAuditUrl - 获取事实性审核 URL](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-factual-review/api-aimiaobi-2023-08-01-getfactauditurl.md) - [DeleteFactAuditUrl - 删除事实性审核 URL](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaobi-article-review-factual-review/api-aimiaobi-2023-08-01-deletefactauditurl.md) - **妙笔-文章审校** - - [QueryAuditTask - 查询审核结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-queryaudittask.md) - [SubmitSmartAudit - 提交智能审校任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-submitsmartaudit.md) - - [SubmitAuditTask - 提交审核任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-submitaudittask.md) - - [CancelAuditTask - 取消审核任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-cancelaudittask.md) - [GetSmartAuditResult - 查询智能审校结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-getsmartauditresult.md) - [ListAuditContentErrorTypes - 获取审校维度列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-listauditcontenterrortypes.md) - [ExportAuditContentResult - 导出智能审校报告](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-exportauditcontentresult.md) - - **妙笔-素材库** - - [SaveMaterialDocument - 保存素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-savematerialdocument.md) - - [DeleteMaterialById - 删除素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-deletematerialbyid.md) - - [UpdateMaterialDocument - 更新素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-updatematerialdocument.md) - - [GetMaterialById - 获取素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-getmaterialbyid.md) - - [ListMaterialDocuments - 获取素材列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-listmaterialdocuments.md) - - **妙笔-素材库-自定义文本** - - [GetCustomText - 获取自定义文本](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-getcustomtext.md) - - [UpdateCustomText - 更新自定义文本](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-updatecustomtext.md) - - [ListCustomText - 获取自定义文本列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-listcustomtext.md) - - [SaveCustomText - 保存自定义文本](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-savecustomtext.md) - - [DocumentExtraction - 文档提取](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-documentextraction.md) - - [DeleteCustomText - 删除自定义文本](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-deletecustomtext.md) - **妙笔-文档管理** - [GenerateExportWordTask - 生成导出文档任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-document-management/api-aimiaobi-2023-08-01-generateexportwordtask.md) - [FetchExportWordTask - 获取导出文档任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-document-management/api-aimiaobi-2023-08-01-fetchexportwordtask.md) @@ -766,99 +1095,110 @@ - [DeleteGeneratedContent - 删除文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-document-management/api-aimiaobi-2023-08-01-deletegeneratedcontent.md) - [UpdateGeneratedContent - 更新文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-document-management/api-aimiaobi-2023-08-01-updategeneratedcontent.md) - [GetGeneratedContent - 获取文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-document-management/api-aimiaobi-2023-08-01-getgeneratedcontent.md) - - [ListGeneratedContents - 获取文档列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-document-management/api-aimiaobi-2023-08-01-listgeneratedcontents.md) - [ExportGeneratedContent - 导出文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-document-management/api-aimiaobi-2023-08-01-exportgeneratedcontent.md) + - [ListGeneratedContents - 获取文档列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-document-management/api-aimiaobi-2023-08-01-listgeneratedcontents.md) + - **妙笔-素材库** + - [SaveMaterialDocument - 保存素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-savematerialdocument.md) + - [UpdateMaterialDocument - 更新素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-updatematerialdocument.md) + - [DeleteMaterialById - 删除素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-deletematerialbyid.md) + - [ListMaterialDocuments - 获取素材列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-listmaterialdocuments.md) + - [GetMaterialById - 获取素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library/api-aimiaobi-2023-08-01-getmaterialbyid.md) - **妙笔-视频混剪** - [GetClipsBuildInResource - 获取智能混剪内置资源](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-getclipsbuildinresource.md) - - [AsyncCreateClipsTimeLine - 创建剪辑口播时间线](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-asynccreateclipstimeline.md) - - [AsyncUploadVideo - 异步上传视频剪辑素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-asyncuploadvideo.md) - [AsyncEditTimeline - 编辑剪辑口播时间线](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-asyncedittimeline.md) + - [AsyncCreateClipsTimeLine - 创建剪辑口播时间线](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-asynccreateclipstimeline.md) - [GetAutoClipsTaskInfo - 获得剪辑任务信息](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-getautoclipstaskinfo.md) - - [AsyncCreateClipsTask - 创建剪辑任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-asynccreateclipstask.md) + - [AsyncUploadVideo - 异步上传视频剪辑素材](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-asyncuploadvideo.md) - [ListAutoClipsTask - 智能混剪任务列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-listautoclipstask.md) + - [AsyncCreateClipsTask - 创建剪辑任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-video-mixed-cut/api-aimiaobi-2023-08-01-asynccreateclipstask.md) + - **妙笔-素材库-自定义文本** + - [GetCustomText - 获取自定义文本](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-getcustomtext.md) + - [UpdateCustomText - 更新自定义文本](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-updatecustomtext.md) + - [ListCustomText - 获取自定义文本列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-listcustomtext.md) + - [DeleteCustomText - 删除自定义文本](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-deletecustomtext.md) + - [SaveCustomText - 保存自定义文本](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-savecustomtext.md) + - [DocumentExtraction - 文档提取](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-material-library-custom-text/api-aimiaobi-2023-08-01-documentextraction.md) + - **妙策-自定义数据源** + - [GetCustomSourceTopicAnalysisTask - 获取自定义源话题分析任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-custom-data-source/api-aimiaobi-2023-08-01-getcustomsourcetopicanalysistask.md) + - [SubmitCustomSourceTopicAnalysis - 提交自定义源话题选题分析任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-custom-data-source/api-aimiaobi-2023-08-01-submitcustomsourcetopicanalysis.md) + - [ExportCustomSourceAnalysisTask - 导出自定义源-话题分析任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-custom-data-source/api-aimiaobi-2023-08-01-exportcustomsourceanalysistask.md) - **公文库检索** - [ListDocumentRetrieve - 公文库检索](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-public-library-retrieval/api-aimiaobi-2023-08-01-listdocumentretrieve.md) - **妙策-选题热点** - [RunTopicSelectionMerge - 选题热点融合](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-runtopicselectionmerge.md) - - [ListHotNewsWithType - 获取选题热点列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listhotnewswithtype.md) - [ListHotSources - 获取三方热榜源列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listhotsources.md) + - [ListHotNewsWithType - 获取选题热点列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listhotnewswithtype.md) - [ListHotTopics - 获取热点话题列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listhottopics.md) - - [ListHotViewPoints - 获取热门视角列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listhotviewpoints.md) - [GetTopicById - 获取热点对象](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-gettopicbyid.md) - - [ListTimedViewAttitude - 获取时效性视角列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listtimedviewattitude.md) - - [ListWebReviewPoints - 获取网友视角列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listwebreviewpoints.md) + - [ListHotViewPoints - 获取热门视角列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listhotviewpoints.md) - [ListFreshViewPoints - 获取新颖视角列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listfreshviewpoints.md) + - [ListTimedViewAttitude - 获取时效性视角列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listtimedviewattitude.md) - [ListPlanningProposal - 获取选题策划列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listplanningproposal.md) + - [ListWebReviewPoints - 获取网友视角列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-listwebreviewpoints.md) - [ExportHotTopicPlanningProposals - 导出选题策划文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaozi-hot-topics/api-aimiaobi-2023-08-01-exporthottopicplanningproposals.md) - **妙策-自定义话题** - [DeleteCustomTopicByTopic - 删除自定义热点事件](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-custom-topic/api-aimiaobi-2023-08-01-deletecustomtopicbytopic.md) - - [ListTopicRecommendEventList - 获取热点推荐事件列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-custom-topic/api-aimiaobi-2023-08-01-listtopicrecommendeventlist.md) - [ListTopicViewPointRecommendEventList - 获取热点事件推荐观点列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-custom-topic/api-aimiaobi-2023-08-01-listtopicviewpointrecommendeventlist.md) - [RunCustomHotTopicAnalysis - 自定义热点话题分析](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-custom-topic/api-aimiaobi-2023-08-01-runcustomhottopicanalysis.md) + - [ListTopicRecommendEventList - 获取热点推荐事件列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-custom-topic/api-aimiaobi-2023-08-01-listtopicrecommendeventlist.md) - [RunCustomHotTopicViewPointAnalysis - 自定义选题视角分析](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-custom-topic/api-aimiaobi-2023-08-01-runcustomhottopicviewpointanalysis.md) - [ListCustomViewPoints - 获取自定义视角列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-custom-topic/api-aimiaobi-2023-08-01-listcustomviewpoints.md) - [DeleteCustomTopicViewPointById - 删除自定义选题视角](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-custom-topic/api-aimiaobi-2023-08-01-deletecustomtopicviewpointbyid.md) - - **妙策-自定义数据源** - - [GetCustomSourceTopicAnalysisTask - 获取自定义源话题分析任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-custom-data-source/api-aimiaobi-2023-08-01-getcustomsourcetopicanalysistask.md) - - [SubmitCustomSourceTopicAnalysis - 提交自定义源话题选题分析任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-custom-data-source/api-aimiaobi-2023-08-01-submitcustomsourcetopicanalysis.md) - - [ExportCustomSourceAnalysisTask - 导出自定义源-话题分析任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-custom-data-source/api-aimiaobi-2023-08-01-exportcustomsourceanalysistask.md) - **妙策-openapi** - [SubmitDocClusterTask - 提交内容聚合任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-openapi/api-aimiaobi-2023-08-01-submitdocclustertask.md) - [GetDocClusterTask - 获取内容聚合任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-openapi/api-aimiaobi-2023-08-01-getdocclustertask.md) - [SubmitTopicSelectionPerspectiveAnalysisTask - 提交选题热点分析任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-openapi/api-aimiaobi-2023-08-01-submittopicselectionperspectiveanalysistask.md) - - [GetTopicSelectionPerspectiveAnalysisTask - 获取选题视角分析任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-openapi/api-aimiaobi-2023-08-01-gettopicselectionperspectiveanalysistask.md) - [SubmitCustomTopicSelectionPerspectiveAnalysisTask - 提交自定义热点选题视角分析任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-openapi/api-aimiaobi-2023-08-01-submitcustomtopicselectionperspectiveanalysistask.md) - [GetCustomTopicSelectionPerspectiveAnalysisTask - 获取自定义选题视角分析任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-openapi/api-aimiaobi-2023-08-01-getcustomtopicselectionperspectiveanalysistask.md) + - [GetTopicSelectionPerspectiveAnalysisTask - 获取选题视角分析任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tips-openapi/api-aimiaobi-2023-08-01-gettopicselectionperspectiveanalysistask.md) - **妙策-新闻播报** - [SubmitCustomHotTopicBroadcastJob - 提交自定义播报单任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-news-broadcast/api-aimiaobi-2023-08-01-submitcustomhottopicbroadcastjob.md) - [GetHotTopicBroadcast - 查询完整播报单（热榜）](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-news-broadcast/api-aimiaobi-2023-08-01-gethottopicbroadcast.md) - [GetCustomHotTopicBroadcastJob - 获取自定义播报单任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-policy-news-broadcast/api-aimiaobi-2023-08-01-getcustomhottopicbroadcastjob.md) - - **妙搜-智能搜索** - - [ListSearchTasks - 查询妙搜搜索生成历史任务列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-listsearchtasks.md) - - [ListSearchTaskDialogues - 查询妙搜搜索生成任务详情列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-listsearchtaskdialogues.md) - - [ListSearchTaskDialogueDatas - 查询搜索生成任务对话详情中数据列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-listsearchtaskdialoguedatas.md) - - [RunSearchGeneration - 妙搜-智能搜索](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-runsearchgeneration.md) - - [RunSearchSimilarArticles - 妙搜-文搜文](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-runsearchsimilararticles.md) - **妙策-企业VOC挖掘** - - [ExportAnalysisTagDetailByTaskId - 导出标签挖掘结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaoce-enterprise-voc-mining/api-aimiaobi-2023-08-01-exportanalysistagdetailbytaskid.md) - [ValidateUploadTemplate - 校验VOC上传模板](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaoce-enterprise-voc-mining/api-aimiaobi-2023-08-01-validateuploadtemplate.md) + - [ExportAnalysisTagDetailByTaskId - 导出标签挖掘结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaoce-enterprise-voc-mining/api-aimiaobi-2023-08-01-exportanalysistagdetailbytaskid.md) - [SubmitEnterpriseVocAnalysisTask - 提交企业VOC分析任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaoce-enterprise-voc-mining/api-aimiaobi-2023-08-01-submitenterprisevocanalysistask.md) - [ListAnalysisTagDetailByTaskId - 根据任务ID获取标签分析明细列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaoce-enterprise-voc-mining/api-aimiaobi-2023-08-01-listanalysistagdetailbytaskid.md) - [GetEnterpriseVocAnalysisTask - 获取企业VOC挖掘任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaoce-enterprise-voc-mining/api-aimiaobi-2023-08-01-getenterprisevocanalysistask.md) - [GetCategoriesByTaskId - 根据任务ID获取分类列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-miaoce-enterprise-voc-mining/api-aimiaobi-2023-08-01-getcategoriesbytaskid.md) - **妙搜-数据源** - [CreateDataset - 数据源-创建](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-createdataset.md) - - [UpdateDataset - 数据源-修改](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-updatedataset.md) - [GetDataset - 数据源-详情](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-getdataset.md) - - [DeleteDataset - 数据源-删除](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-deletedataset.md) + - [UpdateDataset - 数据源-修改](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-updatedataset.md) - [ListDatasets - 数据源-列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-listdatasets.md) - - [AddDatasetDocument - 数据源-添加文档到数据集](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-adddatasetdocument.md) + - [DeleteDataset - 数据源-删除](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-deletedataset.md) - [GetDatasetDocument - 数据源-获取文档详情](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-getdatasetdocument.md) + - [AddDatasetDocument - 数据源-添加文档到数据集](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-adddatasetdocument.md) - [UpdateDatasetDocument - 数据源-修改文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-updatedatasetdocument.md) - [ListDatasetDocuments - 数据源-文档列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-listdatasetdocuments.md) - - [SearchDatasetDocuments - 数据源-搜索文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-searchdatasetdocuments.md) - [DeleteDatasetDocument - 数据源-删除数据集文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-deletedatasetdocument.md) + - [SearchDatasetDocuments - 数据源-搜索文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-data-source/api-aimiaobi-2023-08-01-searchdatasetdocuments.md) + - **妙搜-智能搜索** + - [ListSearchTasks - 查询妙搜搜索生成历史任务列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-listsearchtasks.md) + - [ListSearchTaskDialogues - 查询妙搜搜索生成任务详情列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-listsearchtaskdialogues.md) + - [RunSearchGeneration - 妙搜-智能搜索](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-runsearchgeneration.md) + - [ListSearchTaskDialogueDatas - 查询搜索生成任务对话详情中数据列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-listsearchtaskdialoguedatas.md) + - [RunSearchSimilarArticles - 妙搜-文搜文](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-search-smart-search/api-aimiaobi-2023-08-01-runsearchsimilararticles.md) - **系统配置-干预配置** - [ListInterveneCnt - 获得所有干预项的数量](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-listintervenecnt.md) - [ListIntervenes - 列出干预项](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-listintervenes.md) - - [InsertInterveneGlobalReply - 插入干预全局回复项](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-insertinterveneglobalreply.md) - [ImportInterveneFile - 同步导入干预项文件](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-importintervenefile.md) - - [GetInterveneTemplateFileUrl - 获得干预导入模版文件地址](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-getintervenetemplatefileurl.md) + - [InsertInterveneGlobalReply - 插入干预全局回复项](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-insertinterveneglobalreply.md) - [ImportInterveneFileAsync - 异步导入干预项文件](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-importintervenefileasync.md) + - [GetInterveneTemplateFileUrl - 获得干预导入模版文件地址](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-getintervenetemplatefileurl.md) - [ClearIntervenes - 清除所有干预项](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-clearintervenes.md) - - [ListInterveneRules - 列出干预规则](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-listintervenerules.md) - [GetInterveneGlobalReply - 获得干预全局回复内容](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-getinterveneglobalreply.md) - - [GetInterveneRuleDetail - 获得干预规则的详情](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-getinterveneruledetail.md) - - [InsertInterveneRule - 插入干预规则](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-insertintervenerule.md) + - [ListInterveneRules - 列出干预规则](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-listintervenerules.md) - [ListInterveneImportTasks - 列出干预项导入任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-listinterveneimporttasks.md) + - [InsertInterveneRule - 插入干预规则](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-insertintervenerule.md) + - [GetInterveneRuleDetail - 获得干预规则的详情](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-getinterveneruledetail.md) - [DeleteInterveneRule - 删除干预规则](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-deleteintervenerule.md) - [ExportIntervenes - 导出干预项内容](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-exportintervenes.md) - [GetInterveneImportTaskInfo - 获得干预项目导入任务信息](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-intervention-configuration/api-aimiaobi-2023-08-01-getinterveneimporttaskinfo.md) - - **妙读-抽取类** - - [RunHotword - 抽取关键词](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-extraction-class/api-aimiaobi-2023-08-01-runhotword.md) - **系统配置-信源管理** - - [GetDataSourceOrderConfig - 获取信源配置权重数据](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-source-management/api-aimiaobi-2023-08-01-getdatasourceorderconfig.md) - [SaveDataSourceOrderConfig - 保存信源权重配置](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-source-management/api-aimiaobi-2023-08-01-savedatasourceorderconfig.md) + - [GetDataSourceOrderConfig - 获取信源配置权重数据](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-system-configuration-source-management/api-aimiaobi-2023-08-01-getdatasourceorderconfig.md) - **妙读-基础操作类** - [GetDocInfo - 获取文档信息](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-basic-operation-class/api-aimiaobi-2023-08-01-getdocinfo.md) - [GetFileContentLength - 获取文件长度](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-basic-operation-class/api-aimiaobi-2023-08-01-getfilecontentlength.md) @@ -867,17 +1207,19 @@ - [ListDocs - 获取文档列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-basic-operation-class/api-aimiaobi-2023-08-01-listdocs.md) - [DeleteDocs - 批量删除文档](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-basic-operation-class/api-aimiaobi-2023-08-01-deletedocs.md) - **妙读-生成类** - - [RunDocIntroduction - 文档导读](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-rundocintroduction.md) - [RunMultiDocIntroduction - 多文档聚合摘要](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-runmultidocintroduction.md) - - [RunDocSummary - 文档摘要](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-rundocsummary.md) + - [RunDocIntroduction - 文档导读](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-rundocintroduction.md) - [RunDocBrainmap - 全文脑图](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-rundocbrainmap.md) + - [RunDocSummary - 文档摘要](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-rundocsummary.md) - [RunDocWashing - 改写](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-rundocwashing.md) - [RunBookIntroduction - 书籍导读（抽取书籍卖点/书籍摘要）](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-runbookintroduction.md) - - [RunBookBrainmap - 书籍脑图](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-runbookbrainmap.md) - [RunCommentGeneration - 客户之声预测](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-runcommentgeneration.md) + - [RunBookBrainmap - 书籍脑图](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-generate-class/api-aimiaobi-2023-08-01-runbookbrainmap.md) + - **妙读-抽取类** + - [RunHotword - 抽取关键词](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-extraction-class/api-aimiaobi-2023-08-01-runhotword.md) - **妙读-问答类** - - [RunDocQa - 文档问答（文章问答/多模态文件问答）](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-question-and-answer-class/api-aimiaobi-2023-08-01-rundocqa.md) - [RunGenerateQuestions - 猜你想问](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-question-and-answer-class/api-aimiaobi-2023-08-01-rungeneratequestions.md) + - [RunDocQa - 文档问答（文章问答/多模态文件问答）](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-question-and-answer-class/api-aimiaobi-2023-08-01-rundocqa.md) - **妙读-其他** - [RunDocSmartCard - 文档智能卡片](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-other/api-aimiaobi-2023-08-01-rundocsmartcard.md) - [RunDocTranslation - 文档翻译](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-reading-other/api-aimiaobi-2023-08-01-rundoctranslation.md) @@ -888,39 +1230,42 @@ - [CancelDeepWriteTask - 取消深度写作任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-deep-writing/api-aimiaobi-2023-08-01-canceldeepwritetask.md) - [GetDeepWriteTaskResult - 查询深度写作任务的结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-deep-writing/api-aimiaobi-2023-08-01-getdeepwritetaskresult.md) - [RunDeepWriting - 查询深度写作事件](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-deep-writing/api-aimiaobi-2023-08-01-rundeepwriting.md) + - **标书生成** + - [AsyncUploadTenderDoc - 招标文档解析](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-asyncuploadtenderdoc.md) + - [GetBiddingRemainLimitNum - 获得标书写作剩余额度](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-getbiddingremainlimitnum.md) + - [GetBiddingDocInfo - 获得标书写作结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-getbiddingdocinfo.md) + - [EditBiddingDoc - 编辑标书内容](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-editbiddingdoc.md) + - [DownloadBiddingDoc - 下载标书文件](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-downloadbiddingdoc.md) + - [AsyncWritingBiddingDoc - 标书写作](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-asyncwritingbiddingdoc.md) + - [ListBiddingDoc - 列出标书写作任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-listbiddingdoc.md) - **PPT生成** - [ListEnterprisePptTemplates - 查询企业专属PPT模板列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-listenterpriseppttemplates.md) - [InitiatePptCreationV2 - 初始化PPT创建操作](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-initiatepptcreationv2.md) - - [GetPptTemplateSelector - 查询PPT模板筛选器](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-getppttemplateselector.md) - [ListPptTemplates - 查询PPT模板列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-listppttemplates.md) - [GetPptArtifactExportResult - 查询PPT导出任务的结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-getpptartifactexportresult.md) + - [GetPptTemplateSelector - 查询PPT模板筛选器](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-getppttemplateselector.md) - [ExportPptArtifact - 导出PPT作品](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-exportpptartifact.md) - [GetPptArtifact - 查询PPT作品信息](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-getpptartifact.md) - - [RunPptOutlineGeneration - 生成PPT大纲内容](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-runpptoutlinegeneration.md) - [ListPptArtifacts - 查询PPT作品列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-listpptartifacts.md) + - [RunPptOutlineGeneration - 生成PPT大纲内容](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-runpptoutlinegeneration.md) - [InitiatePptCreation - 初始化用来创建PPT的会话](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-initiatepptcreation.md) - [GetPptConfig - 获取PPT组件配置](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-getpptconfig.md) - [BindPptArtifact - 绑定PPT作品信息](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-bindpptartifact.md) - [DeletePptArtifact - 删除PPT作品](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-ppt-generation/api-aimiaobi-2023-08-01-deletepptartifact.md) - - **标书生成** - - [AsyncUploadTenderDoc - 招标文档解析](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-asyncuploadtenderdoc.md) - - [GetBiddingRemainLimitNum - 获得标书写作剩余额度](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-getbiddingremainlimitnum.md) - - [GetBiddingDocInfo - 获得标书写作结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-getbiddingdocinfo.md) - - [EditBiddingDoc - 编辑标书内容](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-editbiddingdoc.md) - - [DownloadBiddingDoc - 下载标书文件](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-downloadbiddingdoc.md) - - [AsyncWritingBiddingDoc - 标书写作](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-asyncwritingbiddingdoc.md) - - [ListBiddingDoc - 列出标书写作任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-tender-generation/api-aimiaobi-2023-08-01-listbiddingdoc.md) - **其他** - [RunVideoScriptGenerate - AI生成视频剪辑脚本](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-runvideoscriptgenerate.md) - [GetSmartClipTask - 获取智能剪辑任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-getsmartcliptask.md) - [SubmitSmartClipTask - 提交智能一键成片任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-submitsmartcliptask.md) - [SaveOrUpdateOssConfig - 配置-云存储-参数配置](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-saveorupdateossconfig.md) - [CreateDataPermissions - 权限-批量添加](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-createdatapermissions.md) - - [DeleteDataPermissions - 权限-删除](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-deletedatapermissions.md) + - [GenerateViewPoint - 生成选题视角（已过时，不推荐使用）](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-generateviewpoint.md) - [ListDataPermissions - 权限-列表](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-listdatapermissions.md) + - [DeleteDataPermissions - 权限-删除](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-deletedatapermissions.md) - [GetPptInfo - 查询PPT任务信息](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-getpptinfo.md) - - [GenerateViewPoint - 生成选题视角（已过时，不推荐使用）](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-generateviewpoint.md) - [FetchParseDocumentLayoutTask - 获取排版任务结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-other/api-aimiaobi-2023-08-01-fetchparsedocumentlayouttask.md) + - [CancelAuditTask - 取消审核任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-cancelaudittask.md) + - [QueryAuditTask - 查询审核结果](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-queryaudittask.md) + - [SubmitAuditTask - 提交审核任务](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-dir/api-aimiaobi-2023-08-01-dir-wonderful-pen-article-reviser/api-aimiaobi-2023-08-01-submitaudittask.md) - [服务接入点](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-endpoint.md) - [API概览](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-overview.md) - [版本说明](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-changeset.md) @@ -930,14 +1275,24 @@ - [妙笔写作信源对接](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/quanmiao-more/miaobi-writing-source-docking.md) - [妙搜数据集管理通过API引入数据源](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/quanmiao-more/miaosou-introduce-data-source-through-api.md) - [全妙iframe嵌入方案](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/quanmiao-more/iframe-embedding-scheme.md) - - [全妙Logo定制规范及部署方式](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/quanmiao-more/logo-customization-specification-and-deployment-method.md) - [全妙云存储（OSS）设置指南](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/quanmiao-more/oss-setup-guide.md) + - [全妙Logo定制规范及部署方式](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/quanmiao-more/logo-customization-specification-and-deployment-method.md) - [全妙PaaS AgentKey 获取指南](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/quanmiao-more/quanmiao-paas-agentkey-get-guide.md) + - **最佳实践** + - [妙笔API最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-miaobi-api.md) + - [智能审校最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-smart-audit.md) + - [妙搜API最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-miaosou-api.md) + - [妙策API最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-miaoce-api.md) + - [妙读最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/miaodu-best-practices.md) + - [视频混剪最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/best-practices-for-video-mixing-and-cutting.md) + - [PPT生成最佳实践](raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/miaobi-and-miaoce-best-practices/ppt-generation-best-practices.md) - [官方应用-通义听悟Agent](raw/application-user-guide/application-gallery/official-application-tingwu-agent.md) - - [通义法睿](raw/application-user-guide/application-gallery/tongyi-farui.md) - [官方应用-析言GBI](raw/application-user-guide/application-gallery/xiyan-gbi.md) -- **应用观测** - - [应用观测](raw/application-user-guide/application-monitoring/application-observation.md) + - [通义法睿](raw/application-user-guide/application-gallery/tongyi-farui.md) +- **服务支持** + - [相关协议](raw/application-user-guide/application-support/application-related-agreements.md) + - [常见问题](raw/application-user-guide/application-support/application-faq.md) + - [阿里云百炼平台售后服务范围说明](raw/application-user-guide/application-support/application-after-sales-service-scope.md) - **权限管理** - [权限管理](raw/application-user-guide/application-permission-management/application-permission-management-overview.md) - **实践教程** @@ -946,9 +1301,6 @@ - [在企业微信中集成一个 AI 助手](raw/application-user-guide/application-use-cases/add-an-ai-assistant-to-your-work-wechat.md) - [在钉钉上增加一个AI机器人](raw/application-user-guide/application-use-cases/add-an-ai-assistant-to-your-dingtalk.md) - [基于本地知识库构建RAG应用](raw/application-user-guide/application-use-cases/build-rag-application-based-on-local-retrieval.md) -- **服务支持** - - [常见问题](raw/application-user-guide/application-support/application-faq.md) - - [相关协议](raw/application-user-guide/application-support/application-related-agreements.md) ## 模型 API 参考 @@ -957,14 +1309,17 @@ - [安装SDK](raw/model-api-reference/preparations/install-sdk.md) - [使用百炼 CLI](raw/model-api-reference/preparations/use-model-studio-cli.md) - [错误码](raw/model-api-reference/preparations/error-code.md) +- **3D模型生成** + - [Tripo-3D模型生成](raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md) - **图像生成** - **千问** - - [千问-文生图API参考](raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-api.md) + - [千问-图像生成与编辑3.0 API参考](raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-generation-and-editing-api-reference.md) - [千问-图像编辑API参考](raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-edit-api.md) + - [千问-文生图API参考](raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-api.md) - [千问-图像翻译API参考](raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-mt-image-api.md) - **万相** - - [万相-文生图V2版API参考](raw/model-api-reference/image-generation/wan-image-api-reference/text-to-image-v2-api-reference.md) - [万相-文生图V1版API参考](raw/model-api-reference/image-generation/wan-image-api-reference/text-to-image-api-reference.md) + - [万相-文生图V2版API参考](raw/model-api-reference/image-generation/wan-image-api-reference/text-to-image-v2-api-reference.md) - [万相-图像生成与编辑2.7 API参考](raw/model-api-reference/image-generation/wan-image-api-reference/wan-image-generation-and-editing-api-reference.md) - [万相-通用图像编辑2.5](raw/model-api-reference/image-generation/wan-image-api-reference/wan2-5-image-edit-api-reference.md) - [万相-图像生成与编辑2.6 API参考](raw/model-api-reference/image-generation/wan-image-api-reference/wan-image-generation-api-reference.md) @@ -973,48 +1328,66 @@ - [万相-图像局部重绘API参考](raw/model-api-reference/image-generation/wan-image-api-reference/vary-region-api-reference.md) - **Z-Image** - [Z-Image API参考](raw/model-api-reference/image-generation/z-image-generation-api-reference/z-image-api-reference.md) - - **可灵** - - [可灵-图像生成API参考](raw/model-api-reference/image-generation/kling-image-api-reference/kling-image-generation-api-reference.md) - **Vidu** - [Vidu-图像生成API参考](raw/model-api-reference/image-generation/vidu-image-models/vidu-image-generation-api-reference.md) + - **可灵** + - [可灵-图像生成API参考](raw/model-api-reference/image-generation/kling-image-api-reference/kling-image-generation-api-reference.md) - **创意工具** - [人像风格重绘API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/portrait-style-redraw-api-reference.md) - - [图像画面扩展API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/image-scaling-api.md) - [虚拟模特API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/virtual-model-api-details.md) + - [图像画面扩展API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/image-scaling-api.md) - [鞋靴模特API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/shoe-model-api.md) - [创意海报生成API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/creative-poster-generation-api.md) - [人物实例分割API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/image-instance-segmentation-api-reference.md) + - [图像背景生成API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/wanx-background-generation-api-reference.md) - [图像擦除补全API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/image-erase-completion-api-reference.md) - [AI试衣OutfitAnyone](raw/model-api-reference/image-generation/image-creative-tools-api-reference/outfitanyone.md) - - [图像背景生成API参考](raw/model-api-reference/image-generation/image-creative-tools-api-reference/wanx-background-generation-api-reference.md) - [人物写真生成FaceChain](raw/model-api-reference/image-generation/image-creative-tools-api-reference/facechain-portrait-generation.md) - [创意文字WordArt锦书](raw/model-api-reference/image-generation/image-creative-tools-api-reference/wordart-quick-start.md) - [常见问题](raw/model-api-reference/image-generation/image-faq.md) -- **3D模型生成** - - [Tripo-3D模型生成](raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md) - **实时多模态** - [客户端事件](raw/model-api-reference/omni-realtime-api/client-events.md) + - [服务端事件](raw/model-api-reference/omni-realtime-api/server-events.md) - [Python SDK](raw/model-api-reference/omni-realtime-api/omni-realtime-python-sdk.md) - [Java SDK](raw/model-api-reference/omni-realtime-api/omni-realtime-java-sdk.md) - - [服务端事件](raw/model-api-reference/omni-realtime-api/server-events.md) - [实时多模态交互流程](raw/model-api-reference/omni-realtime-api/omni-realtime-interaction-process.md) - [声音复刻API参考](raw/model-api-reference/omni-realtime-api/qwen-omni-voice-cloning.md) +- **Realtime API** + - **快速开始** + - [SDK下载](raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-sdk-download.md) + - [Token鉴权](raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-token-authentication.md) + - [实现接通模型/应用](raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-connect-model.md) + - **最佳实践** + - [通过WebRTC使用多模态交互套件实现实时通话](raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-multimodal-dialog.md) + - [通过WebRTC使用qwen3.5-omni-plus-realtime实现实时通话](raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-omni-realtime.md) + - [通过AOQ使用qwen3.5-omni-plus-realtime实现实时通话](raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-aoq-omni-realtime.md) + - **AOQ客户端API** + - **AOQ SDK功能介绍** + - [连接状态管理](raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-connection-management.md) + - [媒体流发送管理](raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-media-stream-control.md) + - [音频常用功能介绍](raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-audio-features.md) + - [自定义音频播放](raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-playback.md) + - [自定义音频采集](raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-capture.md) + - [视频常用功能介绍](raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-video-features.md) + - [自定义视频输入](raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-video-input.md) + - [AOQ SDK简介](raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-desc.md) + - [Realtime API简介](raw/model-user-guide/realtime-api-user-guide/realtime-api-overview.md) - **更多模型** - - [通义法睿大语言模型](raw/model-api-reference/more-models/tongyi-farui-api.md) - [意图理解能力](raw/model-api-reference/more-models/intent-detect-capability.md) - - [Qwen-Deep-Research API 参考](raw/model-api-reference/more-models/qwen-deep-research-api.md) + - [通义法睿大语言模型](raw/model-api-reference/more-models/tongyi-farui-api.md) - [Qwen-MT API参考](raw/model-api-reference/more-models/qwen-mt-api.md) + - [Qwen-Deep-Research API 参考](raw/model-api-reference/more-models/qwen-deep-research-api.md) - [GUI-Plus API参考](raw/model-api-reference/more-models/gui-plus-interface-interaction-model.md) - [Qwen-OCR API参考](raw/model-api-reference/more-models/qwen-vl-ocr-api-reference.md) - **工具包/框架** - [OpenAI Chat接口兼容](raw/model-api-reference/toolkits-and-frameworks/compatibility-of-openai-with-dashscope.md) - - [OpenAI Responses接口兼容](raw/model-api-reference/toolkits-and-frameworks/compatibility-with-openai-responses-api.md) - [completions 接口](raw/model-api-reference/toolkits-and-frameworks/completions.md) - [OpenAI Vision接口兼容](raw/model-api-reference/toolkits-and-frameworks/qwen-vl-compatible-with-openai.md) + - [OpenAI Responses接口兼容](raw/model-api-reference/toolkits-and-frameworks/compatibility-with-openai-responses-api.md) - [OpenAI文件接口兼容](raw/model-api-reference/toolkits-and-frameworks/openai-file-interface.md) - - [OpenAI兼容-Batch Chat](raw/model-api-reference/toolkits-and-frameworks/openai-compatible-batch-chat.md) - - [OpenAI Embedding接口兼容](raw/model-api-reference/toolkits-and-frameworks/embedding-interfaces-compatible-with-openai.md) - [OpenAI兼容-Batch（文件输入）](raw/model-api-reference/toolkits-and-frameworks/batch-interfaces-compatible-with-openai.md) + - [OpenAI Embedding接口兼容](raw/model-api-reference/toolkits-and-frameworks/embedding-interfaces-compatible-with-openai.md) + - [OpenAI兼容-Batch Chat](raw/model-api-reference/toolkits-and-frameworks/openai-compatible-batch-chat.md) - [OpenAI Conversations接口兼容](raw/model-api-reference/toolkits-and-frameworks/openai-compatible-conversations.md) - [在LangChain中使用阿里云百炼](raw/model-api-reference/toolkits-and-frameworks/use-bailian-in-langchain.md) - **模型生产** @@ -1023,15 +1396,15 @@ - **更多** - [生成临时API Key](raw/model-api-reference/more-about-models/generate-temporary-api-key.md) - [异步任务管理 API](raw/model-api-reference/more-about-models/manage-asynchronous-tasks.md) - - [通过HTTP回调URL或MQ接收异步任务完成通知](raw/model-api-reference/more-about-models/async-task-api.md) - [子业务空间的模型调用](raw/model-api-reference/more-about-models/model-calling-in-sub-workspace.md) - - [DashScope SDK连接复用配置](raw/model-api-reference/more-about-models/connection-multiplexing-configuration.md) + - [通过HTTP回调URL或MQ接收异步任务完成通知](raw/model-api-reference/more-about-models/async-task-api.md) - [上传本地文件获取临时URL](raw/model-api-reference/more-about-models/get-temporary-file-url.md) + - [DashScope SDK连接复用配置](raw/model-api-reference/more-about-models/connection-multiplexing-configuration.md) - **视频生成** - **HappyHorse** - [HappyHorse-文生视频API参考](raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-text-to-video-api-reference.md) - - [HappyHorse-参考生视频API参考](raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-reference-to-video-api-reference.md) - [HappyHorse-图生视频-基于首帧API参考](raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-image-to-video-api-reference.md) + - [HappyHorse-参考生视频API参考](raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-reference-to-video-api-reference.md) - [HappyHorse-视频编辑API参考](raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-video-edit-api-reference.md) - **万相** - **万相-早期视频模型（2.1-2.6）** @@ -1045,46 +1418,40 @@ - [万相2.7-参考生视频API参考](raw/model-api-reference/video-generation-api/wan-api-reference/wan-video-to-video-api-reference.md) - [万相2.7-视频编辑API参考](raw/model-api-reference/video-generation-api/wan-api-reference/wan-video-editing-api-reference.md) - [万相-图生动作API参考](raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-move-api.md) - - [万相-数字人](raw/model-api-reference/video-generation-api/wan-api-reference/wan-s2v-overview.md) - [万相-视频换人API参考](raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-mix-api.md) - - **爱诗** - - [爱诗-图生视频-基于首帧API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-image-to-video-api-reference.md) - - [爱诗-文生视频API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-text-to-video-api-reference.md) - - [爱诗-图生视频-基于首尾帧API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-keyframe-to-video-api-reference.md) - - [爱诗-参考生视频API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-reference-to-video-api-reference.md) - - [爱诗-视频超清API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-upscale-api-reference.md) - - [爱诗-视频对口型API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-lipsync-api-reference.md) - - [爱诗-视频动作模仿API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-motioncontrol-api-reference.md) + - [万相-数字人](raw/model-api-reference/video-generation-api/wan-api-reference/wan-s2v-overview.md) - **人像驱动** - [图生舞蹈视频-舞动人像AnimateAnyone](raw/model-api-reference/video-generation-api/portrait-animation-api-reference/animateanyone-quick-start.md) - [图生唱演视频-悦动人像EMO](raw/model-api-reference/video-generation-api/portrait-animation-api-reference/emo-quick-start.md) - [图生播报视频-灵动人像LivePortrait](raw/model-api-reference/video-generation-api/portrait-animation-api-reference/liveportrait-quick-start.md) - - [图生表情包视频-表情包Emoji](raw/model-api-reference/video-generation-api/portrait-animation-api-reference/emoji-quick-start.md) - [视频口型替换-声动人像VideoRetalk](raw/model-api-reference/video-generation-api/portrait-animation-api-reference/videoretalk.md) + - [图生表情包视频-表情包Emoji](raw/model-api-reference/video-generation-api/portrait-animation-api-reference/emoji-quick-start.md) - [视频风格重绘API参考](raw/model-api-reference/video-generation-api/portrait-animation-api-reference/video-style-transform-api-reference.md) + - **爱诗** + - [爱诗-文生视频API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-text-to-video-api-reference.md) + - [爱诗-图生视频-基于首帧API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-image-to-video-api-reference.md) + - [爱诗-图生视频-基于首尾帧API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-keyframe-to-video-api-reference.md) + - [爱诗-视频对口型API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-lipsync-api-reference.md) + - [爱诗-视频动作模仿API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-motioncontrol-api-reference.md) + - [爱诗-参考生视频API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-reference-to-video-api-reference.md) + - [爱诗-视频超清API参考](raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-upscale-api-reference.md) + - **可灵** + - [可灵-视频生成API文档](raw/model-api-reference/video-generation-api/kling-api-reference/kling-video-generation-api-reference.md) - **Vidu** - - [Vidu-文生视频API参考](raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-text-to-video-api-reference.md) - [Vidu-图生视频-基于首帧API参考](raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-image-to-video-api-reference.md) + - [Vidu-文生视频API参考](raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-text-to-video-api-reference.md) - [Vidu-图生视频-基于首尾帧API参考](raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-keyframe-to-video-api-reference.md) - [Vidu-参考生视频 API 参考](raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-reference-to-video-api-reference.md) - - **可灵** - - [可灵-视频生成API文档](raw/model-api-reference/video-generation-api/kling-api-reference/kling-video-generation-api-reference.md) - **音频** - **语音识别** - **实时语音识别（Fun-ASR）** - [Fun-ASR实时语音识别WebSocket API](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-realtime-websocket-api.md) - [实时语音识别（Fun-ASR）客户端事件](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-client-events.md) - [实时语音识别（Fun-ASR）服务端事件](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-server-events.md) - - [Python SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-realtime-python-sdk.md) - [Java SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-realtime-java-sdk.md) + - [Python SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-realtime-python-sdk.md) - [Fun-ASR实时语音识别Android SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/android-sdk-for-fun-asr-real-time-service.md) - [Fun-ASR实时语音识别iOS SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/ios-sdk-for-fun-asr-real-time-service.md) - - **实时语音识别（Qwen-ASR-Realtime）** - - [Qwen-ASR实时语音识别WebSocket API](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-interaction-process.md) - - [实时语音识别（Qwen-ASR-Realtime）客户端事件](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-client-events.md) - - [实时语音识别（Qwen-ASR-Realtime）服务端事件](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-server-events.md) - - [实时语音识别（Qwen-ASR-Realtime）Python SDK-API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-python-sdk.md) - - [实时语音识别（Qwen-ASR-Realtime）Java SDK-API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-java-sdk.md) - **实时语音识别（Paraformer）** - [Paraformer实时语音识别WebSocket API](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/websocket-for-paraformer-real-time-service.md) - [实时语音识别（Paraformer）客户端事件](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-client-events.md) @@ -1093,24 +1460,32 @@ - [Paraformer实时语音识别Android SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/android-sdk-for-paraformer-real-time-service.md) - [Paraformer实时语音识别Java SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-real-time-speech-recognition-java-sdk.md) - [Paraformer实时语音识别iOS SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/ios-sdk-for-paraformer-real-time-service.md) - - **录音文件识别（Fun-ASR）** - - [Fun-ASR录音文件识别HTTP API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-http-api.md) - - [Fun-ASR录音文件识别Python SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/funauidio-asr-recorded-speech-recognition-python-sdk.md) - - [Fun-ASR录音文件识别iOS SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-ios-sdk.md) - - [Fun-ASR录音文件识别Android SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-android-sdk.md) - - [Fun-ASR录音文件识别Java SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-java-sdk.md) - - **录音文件识别（Paraformer）** - - [Paraformer录音文件识别RESTful API](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-restful-api.md) - - [Paraformer录音文件识别Java SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-java-sdk.md) - - [Paraformer录音文件识别Python SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-python-sdk.md) - - [Paraformer录音文件识别Android SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-android-sdk.md) - - [Paraformer录音文件识别iOS SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-ios-sdk.md) + - **实时语音识别（Qwen-ASR-Realtime）** + - [Qwen-ASR实时语音识别WebSocket API](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-interaction-process.md) + - [实时语音识别（Qwen-ASR-Realtime）客户端事件](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-client-events.md) + - [实时语音识别（Qwen-ASR-Realtime）Python SDK-API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-python-sdk.md) + - [实时语音识别（Qwen-ASR-Realtime）Java SDK-API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-java-sdk.md) + - [实时语音识别（Qwen-ASR-Realtime）服务端事件](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-server-events.md) + - **非实时语音识别（Fun-ASR）** + - [Fun-ASR非实时语音识别HTTP API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-http-api.md) + - [Fun-ASR非实时语音识别Python SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/funauidio-asr-recorded-speech-recognition-python-sdk.md) + - [Fun-ASR非实时语音识别Java SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-java-sdk.md) + - [Fun-ASR非实时语音识别iOS SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-ios-sdk.md) + - [Fun-ASR非实时语音识别Android SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-android-sdk.md) + - **非实时语音识别（Paraformer）** + - [Paraformer非实时语音识别HTTP API](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-restful-api.md) + - [Paraformer非实时语音识别Python SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-python-sdk.md) - [最佳实践](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-best-practices.md) + - [Paraformer非实时语音识别Java SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-java-sdk.md) + - [Paraformer非实时语音识别iOS SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-ios-sdk.md) + - [Paraformer非实时语音识别Android SDK](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-android-sdk.md) - **定制热词** - - [定制热词Python SDK参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/custom-hot-words/vocabulary-python-sdk.md) - [定制热词HTTP API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/custom-hot-words/vocabulary-http-api.md) + - [定制热词Python SDK参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/custom-hot-words/vocabulary-python-sdk.md) - [定制热词Java SDK参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/custom-hot-words/vocabulary-java-sdk.md) - - [录音文件识别（Qwen-ASR）API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-api-reference.md) + - [非实时语音识别（Fun-ASR-Flash）API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-flash.md) + - [非实时语音识别（Fun-ASR-Realtime）API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-realtime.md) + - [非实时语音识别（Qwen-ASR）API参考](raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-api-reference.md) - **语音合成** - **实时语音合成（Qwen-Audio-TTS/CosyVoice）** - [Qwen-Audio-TTS/CosyVoice WebSocket API参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-websocket-api.md) @@ -1123,8 +1498,8 @@ - **实时语音合成（Qwen-TTS-Realtime）** - [Qwen-TTS-Realtime WebSocket API 参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-realtime-api-reference/interactive-process-of-qwen-tts-realtime-synthesis.md) - [客户端事件](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-realtime-api-reference/qwen-tts-realtime-client-events.md) - - [服务端事件](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-realtime-api-reference/qwen-tts-realtime-server-events.md) - [Python SDK](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-realtime-api-reference/qwen-tts-realtime-python-sdk.md) + - [服务端事件](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-realtime-api-reference/qwen-tts-realtime-server-events.md) - [Java SDK](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-realtime-api-reference/qwen-tts-realtime-java-sdk.md) - **实时语音合成（Sambert）** - [Sambert WebSocket API 参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sambert-speech-synthesis/sambert-websocket-api.md) @@ -1134,16 +1509,16 @@ - [语音合成Sambert Android SDK](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sambert-speech-synthesis/sambert-android-sdk.md) - [语音合成Sambert Python SDK](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sambert-speech-synthesis/sambert-python-sdk.md) - [语音合成Sambert iOS SDK](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sambert-speech-synthesis/sambert-ios-sdk.md) - - **非实时语音合成（MiniMax）** - - [MiniMax同步语音合成API参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/minimax-speech-synthesis/minimax-synchronous-speech-synthesis-api.md) - **非实时语音合成（Qwen-Audio-TTS/CosyVoice）** - - [非实时语音合成Qwen-Audio-TTS/CosyVoice Python SDK参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-python-sdk.md) - [非实时语音合成Qwen-Audio-TTS/CosyVoice HTTP API参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-http-api.md) - [非实时语音合成Qwen-Audio-TTS/CosyVoice Java SDK参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-java-sdk.md) + - [非实时语音合成Qwen-Audio-TTS/CosyVoice Python SDK参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-python-sdk.md) + - **非实时语音合成（MiniMax）** + - [MiniMax同步语音合成API参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/minimax-speech-synthesis/minimax-synchronous-speech-synthesis-api.md) - **声音复刻** + - [声音复刻Python SDK参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-python-sdk.md) - [声音复刻HTTP API参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-design-http-api.md) - [声音复刻Java SDK参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-java-sdk.md) - - [声音复刻Python SDK参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-python-sdk.md) - [非实时语音合成（Qwen-TTS）API参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-api.md) - [声音设计API参考](raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/voice-design-api-references.md) - **音乐生成** @@ -1157,8 +1532,8 @@ - [音视频翻译-通义千问 API 参考](raw/model-api-reference/audio-api-references/speech-translation-api-reference/qwen3-livetranslate-flash-api.md) - **语音对话** - **实时语音对话** + - [Qwen-Audio 实时语音对话WebSocket API参考](raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/fun-audiochat-realtime-websocket-api.md) - [Qwen-Audio 实时语音对话客户端事件](raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/fun-audiochat-client-events.md) - - [Qwen-Audio 实时语音对话WebSocket API参考](raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/qwen-audio-realtime-websocket-api.md) - [Qwen-Audio 实时语音对话服务端事件](raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/qwen-audio-realtime-server-events.md) - **向量与排序** - **通用文本向量** @@ -1174,95 +1549,96 @@ ## 应用 API 参考 - **Managed Agents** - - [API 总览与认证](raw/application-api-reference/managed-agents-api/managed-agents-api-overview.md) - [快速开始](raw/application-api-reference/managed-agents-api/managed-agents-quickstart.md) - [Agent](raw/application-api-reference/managed-agents-api/agent-api.md) + - [API 总览与认证](raw/application-api-reference/managed-agents-api/managed-agents-api-overview.md) - [Environment](raw/application-api-reference/managed-agents-api/environment-api.md) - [Session and Event](raw/application-api-reference/managed-agents-api/session-api.md) - [File](raw/application-api-reference/managed-agents-api/files-api.md) - [Skill](raw/application-api-reference/managed-agents-api/skills-api.md) +- **应用调用** + - **DashScope API** + - [新版智能体应用 API 参考](raw/application-api-reference/application-call/application-dashscope-api-reference/new-agent-application-api-reference.md) + - [应用 DashScope API 参考](raw/application-api-reference/application-call/application-dashscope-api-reference/agent-and-workflow-application-api-reference.md) + - **Responses API** + - [同步调用 API 参考](raw/application-api-reference/application-call/openai-responses-api/synchronous-call-api-reference.md) + - [异步调用API参考](raw/application-api-reference/application-call/openai-responses-api/asynchronous-call-api-reference.md) + - [获取APP ID和Workspace ID](raw/application-api-reference/application-call/obtain-the-app-id-and-workspace-id.md) - **应用组件** - **API目录** + - **知识库** + - [CreateIndex - 创建知识库](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-createindex.md) + - [GetIndexJobStatus - 查询知识库创建任务状态](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-getindexjobstatus.md) + - [SubmitIndexJob - 提交知识库创建任务](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-submitindexjob.md) + - [SubmitIndexAddDocumentsJob - 提交知识库追加任务](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-submitindexadddocumentsjob.md) + - [Retrieve - 检索知识库](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-retrieve.md) + - [ListIndexDocuments - 查询知识库下的文件列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindexdocuments.md) + - [ListIndexFileDetails - 查询知识库下的文件详情](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindexfiledetails.md) + - [UpdateIndex - 更新知识库](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-updateindex.md) + - [DeleteIndexDocument - 删除知识库下的文件](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-deleteindexdocument.md) + - [DeleteIndex - 删除知识库](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-deleteindex.md) + - [ListIndices - 查询知识库列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindices.md) + - [ListChunks - 查询索引下的分片列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listchunks.md) + - [DeleteChunk - 删除切片](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-deletechunk.md) + - [GetIndexMonitor - 获取知识库监控数据](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-getindexmonitor.md) + - [UpdateChunk - 修改切片](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-updatechunk.md) - **数据连接（原应用数据）** - [AddCategory - 新增类目](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-addcategory.md) - [ListCategory - 类目列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-listcategory.md) - - [DeleteCategory - 删除类目](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-deletecategory.md) - [ApplyFileUploadLease - 申请文件上传租约](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-applyfileuploadlease.md) + - [DeleteCategory - 删除类目](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-deletecategory.md) - [AddFile - 添加文件](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-addfile.md) - [AddFilesFromAuthorizedOss - 从已授权OSS Bucket中导入文件](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-addfilesfromauthorizedoss.md) - - [DescribeFile - 查询文件状态](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-describefile.md) - [ListFile - 文件列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-listfile.md) + - [DescribeFile - 查询文件状态](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-describefile.md) - [UpdateFileTag - 更新文件标签](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-updatefiletag.md) - - [BatchUpdateFileTag - 批量更新文档标签](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-batchupdatefiletag.md) - [DeleteFile - 删除文件](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-deletefile.md) + - [BatchUpdateFileTag - 批量更新文档标签](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-batchupdatefiletag.md) - [DeleteFiles - 批量删除文件](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-deletefiles.md) - - [GetAvailableParserTypes - 获取文件支持的解析器类型](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-getavailableparsertypes.md) - [GetParseSettings - 获取类目解析设置](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-getparsesettings.md) + - [GetAvailableParserTypes - 获取文件支持的解析器类型](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-getavailableparsertypes.md) - [ChangeParseSetting - 修改类目解析设置](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-changeparsesetting.md) - [AddTable - 添加表格](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-addtable.md) - [UpdateTableFromAuthorizedOss - 从已授权OSS Bucket中选择文件更新表格](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-updatetablefromauthorizedoss.md) - [AddConnector - 新增连接器](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-addconnector.md) - [GetConnector - 获取连接器信息](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-getconnector.md) - [UpdateConnector - 编辑连接器](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-updateconnector.md) - - **知识库** - - [CreateIndex - 创建知识库](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-createindex.md) - - [GetIndexJobStatus - 查询知识库创建任务状态](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-getindexjobstatus.md) - - [SubmitIndexAddDocumentsJob - 提交知识库追加任务](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-submitindexadddocumentsjob.md) - - [SubmitIndexJob - 提交知识库创建任务](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-submitindexjob.md) - - [Retrieve - 检索知识库](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-retrieve.md) - - [DeleteIndexDocument - 删除知识库下的文件](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-deleteindexdocument.md) - - [ListIndexFileDetails - 查询知识库下的文件详情](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindexfiledetails.md) - - [ListIndexDocuments - 查询知识库下的文件列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindexdocuments.md) - - [UpdateIndex - 更新知识库](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-updateindex.md) - - [ListIndices - 查询知识库列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindices.md) - - [DeleteIndex - 删除知识库](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-deleteindex.md) - - [UpdateChunk - 修改切片](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-updatechunk.md) - - [ListChunks - 查询索引下的分片列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listchunks.md) - - [DeleteChunk - 删除切片](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-deletechunk.md) - - [GetIndexMonitor - 获取知识库监控数据](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-getindexmonitor.md) - **Prompt工程** - [CreatePromptTemplate - 创建Prompt模板](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-prompt-engineering/api-bailian-2023-12-29-createprompttemplate.md) - [GetPromptTemplate - 获取Prompt模板](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-prompt-engineering/api-bailian-2023-12-29-getprompttemplate.md) - [UpdatePromptTemplate - 更新Prompt模板](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-prompt-engineering/api-bailian-2023-12-29-updateprompttemplate.md) - - [ListPromptTemplates - 获取Prompt模板列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-prompt-engineering/api-bailian-2023-12-29-listprompttemplates.md) - [DeletePromptTemplate - 删除Prompt模板](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-prompt-engineering/api-bailian-2023-12-29-deleteprompttemplate.md) + - [ListPromptTemplates - 获取Prompt模板列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-prompt-engineering/api-bailian-2023-12-29-listprompttemplates.md) - **其他** - **长期记忆（旧）** - [CreateMemory - 创建长期记忆体](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-creatememory.md) - - [UpdateMemory - 更新长期记忆体](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-updatememory.md) - [GetMemory - 获取长期记忆体](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-getmemory.md) + - [UpdateMemory - 更新长期记忆体](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-updatememory.md) - [DeleteMemory - 删除长期记忆体](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-deletememory.md) + - [CreateMemoryNode - 创建记忆片段](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-creatememorynode.md) - [ListMemories - 获取长期记忆体列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-listmemories.md) - - [UpdateMemoryNode - 更新记忆片段](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-updatememorynode.md) - [GetMemoryNode - 获取记忆片段](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-getmemorynode.md) - - [CreateMemoryNode - 创建记忆片段](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-creatememorynode.md) - - [DeleteMemoryNode - 删除记忆片段](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-deletememorynode.md) + - [UpdateMemoryNode - 更新记忆片段](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-updatememorynode.md) - [ListMemoryNodes - 获取记忆片段列表](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-listmemorynodes.md) - - [GetAlipayTransferStatus - 查询支付宝打赏状态](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-getalipaytransferstatus.md) + - [DeleteMemoryNode - 删除记忆片段](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-deletememorynode.md) - [GetAlipayUrl - 获取支付宝打赏URL](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-getalipayurl.md) + - [GetAlipayTransferStatus - 查询支付宝打赏状态](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-getalipaytransferstatus.md) - [ApplyTempStorageLease - 申请临时文件上传许可](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-applytempstoragelease.md) + - [AddChunk - 新增切片](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-addchunk.md) - [API概览](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-overview.md) - [服务接入点](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-endpoint.md) - [授权信息](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-ram.md) - [版本说明](raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-changeset.md) -- **应用调用** - - **DashScope API** - - [新版智能体应用 API 参考](raw/application-api-reference/application-call/application-dashscope-api-reference/new-agent-application-api-reference.md) - - [应用 DashScope API 参考](raw/application-api-reference/application-call/application-dashscope-api-reference/agent-and-workflow-application-api-reference.md) - - **Responses API** - - [同步调用 API 参考](raw/application-api-reference/application-call/openai-responses-api/synchronous-call-api-reference.md) - - [异步调用API参考](raw/application-api-reference/application-call/openai-responses-api/asynchronous-call-api-reference.md) - - [获取APP ID和Workspace ID](raw/application-api-reference/application-call/obtain-the-app-id-and-workspace-id.md) +- **长期记忆** + - [长期记忆（新）API 参考](raw/application-api-reference/long-term-memory-new/long-term-memory-api-reference.md) - **框架** - **Spring AI Alibaba** - [使用Spring AI Alibaba集成阿里云百炼大模型应用](raw/application-api-reference/frameworks/spring-ai-alibaba/spring-ai-alibaba-integrate-llm-application.md) - [通过Spring AI Alibaba检索阿里云百炼知识库](raw/application-api-reference/frameworks/spring-ai-alibaba/spring-ai-alibaba-integrate-knowledge-base.md) - [通过LlamaIndex API构建RAG应用](raw/application-api-reference/frameworks/llamaindex.md) -- **长期记忆** - - [长期记忆（新）API 参考](raw/application-api-reference/long-term-memory-new/long-term-memory-api-reference.md) - **更多** - - [服务关联角色](raw/application-api-reference/more/bailian-service-linked-role.md) - [生成临时API Key](raw/application-api-reference/more/application-obtain-temporary-authentication-token.md) + - [服务关联角色](raw/application-api-reference/more/bailian-service-linked-role.md) - [知识库SearchFilters](raw/application-api-reference/more/how-to-use-search-filters.md) - [知识检索与问答](raw/application-api-reference/knowledge.md) diff --git a/skills/bailian-docs-llm-wiki/models/families.jsonl b/skills/bailian-docs-llm-wiki/models/families.jsonl index 8bcd4a65..1ef95398 100644 --- a/skills/bailian-docs-llm-wiki/models/families.jsonl +++ b/skills/bailian-docs-llm-wiki/models/families.jsonl @@ -34,7 +34,7 @@ {"slug":"image-erase-completion","name":"图像擦除补全","description":"图像擦除补全通过指定图像mask中要删除的人体、宠物、物品、文字、水印等图像区域，在保留背景的同时移除图像中的一个或多个人物、物体、文字等元素，此功能不支持输入prompt的消除。擦除补全技术结合了计算机视觉、AIGC inpainting等先进技术，可以在多种场景下应用，从而满足用户对隐私保护、内容创作和图像编辑等方面需求。","primaryCapability":"IG","capabilities":["IG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"image-erase-completion","name":"图像擦除补全","capabilities":["IG"]}],"detailPath":"groups/image-erase-completion.json"} {"slug":"image-instance-segmentation","name":"人物实例分割","description":"人物实例分割运用了检测和分割技术，不仅能够在图像中识别出不同的对象，而且还能准确地画出每一个对象边界的像素级掩码（mask）。","primaryCapability":"IG","capabilities":["IG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"image-instance-segmentation","name":"人物实例分割","capabilities":["IG"]}],"detailPath":"groups/image-instance-segmentation.json"} {"slug":"image-out-painting","name":"图像画面扩展","description":"图像画面大模型，对输入图像进行画面自由扩展，支持旋转画面，支持按照扩展系数和扩展像素数两种方式进行扩图。用户可以通过指定宽度、高度画面扩展比例或者左、右、上、下的扩展的像素值来控制画面扩展，可用于创意娱乐、辅助作图、画面设计、影视后期制作等场景。","primaryCapability":"IG","capabilities":["IG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"image-out-painting","name":"图像画面扩展","capabilities":["IG"]}],"detailPath":"groups/image-out-painting.json"} -{"slug":"kimi-models-market-place","name":"Kimi","description":"由月之暗面提供的Kimi系列模型的API服务。","primaryCapability":"TG","capabilities":["TG","Reasoning","VU"],"providers":["moonshot-ai"],"itemCount":4,"items":[{"model":"kimi/kimi-k2.5","name":"Kimi/Kimi K2.5","contextWindow":262144,"capabilities":["TG","Reasoning","VU"]},{"model":"kimi/kimi-k2.6","name":"Kimi/Kimi K2.6","contextWindow":262144,"capabilities":["TG","Reasoning","VU"]},{"model":"kimi/kimi-k2.7-code","name":"kimi/kimi-k2.7-code","contextWindow":262144,"capabilities":["TG","VU","Reasoning"]},{"model":"kimi/kimi-k2.7-code-highspeed","name":"kimi/kimi-k2.7-code-highspeed","contextWindow":262144,"capabilities":["TG","Reasoning","VU"]}],"detailPath":"groups/kimi-models-market-place.json","maxContextWindow":262144} +{"slug":"kimi-models-market-place","name":"Kimi","description":"由月之暗面提供的Kimi系列模型的API服务。","primaryCapability":"TG","capabilities":["TG","VU","Reasoning"],"providers":["moonshot-ai"],"itemCount":5,"items":[{"model":"kimi/kimi-k2.5","name":"Kimi/Kimi K2.5","contextWindow":262144,"capabilities":["TG","Reasoning","VU"]},{"model":"kimi/kimi-k2.6","name":"Kimi/Kimi K2.6","contextWindow":262144,"capabilities":["TG","Reasoning","VU"]},{"model":"kimi/kimi-k2.7-code","name":"kimi/kimi-k2.7-code","contextWindow":262144,"capabilities":["TG","VU","Reasoning"]},{"model":"kimi/kimi-k2.7-code-highspeed","name":"kimi/kimi-k2.7-code-highspeed","contextWindow":262144,"capabilities":["TG","Reasoning","VU"]},{"model":"kimi/kimi-k3","name":"kimi/kimi-k3","contextWindow":1048576,"capabilities":["TG","VU","Reasoning"]}],"detailPath":"groups/kimi-models-market-place.json","maxContextWindow":1048576} {"slug":"kling-models-market-place","name":"可灵AI","description":"由可灵AI提供的高质量视频与图像生成及编辑模型。","primaryCapability":"VG","capabilities":["VG","IG"],"providers":["kling"],"itemCount":4,"items":[{"model":"kling/kling-v3-image-generation","name":"Kling Image 3.0","capabilities":["IG"]},{"model":"kling/kling-v3-omni-image-generation","name":"Kling Image 3.0 Omni","capabilities":["IG"]},{"model":"kling/kling-v3-omni-video-generation","name":"Kling Video 3.0 Omni","capabilities":["VG"]},{"model":"kling/kling-v3-video-generation","name":"Kling Video 3.0","capabilities":["VG"]}],"detailPath":"groups/kling-models-market-place.json"} {"slug":"liveportrait-detect","name":"灵动人像LivePortrait-detect","description":"LivePortrait-detect是辅助LivePortrait的图像检测模型，用于检测图片中的人物形象是否符合视频生成要求。","primaryCapability":"VG","capabilities":["VG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"liveportrait-detect","name":"灵动人像LivePortrait-detect","capabilities":["VG"]}],"detailPath":"groups/liveportrait-detect.json"} {"slug":"liveportrait","name":"灵动人像LivePortrait","description":"LivePortrait是一款视频生成模型，可基于人物图片生成轻量化的人物肖像动态视频。","primaryCapability":"VG","capabilities":["VG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"liveportrait","name":"灵动人像LivePortrait","capabilities":["VG"]}],"detailPath":"groups/liveportrait.json"} @@ -54,18 +54,19 @@ {"slug":"pixverse-v6-market-place","name":"PixVerse V6","description":"由爱诗科技提供的PixVerse V系列视频大模型API服务。","primaryCapability":"VG","capabilities":["VG"],"providers":["pixverse"],"itemCount":4,"items":[{"model":"pixverse/pixverse-v6-it2v","name":"PixVerse-V6-it2v","capabilities":["VG"]},{"model":"pixverse/pixverse-v6-kf2v","name":"PixVerse-V6-kf2v","capabilities":["VG"]},{"model":"pixverse/pixverse-v6-r2v","name":"PixVerse-V6-r2v","capabilities":["VG"]},{"model":"pixverse/pixverse-v6-t2v","name":"PixVerse-V6-t2v","capabilities":["VG"]}],"detailPath":"groups/pixverse-v6-market-place.json"} {"slug":"qvq-max","name":"QVQ-Max","description":"千问QVQ视觉推理模型，支持视觉输入及思维链输出，在数学、编程、视觉分析、创作以及通用任务上都表现了更强的能力。","primaryCapability":"Reasoning","capabilities":["Reasoning","VU"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qvq-max","name":"QVQ-Max","contextWindow":131072,"capabilities":["Reasoning","VU"]}],"detailPath":"groups/qvq-max.json","maxContextWindow":131072} {"slug":"qvq-plus","name":"Qwen-QVQ-Plus","description":"千问QVQ视觉推理模型增强版，支持视觉输入及思维链输出，在数学、编程、视觉分析、创作以及通用任务上都表现了更强的能力。","primaryCapability":"Reasoning","capabilities":["Reasoning","VU"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qvq-plus","name":"QVQ-Plus","contextWindow":131072,"capabilities":["Reasoning","VU"]}],"detailPath":"groups/qvq-plus.json","maxContextWindow":131072} -{"slug":"qwen-audio-realtime-flash","name":"Qwen-Audio-Realtime-Flash","description":"Qwen-Audio-Realtime 是一款登顶全球权威评测的下一代实时双工语音大模型，模型兼顾了模型智商与双工对话节奏,在保持流畅、自然的实时交互体验的同时,语音推理能力不打折扣;并通过并行推理和全向流式等工程优化,将端到端响应时延控制在低水平,实现\"又快又聪明\"的对话体验。Flash版更注重极致的响应速度","primaryCapability":"Realtime-Chatting","capabilities":["Realtime-Chatting"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-audio-3.0-realtime-flash","name":"千问实时语音对话大模型3.0（极速版）","contextWindow":8192,"capabilities":["Realtime-Chatting"]}],"detailPath":"groups/qwen-audio-realtime-flash.json","maxContextWindow":8192} -{"slug":"qwen-audio-realtime-plus","name":"Qwen-Audio-Realtime-Plus","description":"Qwen-Audio-Realtime 是一款登顶全球权威评测的下一代实时双工语音大模型，模型兼顾了模型智商与双工对话节奏,在保持流畅、自然的实时交互体验的同时,语音推理能力不打折扣;并通过并行推理和全向流式等工程优化,将端到端响应时延控制在低水平,实现\"又快又聪明\"的对话体验。Plus版本更注重高质量的回复结果。","primaryCapability":"Realtime-Chatting","capabilities":["Realtime-Chatting"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-audio-3.0-realtime-plus","name":"千问实时语音大模型 （标准版）","contextWindow":8192,"capabilities":["Realtime-Chatting"]}],"detailPath":"groups/qwen-audio-realtime-plus.json","maxContextWindow":8192} +{"slug":"qwen-audio-realtime-flash","name":"Qwen-Audio-Realtime-Flash","description":"Qwen-Audio-Realtime 是一款登顶全球权威评测的下一代实时双工语音大模型，模型兼顾了模型智商与双工对话节奏,在保持流畅、自然的实时交互体验的同时,语音推理能力不打折扣;并通过并行推理和全向流式等工程优化,将端到端响应时延控制在低水平,实现\"又快又聪明\"的对话体验。Flash版更注重极致的响应速度","primaryCapability":"Realtime-Chatting","capabilities":["Realtime-Chatting"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-audio-3.0-realtime-flash","name":"千问实时语音对话大模型3.0（极速版）","contextWindow":40960,"capabilities":["Realtime-Chatting"]}],"detailPath":"groups/qwen-audio-realtime-flash.json","maxContextWindow":40960} +{"slug":"qwen-audio-realtime-plus","name":"Qwen-Audio-Realtime-Plus","description":"Qwen-Audio-Realtime 是一款登顶全球权威评测的下一代实时双工语音大模型，模型兼顾了模型智商与双工对话节奏,在保持流畅、自然的实时交互体验的同时,语音推理能力不打折扣;并通过并行推理和全向流式等工程优化,将端到端响应时延控制在低水平,实现\"又快又聪明\"的对话体验。Plus版本更注重高质量的回复结果。","primaryCapability":"Realtime-Chatting","capabilities":["Realtime-Chatting"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-audio-3.0-realtime-plus","name":"千问实时语音大模型 （标准版）","contextWindow":40960,"capabilities":["Realtime-Chatting"]}],"detailPath":"groups/qwen-audio-realtime-plus.json","maxContextWindow":40960} {"slug":"qwen-audio-tts","name":"Qwen-Audio-TTS","description":"Qwen-Audio-TTS是一款面向实时交互场景和高质量语音生成场景的语音合成大模型。模型支持多种小语种和中文方言，并具有Free-style 指令遵循能力、Context能力和细粒度标签控制能力，可更灵活地控制情绪、语气、角色、语速、音量等表达方式。","primaryCapability":"Realtime-Text-to-Speech","capabilities":["Realtime-Text-to-Speech"],"providers":["qwen"],"itemCount":2,"items":[{"model":"qwen-audio-3.0-tts-flash","name":"qwen-audio-3.0-tts-flash","capabilities":["Realtime-Text-to-Speech"]},{"model":"qwen-audio-3.0-tts-plus","name":"qwen-audio-3.0-tts-plus","capabilities":["Realtime-Text-to-Speech"]}],"detailPath":"groups/qwen-audio-tts.json"} {"slug":"qwen-coder-plus","name":"Qwen-Coder-Plus","description":"千问系列代码及编程模型是专门用于编程和代码生成的语言模型，性能出色，效果突出。","primaryCapability":"TG","capabilities":["TG"],"providers":["qwen-domain-model"],"itemCount":1,"items":[{"model":"qwen-coder-plus","name":"Qwen-Coder-Plus","contextWindow":131072,"capabilities":["TG"]}],"detailPath":"groups/qwen-coder-plus.json","maxContextWindow":131072} {"slug":"qwen-coder-turbo","name":"Qwen-Coder-Turbo","description":"Qwen-Coder-Turbo模型是专门用于编程和代码生成的语言模型，推理速度快，成本低。","primaryCapability":"TG","capabilities":["TG"],"providers":["qwen-domain-model"],"itemCount":1,"items":[{"model":"qwen-coder-turbo","name":"Qwen-Coder-Turbo","contextWindow":131072,"capabilities":["TG"]}],"detailPath":"groups/qwen-coder-turbo.json","maxContextWindow":131072} {"slug":"qwen-deep-research","name":"qwen-deep-research","description":"千问深入研究是一款面向复杂研究任务的高级智能体系统，具备多轮推理与全局规划能力，能够运用互联网搜索等多种工具，对任务进行精细化拆解，开展推理与分析，最终为用户生成可溯源、逻辑严谨的研究型报告。","primaryCapability":"TG","capabilities":["TG"],"providers":["qwen-domain-model"],"itemCount":1,"items":[{"model":"qwen-deep-research","name":"qwen-deep-research","contextWindow":1000000,"capabilities":["TG"]}],"detailPath":"groups/qwen-deep-research.json","maxContextWindow":1000000} {"slug":"qwen-doc-turbo","name":"Qwen-Doc-Turbo","description":"快速对文档进行精准信息抽取，打标分类，内容审核及摘要总结。","primaryCapability":"TG","capabilities":["TG"],"providers":["qwen-domain-model"],"itemCount":1,"items":[{"model":"qwen-doc-turbo","name":"Qwen-Doc-Turbo","contextWindow":262144,"capabilities":["TG"]}],"detailPath":"groups/qwen-doc-turbo.json","maxContextWindow":262144} -{"slug":"qwen-embedding","name":"Qwen-Embedding","description":"基于Qwen模型基座训练的多语言文本统一向量模型，文本检索、聚类、分类性能大幅提升，多语言支持，适用于向量检索、向量化等等场景，可搭配检索增强、文档处理场景使用，支持64~2048维用户自定义向量维度。","primaryCapability":"TR","capabilities":["TR"],"providers":["qwen-domain-model"],"itemCount":6,"items":[{"model":"text-embedding-async-v1","name":"通用文本向量-async-v1","capabilities":["TR"]},{"model":"text-embedding-async-v2","name":"通用文本向量-async-v2","capabilities":["TR"]},{"model":"text-embedding-v1","name":"通用文本向量-v1","capabilities":["TR"]},{"model":"text-embedding-v2","name":"通用文本向量-v2","capabilities":["TR"]},{"model":"text-embedding-v3","name":"通用文本向量-v3","capabilities":["TR"]},{"model":"text-embedding-v4","name":"通用文本向量-v4","capabilities":["TR"]}],"detailPath":"groups/qwen-embedding.json"} +{"slug":"qwen-embedding","name":"Qwen-Embedding","description":"基于Qwen模型基座训练的多语言文本统一向量模型，文本检索、聚类、分类性能大幅提升，多语言支持，适用于向量检索、向量化等等场景，可搭配检索增强、文档处理场景使用，支持64~2048维用户自定义向量维度。","primaryCapability":"TR","capabilities":["TR"],"providers":["qwen","qwen-domain-model"],"itemCount":7,"items":[{"model":"qwen3.7-text-embedding","name":"Qwen3.7-Text-Embedding","contextWindow":131072,"capabilities":["TR"]},{"model":"text-embedding-async-v1","name":"通用文本向量-async-v1","capabilities":["TR"]},{"model":"text-embedding-async-v2","name":"通用文本向量-async-v2","capabilities":["TR"]},{"model":"text-embedding-v1","name":"通用文本向量-v1","capabilities":["TR"]},{"model":"text-embedding-v2","name":"通用文本向量-v2","capabilities":["TR"]},{"model":"text-embedding-v3","name":"通用文本向量-v3","capabilities":["TR"]},{"model":"text-embedding-v4","name":"通用文本向量-v4","capabilities":["TR"]}],"detailPath":"groups/qwen-embedding.json","maxContextWindow":131072} {"slug":"qwen-flash-character","name":"Qwen-Flash-Character","description":"千问系列多语言角色扮演模型，本模型是动态更新版本，模型更新会提前通知，适合拟人化的角色扮演，同时优化了限定人设指令遵循、话题推进、倾听共情等能力，支持个性化角色的深度还原。","primaryCapability":"TG","capabilities":["TG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-flash-character","name":"Qwen-Flash-Character","contextWindow":8192,"capabilities":["TG"]}],"detailPath":"groups/qwen-flash-character.json","maxContextWindow":8192} {"slug":"qwen-flash","name":"Qwen-Flash","description":"Qwen3系列Flash模型，实现思考模式和非思考模式的有效融合，可在对话中切换模式。复杂推理类任务性能优秀，指令遵循、文本理解等能力显著提高。支持1M上下文长度，按照上下文长度进行阶梯计费。","primaryCapability":"Reasoning","capabilities":["Reasoning","TG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-flash","name":"Qwen-Flash","contextWindow":1000000,"capabilities":["Reasoning","TG"]}],"detailPath":"groups/qwen-flash.json","maxContextWindow":1000000} {"slug":"qwen-image-2.0-pro","name":"Qwen-Image-2.0-Pro","description":"Qwen-Image-2.0系列满血版模型，实现了图片生成和图片编辑的融合；具备更专业的文字渲染1k token指令支持能力、更细腻的真实质感，细腻刻画写实场景、更强的语义遵循能力。满血版具备2.0系列最强的文字渲染能力和真实质感。","primaryCapability":"IG","capabilities":["IG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-image-2.0-pro","name":"Qwen-Image-2.0-Pro","capabilities":["IG"]}],"detailPath":"groups/qwen-image-2.0-pro.json"} {"slug":"qwen-image-2.0","name":"Qwen-Image-2.0","description":"Qwen-Image-2.0系列加速版模型，实现了图片生成和图片编辑的融合；具备更专业的文字渲染1k token指令支持能力、更细腻的真实质感，细腻刻画写实场景、更强的语义遵循能力。加速版有效实现了模型效果和性能的最佳平衡。","primaryCapability":"IG","capabilities":["IG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-image-2.0","name":"Qwen-Image-2.0","capabilities":["IG"]}],"detailPath":"groups/qwen-image-2.0.json"} +{"slug":"qwen-image-3.0-pro","name":"Qwen-Image-3.0-Pro","description":"内容丰实：支持最大 4.5k token 输入，支持图中图密集信息排版，让报纸、分镜、菜单、试卷等复杂版面一次生成。\n细节真实：支持 10px 小字精准渲染，微表情、毛孔、发丝等细节生动还原，逼近真实摄影的质感。\n知识厚实：支持 12 国语言、20+ 字体原生渲染，主流网页、游戏、直播等界面仿真，外部知识全纳入。\nQwen-Image-3.0-Pro 不只是在追求\"好看\"，更在追求**“好用”**——让图像生成真正成为可落地的生产力工具。","primaryCapability":"IG","capabilities":["IG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-image-3.0-pro","name":"Qwen-Image-3.0-Pro","capabilities":["IG"]}],"detailPath":"groups/qwen-image-3.0-pro.json"} {"slug":"qwen-image-edit-max","name":"Qwen-Image-Edit-Max","description":"千问图像编辑模型Max系列，提供更稳定、更丰富的编辑能力：提升工业设计与几何推理能力；提升角色一致性；减轻偏移问题；集成Lora能力，可以进行更多功能的图像编辑。此版本为2026年1月16日快照。","primaryCapability":"IG","capabilities":["IG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-image-edit-max","name":"Qwen-Image-Edit-Max","capabilities":["IG"]}],"detailPath":"groups/qwen-image-edit-max.json"} {"slug":"qwen-image-edit","name":"Qwen-Image-Edit-Plus","description":"千问系列图像编辑Plus模型，在首版Edit模型基础上进一步优化了推理性能与系统稳定性，大幅缩短图像生成与编辑的响应时间；支持单次请求返回多张图片，显著提升用户体验。","primaryCapability":"IG","capabilities":["IG"],"providers":["qwen"],"itemCount":2,"items":[{"model":"qwen-image-edit","name":"Qwen-Image-Edit","capabilities":["IG"]},{"model":"qwen-image-edit-plus","name":"Qwen-Image-Edit-Plus","capabilities":["IG"]}],"detailPath":"groups/qwen-image-edit.json"} {"slug":"qwen-image-max","name":"Qwen-Image-Max","description":"千问图像生成模型Max系列，在各类生成任务中表现出色，相较Plus系列大幅度降低生成图片的AI感，提升图像真实性；具备更真实的人物质感、更细腻的自然纹理、更美观的文字渲染。","primaryCapability":"IG","capabilities":["IG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen-image-max","name":"Qwen-Image-Max","capabilities":["IG"]}],"detailPath":"groups/qwen-image-max.json"} @@ -130,6 +131,7 @@ {"slug":"qwen3.6-max","name":"Qwen3.6-Max","description":"Qwen3.6原生Max模型，相较于此前发布的Qwen3-Max和Qwen3.6-Plus，本模型在vibe coding能力上进一步提升、coding agent执行更加高效、前端编程开发能力显著提升；长尾知识能力进一步升级。","primaryCapability":"Reasoning","capabilities":["Reasoning","TG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen3.6-max-preview","name":"Qwen3.6-Max-Preview","contextWindow":262144,"capabilities":["Reasoning","TG"]}],"detailPath":"groups/qwen3.6-max.json","maxContextWindow":262144} {"slug":"qwen3.6-plus","name":"Qwen3.6-Plus","description":"Qwen3.6原生视觉语言系列Plus模型，展现出与当前顶尖前沿模型相媲美的卓越性能，模型效果相较3.5系列显著提升。模型在Agentic coding、前端编程、Vibe coding等代码能力、多模态万物识别、OCR、物体定位等能力上显著增强。","primaryCapability":"Reasoning","capabilities":["Reasoning","VU","TG"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen3.6-plus","name":"Qwen3.6-Plus","contextWindow":1000000,"capabilities":["Reasoning","VU","TG"]}],"detailPath":"groups/qwen3.6-plus.json","maxContextWindow":1000000} {"slug":"qwen3.6","name":"Qwen3.6开源模型","description":"Qwen3.6系列开源模型，基于混合架构设计的原生视觉语言模型，模型效果相较于3.5系列同尺寸有大幅提升。","primaryCapability":"VU","capabilities":["VU","TG","Reasoning"],"providers":["qwen"],"itemCount":2,"items":[{"model":"qwen3.6-27b","name":"Qwen3.6-27B","contextWindow":262144,"capabilities":["Reasoning","VU","TG"]},{"model":"qwen3.6-35b-a3b","name":"Qwen3.6-35B-A3B","contextWindow":262144,"capabilities":["VU","TG","Reasoning"]}],"detailPath":"groups/qwen3.6.json","maxContextWindow":262144} +{"slug":"qwen3.7-flash","name":"Qwen3.7-Flash","description":"Qwen3.7原生视觉语言系列Flash模型，相较3.6-Flash全面提升多模态理解与Agent执行能力。重点强化多模态基础能力、万物识别能力更强，真实世界感知与空间智能进一步提升，Search Agent、CI Agent等多模态Agent场景能力显著升级、端到端任务执行更稳定，多模态Coding能力优化、vibe coding 体验更加流畅。","primaryCapability":"TG","capabilities":["TG","VU","Reasoning"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen3.7-flash","name":"Qwen3.7-Flash","contextWindow":1000000,"capabilities":["TG","VU","Reasoning"]}],"detailPath":"groups/qwen3.7-flash.json","maxContextWindow":1000000} {"slug":"qwen3.7-max","name":"Qwen3.7-Max","description":"Qwen3.7系列中规模最大、综合能力最强的Max模型，当前开放纯文本模型能力供体验。Qwen3.7是面向智能体时代的新一代旗舰模型，核心优势在于智能体能力的广度与深度：在编程、办公与生产力、长周期自主执行方面均能出色胜任各项任务。","primaryCapability":"Reasoning","capabilities":["Reasoning","TG"],"providers":["qwen"],"itemCount":2,"items":[{"model":"qwen3.7-max","name":"Qwen3.7-Max","contextWindow":1000000,"capabilities":["Reasoning","TG"]},{"model":"qwen3.7-max-preview","name":"Qwen3.7-Max-Preview","contextWindow":1000000,"capabilities":["TG","Reasoning"]}],"detailPath":"groups/qwen3.7-max.json","maxContextWindow":1000000} {"slug":"qwen3.7-plus","name":"Qwen3.7-Plus","description":"Qwen3.7系列中高性价比Plus模型，在强大文本能力的基础上全面升级了视觉-语言能力，同时保持了在编码、工具使用和生产力工作流方面的完整智能体能力。其核心特色为多模态交互混合智能体能力，能够感知真实世界场景、读取屏幕并操作 GUI、基于视觉参考生成代码、端到端导航移动应用。","primaryCapability":"TG","capabilities":["TG","Reasoning","VU"],"providers":["qwen"],"itemCount":1,"items":[{"model":"qwen3.7-plus","name":"Qwen3.7-Plus","contextWindow":1000000,"capabilities":["TG","Reasoning","VU"]}],"detailPath":"groups/qwen3.7-plus.json","maxContextWindow":1000000} {"slug":"qwen3","name":"Qwen3开源模型","description":"Qwen3系列开源模型，包含混合模型、思考模型与非思考模型，思考能力与通用能力均达到同规模业界SOTA水平。","primaryCapability":"VU","capabilities":["VU","Reasoning","TG"],"providers":["qwen"],"itemCount":20,"items":[{"model":"qwen3-14b","name":"Qwen3-14B","contextWindow":131072,"capabilities":["Reasoning","TG"]},{"model":"qwen3-235b-a22b","name":"Qwen3-235B-A22B","contextWindow":131072,"capabilities":["Reasoning","TG"]},{"model":"qwen3-235b-a22b-instruct-2507","name":"Qwen3-235B-A22B-Instruct-2507","contextWindow":131072,"capabilities":["TG"]},{"model":"qwen3-235b-a22b-thinking-2507","name":"Qwen3-235B-A22B-Thinking-2507","contextWindow":131072,"capabilities":["Reasoning","TG"]},{"model":"qwen3-30b-a3b","name":"Qwen3-30B-A3B","contextWindow":131072,"capabilities":["Reasoning","TG"]},{"model":"qwen3-30b-a3b-instruct-2507","name":"Qwen3-30B-A3B-Instruct-2507","contextWindow":131072,"capabilities":["TG"]},{"model":"qwen3-30b-a3b-thinking-2507","name":"Qwen3-30B-A3B-Thinking-2507","contextWindow":81920,"capabilities":["Reasoning","TG"]},{"model":"qwen3-32b","name":"Qwen3-32B","contextWindow":131072,"capabilities":["Reasoning","TG"]},{"model":"qwen3-8b","name":"Qwen3-8B","contextWindow":131072,"capabilities":["Reasoning","TG"]},{"model":"qwen3-coder-next","name":"通义千问3-Coder-Next","contextWindow":262144,"capabilities":["TG"]},{"model":"qwen3-next-80b-a3b-instruct","name":"Qwen3-Next-80B-A3B-Instruct","contextWindow":131072,"capabilities":["TG"]},{"model":"qwen3-next-80b-a3b-thinking","name":"Qwen3-Next-80B-A3B-Thinking","contextWindow":131072,"capabilities":["Reasoning","TG"]},{"model":"qwen3-vl-235b-a22b-instruct","name":"Qwen3-VL-235B-A22B-Instruct","contextWindow":131072,"capabilities":["VU"]},{"model":"qwen3-vl-235b-a22b-thinking","name":"Qwen3-VL-235B-A22B-Thinking","contextWindow":131072,"capabilities":["VU","Reasoning"]},{"model":"qwen3-vl-30b-a3b-instruct","name":"Qwen3-VL-30B-A3B-Instruct","contextWindow":131072,"capabilities":["VU"]},{"model":"qwen3-vl-30b-a3b-thinking","name":"Qwen3-VL-30B-A3B-Thinking","contextWindow":131072,"capabilities":["VU","Reasoning"]},{"model":"qwen3-vl-32b-instruct","name":"Qwen3-VL-32B-Instruct","contextWindow":131072,"capabilities":["VU"]},{"model":"qwen3-vl-32b-thinking","name":"Qwen3-VL-32B-Thinking","contextWindow":131072,"capabilities":["VU","Reasoning"]},{"model":"qwen3-vl-8b-instruct","name":"Qwen3-VL-8B-Instruct","contextWindow":131072,"capabilities":["VU"]},{"model":"qwen3-vl-8b-thinking","name":"Qwen3-VL-8B-Thinking","contextWindow":131072,"capabilities":["VU","Reasoning"]}],"detailPath":"groups/qwen3.json","maxContextWindow":262144} diff --git a/skills/bailian-docs-llm-wiki/models/groups/kimi-models-market-place.json b/skills/bailian-docs-llm-wiki/models/groups/kimi-models-market-place.json index 1646ab45..9523e538 100644 --- a/skills/bailian-docs-llm-wiki/models/groups/kimi-models-market-place.json +++ b/skills/bailian-docs-llm-wiki/models/groups/kimi-models-market-place.json @@ -2,6 +2,117 @@ "name": "Kimi", "description": "由月之暗面提供的Kimi系列模型的API服务。", "items": [ + { + "inferenceMetadata": { + "response_modality": [ + "Text" + ], + "request_modality": [ + "Text", + "Image", + "Video" + ] + }, + "description": "Kimi K3 是 Kimi 迄今能力最强的旗舰模型，拥有 2.8 万亿参数，基于 KDA 混合线性注意力机制（Kimi Delta Attention）和注意力残差（Attention Residuals）技术构建，原生支持视觉理解，并拥有 100 万 token 上下文窗口。它是全球首个开源的 3 万亿级别模型，面向长程编程、知识工作和推理等前沿智能场景而设计。", + "features": [ + "function-calling", + "structured-outputs", + "cache", + "prefix-completion" + ], + "provider": "moonshot-ai", + "model": "kimi/kimi-k3", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "20", + "type": "input_token", + "priceName": "输入" + }, + { + "priceUnit": "每百万tokens", + "price": "100", + "type": "output_token", + "priceName": "输出" + }, + { + "priceUnit": "每百万tokens", + "price": "2", + "type": "input_token_cache", + "priceName": "输入（缓存命中）" + } + ], + "qpmInfo": { + "model-default-actual": { + "count_limit_period": 60, + "usage_limit": 3000000, + "usage_limit_field": "total_tokens", + "count_limit": 500, + "usage_limit_period": 60, + "type": "model-default" + }, + "model-default": { + "count_limit_period": 60, + "usage_limit": 3000000, + "usage_limit_field": "total_tokens", + "count_limit": 500, + "usage_limit_period": 60, + "type": "model-default" + } + }, + "capabilities": [ + "TG", + "VU", + "Reasoning" + ], + "modelAlias": "", + "versionTag": "MAJOR", + "maxOutputTokens": 1048576, + "latestOnlineAt": "2026-07-17T08:23:12.000+00:00", + "contextWindow": 1048576, + "maxInputTokens": 1048576, + "inferenceProvider": "moonshot-ai", + "name": "kimi/kimi-k3", + "docUrl": "https://bailian.console.aliyun.com/cn-beijing?tab=doc#/doc/?type=model&url=3021620", + "category": "Third-party", + "predictConfig": [ + { + "name": "system", + "key": "systemMessage", + "tip": "系统人设，例如“你是一个AI助手”。" + }, + { + "name": "top_p", + "key": "top_p", + "default": 0.8, + "tip": "控制核采样方法的概率阈值，取值越大，生成的随机性越高。", + "range": [ + 0.0001, + 1 + ] + }, + { + "name": "temperature", + "key": "temperature", + "default": 0.7, + "tip": "控制生成随机性和多样性，数值越高多样性越强，数值越低一致性越强，范围(0,2)。建议该参数和top_p只设置1个。", + "range": [ + 0, + 1.9999 + ] + } + ], + "samples": { + "openai": { + "completionsAPI": { + "curl": "curl --location 'https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n \"model\": \"kimi/kimi-k3\",\n \"messages\": [\n {\n \"role\": \"user\",\n \"content\": [\n {\"type\": \"image_url\", \"image_url\": {\"url\": \"https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241108/ctdzex/biaozhun.jpg\"}},\n {\"type\": \"text\", \"text\": \"请仅输出图像中的文本内容。\"}\n ]\n }]\n}'", + "python": "import os\nfrom openai import OpenAI\n\nclient = OpenAI(\n api_key=os.getenv(\"DASHSCOPE_API_KEY\"),\n base_url=\"https://dashscope.aliyuncs.com/compatible-mode/v1\",\n)\n\ncompletion = client.chat.completions.create(\n model=\"kimi/kimi-k3\",\n messages=[\n {\n \"role\": \"user\",\n \"content\": [\n {\n \"type\": \"image_url\",\n \"image_url\": {\n \"url\": \"https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241108/ctdzex/biaozhun.jpg\"\n },\n },\n {\"type\": \"text\", \"text\": \"请仅输出图像中的文本内容。\"},\n ],\n },\n ],\n)\nprint(completion.choices[0].message.content)", + "nodejs": "import OpenAI from \"openai\";\n\nconst openai = new OpenAI({\n // 若没有配置环境变量，请用百炼API Key将下行替换为：apiKey: \"sk-xxx\"\n apiKey: process.env.DASHSCOPE_API_KEY,\n baseURL: \"https://dashscope.aliyuncs.com/compatible-mode/v1\"\n});\n\nasync function main() {\n const response = await openai.chat.completions.create({\n model: \"kimi/kimi-k3\",\n messages: [\n {\n role: \"user\",\n content: [{\n type: \"image_url\",\n image_url: {\n \"url\": \"https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241108/ctdzex/biaozhun.jpg\"\n }\n },\n {\n type: \"text\",\n text: \"请仅输出图像中的文本内容。\"\n }\n ]\n }\n ]\n });\n console.log(response.choices[0].message.content);\n}\nmain()", + "docUrl": "https://help.aliyun.com/document_detail/3021620.html" + } + } + } + }, { "inferenceMetadata": { "response_modality": [ @@ -21,10 +132,27 @@ "cache" ], "provider": "moonshot-ai", - "limit": { - "message": "model not exist" - }, "model": "kimi/kimi-k2.7-code-highspeed", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "13", + "type": "input_token", + "priceName": "输入" + }, + { + "priceUnit": "每百万tokens", + "price": "54", + "type": "output_token", + "priceName": "输出" + }, + { + "priceUnit": "每百万tokens", + "price": "2.6", + "type": "input_token_cache", + "priceName": "输入（缓存命中）" + } + ], "qpmInfo": { "model-default-actual": { "count_limit_period": 60, @@ -114,10 +242,27 @@ "cache" ], "provider": "moonshot-ai", - "limit": { - "message": "model not exist" - }, "model": "kimi/kimi-k2.7-code", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "6.5", + "type": "input_token", + "priceName": "输入" + }, + { + "priceUnit": "每百万tokens", + "price": "27", + "type": "output_token", + "priceName": "输出" + }, + { + "priceUnit": "每百万tokens", + "price": "1.3", + "type": "input_token_cache", + "priceName": "输入（缓存命中）" + } + ], "qpmInfo": { "model-default-actual": { "count_limit_period": 60, @@ -207,10 +352,27 @@ "cache" ], "provider": "moonshot-ai", - "limit": { - "message": "model not exist" - }, "model": "kimi/kimi-k2.6", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "6.5", + "type": "input_token", + "priceName": "输入" + }, + { + "priceUnit": "每百万tokens", + "price": "27", + "type": "output_token", + "priceName": "输出" + }, + { + "priceUnit": "每百万tokens", + "price": "1.1", + "type": "input_token_cache", + "priceName": "输入（缓存命中）" + } + ], "qpmInfo": { "model-default-actual": { "count_limit_period": 6, @@ -287,11 +449,28 @@ "cache" ], "provider": "moonshot-ai", - "limit": { - "message": "model not exist" - }, "model": "kimi/kimi-k2.5", "iconUrl": "https://img.alicdn.com/imgextra/i4/O1CN01KzHLBW1LISVEUaotl_!!6000000001276-2-tps-56-56.png", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "4", + "type": "input_token", + "priceName": "输入" + }, + { + "priceUnit": "每百万tokens", + "price": "21", + "type": "output_token", + "priceName": "输出" + }, + { + "priceUnit": "每百万tokens", + "price": "0.7", + "type": "input_token_cache", + "priceName": "输入（缓存命中）" + } + ], "qpmInfo": { "model-default-actual": { "count_limit_period": 6, diff --git a/skills/bailian-docs-llm-wiki/models/groups/qwen-audio-realtime-flash.json b/skills/bailian-docs-llm-wiki/models/groups/qwen-audio-realtime-flash.json index 52c4ecf2..f7459176 100644 --- a/skills/bailian-docs-llm-wiki/models/groups/qwen-audio-realtime-flash.json +++ b/skills/bailian-docs-llm-wiki/models/groups/qwen-audio-realtime-flash.json @@ -68,10 +68,10 @@ ], "modelAlias": "", "versionTag": "MAJOR", - "maxOutputTokens": 4096, - "latestOnlineAt": "2026-07-14T06:59:47.013+00:00", - "contextWindow": 8192, - "maxInputTokens": 4096, + "maxOutputTokens": 8192, + "latestOnlineAt": "2026-07-14T06:59:47.000+00:00", + "contextWindow": 40960, + "maxInputTokens": 16384, "inferenceProvider": "aliyun-bailian", "name": "千问实时语音对话大模型3.0（极速版）", "docUrl": "https://help.aliyun.com/document_detail/3041584.html", diff --git a/skills/bailian-docs-llm-wiki/models/groups/qwen-audio-realtime-plus.json b/skills/bailian-docs-llm-wiki/models/groups/qwen-audio-realtime-plus.json index e02c2e41..e0e67a7c 100644 --- a/skills/bailian-docs-llm-wiki/models/groups/qwen-audio-realtime-plus.json +++ b/skills/bailian-docs-llm-wiki/models/groups/qwen-audio-realtime-plus.json @@ -68,10 +68,10 @@ ], "modelAlias": "", "versionTag": "MAJOR", - "maxOutputTokens": 4096, - "latestOnlineAt": "2026-07-14T06:59:43.526+00:00", - "contextWindow": 8192, - "maxInputTokens": 4096, + "maxOutputTokens": 8192, + "latestOnlineAt": "2026-07-14T06:59:43.000+00:00", + "contextWindow": 40960, + "maxInputTokens": 16384, "inferenceProvider": "aliyun-bailian", "name": "千问实时语音大模型 （标准版）", "docUrl": "https://help.aliyun.com/document_detail/3041584.html", diff --git a/skills/bailian-docs-llm-wiki/models/groups/qwen-embedding.json b/skills/bailian-docs-llm-wiki/models/groups/qwen-embedding.json index f4d5508c..0d655f22 100644 --- a/skills/bailian-docs-llm-wiki/models/groups/qwen-embedding.json +++ b/skills/bailian-docs-llm-wiki/models/groups/qwen-embedding.json @@ -2,6 +2,75 @@ "name": "Qwen-Embedding", "description": "基于Qwen模型基座训练的多语言文本统一向量模型，文本检索、聚类、分类性能大幅提升，多语言支持，适用于向量检索、向量化等等场景，可搭配检索增强、文档处理场景使用，支持64~2048维用户自定义向量维度。", "items": [ + { + "inferenceMetadata": { + "response_modality": [ + "Text" + ], + "request_modality": [ + "Text" + ] + }, + "description": "是通义实验室基于Qwen3.7训练的多语言文本统一向量模型，相较text-embedding-v4版本在文本检索、聚类、分类性能大幅提升；在MTEB多语言、中英、Code检索等评测任务上效果提升20%；支持256~2560维用户自定义向量维度。", + "collectionTag": "qwen3.7", + "features": [ + "model-experience" + ], + "provider": "qwen", + "model": "qwen3.7-text-embedding", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "0.5", + "type": "embedding_token", + "priceName": "文本输入" + } + ], + "qpmInfo": { + "model-default-actual": { + "count_limit_period": 6, + "usage_limit": 100000, + "usage_limit_field": "total_tokens", + "count_limit": 2400, + "usage_limit_period": 6, + "type": "model-default" + }, + "model-default": { + "count_limit_period": 6, + "usage_limit": 100000, + "usage_limit_field": "total_tokens", + "count_limit": 2400, + "usage_limit_period": 6, + "type": "model-default" + } + }, + "capabilities": [ + "TR" + ], + "modelAlias": "", + "versionTag": "MAJOR", + "latestOnlineAt": "2026-07-15T02:24:33.000+00:00", + "contextWindow": 131072, + "maxInputTokens": 131072, + "inferenceProvider": "aliyun-bailian", + "name": "Qwen3.7-Text-Embedding", + "docUrl": "https://help.aliyun.com/document_detail/2842587.html", + "category": "Embeddings", + "predictConfig": [ + { + "name": "topK" + } + ], + "samples": { + "dashscope": { + "default": { + "curl": "curl --location 'https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"qwen3.7-text-embedding\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", + "python": "import dashscope\nfrom http import HTTPStatus\ninput_texts = \"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"\n\nresp = dashscope.TextEmbedding.call(\nmodel=\"qwen3.7-text-embedding\",\ninput=input_texts\n)\nprint(resp)", + "java": "import java.util.Arrays;\nimport java.util.concurrent.Semaphore;\nimport com.alibaba.dashscope.common.ResultCallback;\nimport com.alibaba.dashscope.embeddings.TextEmbedding;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingParam;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingResult;\nimport com.alibaba.dashscope.exception.ApiException;\nimport com.alibaba.dashscope.exception.NoApiKeyException;\nimport com.alibaba.dashscope.utils.Constants;\n\npublic final class Main {\npublic static void main(String[] args) {\ntry {\nTextEmbeddingParam param = TextEmbeddingParam\n.builder()\n.model(\"qwen3.7-text-embedding\") \n.texts(Arrays.asList(\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\")) \n.build();\nTextEmbedding textEmbedding = new TextEmbedding();\nTextEmbeddingResult result = textEmbedding.call(param);\n\nSystem.out.println(result);\n\n} catch (ApiException | NoApiKeyException e) {\nSystem.out.println(e.getMessage());\n}\n}\n}" + } + } + } + }, { "inferenceMetadata": { "response_modality": [ @@ -16,10 +85,21 @@ "model-experience" ], "provider": "qwen-domain-model", - "limit": { - "message": "model not exist" - }, "model": "text-embedding-v4", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "0.25", + "type": "embedding_token_batch", + "priceName": "向量输入（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.5", + "type": "embedding_token", + "priceName": "文本输入" + } + ], "qpmInfo": { "model-default-actual": { "count_limit_period": 1, @@ -44,7 +124,7 @@ "versionTag": "MAJOR", "latestOnlineAt": "2025-06-05T03:07:20.000+00:00", "offlineInfo": {}, - "inferenceProvider": "bailian", + "inferenceProvider": "aliyun-bailian", "name": "通用文本向量-v4", "docUrl": "https://help.aliyun.com/document_detail/2842587.html", "category": "Embeddings", @@ -56,7 +136,7 @@ "samples": { "dashscope": { "default": { - "curl": "curl --location 'https://ws-nckitja1d28cec5v.cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-v4\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", + "curl": "curl --location 'https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-v4\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", "python": "import dashscope\nfrom http import HTTPStatus\ninput_texts = \"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"\n\nresp = dashscope.TextEmbedding.call(\nmodel=\"text-embedding-v4\",\ninput=input_texts\n)\nprint(resp)", "java": "import java.util.Arrays;\nimport java.util.concurrent.Semaphore;\nimport com.alibaba.dashscope.common.ResultCallback;\nimport com.alibaba.dashscope.embeddings.TextEmbedding;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingParam;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingResult;\nimport com.alibaba.dashscope.exception.ApiException;\nimport com.alibaba.dashscope.exception.NoApiKeyException;\nimport com.alibaba.dashscope.utils.Constants;\n\npublic final class Main {\npublic static void main(String[] args) {\ntry {\nTextEmbeddingParam param = TextEmbeddingParam\n.builder()\n.model(\"text-embedding-v4\") \n.texts(Arrays.asList(\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\")) \n.build();\nTextEmbedding textEmbedding = new TextEmbedding();\nTextEmbeddingResult result = textEmbedding.call(param);\n\nSystem.out.println(result);\n\n} catch (ApiException | NoApiKeyException e) {\nSystem.out.println(e.getMessage());\n}\n}\n}" } @@ -73,10 +153,21 @@ "description": "通用文本向量，是通义实验室基于LLM底座的多语言文本统一向量模型，面向全球多个主流语种，提供高水准的向量服务，帮助开发者将文本数据快速转换为高质量的向量数据。", "features": [], "provider": "qwen-domain-model", - "limit": { - "message": "model not exist" - }, "model": "text-embedding-v3", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "0.25", + "type": "embedding_token_batch", + "priceName": "向量输入（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.5", + "type": "embedding_token", + "priceName": "文本输入" + } + ], "qpmInfo": { "model-default-actual": { "count_limit_period": 1, @@ -100,7 +191,7 @@ ], "versionTag": "MAJOR", "latestOnlineAt": "2024-07-12T09:44:51.000+00:00", - "inferenceProvider": "bailian", + "inferenceProvider": "aliyun-bailian", "name": "通用文本向量-v3", "docUrl": "https://help.aliyun.com/document_detail/2712515.html", "predictConfig": [ @@ -133,7 +224,7 @@ "samples": { "dashscope": { "default": { - "curl": "curl --location 'https://ws-nckitja1d28cec5v.cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-v3\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", + "curl": "curl --location 'https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-v3\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", "python": "import dashscope\nfrom http import HTTPStatus\ninput_texts = \"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"\n\nresp = dashscope.TextEmbedding.call(\nmodel=\"text-embedding-v3\",\ninput=input_texts\n)\nprint(resp)", "java": "import java.util.Arrays;\nimport java.util.concurrent.Semaphore;\nimport com.alibaba.dashscope.common.ResultCallback;\nimport com.alibaba.dashscope.embeddings.TextEmbedding;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingParam;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingResult;\nimport com.alibaba.dashscope.exception.ApiException;\nimport com.alibaba.dashscope.exception.NoApiKeyException;\nimport com.alibaba.dashscope.utils.Constants;\n\npublic final class Main {\npublic static void main(String[] args) {\ntry {\nTextEmbeddingParam param = TextEmbeddingParam\n.builder()\n.model(\"text-embedding-v3\") \n.texts(Arrays.asList(\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\")) \n.build();\nTextEmbedding textEmbedding = new TextEmbedding();\nTextEmbeddingResult result = textEmbedding.call(param);\n\nSystem.out.println(result);\n\n} catch (ApiException | NoApiKeyException e) {\nSystem.out.println(e.getMessage());\n}\n}\n}" } @@ -150,10 +241,21 @@ "description": "通用文本向量，是通义实验室基于LLM底座的多语言文本统一向量模型，面向全球多个主流语种，提供高水准的向量服务，帮助开发者将文本数据快速转换为高质量的向量数据。", "features": [], "provider": "qwen-domain-model", - "limit": { - "message": "model not exist" - }, "model": "text-embedding-v2", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "0.35", + "type": "embedding_token_batch", + "priceName": "向量输入（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.7", + "type": "embedding_token", + "priceName": "文本输入" + } + ], "qpmInfo": { "model-default-actual": { "count_limit_period": 1, @@ -177,7 +279,7 @@ ], "versionTag": "MAJOR", "latestOnlineAt": "2024-04-09T09:03:19.000+00:00", - "inferenceProvider": "bailian", + "inferenceProvider": "aliyun-bailian", "name": "通用文本向量-v2", "docUrl": "https://help.aliyun.com/document_detail/2712515.html", "predictConfig": [ @@ -210,7 +312,7 @@ "samples": { "dashscope": { "default": { - "curl": "curl --location 'https://ws-nckitja1d28cec5v.cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-v2\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", + "curl": "curl --location 'https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-v2\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", "python": "import dashscope\nfrom http import HTTPStatus\ninput_texts = \"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"\n\nresp = dashscope.TextEmbedding.call(\nmodel=\"text-embedding-v2\",\ninput=input_texts\n)\nprint(resp)", "java": "import java.util.Arrays;\nimport java.util.concurrent.Semaphore;\nimport com.alibaba.dashscope.common.ResultCallback;\nimport com.alibaba.dashscope.embeddings.TextEmbedding;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingParam;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingResult;\nimport com.alibaba.dashscope.exception.ApiException;\nimport com.alibaba.dashscope.exception.NoApiKeyException;\nimport com.alibaba.dashscope.utils.Constants;\n\npublic final class Main {\npublic static void main(String[] args) {\ntry {\nTextEmbeddingParam param = TextEmbeddingParam\n.builder()\n.model(\"text-embedding-v2\") \n.texts(Arrays.asList(\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\")) \n.build();\nTextEmbedding textEmbedding = new TextEmbedding();\nTextEmbeddingResult result = textEmbedding.call(param);\n\nSystem.out.println(result);\n\n} catch (ApiException | NoApiKeyException e) {\nSystem.out.println(e.getMessage());\n}\n}\n}" } @@ -227,10 +329,21 @@ "description": "通用文本向量，是通义实验室基于LLM底座的多语言文本统一向量模型，面向全球多个主流语种，提供高水准的向量服务，帮助开发者将文本数据快速转换为高质量的向量数据。", "features": [], "provider": "qwen-domain-model", - "limit": { - "message": "model not exist" - }, "model": "text-embedding-v1", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "0.35", + "type": "embedding_token_batch", + "priceName": "向量输入（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.7", + "type": "embedding_token", + "priceName": "文本输入" + } + ], "qpmInfo": { "model-default-actual": { "count_limit_period": 1, @@ -254,7 +367,7 @@ ], "versionTag": "MAJOR", "latestOnlineAt": "2024-04-09T09:02:12.000+00:00", - "inferenceProvider": "bailian", + "inferenceProvider": "aliyun-bailian", "name": "通用文本向量-v1", "docUrl": "https://help.aliyun.com/document_detail/2712515.html", "predictConfig": [ @@ -287,7 +400,7 @@ "samples": { "dashscope": { "default": { - "curl": "curl --location 'https://ws-nckitja1d28cec5v.cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-v1\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", + "curl": "curl --location 'https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-v1\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", "python": "import dashscope\nfrom http import HTTPStatus\ninput_texts = \"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"\n\nresp = dashscope.TextEmbedding.call(\nmodel=\"text-embedding-v1\",\ninput=input_texts\n)\nprint(resp)", "java": "import java.util.Arrays;\nimport java.util.concurrent.Semaphore;\nimport com.alibaba.dashscope.common.ResultCallback;\nimport com.alibaba.dashscope.embeddings.TextEmbedding;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingParam;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingResult;\nimport com.alibaba.dashscope.exception.ApiException;\nimport com.alibaba.dashscope.exception.NoApiKeyException;\nimport com.alibaba.dashscope.utils.Constants;\n\npublic final class Main {\npublic static void main(String[] args) {\ntry {\nTextEmbeddingParam param = TextEmbeddingParam\n.builder()\n.model(\"text-embedding-v1\") \n.texts(Arrays.asList(\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\")) \n.build();\nTextEmbedding textEmbedding = new TextEmbedding();\nTextEmbeddingResult result = textEmbedding.call(param);\n\nSystem.out.println(result);\n\n} catch (ApiException | NoApiKeyException e) {\nSystem.out.println(e.getMessage());\n}\n}\n}" } @@ -304,16 +417,21 @@ "description": "通用文本向量的批处理接口，通过这个接口客户可以以文本方式一次性的提交大批量的向量计算请求，在系统完成所有的计算之后，大模型服务平台会将结果信息存储在结果文件中供客户下载解析。", "features": [], "provider": "qwen-domain-model", - "limit": { - "message": "model not exist" - }, "model": "text-embedding-async-v2", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "0.7", + "type": "embedding_token", + "priceName": "文本输入" + } + ], "capabilities": [ "TR" ], "versionTag": "MAJOR", "latestOnlineAt": "2024-04-09T09:05:28.000+00:00", - "inferenceProvider": "bailian", + "inferenceProvider": "aliyun-bailian", "name": "通用文本向量-async-v2", "docUrl": "https://help.aliyun.com/document_detail/2712516.html", "predictConfig": [ @@ -346,7 +464,7 @@ "samples": { "dashscope": { "default": { - "curl": "curl --location 'https://ws-nckitja1d28cec5v.cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-async-v2\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", + "curl": "curl --location 'https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-async-v2\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", "python": "import dashscope\nfrom http import HTTPStatus\ninput_texts = \"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"\n\nresp = dashscope.TextEmbedding.call(\nmodel=\"text-embedding-async-v2\",\ninput=input_texts\n)\nprint(resp)", "java": "import java.util.Arrays;\nimport java.util.concurrent.Semaphore;\nimport com.alibaba.dashscope.common.ResultCallback;\nimport com.alibaba.dashscope.embeddings.TextEmbedding;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingParam;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingResult;\nimport com.alibaba.dashscope.exception.ApiException;\nimport com.alibaba.dashscope.exception.NoApiKeyException;\nimport com.alibaba.dashscope.utils.Constants;\n\npublic final class Main {\npublic static void main(String[] args) {\ntry {\nTextEmbeddingParam param = TextEmbeddingParam\n.builder()\n.model(\"text-embedding-async-v2\") \n.texts(Arrays.asList(\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\")) \n.build();\nTextEmbedding textEmbedding = new TextEmbedding();\nTextEmbeddingResult result = textEmbedding.call(param);\n\nSystem.out.println(result);\n\n} catch (ApiException | NoApiKeyException e) {\nSystem.out.println(e.getMessage());\n}\n}\n}" } @@ -363,16 +481,21 @@ "description": "通用文本向量的批处理接口，通过这个接口客户可以以文本方式一次性的提交大批量的向量计算请求，在系统完成所有的计算之后，大模型服务平台会将结果信息存储在结果文件中供客户下载解析。", "features": [], "provider": "qwen-domain-model", - "limit": { - "message": "model not exist" - }, "model": "text-embedding-async-v1", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "0.7", + "type": "embedding_token", + "priceName": "文本输入" + } + ], "capabilities": [ "TR" ], "versionTag": "MAJOR", "latestOnlineAt": "2024-04-09T09:04:41.000+00:00", - "inferenceProvider": "bailian", + "inferenceProvider": "aliyun-bailian", "name": "通用文本向量-async-v1", "docUrl": "https://help.aliyun.com/document_detail/2712516.html", "predictConfig": [ @@ -405,7 +528,7 @@ "samples": { "dashscope": { "default": { - "curl": "curl --location 'https://ws-nckitja1d28cec5v.cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-async-v1\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", + "curl": "curl --location 'https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--header 'Content-Type: application/json' \\\n--data '{\n\"model\": \"text-embedding-async-v1\",\n\"input\": {\n\"texts\":[\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"] \n}'", "python": "import dashscope\nfrom http import HTTPStatus\ninput_texts = \"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\"\n\nresp = dashscope.TextEmbedding.call(\nmodel=\"text-embedding-async-v1\",\ninput=input_texts\n)\nprint(resp)", "java": "import java.util.Arrays;\nimport java.util.concurrent.Semaphore;\nimport com.alibaba.dashscope.common.ResultCallback;\nimport com.alibaba.dashscope.embeddings.TextEmbedding;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingParam;\nimport com.alibaba.dashscope.embeddings.TextEmbeddingResult;\nimport com.alibaba.dashscope.exception.ApiException;\nimport com.alibaba.dashscope.exception.NoApiKeyException;\nimport com.alibaba.dashscope.utils.Constants;\n\npublic final class Main {\npublic static void main(String[] args) {\ntry {\nTextEmbeddingParam param = TextEmbeddingParam\n.builder()\n.model(\"text-embedding-async-v1\") \n.texts(Arrays.asList(\"衣服的质量杠杠的，很漂亮，不枉我等了这么久啊，喜欢，以后还来这里买\")) \n.build();\nTextEmbedding textEmbedding = new TextEmbedding();\nTextEmbeddingResult result = textEmbedding.call(param);\n\nSystem.out.println(result);\n\n} catch (ApiException | NoApiKeyException e) {\nSystem.out.println(e.getMessage());\n}\n}\n}" } diff --git a/skills/bailian-docs-llm-wiki/models/groups/qwen-image-3.0-pro.json b/skills/bailian-docs-llm-wiki/models/groups/qwen-image-3.0-pro.json new file mode 100644 index 00000000..e25d65a4 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/models/groups/qwen-image-3.0-pro.json @@ -0,0 +1,78 @@ +{ + "name": "Qwen-Image-3.0-Pro", + "description": "内容丰实：支持最大 4.5k token 输入，支持图中图密集信息排版，让报纸、分镜、菜单、试卷等复杂版面一次生成。\n细节真实：支持 10px 小字精准渲染，微表情、毛孔、发丝等细节生动还原，逼近真实摄影的质感。\n知识厚实：支持 12 国语言、20+ 字体原生渲染，主流网页、游戏、直播等界面仿真，外部知识全纳入。\nQwen-Image-3.0-Pro 不只是在追求\"好看\"，更在追求**“好用”**——让图像生成真正成为可落地的生产力工具。", + "items": [ + { + "inferenceMetadata": { + "response_modality": [ + "Image" + ], + "request_modality": [ + "Image", + "Text" + ] + }, + "description": "内容丰实：支持最大 4.5k token 输入，支持图中图密集信息排版，让报纸、分镜、菜单、试卷等复杂版面一次生成。\n细节真实：支持 10px 小字精准渲染，微表情、毛孔、发丝等细节生动还原，逼近真实摄影的质感。\n知识厚实：支持 12 国语言、20+ 字体原生渲染，主流网页、游戏、直播等界面仿真，外部知识全纳入。\nQwen-Image-3.0-Pro 不只是在追求\"好看\"，更在追求**“好用”**——让图像生成真正成为可落地的生产力工具。", + "features": [ + "model-experience" + ], + "provider": "qwen", + "model": "qwen-image-3.0-pro", + "qpmInfo": { + "model-default-actual": { + "count_limit_period": 60, + "count_limit": 1, + "type": "model-default" + }, + "model-default": { + "count_limit_period": 60, + "count_limit": 1, + "type": "model-default" + } + }, + "capabilities": [ + "IG" + ], + "modelAlias": "", + "versionTag": "MAJOR", + "latestOnlineAt": "2026-07-20T14:02:23.000+00:00", + "inferenceProvider": "aliyun-bailian", + "name": "Qwen-Image-3.0-Pro", + "docUrl": "https://help.aliyun.com/document_detail/3047054.html", + "category": "Visual", + "predictConfig": [ + { + "name": "negative_prompt", + "key": "negativePrompt", + "tip": "通过指定用户不想看到的内容来优化模型输出，使模型产生更有针对性和理想的结果。" + }, + { + "name": "size", + "key": "size", + "default": "2048*2048", + "tip": "输出分辨率" + }, + { + "name": "n", + "key": "n", + "default": 1, + "tip": "本次请求生成的图片数量" + }, + { + "name": "智能改写", + "key": "prompt_extend", + "default": false, + "tip": "开启后会使用大模型对输入prompt进行智能改写，仅对正向提示词有效。对于较短的输prompt生成效果提升明显，但会增加3-4秒耗时。" + } + ], + "samples": { + "dashscope": { + "default": { + "curl": "curl --location 'https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \\\n--header 'Content-Type: application/json' \\\n--header \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n--data '{\n \"model\": \"qwen-image-3.0-pro\",\n \"input\": {\n \"messages\": [\n {\n \"role\": \"user\",\n \"content\": [\n {\n \"text\": \"画面是一张竖幅户外人像摄影，整体从上到下呈现温暖的午后街景氛围。顶部左侧到上方大面积被深绿色藤蔓和橙色小花覆盖，花叶从建筑檐口自然垂落，受阳光照射的叶片呈黄绿色高光，阴影处则偏深绿，形成浓密而柔和的背景层次。左上至中上区域是一块深蓝色横向招牌，招牌表面较暗、略带磨砂质感，上面以白色哥特体大字写着 Il Messaggero，文字位于画面左侧偏上，部分被前景花叶轻微遮挡，字体高对比、带装饰性尖角和粗细变化。招牌下方是报刊亭或书报摊的玻璃展示窗，黑色金属框架将橱窗分隔成多个矩形区域，内部陈列着许多报纸、杂志和书刊封面，但大多因景深虚化和光线反射而难以辨读，形成浅色纸张与深色边框交错的背景纹理。画面右上方是强烈的逆光区域，阳光从街道尽头照入，背景建筑被虚化成米灰色块面，边缘柔和，呈现明显的浅景深效果。画面中部偏右是一名年轻成年女性的半身至膝上人像，她回头面向镜头微笑，身体略向右转，肩背朝向观者，姿态自然放松。她有长而浓密的黑色波浪卷发，发丝被逆光勾勒出金色轮廓光，发梢在右侧向外散开，显得轻盈蓬松。她肤色白皙，脸型柔和偏鹅蛋形，眉形细致，眼睛明亮，眼妆清透，睫毛明显，面部带有自然高光，唇部为柔和珊瑚红色，笑容露齿，表情亲切明朗。她佩戴小巧耳饰，身穿黑色细肩带露背连衣裙，面料颜色深黑、轮廓简洁，细肩带从肩部向背部延伸，背部线条清晰。画面下部偏左到中部，她双手抱着一束玫瑰花，花束体积较大，主要由橙色、杏色、粉色和浅桃色玫瑰组成，花瓣层层卷曲，边缘被阳光照亮，绿色叶片和长花茎从花束下方垂出，花束与黑色裙装形成鲜明色彩对比。右侧背景是一条被阳光照亮的城市街道，地面呈暖灰与金黄色调，远处建筑、街边设施和一个模糊的红色圆形交通标志位于右下远景，均因焦外虚化而只保留色块和轮廓。整张照片采用暖色胶片感处理，带有细腻颗粒、柔和对比和明显逆光边缘光，人物位于视觉焦点，背景报刊亭、花藤、街道和阳光共同营造出浪漫、明亮、都市漫步式的氛围。\"\n }\n ]\n }\n ]\n },\n \"parameters\": {\n \"prompt_extend\": true\n }\n}'", + "docUrl": "https://help.aliyun.com/document_detail/3047054.html" + } + } + } + } + ] +} diff --git a/skills/bailian-docs-llm-wiki/models/groups/qwen3.7-flash.json b/skills/bailian-docs-llm-wiki/models/groups/qwen3.7-flash.json new file mode 100644 index 00000000..bc12b033 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/models/groups/qwen3.7-flash.json @@ -0,0 +1,406 @@ +{ + "name": "Qwen3.7-Flash", + "description": "Qwen3.7原生视觉语言系列Flash模型，相较3.6-Flash全面提升多模态理解与Agent执行能力。重点强化多模态基础能力、万物识别能力更强，真实世界感知与空间智能进一步提升，Search Agent、CI Agent等多模态Agent场景能力显著升级、端到端任务执行更稳定，多模态Coding能力优化、vibe coding 体验更加流畅。", + "items": [ + { + "inferenceMetadata": { + "response_modality": [ + "Text" + ], + "request_modality": [ + "Image", + "Text", + "Video" + ] + }, + "builtInToolMultiPrices": [ + { + "supportedApi": "Responses API", + "name": "code_interpreter", + "docUrl": "https://help.aliyun.com/document_detail/2990719.html", + "type": "code_interpreter", + "prices": [ + { + "priceUnit": "", + "price": "0", + "currency": "CNY" + } + ] + }, + { + "supportedApi": "Responses API", + "name": "i2i_search", + "docUrl": "https://help.aliyun.com/document_detail/3022096.html", + "type": "i2i_search", + "prices": [ + { + "priceUnit": "千次调用", + "price": "48", + "currency": "CNY" + } + ] + }, + { + "supportedApi": "Responses API", + "name": "t2i_search", + "docUrl": "https://help.aliyun.com/document_detail/3022091.html", + "type": "t2i_search", + "prices": [ + { + "priceUnit": "千次调用", + "price": "24", + "currency": "CNY" + } + ] + }, + { + "supportedApi": "Responses API", + "name": "web_extractor", + "docUrl": "https://help.aliyun.com/document_detail/3016816.html", + "type": "web_extractor", + "prices": [ + { + "priceUnit": "", + "price": "0", + "currency": "CNY" + } + ] + }, + { + "supportedApi": "Responses API", + "name": "web_search", + "docUrl": "https://help.aliyun.com/document_detail/2867560.html", + "type": "web_search", + "prices": [ + { + "priceUnit": "千次调用", + "price": "4", + "currency": "CNY" + } + ] + } + ], + "description": "Qwen3.7原生视觉语言系列Flash模型，相较3.6-Flash全面提升多模态理解与Agent执行能力。重点强化多模态基础能力、万物识别能力更强，真实世界感知与空间智能进一步提升，Search Agent、CI Agent等多模态Agent场景能力显著升级、端到端任务执行更稳定，多模态Coding能力优化、vibe coding 体验更加流畅。", + "collectionTag": "qwen3.7", + "features": [ + "batch", + "cache", + "function-calling", + "model-experience", + "prefix-completion", + "structured-outputs", + "web-search" + ], + "provider": "qwen", + "model": "qwen3.7-flash", + "qpmInfo": { + "model-default-actual": { + "count_limit_period": 6, + "usage_limit": 5000000, + "usage_limit_field": "total_tokens", + "count_limit": 3000, + "usage_limit_period": 60, + "type": "model-default" + }, + "model-default": { + "count_limit_period": 6, + "usage_limit": 5000000, + "usage_limit_field": "total_tokens", + "count_limit": 3000, + "usage_limit_period": 60, + "type": "model-default" + } + }, + "priceTimeBands": [], + "multiPrices": [ + { + "rangeStart": 0, + "rangeName": "输入<=32k", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "0.2", + "type": "input_token", + "priceName": "输入" + }, + { + "priceUnit": "每百万tokens", + "price": "0.8", + "type": "output_token", + "priceName": "输出" + }, + { + "priceUnit": "每百万tokens", + "price": "0.04", + "type": "input_token_cache", + "priceName": "输入（缓存命中）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.1", + "type": "input_token_batch", + "priceName": "输入（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.4", + "type": "output_token_batch", + "priceName": "输出（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.25", + "type": "input_token_cache_creation_5m", + "priceName": "显式缓存创建" + }, + { + "priceUnit": "每百万tokens", + "price": "0.02", + "type": "input_token_cache_read", + "priceName": "显式缓存命中" + }, + { + "priceUnit": "每百万tokens", + "price": "0.2", + "discount": 0.5, + "type": "input_token_batch_chat", + "priceName": "输入（Batch Chat)" + }, + { + "priceUnit": "每百万tokens", + "price": "0.8", + "discount": 0.5, + "type": "output_token_batch_chat", + "priceName": "输出（Batch Chat)" + } + ], + "rangeEnd": 32768 + }, + { + "rangeStart": 32768, + "rangeName": "32k<输入<=256k", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "0.6", + "type": "input_token", + "priceName": "输入" + }, + { + "priceUnit": "每百万tokens", + "price": "2.4", + "type": "output_token", + "priceName": "输出" + }, + { + "priceUnit": "每百万tokens", + "price": "0.12", + "type": "input_token_cache", + "priceName": "输入（缓存命中）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.3", + "type": "input_token_batch", + "priceName": "输入（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "1.2", + "type": "output_token_batch", + "priceName": "输出（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.75", + "type": "input_token_cache_creation_5m", + "priceName": "显式缓存创建" + }, + { + "priceUnit": "每百万tokens", + "price": "0.06", + "type": "input_token_cache_read", + "priceName": "显式缓存命中" + }, + { + "priceUnit": "每百万tokens", + "price": "0.6", + "discount": 0.5, + "type": "input_token_batch_chat", + "priceName": "输入（Batch Chat)" + }, + { + "priceUnit": "每百万tokens", + "price": "2.4", + "discount": 0.5, + "type": "output_token_batch_chat", + "priceName": "输出（Batch Chat)" + } + ], + "rangeEnd": 262144 + }, + { + "rangeStart": 262144, + "rangeName": "256k<输入<=1m", + "prices": [ + { + "priceUnit": "每百万tokens", + "price": "1.2", + "type": "input_token", + "priceName": "输入" + }, + { + "priceUnit": "每百万tokens", + "price": "4.8", + "type": "output_token", + "priceName": "输出" + }, + { + "priceUnit": "每百万tokens", + "price": "0.24", + "type": "input_token_cache", + "priceName": "输入（缓存命中）" + }, + { + "priceUnit": "每百万tokens", + "price": "0.6", + "type": "input_token_batch", + "priceName": "输入（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "2.4", + "type": "output_token_batch", + "priceName": "输出（Batch File）" + }, + { + "priceUnit": "每百万tokens", + "price": "1.5", + "type": "input_token_cache_creation_5m", + "priceName": "显式缓存创建" + }, + { + "priceUnit": "每百万tokens", + "price": "0.12", + "type": "input_token_cache_read", + "priceName": "显式缓存命中" + }, + { + "priceUnit": "每百万tokens", + "price": "1.2", + "discount": 0.5, + "type": "input_token_batch_chat", + "priceName": "输入（Batch Chat)" + }, + { + "priceUnit": "每百万tokens", + "price": "4.8", + "discount": 0.5, + "type": "output_token_batch_chat", + "priceName": "输出（Batch Chat)" + } + ], + "rangeEnd": 1000000 + } + ], + "capabilities": [ + "TG", + "VU", + "Reasoning" + ], + "modelAlias": "qwen3.7-flash", + "versionTag": "MAJOR", + "maxOutputTokens": 65536, + "latestOnlineAt": "2026-07-20T17:02:37.000+00:00", + "contextWindow": 1000000, + "maxInputTokens": 991808, + "inferenceProvider": "aliyun-bailian", + "name": "Qwen3.7-Flash", + "docUrl": "https://help.aliyun.com/document_detail/2712576.html", + "category": "Cost-optimized", + "predictConfig": [ + { + "name": "system", + "key": "systemMessage", + "tip": "系统人设，例如“你是一个AI助手”。" + }, + { + "name": "top_p", + "key": "top_p", + "default": 0.8, + "tip": "控制核采样方法的概率阈值，取值越大，生成的随机性越高。", + "range": [ + 0.0001, + 1 + ] + }, + { + "name": "stop", + "key": "stop", + "tip": "用于控制生成时遇到某些内容则停止。您可传入多个字符串。" + }, + { + "name": "temperature", + "key": "temperature", + "default": 0.7, + "tip": "控制生成随机性和多样性，数值越高多样性越强，数值越低一致性越强，范围(0,2)。建议该参数和top_p只设置1个。", + "range": [ + 0, + 1.9999 + ] + }, + { + "name": "enable_search", + "key": "enable_search", + "tip": "联网搜索补充互联网知识" + }, + { + "name": "enable_thinking", + "key": "enable_thinking", + "default": true, + "tip": "推理模式" + }, + { + "name": "thinking_budget", + "key": "thinking_budget", + "default": 4000, + "tip": "思维链的最大输出tokens数量", + "range": [ + 1, + 32768 + ] + }, + { + "name": "result_format", + "key": "result_format", + "default": "message", + "tip": "返回结果格式" + } + ], + "samples": { + "openai": { + "completionsAPI": { + "node": "import OpenAI from \"openai\";\nimport process from 'process';\n\n// 初始化 openai 客户端\nconst openai = new OpenAI({\n apiKey: process.env.DASHSCOPE_API_KEY, // 从环境变量读取\n baseURL: 'https://[workspace-id].cn-beijing.maas.aliyuncs.com/compatible-mode/v1'\n});\nlet isAnswering = false;\nasync function main() {\n try {\n const messages = [{ role: 'user', content: '你是谁' }];\n const stream = await openai.chat.completions.create({\n model: 'qwen3.7-flash',\n messages,\n stream: true,\n enable_thinking: true\n });\n console.log('\\n' + '='.repeat(20) + '思考过程' + '='.repeat(20));\n for await (const chunk of stream) {\n if (!chunk.choices || chunk.choices.length === 0) continue;\n const delta = chunk.choices[0].delta;\n if (delta.reasoning_content !== undefined && delta.reasoning_content !== null) {\n if (!isAnswering) {\n process.stdout.write(delta.reasoning_content);\n }\n }\n if (delta.content !== undefined && delta.content) {\n if (!isAnswering) {\n console.log('\\n' + '='.repeat(20) + '完整回复' + '='.repeat(20));\n isAnswering = true;\n }\n process.stdout.write(delta.content);\n }\n }\n } catch (error) {\n console.error('Error:', error);\n }\n}\nmain();", + "curl": "curl -X POST https://[workspace-id].cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \\\n-H \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n-H \"Content-Type: application/json\" \\\n-d '{\n \"model\": \"qwen3.7-flash\",\n \"messages\": [\n {\n \"role\": \"user\", \n \"content\": \"你是谁\"\n }\n ],\n \"stream\": true,\n \"enable_thinking\": true\n}'", + "python": "from openai import OpenAI\nimport os\n\nclient = OpenAI(\n # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key=\"sk-xxx\"\n api_key=os.getenv(\"DASHSCOPE_API_KEY\"),\n base_url=\"https://[workspace-id].cn-beijing.maas.aliyuncs.com/compatible-mode/v1\",\n)\n\nmessages = [{\"role\": \"user\", \"content\": \"你是谁\"}]\ncompletion = client.chat.completions.create(\n model=\"qwen3.7-flash\", # 您可以按需更换为其它深度思考模型\n messages=messages,\n extra_body={\"enable_thinking\": True},\n stream=True\n)\nis_answering = False # 是否进入回复阶段\nprint(\"\\n\" + \"=\" * 20 + \"思考过程\" + \"=\" * 20)\nfor chunk in completion:\n if not chunk.choices:\n continue\n delta = chunk.choices[0].delta\n if hasattr(delta, \"reasoning_content\") and delta.reasoning_content is not None:\n if not is_answering:\n print(delta.reasoning_content, end=\"\", flush=True)\n if hasattr(delta, \"content\") and delta.content:\n if not is_answering:\n print(\"\\n\" + \"=\" * 20 + \"完整回复\" + \"=\" * 20)\n is_answering = True\n print(delta.content, end=\"\", flush=True)", + "docUrl": "https://help.aliyun.com/document_detail/3016807.html" + }, + "responsesAPI": { + "node": "import OpenAI from \"openai\";\n\nconst openai = new OpenAI({\n apiKey: process.env.DASHSCOPE_API_KEY,\n baseURL: \"https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v2/apps/protocols/compatible-mode/v1\"\n});\n\nasync function main() {\n const response = await openai.responses.create({\n model: \"qwen3.7-flash\",\n input: \"9.9和9.11哪个大？\",\n enable_thinking: true // 启用思考模式\n });\n\n // 遍历输出项\n for (const item of response.output) {\n if (item.type === \"reasoning\") {\n console.log(\"【推理过程】\");\n for (const summary of item.summary) {\n console.log(summary.text.substring(0, 500));\n }\n console.log();\n } else if (item.type === \"message\") {\n console.log(\"【最终答案】\");\n console.log(item.content[0].text);\n }\n }\n}\n\nmain();", + "curl": "curl -X POST https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v2/apps/protocols/compatible-mode/v1/responses \\\n-H \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n-H \"Content-Type: application/json\" \\\n-d '{\n \"model\": \"qwen3.7-flash\",\n \"input\": \"9.9和9.11哪个大？\",\n \"enable_thinking\": true\n}'", + "python": "import os\nfrom openai import OpenAI\n\nclient = OpenAI(\n api_key=os.getenv(\"DASHSCOPE_API_KEY\"),\n base_url=\"https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v2/apps/protocols/compatible-mode/v1\",\n)\n\nresponse = client.responses.create(\n model=\"qwen3.7-flash\",\n input=\"9.9和9.11哪个大？\",\n extra_body={\n \"enable_thinking\": True # 启用思考模式\n }\n)\n\n# 遍历输出项\nfor item in response.output:\n if item.type == \"reasoning\":\n # 打印推理过程摘要\n print(\"【推理过程】\")\n for summary in item.summary:\n print(summary.text[:500]) # 截取前500字符\n print()\n elif item.type == \"message\":\n # 打印最终答案\n print(\"【最终答案】\")\n print(item.content[0].text)", + "docUrl": "https://help.aliyun.com/document_detail/3016808.html" + } + }, + "dashscope": { + "default": { + "curl": "curl -X POST https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation \\\n-H \"Authorization: Bearer $DASHSCOPE_API_KEY\" \\\n-H 'Content-Type: application/json' \\\n-d '{\n \"model\": \"qwen3.7-flash\",\n \"input\":{\n \"messages\":[\n {\n \"role\": \"user\",\n \"content\": [\n {\"image\": \"https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241022/emyrja/dog_and_girl.jpeg\"},\n {\"text\": \"图中描绘的是什么景象?\"}\n ]\n }\n ]\n }\n}'", + "python": "import os\nimport dashscope\ndashscope.base_http_api_url = \"https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1\"\n\nmessages = [\n {\n \"role\": \"user\",\n \"content\": [\n {\"image\": \"https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241022/emyrja/dog_and_girl.jpeg\"},\n {\"text\": \"图中描绘的是什么景象?\"}]\n }]\nresponse = dashscope.MultiModalConversation.call(\n api_key=os.getenv('DASHSCOPE_API_KEY'),\n model='qwen3.7-flash',\n messages=messages\n)\nprint(response.output.choices[0].message.content[0][\"text\"])\n", + "java": "import java.util.Arrays;\nimport java.util.Collections;\nimport com.alibaba.dashscope.aigc.multimodalconversation.MultiModalConversation;\nimport com.alibaba.dashscope.aigc.multimodalconversation.MultiModalConversationParam;\nimport com.alibaba.dashscope.aigc.multimodalconversation.MultiModalConversationResult;\nimport com.alibaba.dashscope.common.MultiModalMessage;\nimport com.alibaba.dashscope.common.Role;\nimport com.alibaba.dashscope.exception.ApiException;\nimport com.alibaba.dashscope.exception.NoApiKeyException;\nimport com.alibaba.dashscope.exception.UploadFileException;\nimport com.alibaba.dashscope.utils.Constants;\n\npublic class Main {\n \n static {Constants.baseHttpApiUrl=\"https://[workspace-id].cn-beijing.maas.aliyuncs.com/api/v1\";}\n\n public static void simpleMultiModalConversationCall()\n throws ApiException, NoApiKeyException, UploadFileException {\n MultiModalConversation conv = new MultiModalConversation();\n MultiModalMessage userMessage = MultiModalMessage.builder().role(Role.USER.getValue())\n .content(Arrays.asList(\n Collections.singletonMap(\"image\", \"https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241022/emyrja/dog_and_girl.jpeg\"),\n Collections.singletonMap(\"text\", \"图中描绘的是什么景象?\"))).build();\n MultiModalConversationParam param = MultiModalConversationParam.builder()\n .apiKey(System.getenv(\"DASHSCOPE_API_KEY\"))\n .model(\"qwen3.7-flash\")\n .messages(Arrays.asList(userMessage))\n .build();\n MultiModalConversationResult result = conv.call(param);\n System.out.println(result.getOutput().getChoices().get(0).getMessage().getContent().get(0).get(\"text\"));\n }\n public static void main(String[] args) {\n try {\n simpleMultiModalConversationCall();\n } catch (ApiException | NoApiKeyException | UploadFileException e) {\n System.out.println(e.getMessage());\n }\n System.exit(0);\n }\n}", + "docUrl": "https://help.aliyun.com/document_detail/3016807.html" + } + } + } + } + ] +} diff --git a/skills/bailian-docs-llm-wiki/models/index.json b/skills/bailian-docs-llm-wiki/models/index.json index c5024785..1763bdf2 100644 --- a/skills/bailian-docs-llm-wiki/models/index.json +++ b/skills/bailian-docs-llm-wiki/models/index.json @@ -1,10 +1,10 @@ { - "updatedAt": "2026-07-16", - "totalFamilies": 170, - "totalModels": 385, + "updatedAt": "2026-07-28", + "totalFamilies": 172, + "totalModels": 389, "capabilityDistribution": { - "TG": 35, - "IG": 30, + "TG": 36, + "IG": 31, "VG": 26, "TTS": 16, "Reasoning": 14, @@ -21,7 +21,7 @@ "3D-generation": 1 }, "providerDistribution": { - "qwen": 100, + "qwen": 103, "qwen-domain-model": 34, "wan": 13, "happyhorse": 4, @@ -635,20 +635,21 @@ "primaryCapability": "TG", "capabilities": [ "TG", - "Reasoning", - "VU" + "VU", + "Reasoning" ], "providers": [ "moonshot-ai" ], - "itemCount": 4, + "itemCount": 5, "items": [ "kimi/kimi-k2.5", "kimi/kimi-k2.6", "kimi/kimi-k2.7-code", - "kimi/kimi-k2.7-code-highspeed" + "kimi/kimi-k2.7-code-highspeed", + "kimi/kimi-k3" ], - "maxContextWindow": 262144 + "maxContextWindow": 1048576 }, { "slug": "kling-models-market-place", @@ -974,7 +975,7 @@ "items": [ "qwen-audio-3.0-realtime-flash" ], - "maxContextWindow": 8192 + "maxContextWindow": 40960 }, { "slug": "qwen-audio-realtime-plus", @@ -990,7 +991,7 @@ "items": [ "qwen-audio-3.0-realtime-plus" ], - "maxContextWindow": 8192 + "maxContextWindow": 40960 }, { "slug": "qwen-audio-tts", @@ -1080,17 +1081,20 @@ "TR" ], "providers": [ + "qwen", "qwen-domain-model" ], - "itemCount": 6, + "itemCount": 7, "items": [ + "qwen3.7-text-embedding", "text-embedding-async-v1", "text-embedding-async-v2", "text-embedding-v1", "text-embedding-v2", "text-embedding-v3", "text-embedding-v4" - ] + ], + "maxContextWindow": 131072 }, { "slug": "qwen-flash-character", @@ -1155,6 +1159,21 @@ "qwen-image-2.0" ] }, + { + "slug": "qwen-image-3.0-pro", + "name": "Qwen-Image-3.0-Pro", + "primaryCapability": "IG", + "capabilities": [ + "IG" + ], + "providers": [ + "qwen" + ], + "itemCount": 1, + "items": [ + "qwen-image-3.0-pro" + ] + }, { "slug": "qwen-image-edit-max", "name": "Qwen-Image-Edit-Max", @@ -2202,6 +2221,24 @@ ], "maxContextWindow": 262144 }, + { + "slug": "qwen3.7-flash", + "name": "Qwen3.7-Flash", + "primaryCapability": "TG", + "capabilities": [ + "TG", + "VU", + "Reasoning" + ], + "providers": [ + "qwen" + ], + "itemCount": 1, + "items": [ + "qwen3.7-flash" + ], + "maxContextWindow": 1000000 + }, { "slug": "qwen3.7-max", "name": "Qwen3.7-Max", diff --git a/skills/bailian-docs-llm-wiki/models/index.md b/skills/bailian-docs-llm-wiki/models/index.md index 16b4073f..42738968 100644 --- a/skills/bailian-docs-llm-wiki/models/index.md +++ b/skills/bailian-docs-llm-wiki/models/index.md @@ -1,6 +1,6 @@ # 百炼模型市场索引 -> 自动生成 · 共 170 个模型家族 · 385 个主干模型 · 更新于 2026-07-16 +> 自动生成 · 共 172 个模型家族 · 389 个主干模型 · 更新于 2026-07-28 **机器查询走结构化文件**： @@ -11,7 +11,7 @@ join：`models.jsonl[].family == families.jsonl[].slug == index.json.families[].slug`。 -## 文本生成 `TG` — 35 个家族 +## 文本生成 `TG` — 36 个家族 - [GLM](groups/glm-4.5.json) — GLM是由智谱提供的开源模型。 - 模型：`glm-4.5`, `glm-4.5-air`, `glm-4.6`, `glm-4.7`, `glm-5`, `glm-5.1`, `glm-5.2` @@ -20,7 +20,7 @@ join：`models.jsonl[].family == families.jsonl[].slug == index.json.families[]. - [Kimi](groups/Kimi-K2.json) — Kimi是由月之暗面提供的开源模型，包含k2.7-code、k2.6、k2.5、k2-thinking、k2-instruct等多模态和大语言模型。 - 模型：`kimi-k2-thinking`, `kimi-k2.5`, `kimi-k2.6`, `kimi-k2.7-code`, `Moonshot-Kimi-K2-Instruct` - [Kimi](groups/kimi-models-market-place.json) — 由月之暗面提供的Kimi系列模型的API服务。 - - 模型：`kimi/kimi-k2.5`, `kimi/kimi-k2.6`, `kimi/kimi-k2.7-code`, `kimi/kimi-k2.7-code-highspeed` + - 模型：`kimi/kimi-k2.5`, `kimi/kimi-k2.6`, `kimi/kimi-k2.7-code`, `kimi/kimi-k2.7-code-highspeed`, `kimi/kimi-k3` - [MiMo文本模型](groups/xiaomi-models-market-place.json) — 由小米MiMo提供的MiMo文本模型API服务 - 模型：`xiaomi/mimo-v2.5-pro` - [MiniMax文本模型](groups/minimax-models-market-place.json) — 由MiniMax提供的MiniMax-M系列文本模型API服务。 @@ -65,6 +65,8 @@ join：`models.jsonl[].family == families.jsonl[].slug == index.json.families[]. - 模型：`qwen3-max`, `qwen3-max-preview` - [Qwen3.5-Plus](groups/qwen3.5-plus.json) — Qwen3.5原生视觉语言系列Plus模型，展现出与当前顶尖前沿模型相媲美的卓越性能，模型效果在纯文本与多模态方面相较3系列均实现飞跃式进步。 - 模型：`qwen3.5-plus` +- [Qwen3.7-Flash](groups/qwen3.7-flash.json) — Qwen3.7原生视觉语言系列Flash模型，相较3.6-Flash全面提升多模态理解与Agent执行能力。重点强化多模态基础能力、万物识别能力更强，真实世界感知与空间智能进一步提升，Search A… + - 模型：`qwen3.7-flash` - [Qwen3.7-Plus](groups/qwen3.7-plus.json) — Qwen3.7系列中高性价比Plus模型，在强大文本能力的基础上全面升级了视觉-语言能力，同时保持了在编码、工具使用和生产力工作流方面的完整智能体能力。其核心特色为多模态交互混合智能体能力，能够感知真… - 模型：`qwen3.7-plus` - [SiliconFlow DeepSeek](groups/siliconflow-models.json) — 由硅基流动提供的DeepSeek系列模型API服务。 @@ -84,7 +86,7 @@ join：`models.jsonl[].family == families.jsonl[].slug == index.json.families[]. - [通义法睿-Plus-32K](groups/farui-plus.json) — 通义法睿是以通义千问为基座经法律行业数据和知识专门训练的法律行业大模型产品，综合运用了模型精调、强化学习、 RAG检索增强、法律Agent技术，具有回答法律问题、推理法律适用、推荐裁判类案、辅助案情分… - 模型：`farui-plus` -## 图像生成 `IG` — 30 个家族 +## 图像生成 `IG` — 31 个家族 - [AI试衣-Plus版](groups/aitryon-plus.json) — aitryon-plus是一款效果出众的虚拟试衣图片生成模型，可基于服饰平拍图片以及人物正面全身照，输出服饰的人物试衣效果图片。 相较于aitryon模型，aitryon-plus模型在图片清晰度、服… - 模型：`aitryon-plus` @@ -102,6 +104,8 @@ join：`models.jsonl[].family == families.jsonl[].slug == index.json.families[]. - 模型：`qwen-image-2.0` - [Qwen-Image-2.0-Pro](groups/qwen-image-2.0-pro.json) — Qwen-Image-2.0系列满血版模型，实现了图片生成和图片编辑的融合；具备更专业的文字渲染1k token指令支持能力、更细腻的真实质感，细腻刻画写实场景、更强的语义遵循能力。满血版具备2.0系… - 模型：`qwen-image-2.0-pro` +- [Qwen-Image-3.0-Pro](groups/qwen-image-3.0-pro.json) — 内容丰实：支持最大 4.5k token 输入，支持图中图密集信息排版，让报纸、分镜、菜单、试卷等复杂版面一次生成。 细节真实：支持 10px 小字精准渲染，微表情、毛孔、发丝等细节生动还原，逼近真实… + - 模型：`qwen-image-3.0-pro` - [Qwen-Image-Edit-Max](groups/qwen-image-edit-max.json) — 千问图像编辑模型Max系列，提供更稳定、更丰富的编辑能力：提升工业设计与几何推理能力；提升角色一致性；减轻偏移问题；集成Lora能力，可以进行更多功能的图像编辑。此版本为2026年1月16日快照。 - 模型：`qwen-image-edit-max` - [Qwen-Image-Edit-Plus](groups/qwen-image-edit.json) — 千问系列图像编辑Plus模型，在首版Edit模型基础上进一步优化了推理性能与系统稳定性，大幅缩短图像生成与编辑的响应时间；支持单次请求返回多张图片，显著提升用户体验。 @@ -390,7 +394,7 @@ join：`models.jsonl[].family == families.jsonl[].slug == index.json.families[]. ## 翻译 `TR` — 2 个家族 - [Qwen-Embedding](groups/qwen-embedding.json) — 基于Qwen模型基座训练的多语言文本统一向量模型，文本检索、聚类、分类性能大幅提升，多语言支持，适用于向量检索、向量化等等场景，可搭配检索增强、文档处理场景使用，支持64~2048维用户自定义向量维度… - - 模型：`text-embedding-async-v1`, `text-embedding-async-v2`, `text-embedding-v1`, `text-embedding-v2`, `text-embedding-v3`, `text-embedding-v4` + - 模型：`qwen3.7-text-embedding`, `text-embedding-async-v1`, `text-embedding-async-v2`, `text-embedding-v1`, `text-embedding-v2`, `text-embedding-v3`, `text-embedding-v4` - [Qwen-Rerank](groups/qwen-rerank.json) — 基于Qwen LLM底座训练的文本排序模型，对输入的Query和候选Docs进行相关性排序，支持100+语种和长文本输入，适用于文本检索、RAG等场景，效果对齐Qwen家族开源Rerank系列模型。 - 模型：`gte-rerank-v2`, `qwen3-rerank`, `qwen3-vl-rerank` diff --git a/skills/bailian-docs-llm-wiki/models/models.jsonl b/skills/bailian-docs-llm-wiki/models/models.jsonl index 5e7d3d5e..bc24f3b8 100644 --- a/skills/bailian-docs-llm-wiki/models/models.jsonl +++ b/skills/bailian-docs-llm-wiki/models/models.jsonl @@ -73,10 +73,11 @@ {"model":"image-erase-completion","name":"图像擦除补全","family":"image-erase-completion","familyName":"图像擦除补全","provider":"qwen","capabilities":["IG"],"features":[],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Image"]},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2840907.html","detailPath":"groups/image-erase-completion.json"} {"model":"image-instance-segmentation","name":"人物实例分割","family":"image-instance-segmentation","familyName":"人物实例分割","provider":"qwen","capabilities":["IG"],"features":[],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Image"]},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2840906.html","detailPath":"groups/image-instance-segmentation.json"} {"model":"image-out-painting","name":"图像画面扩展","family":"image-out-painting","familyName":"图像画面扩展","provider":"qwen","capabilities":["IG"],"features":[],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Image"]},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2796845.html","detailPath":"groups/image-out-painting.json"} -{"model":"kimi/kimi-k2.5","name":"Kimi/Kimi K2.5","family":"kimi-models-market-place","familyName":"Kimi","provider":"moonshot-ai","capabilities":["TG","Reasoning","VU"],"features":["function-calling","structured-outputs","prefix-completion","cache"],"contextWindow":262144,"maxInputTokens":262144,"maxOutputTokens":262144,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":50,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":50,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/3021620.html","detailPath":"groups/kimi-models-market-place.json"} -{"model":"kimi/kimi-k2.6","name":"Kimi/Kimi K2.6","family":"kimi-models-market-place","familyName":"Kimi","provider":"moonshot-ai","capabilities":["TG","Reasoning","VU"],"features":["function-calling","structured-outputs","prefix-completion","cache"],"contextWindow":262144,"maxInputTokens":262144,"maxOutputTokens":262144,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":50,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":50,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/3021620.html","detailPath":"groups/kimi-models-market-place.json"} -{"model":"kimi/kimi-k2.7-code","name":"kimi/kimi-k2.7-code","family":"kimi-models-market-place","familyName":"Kimi","provider":"moonshot-ai","capabilities":["TG","VU","Reasoning"],"features":["function-calling","structured-outputs","prefix-completion","cache"],"contextWindow":262144,"maxInputTokens":262144,"maxOutputTokens":262144,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://bailian.console.aliyun.com/cn-beijing?tab=doc#/doc/?type=model&url=3021620","detailPath":"groups/kimi-models-market-place.json"} -{"model":"kimi/kimi-k2.7-code-highspeed","name":"kimi/kimi-k2.7-code-highspeed","family":"kimi-models-market-place","familyName":"Kimi","provider":"moonshot-ai","capabilities":["TG","Reasoning","VU"],"features":["function-calling","structured-outputs","prefix-completion","cache"],"contextWindow":262144,"maxInputTokens":262144,"maxOutputTokens":262144,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://bailian.console.aliyun.com/cn-beijing?tab=doc#/doc/?type=model&url=3021620","detailPath":"groups/kimi-models-market-place.json"} +{"model":"kimi/kimi-k2.5","name":"Kimi/Kimi K2.5","family":"kimi-models-market-place","familyName":"Kimi","provider":"moonshot-ai","capabilities":["TG","Reasoning","VU"],"features":["function-calling","structured-outputs","prefix-completion","cache"],"contextWindow":262144,"maxInputTokens":262144,"maxOutputTokens":262144,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"prices":[{"type":"input_token","unit":"每百万tokens","price":"4"},{"type":"output_token","unit":"每百万tokens","price":"21"},{"type":"input_token_cache","unit":"每百万tokens","price":"0.7"}],"qpmInfo":{"model-default-actual":{"count_limit":50,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":50,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/3021620.html","detailPath":"groups/kimi-models-market-place.json"} +{"model":"kimi/kimi-k2.6","name":"Kimi/Kimi K2.6","family":"kimi-models-market-place","familyName":"Kimi","provider":"moonshot-ai","capabilities":["TG","Reasoning","VU"],"features":["function-calling","structured-outputs","prefix-completion","cache"],"contextWindow":262144,"maxInputTokens":262144,"maxOutputTokens":262144,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"prices":[{"type":"input_token","unit":"每百万tokens","price":"6.5"},{"type":"output_token","unit":"每百万tokens","price":"27"},{"type":"input_token_cache","unit":"每百万tokens","price":"1.1"}],"qpmInfo":{"model-default-actual":{"count_limit":50,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":50,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/3021620.html","detailPath":"groups/kimi-models-market-place.json"} +{"model":"kimi/kimi-k2.7-code","name":"kimi/kimi-k2.7-code","family":"kimi-models-market-place","familyName":"Kimi","provider":"moonshot-ai","capabilities":["TG","VU","Reasoning"],"features":["function-calling","structured-outputs","prefix-completion","cache"],"contextWindow":262144,"maxInputTokens":262144,"maxOutputTokens":262144,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"prices":[{"type":"input_token","unit":"每百万tokens","price":"6.5"},{"type":"output_token","unit":"每百万tokens","price":"27"},{"type":"input_token_cache","unit":"每百万tokens","price":"1.3"}],"qpmInfo":{"model-default-actual":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://bailian.console.aliyun.com/cn-beijing?tab=doc#/doc/?type=model&url=3021620","detailPath":"groups/kimi-models-market-place.json"} +{"model":"kimi/kimi-k2.7-code-highspeed","name":"kimi/kimi-k2.7-code-highspeed","family":"kimi-models-market-place","familyName":"Kimi","provider":"moonshot-ai","capabilities":["TG","Reasoning","VU"],"features":["function-calling","structured-outputs","prefix-completion","cache"],"contextWindow":262144,"maxInputTokens":262144,"maxOutputTokens":262144,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"prices":[{"type":"input_token","unit":"每百万tokens","price":"13"},{"type":"output_token","unit":"每百万tokens","price":"54"},{"type":"input_token_cache","unit":"每百万tokens","price":"2.6"}],"qpmInfo":{"model-default-actual":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://bailian.console.aliyun.com/cn-beijing?tab=doc#/doc/?type=model&url=3021620","detailPath":"groups/kimi-models-market-place.json"} +{"model":"kimi/kimi-k3","name":"kimi/kimi-k3","family":"kimi-models-market-place","familyName":"Kimi","provider":"moonshot-ai","capabilities":["TG","VU","Reasoning"],"features":["function-calling","structured-outputs","cache","prefix-completion"],"contextWindow":1048576,"maxInputTokens":1048576,"maxOutputTokens":1048576,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"prices":[{"type":"input_token","unit":"每百万tokens","price":"20"},{"type":"output_token","unit":"每百万tokens","price":"100"},{"type":"input_token_cache","unit":"每百万tokens","price":"2"}],"qpmInfo":{"model-default-actual":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":500,"count_limit_period":60,"usage_limit":3000000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://bailian.console.aliyun.com/cn-beijing?tab=doc#/doc/?type=model&url=3021620","detailPath":"groups/kimi-models-market-place.json"} {"model":"kling/kling-v3-image-generation","name":"Kling Image 3.0","family":"kling-models-market-place","familyName":"可灵AI","provider":"kling","capabilities":["IG"],"features":[],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Image","Text"]},"qpmInfo":{"model-default-actual":{"count_limit":5,"count_limit_period":1},"model-default":{"count_limit":5,"count_limit_period":1}},"versionTag":"MAJOR","detailPath":"groups/kling-models-market-place.json"} {"model":"kling/kling-v3-omni-image-generation","name":"Kling Image 3.0 Omni","family":"kling-models-market-place","familyName":"可灵AI","provider":"kling","capabilities":["IG"],"features":[],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Image","Text"]},"qpmInfo":{"model-default-actual":{"count_limit":5,"count_limit_period":1},"model-default":{"count_limit":5,"count_limit_period":1}},"versionTag":"MAJOR","detailPath":"groups/kling-models-market-place.json"} {"model":"kling/kling-v3-omni-video-generation","name":"Kling Video 3.0 Omni","family":"kling-models-market-place","familyName":"可灵AI","provider":"kling","capabilities":["VG"],"features":[],"inferenceMetadata":{"response_modality":["Video"],"request_modality":["Image","Text","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":5,"count_limit_period":1},"model-default":{"count_limit":5,"count_limit_period":1}},"versionTag":"MAJOR","detailPath":"groups/kling-models-market-place.json"} @@ -113,24 +114,26 @@ {"model":"pixverse/pixverse-v6-t2v","name":"PixVerse-V6-t2v","family":"pixverse-v6-market-place","familyName":"PixVerse V6","provider":"pixverse","capabilities":["VG"],"features":[],"inferenceMetadata":{"response_modality":["Video"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":5,"count_limit_period":1},"model-default":{"count_limit":5,"count_limit_period":1}},"versionTag":"MAJOR","docUrl":"○ https://help.aliyun.com/document_detail/3025608.html","detailPath":"groups/pixverse-v6-market-place.json"} {"model":"qvq-max","name":"QVQ-Max","family":"qvq-max","familyName":"QVQ-Max","provider":"qwen","capabilities":["Reasoning","VU"],"features":["model-experience"],"contextWindow":131072,"maxInputTokens":106496,"maxOutputTokens":8192,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":60,"count_limit_period":60,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":60,"count_limit_period":60,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2877996.html","detailPath":"groups/qvq-max.json"} {"model":"qvq-plus","name":"QVQ-Plus","family":"qvq-plus","familyName":"Qwen-QVQ-Plus","provider":"qwen","capabilities":["Reasoning","VU"],"features":["model-experience"],"contextWindow":131072,"maxInputTokens":106496,"maxOutputTokens":8192,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text","Image","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":5,"count_limit_period":5,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":5,"count_limit_period":5,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2877996.html","detailPath":"groups/qvq-plus.json"} -{"model":"qwen-audio-3.0-realtime-flash","name":"千问实时语音对话大模型3.0（极速版）","family":"qwen-audio-realtime-flash","familyName":"Qwen-Audio-Realtime-Flash","provider":"qwen","capabilities":["Realtime-Chatting"],"features":["function-calling"],"contextWindow":8192,"maxInputTokens":4096,"maxOutputTokens":4096,"inferenceMetadata":{"response_modality":["Audio","Text"],"request_modality":["Audio","Text"]},"prices":[{"type":"audio_input_token","unit":"每百万tokens","price":"30"},{"type":"audio_text_input_token","unit":"每百万tokens","price":"3"},{"type":"audio_text_output_token","unit":"每百万tokens","price":"30"},{"type":"audio_output_token","unit":"每百万tokens","price":"100"}],"qpmInfo":{"model-default-actual":{"count_limit":6,"count_limit_period":6,"usage_limit":10000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":6,"count_limit_period":6,"usage_limit":10000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/3041584.html","detailPath":"groups/qwen-audio-realtime-flash.json"} -{"model":"qwen-audio-3.0-realtime-plus","name":"千问实时语音大模型 （标准版）","family":"qwen-audio-realtime-plus","familyName":"Qwen-Audio-Realtime-Plus","provider":"qwen","capabilities":["Realtime-Chatting"],"features":["function-calling"],"contextWindow":8192,"maxInputTokens":4096,"maxOutputTokens":4096,"inferenceMetadata":{"response_modality":["Audio","Text"],"request_modality":["Audio","Text"]},"prices":[{"type":"audio_input_token","unit":"每百万tokens","price":"40"},{"type":"audio_text_input_token","unit":"每百万tokens","price":"5"},{"type":"audio_text_output_token","unit":"每百万tokens","price":"40"},{"type":"audio_output_token","unit":"每百万tokens","price":"150"}],"qpmInfo":{"model-default-actual":{"count_limit":6,"count_limit_period":6,"usage_limit":10000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":6,"count_limit_period":6,"usage_limit":10000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/3041584.html","detailPath":"groups/qwen-audio-realtime-plus.json"} +{"model":"qwen-audio-3.0-realtime-flash","name":"千问实时语音对话大模型3.0（极速版）","family":"qwen-audio-realtime-flash","familyName":"Qwen-Audio-Realtime-Flash","provider":"qwen","capabilities":["Realtime-Chatting"],"features":["function-calling"],"contextWindow":40960,"maxInputTokens":16384,"maxOutputTokens":8192,"inferenceMetadata":{"response_modality":["Audio","Text"],"request_modality":["Audio","Text"]},"prices":[{"type":"audio_input_token","unit":"每百万tokens","price":"30"},{"type":"audio_text_input_token","unit":"每百万tokens","price":"3"},{"type":"audio_text_output_token","unit":"每百万tokens","price":"30"},{"type":"audio_output_token","unit":"每百万tokens","price":"100"}],"qpmInfo":{"model-default-actual":{"count_limit":6,"count_limit_period":6,"usage_limit":10000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":6,"count_limit_period":6,"usage_limit":10000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/3041584.html","detailPath":"groups/qwen-audio-realtime-flash.json"} +{"model":"qwen-audio-3.0-realtime-plus","name":"千问实时语音大模型 （标准版）","family":"qwen-audio-realtime-plus","familyName":"Qwen-Audio-Realtime-Plus","provider":"qwen","capabilities":["Realtime-Chatting"],"features":["function-calling"],"contextWindow":40960,"maxInputTokens":16384,"maxOutputTokens":8192,"inferenceMetadata":{"response_modality":["Audio","Text"],"request_modality":["Audio","Text"]},"prices":[{"type":"audio_input_token","unit":"每百万tokens","price":"40"},{"type":"audio_text_input_token","unit":"每百万tokens","price":"5"},{"type":"audio_text_output_token","unit":"每百万tokens","price":"40"},{"type":"audio_output_token","unit":"每百万tokens","price":"150"}],"qpmInfo":{"model-default-actual":{"count_limit":6,"count_limit_period":6,"usage_limit":10000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":6,"count_limit_period":6,"usage_limit":10000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/3041584.html","detailPath":"groups/qwen-audio-realtime-plus.json"} {"model":"qwen-audio-3.0-tts-flash","name":"qwen-audio-3.0-tts-flash","family":"qwen-audio-tts","familyName":"Qwen-Audio-TTS","provider":"qwen","capabilities":["Realtime-Text-to-Speech"],"features":[],"inferenceMetadata":{"response_modality":["Audio"],"request_modality":["Text"]},"prices":[{"type":"cosy_tts_number","unit":"每万字符","price":"1"}],"qpmInfo":{"model-default-actual":{"count_limit":3,"count_limit_period":1},"model-default":{"count_limit":3,"count_limit_period":1}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2938790.html","detailPath":"groups/qwen-audio-tts.json"} {"model":"qwen-audio-3.0-tts-plus","name":"qwen-audio-3.0-tts-plus","family":"qwen-audio-tts","familyName":"Qwen-Audio-TTS","provider":"qwen","capabilities":["Realtime-Text-to-Speech"],"features":[],"inferenceMetadata":{"response_modality":["Audio"],"request_modality":["Text"]},"prices":[{"type":"cosy_tts_number","unit":"每万字符","price":"1.4"}],"qpmInfo":{"model-default-actual":{"count_limit":3,"count_limit_period":1},"model-default":{"count_limit":3,"count_limit_period":1}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2938790.html","detailPath":"groups/qwen-audio-tts.json"} {"model":"qwen-coder-plus","name":"Qwen-Coder-Plus","family":"qwen-coder-plus","familyName":"Qwen-Coder-Plus","provider":"qwen-domain-model","capabilities":["TG"],"features":["model-experience"],"contextWindow":131072,"maxInputTokens":129024,"maxOutputTokens":8192,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":20,"count_limit_period":1,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":20,"count_limit_period":1,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2850166.html","detailPath":"groups/qwen-coder-plus.json"} {"model":"qwen-coder-turbo","name":"Qwen-Coder-Turbo","family":"qwen-coder-turbo","familyName":"Qwen-Coder-Turbo","provider":"qwen-domain-model","capabilities":["TG"],"features":["model-experience"],"contextWindow":131072,"maxInputTokens":129024,"maxOutputTokens":8192,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":20,"count_limit_period":1,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":20,"count_limit_period":1,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2850166.html","detailPath":"groups/qwen-coder-turbo.json"} {"model":"qwen-deep-research","name":"qwen-deep-research","family":"qwen-deep-research","familyName":"qwen-deep-research","provider":"qwen-domain-model","capabilities":["TG"],"features":[],"contextWindow":1000000,"maxInputTokens":997952,"maxOutputTokens":32768,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":120,"count_limit_period":60,"usage_limit":1200000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":120,"count_limit_period":60,"usage_limit":1200000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2975991.html","detailPath":"groups/qwen-deep-research.json"} {"model":"qwen-doc-turbo","name":"Qwen-Doc-Turbo","family":"qwen-doc-turbo","familyName":"Qwen-Doc-Turbo","provider":"qwen-domain-model","capabilities":["TG"],"features":["cache"],"contextWindow":262144,"maxInputTokens":253952,"maxOutputTokens":8192,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":60,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":60,"count_limit_period":6,"usage_limit":300000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2948885.html","detailPath":"groups/qwen-doc-turbo.json"} -{"model":"text-embedding-async-v1","name":"通用文本向量-async-v1","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712516.html","detailPath":"groups/qwen-embedding.json"} -{"model":"text-embedding-async-v2","name":"通用文本向量-async-v2","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712516.html","detailPath":"groups/qwen-embedding.json"} -{"model":"text-embedding-v1","name":"通用文本向量-v1","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712515.html","detailPath":"groups/qwen-embedding.json"} -{"model":"text-embedding-v2","name":"通用文本向量-v2","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712515.html","detailPath":"groups/qwen-embedding.json"} -{"model":"text-embedding-v3","name":"通用文本向量-v3","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712515.html","detailPath":"groups/qwen-embedding.json"} -{"model":"text-embedding-v4","name":"通用文本向量-v4","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":["model-experience"],"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":30,"count_limit_period":1,"usage_limit":1200000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":30,"count_limit_period":1,"usage_limit":1200000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2842587.html","detailPath":"groups/qwen-embedding.json"} +{"model":"qwen3.7-text-embedding","name":"Qwen3.7-Text-Embedding","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen","capabilities":["TR"],"features":["model-experience"],"contextWindow":131072,"maxInputTokens":131072,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"prices":[{"type":"embedding_token","unit":"每百万tokens","price":"0.5"}],"qpmInfo":{"model-default-actual":{"count_limit":2400,"count_limit_period":6,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":2400,"count_limit_period":6,"usage_limit":100000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2842587.html","detailPath":"groups/qwen-embedding.json"} +{"model":"text-embedding-async-v1","name":"通用文本向量-async-v1","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"prices":[{"type":"embedding_token","unit":"每百万tokens","price":"0.7"}],"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712516.html","detailPath":"groups/qwen-embedding.json"} +{"model":"text-embedding-async-v2","name":"通用文本向量-async-v2","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"prices":[{"type":"embedding_token","unit":"每百万tokens","price":"0.7"}],"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712516.html","detailPath":"groups/qwen-embedding.json"} +{"model":"text-embedding-v1","name":"通用文本向量-v1","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"prices":[{"type":"embedding_token_batch","unit":"每百万tokens","price":"0.35"},{"type":"embedding_token","unit":"每百万tokens","price":"0.7"}],"qpmInfo":{"model-default-actual":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712515.html","detailPath":"groups/qwen-embedding.json"} +{"model":"text-embedding-v2","name":"通用文本向量-v2","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"prices":[{"type":"embedding_token_batch","unit":"每百万tokens","price":"0.35"},{"type":"embedding_token","unit":"每百万tokens","price":"0.7"}],"qpmInfo":{"model-default-actual":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712515.html","detailPath":"groups/qwen-embedding.json"} +{"model":"text-embedding-v3","name":"通用文本向量-v3","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":[],"inferenceMetadata":{"response_modality":[],"request_modality":["Text"]},"prices":[{"type":"embedding_token_batch","unit":"每百万tokens","price":"0.25"},{"type":"embedding_token","unit":"每百万tokens","price":"0.5"}],"qpmInfo":{"model-default-actual":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":30,"count_limit_period":1,"usage_limit":120000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712515.html","detailPath":"groups/qwen-embedding.json"} +{"model":"text-embedding-v4","name":"通用文本向量-v4","family":"qwen-embedding","familyName":"Qwen-Embedding","provider":"qwen-domain-model","capabilities":["TR"],"features":["model-experience"],"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"prices":[{"type":"embedding_token_batch","unit":"每百万tokens","price":"0.25"},{"type":"embedding_token","unit":"每百万tokens","price":"0.5"}],"qpmInfo":{"model-default-actual":{"count_limit":30,"count_limit_period":1,"usage_limit":1200000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":30,"count_limit_period":1,"usage_limit":1200000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2842587.html","detailPath":"groups/qwen-embedding.json"} {"model":"qwen-flash-character","name":"Qwen-Flash-Character","family":"qwen-flash-character","familyName":"Qwen-Flash-Character","provider":"qwen","capabilities":["TG"],"features":["model-experience","cache","web-search"],"contextWindow":8192,"maxInputTokens":8000,"maxOutputTokens":4096,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":120,"count_limit_period":60,"usage_limit":500000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":120,"count_limit_period":60,"usage_limit":500000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2874763.html","detailPath":"groups/qwen-flash-character.json"} {"model":"qwen-flash","name":"Qwen-Flash","family":"qwen-flash","familyName":"Qwen-Flash","provider":"qwen","capabilities":["Reasoning","TG"],"features":["model-experience","function-calling","structured-outputs","web-search","prefix-completion","cache","batch"],"contextWindow":1000000,"maxInputTokens":997952,"maxOutputTokens":32768,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":15000,"count_limit_period":30,"usage_limit":5000000,"usage_limit_field":"total_tokens","usage_limit_period":30},"model-default":{"count_limit":15000,"count_limit_period":30,"usage_limit":5000000,"usage_limit_field":"total_tokens","usage_limit_period":30}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712576.html","detailPath":"groups/qwen-flash.json"} {"model":"qwen-image-2.0-pro","name":"Qwen-Image-2.0-Pro","family":"qwen-image-2.0-pro","familyName":"Qwen-Image-2.0-Pro","provider":"qwen","capabilities":["IG"],"features":["model-experience"],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Text","Image"]},"qpmInfo":{"model-default-actual":{"count_limit":2,"count_limit_period":60},"model-default":{"count_limit":2,"count_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2976416.html","detailPath":"groups/qwen-image-2.0-pro.json"} {"model":"qwen-image-2.0","name":"Qwen-Image-2.0","family":"qwen-image-2.0","familyName":"Qwen-Image-2.0","provider":"qwen","capabilities":["IG"],"features":["model-experience"],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Text","Image"]},"qpmInfo":{"model-default-actual":{"count_limit":2,"count_limit_period":1},"model-default":{"count_limit":2,"count_limit_period":1}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2976416.html","detailPath":"groups/qwen-image-2.0.json"} +{"model":"qwen-image-3.0-pro","name":"Qwen-Image-3.0-Pro","family":"qwen-image-3.0-pro","familyName":"Qwen-Image-3.0-Pro","provider":"qwen","capabilities":["IG"],"features":["model-experience"],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Image","Text"]},"qpmInfo":{"model-default-actual":{"count_limit":1,"count_limit_period":60},"model-default":{"count_limit":1,"count_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/3047054.html","detailPath":"groups/qwen-image-3.0-pro.json"} {"model":"qwen-image-edit-max","name":"Qwen-Image-Edit-Max","family":"qwen-image-edit-max","familyName":"Qwen-Image-Edit-Max","provider":"qwen","capabilities":["IG"],"features":["model-experience"],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Text","Image"]},"qpmInfo":{"model-default-actual":{"count_limit":2,"count_limit_period":60},"model-default":{"count_limit":2,"count_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2976416.html","detailPath":"groups/qwen-image-edit-max.json"} {"model":"qwen-image-edit","name":"Qwen-Image-Edit","family":"qwen-image-edit","familyName":"Qwen-Image-Edit-Plus","provider":"qwen","capabilities":["IG"],"features":["model-experience"],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Text","Image"]},"qpmInfo":{"model-default-actual":{"count_limit":2,"count_limit_period":1},"model-default":{"count_limit":2,"count_limit_period":1}},"versionTag":"SNAPSHOT","docUrl":"https://help.aliyun.com/document_detail/2976416.html","detailPath":"groups/qwen-image-edit.json"} {"model":"qwen-image-edit-plus","name":"Qwen-Image-Edit-Plus","family":"qwen-image-edit","familyName":"Qwen-Image-Edit-Plus","provider":"qwen","capabilities":["IG"],"features":["model-experience"],"inferenceMetadata":{"response_modality":["Image"],"request_modality":["Image"]},"qpmInfo":{"model-default-actual":{"count_limit":2,"count_limit_period":1},"model-default":{"count_limit":2,"count_limit_period":1}},"versionTag":"SNAPSHOT","docUrl":"https://help.aliyun.com/document_detail/2976416.html","detailPath":"groups/qwen-image-edit.json"} @@ -218,6 +221,7 @@ {"model":"qwen3.6-plus","name":"Qwen3.6-Plus","family":"qwen3.6-plus","familyName":"Qwen3.6-Plus","provider":"qwen","capabilities":["Reasoning","VU","TG"],"features":["batch","cache","function-calling","model-experience","prefix-completion","structured-outputs","web-search"],"contextWindow":1000000,"maxInputTokens":991808,"maxOutputTokens":65536,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Image","Text","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":500,"count_limit_period":1,"usage_limit":2500000,"usage_limit_field":"total_tokens","usage_limit_period":30},"model-default":{"count_limit":500,"count_limit_period":1,"usage_limit":2500000,"usage_limit_field":"total_tokens","usage_limit_period":30}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712576.html","detailPath":"groups/qwen3.6-plus.json"} {"model":"qwen3.6-27b","name":"Qwen3.6-27B","family":"qwen3.6","familyName":"Qwen3.6开源模型","provider":"qwen","capabilities":["Reasoning","VU","TG"],"features":["function-calling","model-experience","prefix-completion","structured-outputs","web-search"],"contextWindow":262144,"maxInputTokens":260096,"maxOutputTokens":65536,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Image","Text","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":60,"count_limit_period":6,"usage_limit":1000000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":60,"count_limit_period":6,"usage_limit":1000000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"SNAPSHOT","docUrl":"https://help.aliyun.com/document_detail/2712576.html","detailPath":"groups/qwen3.6.json"} {"model":"qwen3.6-35b-a3b","name":"Qwen3.6-35B-A3B","family":"qwen3.6","familyName":"Qwen3.6开源模型","provider":"qwen","capabilities":["VU","TG","Reasoning"],"features":["function-calling","model-experience","prefix-completion","structured-outputs","web-search"],"contextWindow":262144,"maxInputTokens":260096,"maxOutputTokens":65536,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Image","Text","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":60,"count_limit_period":6,"usage_limit":1000000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":60,"count_limit_period":6,"usage_limit":1000000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"SNAPSHOT","docUrl":"https://help.aliyun.com/document_detail/2712576.html","detailPath":"groups/qwen3.6.json"} +{"model":"qwen3.7-flash","name":"Qwen3.7-Flash","family":"qwen3.7-flash","familyName":"Qwen3.7-Flash","provider":"qwen","capabilities":["TG","VU","Reasoning"],"features":["batch","cache","function-calling","model-experience","prefix-completion","structured-outputs","web-search"],"contextWindow":1000000,"maxInputTokens":991808,"maxOutputTokens":65536,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Image","Text","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":3000,"count_limit_period":6,"usage_limit":5000000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":3000,"count_limit_period":6,"usage_limit":5000000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712576.html","detailPath":"groups/qwen3.7-flash.json"} {"model":"qwen3.7-max","name":"Qwen3.7-Max","family":"qwen3.7-max","familyName":"Qwen3.7-Max","provider":"qwen","capabilities":["Reasoning","TG"],"features":["batch","cache","function-calling","model-experience","prefix-completion","web-search"],"contextWindow":1000000,"maxInputTokens":991808,"maxOutputTokens":65536,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":500,"count_limit_period":1,"usage_limit":500000,"usage_limit_field":"total_tokens","usage_limit_period":6},"model-default":{"count_limit":500,"count_limit_period":1,"usage_limit":500000,"usage_limit_field":"total_tokens","usage_limit_period":6}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712576.html","detailPath":"groups/qwen3.7-max.json"} {"model":"qwen3.7-max-preview","name":"Qwen3.7-Max-Preview","family":"qwen3.7-max","familyName":"Qwen3.7-Max","provider":"qwen","capabilities":["TG","Reasoning"],"features":["function-calling","model-experience","prefix-completion","web-search"],"contextWindow":1000000,"maxInputTokens":991808,"maxOutputTokens":65536,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Text"]},"qpmInfo":{"model-default-actual":{"count_limit":6,"count_limit_period":6,"usage_limit":500000,"usage_limit_field":"total_tokens","usage_limit_period":60},"model-default":{"count_limit":6,"count_limit_period":6,"usage_limit":500000,"usage_limit_field":"total_tokens","usage_limit_period":60}},"versionTag":"SNAPSHOT","docUrl":"https://help.aliyun.com/document_detail/2712576.html","detailPath":"groups/qwen3.7-max.json"} {"model":"qwen3.7-plus","name":"Qwen3.7-Plus","family":"qwen3.7-plus","familyName":"Qwen3.7-Plus","provider":"qwen","capabilities":["TG","Reasoning","VU"],"features":["batch","cache","function-calling","model-experience","prefix-completion","web-search","structured-outputs"],"contextWindow":1000000,"maxInputTokens":991808,"maxOutputTokens":65536,"inferenceMetadata":{"response_modality":["Text"],"request_modality":["Image","Text","Video"]},"qpmInfo":{"model-default-actual":{"count_limit":3000,"count_limit_period":6,"usage_limit":2500000,"usage_limit_field":"total_tokens","usage_limit_period":30},"model-default":{"count_limit":3000,"count_limit_period":6,"usage_limit":2500000,"usage_limit_field":"total_tokens","usage_limit_period":30}},"versionTag":"MAJOR","docUrl":"https://help.aliyun.com/document_detail/2712576.html","detailPath":"groups/qwen3.7-plus.json"} diff --git a/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-createindex.md b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-createindex.md index 3b1ff1ad..04d60432 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-createindex.md +++ b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-createindex.md @@ -163,6 +163,8 @@ string 默认值为空，此时使用 text-embedding-v2 模型。 +text-embedding-v4 + RerankModelName string @@ -322,7 +324,7 @@ array 否 -创建知识库时可同步导入文件。此处通过指定类目 ID，可导入对应类目下的所有文件（建议导入不超过 10000 个。如有剩余文件，后续可调用 **SubmitIndexAddDocumentsJob** 接口继续导入）。 +创建知识库时可同步导入文件。此处通过指定类目 ID，可导入对应类目下的所有文件（建议导入不超过 500 个。如有剩余文件，后续可调用 **SubmitIndexAddDocumentsJob** 接口继续导入）。 string diff --git a/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindexfiledetails.md b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindexfiledetails.md index 05ded095..5bfce045 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindexfiledetails.md +++ b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-listindexfiledetails.md @@ -283,14 +283,6 @@ Excel 文件表头是否支持拼装。 true -OverlapSize - -string - -分段重叠长度。 - -100 - Message string @@ -315,14 +307,6 @@ string cate\_21a407a3372c4ba7aedc649709143f0cxxxxxxxx -GmtModified - -integer - -文件导入知识库的时间，采用 Unix timestamp 格式。 - -1744856423000 - DocumentType string @@ -331,13 +315,13 @@ string pdf -ChunkMode +Id string -自定义切分。 +文件 ID。 -DashSplitter +doc\_c134aa2073204a5d936d870bf960f56axxxxxxxx Code @@ -347,14 +331,6 @@ string 110002 -separator - -string - -分句标识符。 - -" " - Name string @@ -371,13 +347,41 @@ string 600 -Id +OverlapSize string -文件 ID。 +分段重叠长度。 -doc\_c134aa2073204a5d936d870bf960f56axxxxxxxx +100 + +ChunkMode + +string + +自定义切分。 + +DashSplitter + +GmtModified + +integer + +文件导入知识库的时间，采用 Unix timestamp 格式。 + +1744856423000 + +separator + +string + +分句标识符。 + +" " + +MetaExtractInfo + +string IndexId @@ -436,18 +440,19 @@ string { "Status": "RUNNING", "EnableHeaders": "true", - "OverlapSize": "100", "Message": "check fileUrlKey[file_path] / fileNameKey[null] / fileExtensionKey[file_extension] is invalid", "Size": 996764, "SourceId": "cate_21a407a3372c4ba7aedc649709143f0cxxxxxxxx\n", - "GmtModified": 1744856423000, "DocumentType": "pdf", - "ChunkMode": "DashSplitter", + "Id": "doc_c134aa2073204a5d936d870bf960f56axxxxxxxx\n", "Code": "110002", - "separator": "\" \"", "Name": "翻译平台运维文档\n", "ChunkSize": "600", - "Id": "doc_c134aa2073204a5d936d870bf960f56axxxxxxxx\n" + "OverlapSize": "100", + "ChunkMode": "DashSplitter", + "GmtModified": 1744856423000, + "separator": "\" \"", + "MetaExtractInfo": "" } ], "IndexId": "79c0alxxxx", diff --git a/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-addchunk.md b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-addchunk.md new file mode 100644 index 00000000..807c00f2 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-addchunk.md @@ -0,0 +1,233 @@ +# AddChunk - 新增切片 + +使用此API可为文档搜索类（document）、数据查询类（table）、图片问答类（image）知识库添加切片。 + +## 接口说明 + +- 对于文档搜索类（document）、数据查询类（table）、图片问答类（image）知识库，本接口可向指定知识库中添加切片内容；目前尚不支持对音视频搜索类（multimedia）知识库进行相关操作。仅当数据源为表格连接器(excel)时，对数据查询与图片问答类型知识库的操作方可生效。 + +- RAM 用户（子账号）需要首先获取阿里云百炼的 [API 权限](https://help.aliyun.com/zh/model-studio/grant-data-access-permission-to-ram-user)（需要`AliyunBailianDataFullAccess`，已包括 sfm:ChunkList 权限点），并[加入一个业务空间](https://help.aliyun.com/zh/model-studio/grant-the-business-space-permission-to-ram-users)后，方可调用本接口。阿里云账号（主账号）可直接调用无须授权。建议您通过最新版[阿里云百炼 SDK](https://api.aliyun.com/api-tools/sdk/bailian?version=2023-12-29)[阿里云百炼 SDK](https://api.alibabacloud.com/api-tools/sdk/bailian?version=2023-12-29)来调用本接口。 + +- 调用本接口前，请确保您的知识库已经创建完成且未被删除（即知识库 ID`IndexId`有效）。 + +- 本接口具有幂等性。 + + +**限流说明：** 本接口频繁调用会被限流，频率请勿超过 10 次/秒。如遇限流，请稍后重试。 + +## 调试 + +[您可以在OpenAPI Explorer中直接运行该接口，免去您计算签名的困扰。运行成功后，OpenAPI Explorer可以自动生成SDK代码示例。](https://api.aliyun.com/api/bailian/2023-12-29/AddChunk) + + [![](https://img.alicdn.com/tfs/TB16JcyXHr1gK0jSZR0XXbP8XXa-24-26.png) 调试](https://api.aliyun.com/api/bailian/2023-12-29/AddChunk) + +## **授权信息** + +当前API暂无授权信息透出。 + +## 请求语法 + +``` +POST /{WorkspaceId}/chunk/create HTTP/1.1 +``` + +## 路径参数 + +**名称** + +**类型** + +**必填** + +**描述** + +**示例值** + +WorkspaceId + +string + +是 + +工作区标识 + +llm-19hxxxxx7htdf9lh + +## 请求参数 + +**名称** + +**类型** + +**必填** + +**描述** + +**示例值** + +PipelineId + +string + +是 + +知识库 id + +79c0alxxxx + +dataId + +string + +否 + +文件 id + +doc\_xxx + +field + +object + +否 + +插入的切片内容信息，以键值对形式传入。文档搜索类知识库使用固定 key 列表： + +- content（**String**）：**必填**，切片正文内容 + +- title（**String**）**选填**，切片标题 + +- image\_urls（**Array**）：**选填**，切片包含的图片链接，最多 10 张 + + +数据查询类、图片问答类知识库 key 不固定，由该知识库的数据源表格决定：key 为 Excel 列标题，value 为对应列的值。 + +{ "content": "The Bailian platform supports parsing multiple document formats including PDF, Word, and PPT.", "title": "Document Parsing and Chunking", "image\_urls": \[ "https://example.com/images/chunk-flow.png", "https://example.com/images/parsing-result.png" \] } + +any + +否 + +插入切片的表头字段信息，仅数据查询类与图片问答类知识库支持。需要参与检索或参与回复的表头为必填。各类型取值要求： + +- **String 类型** ：最大长度 6000 + +- **时间 类型**：13 位时间戳（毫秒） + +- **Long 类型**：整数，最大 2147483647 + +- **Double 类型**：支持小数 + +- **image\_url 类型**：最多 5 张，多张用英文逗号拼接为一个字符串 + + +{"Product Name": "Wireless Bluetooth Headphones", "Publish Time": 1752624000000, "Stock Quantity": 1580, "Unit Price": 299.99, "image\_url":"https://example.com/images/headphones-front.jpg,https://example.com/images/headphones-side.jpg,https://example.com/images/headphones-package.jpg" } + +## **返回参数** + +**名称** + +**类型** + +**描述** + +**示例值** + +object + +Schema of Response + +RequestId + +string + +请求 id + +35A267BF-xxxx-54DB-8394-AA3B0742D833 + +Code + +string + +错误状态码 + +Index.InvalidParameter + +Message + +string + +错误信息 + +Required parameter(%s) missing or invalid, please check the request parameters. + +Success + +boolean + +接口调用是否成功 + +**枚举值：** + +- true : + + true + +- false : + + false + + +true + +Data + +boolean + +请求成功返回的业务数据 + +**枚举值：** + +- true : + + true + +- false : + + false + + +true + +Status + +string + +接口返回的状态码 + +200 + +## 示例 + +正常返回示例 + +`JSON`格式 + +``` +{ + "RequestId": "35A267BF-xxxx-54DB-8394-AA3B0742D833", + "Code": "Index.InvalidParameter", + "Message": "Required parameter(%s) missing or invalid, please check the request parameters.", + "Success": true, + "Data": true, + "Status": "200" +} +``` + +## 错误码 + +访问[错误中心](https://api.aliyun.com/document/bailian/2023-12-29/errorCode)查看更多错误码。 + +## **变更历史** + +更多信息，参考[变更详情](https://api.aliyun.com/document/bailian/2023-12-29/AddChunk#workbench-doc-change-demo)。 diff --git a/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-overview.md b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-overview.md index 70c29639..8d6507eb 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-overview.md +++ b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-overview.md @@ -373,3 +373,9 @@ API概述 申请临时文件上传许可 该接口用于高代码部署，其他场景暂不支持。用于申请临时文件上传许可，之后需要自己完成文件上传动作。 + +[AddChunk](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addchunk) + +新增切片 + +使用此API可为文档搜索类（document）、数据查询类（table）、图片问答类（image）知识库添加切片。 diff --git a/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-ram.md b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-ram.md index fa10b5cc..ef5ee20d 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-ram.md +++ b/skills/bailian-docs-llm-wiki/raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-ram.md @@ -78,11 +78,11 @@ **关联操作** -sfm:ChangeParseSetting +sfm:ListCategory -[ChangeParseSetting](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-changeparsesetting) +[ListCategory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listcategory) -update +list \*全部资源 @@ -92,11 +92,11 @@ update 无 -sfm:UpdateFileTag +sfm:GetIndexJobStatus -[UpdateFileTag](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatefiletag) +[GetIndexJobStatus](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getindexjobstatus) -update +get \*全部资源 @@ -106,11 +106,11 @@ update 无 -sfm:DeleteCategory +sfm:AddCategory -[DeleteCategory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletecategory) +[AddCategory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addcategory) -delete +create \*全部资源 @@ -120,11 +120,11 @@ delete 无 -sfm:UpdatePromptTemplate +sfm:GetAlipayUrl -[UpdatePromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updateprompttemplate) +[GetAlipayUrl](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getalipayurl) -update +none \*全部资源 @@ -134,11 +134,11 @@ update 无 -sfm:SubmitIndexJob +sfm:DeleteMemoryNode -[SubmitIndexJob](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-submitindexjob) +[DeleteMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletememorynode) -create +delete \*全部资源 @@ -162,11 +162,11 @@ get 无 -sfm:DeleteChunk +sfm:SubmitIndexJob -[DeleteChunk](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletechunk) +[SubmitIndexJob](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-submitindexjob) -delete +create \*全部资源 @@ -176,11 +176,11 @@ delete 无 -sfm:GetAlipayTransferStatus +sfm:DeleteCategory -[GetAlipayTransferStatus](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getalipaytransferstatus) +[DeleteCategory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletecategory) -none +delete \*全部资源 @@ -190,11 +190,11 @@ none 无 -sfm:DeletePromptTemplate +sfm:ListIndexFileDetails -[DeletePromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deleteprompttemplate) +[ListIndexFileDetails](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listindexfiledetails) -delete +list \*全部资源 @@ -204,11 +204,11 @@ delete 无 -sfm:ListIndex +sfm:UpdateConnector -[ListIndices](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listindices) +[UpdateConnector](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updateconnector) -list +update \*全部资源 @@ -218,11 +218,11 @@ list 无 -sfm:CreateMemoryNode +sfm:ChangeParseSetting -[CreateMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-creatememorynode) +[ChangeParseSetting](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-changeparsesetting) -create +update \*全部资源 @@ -232,11 +232,11 @@ create 无 -sfm:UpdateMemory +sfm:GetMemory -[UpdateMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatememory) +[GetMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getmemory) -update +get \*全部资源 @@ -246,11 +246,11 @@ update 无 -sfm:ChunkList +sfm:GetAvailableParserTypes -[ListChunks](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listchunks) +[GetAvailableParserTypes](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getavailableparsertypes) -list +get \*全部资源 @@ -260,11 +260,11 @@ list 无 -sfm:ListIndexFileDetails +sfm:UpdateMemory -[ListIndexFileDetails](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listindexfiledetails) +[UpdateMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatememory) -list +update \*全部资源 @@ -274,9 +274,9 @@ list 无 -sfm:SubmitIndexAddDocumentsJob +sfm:CreateMemory -[SubmitIndexAddDocumentsJob](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-submitindexadddocumentsjob) +[CreateMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-creatememory) create @@ -288,11 +288,11 @@ create 无 -sfm:UpdateIndex +sfm:ApplyFileUploadLease -[UpdateIndex](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updateindex) +[ApplyFileUploadLease](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-applyfileuploadlease) -update +none \*全部资源 @@ -316,25 +316,11 @@ list 无 -sfm:Retrieve - -[Retrieve](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-retrieve) - -none - -\*全部资源 - -`*****` - -无 - -无 - -sfm:AddConnector +sfm:DeleteConnector -[AddConnector](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addconnector) +DeleteConnector -create +delete \*全部资源 @@ -344,11 +330,11 @@ create 无 -sfm:ListMemories +sfm:BatchUpdateFileTag -[ListMemories](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listmemories) +[BatchUpdateFileTag](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-batchupdatefiletag) -list +update \*全部资源 @@ -372,11 +358,11 @@ create 无 -sfm:UpdateConnector +sfm:DeleteChunk -[UpdateConnector](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updateconnector) +[DeleteChunk](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletechunk) -update +delete \*全部资源 @@ -386,9 +372,9 @@ update 无 -sfm:DeleteFile +sfm:DeleteMemory -[DeleteFile](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletefile) +[DeleteMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletememory) delete @@ -400,11 +386,11 @@ delete 无 -sfm:AddCategory +sfm:DeletePromptTemplate -[AddCategory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addcategory) +[DeletePromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deleteprompttemplate) -create +delete \*全部资源 @@ -414,11 +400,11 @@ create 无 -sfm:BatchUpdateFileTag +sfm:ListMemories -[BatchUpdateFileTag](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-batchupdatefiletag) +[ListMemories](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listmemories) -update +list \*全部资源 @@ -428,11 +414,11 @@ update 无 -sfm:ListIndexFiles +sfm:AddConnector -[ListIndexDocuments](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listindexdocuments) +[AddConnector](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addconnector) -list +create \*全部资源 @@ -442,11 +428,11 @@ list 无 -sfm:ListMemoryNodes +sfm:GetConnector -[ListMemoryNodes](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listmemorynodes) +[GetConnector](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getconnector) -list +get \*全部资源 @@ -456,11 +442,11 @@ list 无 -sfm:DescribeFile +sfm:UpdatePromptTemplate -[DescribeFile](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-describefile) +[UpdatePromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updateprompttemplate) -none +update \*全部资源 @@ -470,11 +456,11 @@ none 无 -sfm:CreateIndex +sfm:DeleteFiles -[CreateIndex](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-createindex) +[DeleteFiles](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletefiles) -create +delete \*全部资源 @@ -484,11 +470,11 @@ create 无 -sfm:ListCategory +sfm:GetMemoryNode -[ListCategory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listcategory) +[GetMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getmemorynode) -list +get \*全部资源 @@ -498,11 +484,11 @@ list 无 -sfm:AddTable +sfm:ApplyTempStorageLease -[AddTable](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addtable) +[ApplyTempStorageLease](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-applytempstoragelease) -create +none \*全部资源 @@ -512,11 +498,11 @@ create 无 -sfm:UpdateMemoryNode +sfm:DeleteFile -[UpdateMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatememorynode) +[DeleteFile](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletefile) -update +delete \*全部资源 @@ -526,11 +512,11 @@ update 无 -sfm:GetAvailableParserTypes +sfm:AddFilesFromAuthorizedOss -[GetAvailableParserTypes](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getavailableparsertypes) +[AddFilesFromAuthorizedOss](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addfilesfromauthorizedoss) -get +create \*全部资源 @@ -540,11 +526,11 @@ get 无 -sfm:DeleteMemory +sfm:CreateMemoryNode -[DeleteMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletememory) +[CreateMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-creatememorynode) -delete +create \*全部资源 @@ -554,11 +540,11 @@ delete 无 -sfm:DeleteConnector +sfm:UpdateChunk -DeleteConnector +[UpdateChunk](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatechunk) -delete +update \*全部资源 @@ -568,11 +554,11 @@ delete 无 -sfm:UpdateTableFromAuthorizedOss +sfm:ListCategory -[UpdateTableFromAuthorizedOss](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatetablefromauthorizedoss) +[ListCategory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listcategory) -update +list \*全部资源 @@ -582,11 +568,11 @@ update 无 -sfm:ApplyTempStorageLease +sfm:GetIndexJobStatus -[ApplyTempStorageLease](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-applytempstoragelease) +[GetIndexJobStatus](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getindexjobstatus) -none +get \*全部资源 @@ -596,9 +582,9 @@ none 无 -sfm:GetConnector +sfm:GetParseSettings -[GetConnector](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getconnector) +[GetParseSettings](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getparsesettings) get @@ -610,11 +596,11 @@ get 无 -sfm:DeleteMemoryNode +sfm:GetAlipayUrl -[DeleteMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletememorynode) +[GetAlipayUrl](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getalipayurl) -delete +none \*全部资源 @@ -624,11 +610,11 @@ delete 无 -sfm:GetMemory +sfm:AddCategory -[GetMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getmemory) +[AddCategory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addcategory) -get +create \*全部资源 @@ -638,11 +624,11 @@ get 无 -sfm:DeleteIndex +sfm:SubmitIndexJob -[DeleteIndex](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deleteindex) +[SubmitIndexJob](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-submitindexjob) -none +create \*全部资源 @@ -652,9 +638,9 @@ none 无 -sfm:UpdateChunk +sfm:ChangeParseSetting -[UpdateChunk](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatechunk) +[ChangeParseSetting](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-changeparsesetting) update @@ -666,9 +652,9 @@ update 无 -sfm:DeleteIndexDocument +sfm:DeleteMemoryNode -[DeleteIndexDocument](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deleteindexdocument) +[DeleteMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletememorynode) delete @@ -680,11 +666,11 @@ delete 无 -sfm:GetMemoryNode +sfm:UpdateConnector -[GetMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getmemorynode) +[UpdateConnector](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updateconnector) -get +update \*全部资源 @@ -694,9 +680,9 @@ get 无 -sfm:GetIndexJobStatus +sfm:GetMemory -[GetIndexJobStatus](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getindexjobstatus) +[GetMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getmemory) get @@ -708,11 +694,11 @@ get 无 -sfm:ApplyFileUploadLease +sfm:ListIndexFileDetails -[ApplyFileUploadLease](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-applyfileuploadlease) +[ListIndexFileDetails](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listindexfiledetails) -none +list \*全部资源 @@ -722,11 +708,25 @@ none 无 -sfm:GetPromptTemplate +sfm:DeleteConnector -[GetPromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getprompttemplate) +DeleteConnector -get +delete + +\*全部资源 + +`*****` + +无 + +无 + +sfm:UpdateMemory + +[UpdateMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatememory) + +update \*全部资源 @@ -750,9 +750,79 @@ create 无 -sfm:GetAlipayUrl +sfm:ListFile -[GetAlipayUrl](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getalipayurl) +[ListFile](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listfile) + +list + +\*全部资源 + +`*****` + +无 + +无 + +sfm:DeleteChunk + +[DeleteChunk](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletechunk) + +delete + +\*全部资源 + +`*****` + +无 + +无 + +sfm:GetAvailableParserTypes + +[GetAvailableParserTypes](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getavailableparsertypes) + +get + +\*全部资源 + +`*****` + +无 + +无 + +sfm:DeleteCategory + +[DeleteCategory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletecategory) + +delete + +\*全部资源 + +`*****` + +无 + +无 + +sfm:ListMemories + +[ListMemories](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listmemories) + +list + +\*全部资源 + +`*****` + +无 + +无 + +sfm:ApplyFileUploadLease + +[ApplyFileUploadLease](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-applyfileuploadlease) none @@ -764,11 +834,11 @@ none 无 -sfm:CreatePromptTemplate +sfm:BatchUpdateFileTag -[CreatePromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-createprompttemplate) +[BatchUpdateFileTag](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-batchupdatefiletag) -create +update \*全部资源 @@ -778,9 +848,9 @@ create 无 -sfm:GetIndexMonitor +sfm:GetConnector -[GetIndexMonitor](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getindexmonitor) +[GetConnector](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getconnector) get @@ -792,6 +862,48 @@ get 无 +sfm:GetMemoryNode + +[GetMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getmemorynode) + +get + +\*全部资源 + +`*****` + +无 + +无 + +sfm:AddFile + +[AddFile](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addfile) + +create + +\*全部资源 + +`*****` + +无 + +无 + +sfm:AddConnector + +[AddConnector](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addconnector) + +create + +\*全部资源 + +`*****` + +无 + +无 + sfm:DeleteFiles [DeleteFiles](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletefiles) @@ -806,6 +918,48 @@ delete 无 +sfm:DeleteFile + +[DeleteFile](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletefile) + +delete + +\*全部资源 + +`*****` + +无 + +无 + +sfm:DeletePromptTemplate + +[DeletePromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deleteprompttemplate) + +delete + +\*全部资源 + +`*****` + +无 + +无 + +sfm:UpdatePromptTemplate + +[UpdatePromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updateprompttemplate) + +update + +\*全部资源 + +`*****` + +无 + +无 + sfm:AddFilesFromAuthorizedOss [AddFilesFromAuthorizedOss](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addfilesfromauthorizedoss) @@ -820,6 +974,174 @@ create 无 +sfm:CreateMemoryNode + +[CreateMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-creatememorynode) + +create + +\*全部资源 + +`*****` + +无 + +无 + +sfm:DeleteMemory + +[DeleteMemory](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deletememory) + +delete + +\*全部资源 + +`*****` + +无 + +无 + +sfm:UpdateChunk + +[UpdateChunk](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatechunk) + +update + +\*全部资源 + +`*****` + +无 + +无 + +sfm:ApplyTempStorageLease + +[ApplyTempStorageLease](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-applytempstoragelease) + +none + +\*全部资源 + +`*****` + +无 + +无 + +sfm:GetIndexMonitor + +[GetIndexMonitor](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getindexmonitor) + +get + +\*全部资源 + +`*****` + +无 + +无 + +sfm:Retrieve + +[Retrieve](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-retrieve) + +none + +\*全部资源 + +`*****` + +无 + +无 + +sfm:DeleteIndex + +[DeleteIndex](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deleteindex) + +none + +\*全部资源 + +`*****` + +无 + +无 + +sfm:ListMemoryNodes + +[ListMemoryNodes](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listmemorynodes) + +list + +\*全部资源 + +`*****` + +无 + +无 + +sfm:ListIndex + +[ListIndices](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listindices) + +list + +\*全部资源 + +`*****` + +无 + +无 + +sfm:UpdateMemoryNode + +[UpdateMemoryNode](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatememorynode) + +update + +\*全部资源 + +`*****` + +无 + +无 + +sfm:GetPromptTemplate + +[GetPromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getprompttemplate) + +get + +\*全部资源 + +`*****` + +无 + +无 + +sfm:CreateIndex + +[CreateIndex](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-createindex) + +create + +\*全部资源 + +`*****` + +无 + +无 + sfm:ListPromptTemplates [ListPromptTemplates](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listprompttemplates) @@ -834,6 +1156,160 @@ list 无 +sfm:DeleteIndexDocument + +[DeleteIndexDocument](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-deleteindexdocument) + +delete + +\*全部资源 + +`*****` + +无 + +无 + +sfm:SubmitIndexAddDocumentsJob + +[SubmitIndexAddDocumentsJob](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-submitindexadddocumentsjob) + +create + +\*全部资源 + +`*****` + +无 + +无 + +sfm:UpdateFileTag + +[UpdateFileTag](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatefiletag) + +update + +\*全部资源 + +`*****` + +无 + +无 + +sfm:UpdateTableFromAuthorizedOss + +[UpdateTableFromAuthorizedOss](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatetablefromauthorizedoss) + +update + +\*全部资源 + +`*****` + +无 + +无 + +sfm:ListIndexFiles + +[ListIndexDocuments](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listindexdocuments) + +list + +\*全部资源 + +`*****` + +无 + +无 + +sfm:UpdateIndex + +[UpdateIndex](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updateindex) + +update + +\*全部资源 + +`*****` + +无 + +无 + +sfm:DescribeFile + +[DescribeFile](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-describefile) + +none + +\*全部资源 + +`*****` + +无 + +无 + +sfm:ChunkList + +[ListChunks](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-listchunks) + +list + +\*全部资源 + +`*****` + +无 + +无 + +sfm:AddTable + +[AddTable](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-addtable) + +create + +\*全部资源 + +`*****` + +无 + +无 + +sfm:CreatePromptTemplate + +[CreatePromptTemplate](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-createprompttemplate) + +create + +\*全部资源 + +`*****` + +无 + +无 + +sfm:GetAlipayTransferStatus + +[GetAlipayTransferStatus](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getalipaytransferstatus) + +none + +\*全部资源 + +`*****` + +无 + +无 + ## 资源（Resource） 下表是_大模型服务平台百炼_定义的资源，这些资源可以在 RAM 权限策略语句的`Resource`元素中使用，用来授予对该资源执行具体操作的权限。 其中，资源 ARN 是资源在阿里云上的唯一标识。具体说明如下： diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-error-code.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-error-code.md index 8b077c85..f0d03758 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-error-code.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-error-code.md @@ -156,9 +156,9 @@ （1）官方音色： -- 参考官方文档：cosyvoice-v2 / cosyvoice-v3 / cosyvoice-v3-plus / cosyvoice-v3-flash 支持的官方音色参考[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)，qwen-tts-realtime / qwen3-tts 支持的官方音色参考[支持的音色](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide#422789c49bqqx)，sambert支持的音色参考[模型列表](https://help.aliyun.com/zh/model-studio/sambert-java-sdk#74cedcb97el0b)（去掉开头的"sambert-"和末尾的"-v1"后就是voice的取值）。 +- 参考官方文档：[支持的音色](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide#bac280ddf5a1u)，sambert支持的音色参考[模型列表](https://help.aliyun.com/zh/model-studio/sambert-java-sdk#74cedcb97el0b)（去掉开头的"sambert-"和末尾的"-v1"后就是voice的取值）。 -- 其他语音合成模型的音色都可以在多模态交互控制台上查看：在左侧**语音交互**配置区域选择对应的语音合成模型，点击右侧**语音交互体验**区域的右上角即可查看可用的音色列表。 +- 语音合成模型的音色都可以在多模态交互控制台上查看：在左侧**语音交互**配置区域选择对应的语音合成模型，点击右侧**语音交互体验**区域的右上角即可查看可用的音色列表。 （2）复刻音色，确认音色状态为“OK”后才能使用。查询方法参考[查询特定音色](https://help.aliyun.com/zh/model-studio/cosyvoice-clone-design-api#34490e5a2by7z)。 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-interaction-protocol.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-interaction-protocol.md index 83255676..2b6dec85 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-interaction-protocol.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-api-references/multimodal-interaction-protocol.md @@ -60,11 +60,11 @@ wss://dashscope.aliyuncs.com/api-ws/v1/inference 语音识别支持的模型包括：[Paraformer实时语音识别](https://help.aliyun.com/zh/model-studio/paraformer-real-time-speech-recognition-api-reference/)（Paraformer），[FUN-ASR实时语音识别](https://help.aliyun.com/zh/model-studio/fun-asr-real-time-speech-recognition-api-reference/)（FunASR），[千问3-ASR-Flash-Realtime](https://help.aliyun.com/zh/model-studio/real-time-speech-recognition-user-guide)（qwen3-asr-flash-realtime），多模态交互轻量版语音识别（AppSpecificASR-Realtime）。 -语音合成支持的模型包括：[语音合成CosyVoice-v2大模型](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)（cosyvoice-v2），[语音合成CosyVoice-v3-Flash大模型](https://help.aliyun.com/zh/model-studio/text-to-speech)（cosyvoice-v3-flash），[语音合成CosyVoice-v3-plus大模型](https://help.aliyun.com/zh/model-studio/text-to-speech)（cosyvoice-v3-plus），[语音合成CosyVoice-v3.5-Flash大模型](https://help.aliyun.com/zh/model-studio/text-to-speech)（cosyvoice-v3.5-flash），[语音合成CosyVoice-v3.5-Plus大模型](https://help.aliyun.com/zh/model-studio/text-to-speech)（cosyvoice-v3.5-plus），[千问3-TTS-Flash-Realtime](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide)（qwen3-tts），[千问3-TTS-Instruct-Flash-Realtime](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide)（qwen3-tts-instruct），[千问3-声音设计](https://help.aliyun.com/zh/model-studio/qwen-tts-voice-design)（qwen3-tts-vd），[千问3-声音复刻](https://help.aliyun.com/zh/model-studio/qwen-tts-voice-cloning)（qwen3-tts-vc），[Sambert语音合成](https://help.aliyun.com/zh/model-studio/sambert-speech-synthesis/)（sambert），多模态交互轻量版语音合成（AppSpecificTTS）。 +语音合成支持的模型包括：[语音合成CosyVoice-v2大模型](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)（cosyvoice-v2），[语音合成CosyVoice-v3-Flash大模型](https://help.aliyun.com/zh/model-studio/text-to-speech)（cosyvoice-v3-flash），[语音合成CosyVoice-v3-plus大模型](https://help.aliyun.com/zh/model-studio/text-to-speech)（cosyvoice-v3-plus），[语音合成CosyVoice-v3.5-Flash大模型](https://help.aliyun.com/zh/model-studio/text-to-speech)（cosyvoice-v3.5-flash），[语音合成CosyVoice-v3.5-Plus大模型](https://help.aliyun.com/zh/model-studio/text-to-speech)（cosyvoice-v3.5-plus），[Qwen-Audio-3.0-TTS-Plus](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide)（qwen-audio-3.0-tts-plus）、[Qwen-Audio-3.0-TTS-Flash](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide)（qwen-audio-3.0-tts-flash），[千问3-TTS-Flash-Realtime](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide)（qwen3-tts），[千问3-TTS-Instruct-Flash-Realtime](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide)（qwen3-tts-instruct），[千问3-声音设计](https://help.aliyun.com/zh/model-studio/qwen-tts-voice-design)（qwen3-tts-vd），[千问3-声音复刻](https://help.aliyun.com/zh/model-studio/qwen-tts-voice-cloning)（qwen3-tts-vc），[Sambert语音合成](https://help.aliyun.com/zh/model-studio/sambert-speech-synthesis/)（sambert），多模态交互轻量版语音合成（AppSpecificTTS）。 语音合成支持的音色，可以在控制台上选择了模型后，点击右侧语音交互体验区域的右上角查看音色列表。 -官方音色也可以参考官方文档：cosyvoice-v2 / cosyvoice-v3-plus / cosyvoice-v3-flash 支持的官方音色参考[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)，qwen3-tts 支持的官方音色参考[支持的音色](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide#422789c49bqqx)，sambert支持的音色参考[模型列表](https://help.aliyun.com/zh/model-studio/sambert-java-sdk#74cedcb97el0b)（去掉开头的"sambert-"和末尾的"-v1"后就是voice的取值）。 +官方音色也可以参考官方文档：[支持的音色](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide#bac280ddf5a1u)，sambert支持的音色参考[Sambert音色列表](https://help.aliyun.com/zh/model-studio/sambert-java-sdk)（去掉开头的"sambert-"和末尾的"-v1"后就是voice的取值）。 使用复刻音色时，确认复刻音色状态为"OK"后才能使用。查询方法参考[查询特定音色](https://help.aliyun.com/zh/model-studio/cosyvoice-clone-design-api#34490e5a2by7z)。 @@ -363,6 +363,14 @@ string 热词id，设置该参数时会覆盖管控台热词配置。当管控台提供的热词不能满足客户需求时，可以考虑用Open API程序化管理热词，参见[热词API文档](https://next.api.aliyun.com/document/MultimodalDialog/2025-09-03/Vocabulary)。 +language + +string + +否 + +语音识别语种，默认和控制台选择的语言保持一致。 + **parameters.downstream**的参数说明如下： **一级参数** @@ -520,9 +528,15 @@ string 否 -设置指令，用于控制方言、情感等合成效果。该功能适用于qwen3-tts-instruct-flash-realtime、cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-plus、cosyvoice-v3-flash。 +设置指令，用于控制方言、情感等合成效果。该功能适用的模型以及在不同模型的格式要求请参见[指令控制](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide#12884a10929p9)。 -instruction有固定格式要求，具体格式参考[Java SDK](https://help.aliyun.com/zh/model-studio/cosyvoice-java-sdk)里的"instruction"参数说明。 +language + +string + +否 + +语音合成语种，默认和控制台选择的语言保持一致。 **parameters.client\_info**的参数说明如下： @@ -1966,6 +1980,26 @@ object 与Start消息中biz\_params相同，传递对话系统自定义参数。UpdateInfo指令中biz\_params下面每个子项会全量替换Start指令中biz\_params下面的同名项，并在本次连接后续所有对话中生效。 +upstream + +language + +String + +否 + +语音识别语种更新，不配置则语种不变。 + +downstream + +language + +String + +否 + +语音合成语种更新，不配置则语种不变。 + 示例如下： ``` diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-timbre-list.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-timbre-list.md index 593e3a17..ee39a50f 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-timbre-list.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/multimodal-guidelines/multimodal-timbre-list.md @@ -102,6 +102,8 @@ pcm ## **音色列表** +套件支持龙火火、龙汪汪、龙川叔等专有亮点音色，适配儿童陪伴、情感闲聊等场景，实现沉浸式互动。 + ### **CosyVoice-v3-Flash大模型** **适用场景** @@ -122,55 +124,51 @@ pcm 备注 -**标杆音色** - -龙安欢 - -longanhuan - -20-30岁 - -**欢脱元气女** +消费电子-儿童陪伴 -你看我刚从食堂买的豆沙包！还是热乎的，我特意多买了一个给你，甜而不腻超好吃。对了对了，下午课间咱们一起去操场晒晒太阳吧，今天天气超级好！ +龙火火 -中英双语 +longhuohuo\_v3 -龙安洋 +年龄6-10岁 -longanyang +桀骜不驯男童 -20-30岁 +我才不会被你们束缚，我的自由谁也夺不走。不管你们用什么方法，都别想 -**阳光大男孩** +困住我，我火火可不是好欺负的，你们别想控制我，我就是要做我自己，谁 -我呀，周末打算去书店逛逛，听说新到了一批科普杂志，你要是有空的话，咱们可以一起去，看完还能顺路去旁边的甜品店坐会儿。哦对了，你上次说想买的那支笔，我路过文具店帮你带了，试试好不好用。 +也拦不住，不服就来战。 中英双语 -龙呼呼 +多模态交互开发套件专属音色 -longhuhu\_v3 +龙汪汪 -年龄6-10岁 +longwangwang\_v3 -**天真烂漫女童** +年龄6-15岁 -喂，你有没有想过，如果我们能飞到更高的地方，是不是就能够看到整个大陆的全貌了呢！呼呼想想就觉得好激动啊！ +台湾少年音 + +大家好！我是专为3至12岁儿童设计的智能陪伴机器人，集语音对话、中英双语故事、儿歌播放、习惯养成与情绪互动于一体。我采用安全环保材质，具备护眼屏幕和家长远程管控功能，用温暖声音和趣味内容，全天候陪伴孩子快乐学习、健康成长！ 中英双语 -消费电子-儿童陪伴 +多模态交互开发套件专属音色 -龙汪汪 +方言 -longwangwang\_v3 +龙川叔 -年龄6-15岁 +longchuanshu\_v3 -**台湾少年音** +35-40岁 -大家好！我是专为3至12岁儿童设计的智能陪伴机器人，集语音对话、中英双语故事、儿歌播放、习惯养成与情绪互动于一体。我采用安全环保材质，具备护眼屏幕和家长远程管控功能，用温暖声音和趣味内容，全天候陪伴孩子快乐学习、健康成长！ +油腻搞笑叔 + +看清楚噻，过了这条天梯路，就是传说中的灵秀打卡地了，怕是要搞快点拍。最近天庭在搞交通整顿，乱飘的祥云都要遭扣分，你可别踩错云头了哈，那边是嫦娥的私人跑道。 中英双语 @@ -184,7 +182,7 @@ longanshuo\_v3 20-25岁 -**干净清爽男** +干净清爽男 那个……要是你搬东西需要帮忙，随时叫我，我力气还挺大的，之前帮同学搬宿舍，好几箱书都是我扛上去的，别客气啊。 @@ -200,7 +198,7 @@ longfeifei\_v3 20-25 -**甜美娇气女** +甜美娇气女 你最近是不是又熬夜了？明天我给你带热牛奶，顺便给你看我收藏的护眼食谱，保证让你眼睛亮晶晶的！ @@ -208,6 +206,44 @@ longfeifei\_v3 多模态交互开发套件专属音色 +标杆音色 + +龙安欢 + +longanhuan + +20-30岁 + +欢脱元气女 + +你看我刚从食堂买的豆沙包！还是热乎的，我特意多买了一个给你，甜而不腻超好吃。对了对了，下午课间咱们一起去操场晒晒太阳吧，今天天气超级好！ + +中英双语 + +龙安洋 + +longanyang + +20-30岁 + +阳光大男孩 + +我呀，周末打算去书店逛逛，听说新到了一批科普杂志，你要是有空的话，咱们可以一起去，看完还能顺路去旁边的甜品店坐会儿。哦对了，你上次说想买的那支笔，我路过文具店帮你带了，试试好不好用。 + +中英双语 + +龙呼呼 + +longhuhu\_v3 + +年龄6-10岁 + +天真烂漫女童 + +喂，你有没有想过，如果我们能飞到更高的地方，是不是就能够看到整个大陆的全貌了呢！呼呼想想就觉得好激动啊！ + +中英双语 + 陪伴闲聊 龙安智 @@ -216,7 +252,7 @@ longanzhi\_v3 25-35岁 -**睿智轻熟男** +睿智轻熟男 各位听众朋友们，欢迎收听今天的播客，今天咱们来探讨一下人生选择这个话题，每个人在人生的不同阶段都会面临各种选择，希望我分享的一些观点，能给你们带来一些启发。 @@ -230,7 +266,7 @@ longanqin\_v3 20-25岁 -**亲和活泼女** +亲和活泼女 哎！你们听说了吗？楼下新开了家奶茶店，装修得特别清新，还有好多新奇的口味，像什么杨枝甘露爆珠茶、草莓奶冻茶，听着就想喝。咱们下班一起去试试呗，要是好喝的话，以后下午茶就有着落啦！ @@ -244,7 +280,7 @@ longanling\_v3 20-30岁 -**思维灵动女** +思维灵动女 欢迎大家来到今天的播客节目，今天咱们要聊的话题是关于职场成长的，我会跟大家分享一些实用的小技巧，希望能帮助大家在工作中少走一些弯路，更快地实现自己的目标。 @@ -258,7 +294,7 @@ longanya\_v3 25-35岁 -**高雅气质女** +高雅气质女 周末在家泡一壶茶，选一本喜欢的书，坐在窗边晒晒太阳，阳光洒在身上暖暖的，茶香伴着书香，这样安安静静的时光，感觉心里特别舒服。偶尔还会写几笔毛笔字，让自己的节奏慢下来，享受这份惬意。 @@ -272,7 +308,7 @@ longanwen\_v3 25-35岁 -**优雅知性女** +优雅知性女 晚上好，现在为您播放舒缓的音乐，帮助您放松身心，进入睡眠状态，如果您有其他需求，比如设置明天的闹钟，随时跟我说。 @@ -286,7 +322,7 @@ longanyun\_v3 30-35岁 -**居家暖男** +居家暖男 您已经辛苦一天了，我已经为您调整好室内温度，还准备了您喜欢的饮品，您可以放松休息一下，有任何需要，都可以叫我哦。 @@ -300,7 +336,7 @@ longjiqi\_v3 20-30岁 -**呆萌机器人** +呆萌机器人 主人，我刚才查了天气，明天会下雨哦，雨量还不小，已经帮您把雨伞放在门口的挂钩上啦，可别忘记带呀！另外我还帮您把明天要穿的外套叠好了，放在床头，这样早上出门就能省点时间啦。 @@ -314,7 +350,7 @@ longhouge\_v3 20-25岁 -**经典猴哥** +经典猴哥 放心！这点小事难不倒俺老孙，包在我身上，保证给你办得妥妥当当的！你就等着好消息就行，要是有谁敢从中作梗，看俺老孙的金箍棒不收拾他，保管让他服服帖帖的！ @@ -328,7 +364,7 @@ longanlang\_v3 20-25岁 -**清爽利落男** +清爽利落男 主人，您需要查询的天气信息已经找到，今日气温在20-28摄氏度，适合穿薄外套，另外，您关注的股票今日涨幅为2%，需要为您详细播报吗？ @@ -342,7 +378,7 @@ longyingmu\_v3 25-30岁 -**优雅知性女** +优雅知性女 您好！我是智能电话助手，很高兴为您服务。请问您需要咨询业务、预约办理，还是查询信息？为提升效率，部分事项将通过短信或官方APP同步发送，请注意查收。感谢您的理解与支持，祝您生活愉快！ @@ -356,7 +392,7 @@ longanli\_v3 25-35岁 -**利落从容女** +利落从容女 主人，现在为您播报今日的日程安排，上午九点有一个重要会议，下午三点需要处理文件，我已经为您准备好相关资料，如有调整，随时告诉我。 @@ -370,7 +406,7 @@ longwanjun\_v3 20-30岁 -**细腻柔声女** +细腻柔声女 苏晚坐在窗边，手指轻轻摩挲着那本泛黄的日记，纸页上娟秀的字迹还带着当年的温度。她想起昨天在老宅阁楼里发现它时的场景，日记本被藏在一个旧木盒里，旁边还放着一枚生锈的银簪，簪头刻着小小的‘辰’字。 @@ -384,7 +420,7 @@ longyichen\_v3 20-30岁 -**洒脱活力男** +洒脱活力男 他刚避开三头青鳞兽的围攻，转身就在峡谷深处发现了传说中的赤血晶矿，那矿石在暗处泛着红光，看得人眼睛都直了！可没等他高兴多久，矿洞深处突然传来一声震耳欲聋的咆哮，好家伙，竟是一头沉睡百年的紫金巨兽被惊醒了，林辰这一战，怕是要拿出压箱底的本事了！ @@ -398,7 +434,7 @@ longlaobo\_v3 60岁以上 -**沧桑岁月爷** +沧桑岁月爷 咱们今天聊《红楼梦》里的‘刘姥姥进大观园’，这段写得妙啊，妙就妙在把小人物的通透和豪门的繁华对比得淋漓尽致。刘姥姥第一次进荣国府，见着什么都新鲜，可她不卑不亢，说话既风趣又懂分寸，既讨了贾府上下的喜欢，又没丢了自己的骨气。 @@ -412,7 +448,7 @@ longlaoyi\_v3 60岁以上 -**烟火从容阿姨** +烟火从容阿姨 咱们接着读《城南旧事》，英子跟着妈妈去惠安馆，远远就看见疯女人秀贞站在门口，头发有些乱，可眼睛亮得很。英子不怕她，反而慢慢走过去，从口袋里掏出一颗糖递过去，就像对待普通的阿姨一样。 @@ -426,7 +462,7 @@ longyingxiao\_v3 20-25岁 -**清甜推销女** +清甜推销女 您好呀，请问是赵女士吗？我是这个品牌的客服，咱们品牌最近推出了一款特别实用的产品，特别适合您这样的人群，想跟您简单介绍一下，您看方便吗？ @@ -440,7 +476,7 @@ longyingxun\_v3 20-25岁 -**年轻青涩男** +年轻青涩男 您好！我是您的专属客服专员，工号：106857，很高兴为您服务。请问有什么可以帮您？为提升效率，操作链接将通过短信发送到您的手机，请注意查收。感谢您的耐心与理解，我们将竭诚为您提供专业、高效、贴心的服务！ @@ -482,7 +518,7 @@ longdaiyu\_v3 年龄15-25岁 -**娇率才女音** +娇率才女音 这世间，原就没什么圆满。银幕之上，不过又是一场痴梦罢了。那主角拼尽全力追逐的，不过是一缕抓不住的风；那轰轰烈烈的情爱，终究敌不过命运轻轻一叹。我冷眼瞧着，倒觉心酸——今天我们就一起来看2025年最新的爱情悲剧悬疑片。 @@ -496,7 +532,7 @@ longanyue\_v3 25-35岁 -**欢脱粤语男** +欢脱粤语男 各位老友记，今日同大家讲下呢个有趣嘅故事，里面嘅情节好搞笑，保证大家听咗会笑出声，一齐来听听啦！ @@ -510,7 +546,7 @@ longshange\_v3 25-35岁 -**原味陕北男** +原味陕北男 额今儿个嫽扎咧！咥了一老碗油泼面，再喝口冰峰汽水，美得很！这日子过得跟秦腔一样有味儿，嫽就一个字，不谝闲传，实诚得很！ @@ -524,7 +560,7 @@ longanmin\_v3 18-25 -**清纯萝莉女** +清纯萝莉女 阮今仔日真欢喜，食了一碗热腾腾的卤肉饭，再配一罐黑松沙士，爽快到毋知讲啥！这阵日子过得真有“古意”，亲像听一出歌仔戏，甜中带苦，苦底又透甘。恁若得闲，就来阮兜坐一下，咱来“斗阵”讲讲话、笑笑，毋通客气啦！ @@ -538,7 +574,7 @@ loongbella\_v3 25-30 -**精准干练女** +精准干练女 体育界传来喜讯，中国女子足球队在刚刚结束的世界锦标赛中表现出色，最终夺得了亚军的好成绩。球队成员们凭借顽强拼搏的精神赢得了国内外球迷的高度评价。 @@ -890,7 +926,7 @@ longantai 中英双语 -**嗲甜台湾女** +嗲甜台湾女 柔顺的长发，穿着浅粉色的针织衫，整体造型甜美可爱。 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/product-billing.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/product-billing.md index 587f5b5c..0d68fb2b 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/product-billing.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/multimodal-products/product-billing.md @@ -171,7 +171,7 @@ Fun-ASR、通义千问3-ASR-Flash-Realtime 语音合成 -CosyVoice-v3-Plus、通义千问3-TTS 系列 +Qwen-Audio-3.0-TTS-Plus、Qwen-Audio-3.0-TTS-Flash、CosyVoice-v3.5-Plus、CosyVoice-v3.5-Flash、CosyVoice-v3-Plus、通义千问3-TTS 系列 3x diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/lingque-ccai-dialogue-analytics-ram-subaccount-usage-and-authorization-procedures.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/use-and-authorize-ram-users-for-ccai-dialogue-analysis.md similarity index 100% rename from skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/lingque-ccai-dialogue-analytics-ram-subaccount-usage-and-authorization-procedures.md rename to skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-lingque-ccai-dialogue-analysis-aio/call-tyxm-ccai-aio-api/use-and-authorize-ram-users-for-ccai-dialogue-analysis.md diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-lingque-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listvoices.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-lingque-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listvoices.md new file mode 100644 index 00000000..f04658b4 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-lingque-ccai-voice-dialogue-robot/api-reference-chat6/api-bailianvoicebot-2025-01-01-dir/api-bailianvoicebot-2025-01-01-dir-application-management/api-bailianvoicebot-2025-01-01-listvoices.md @@ -0,0 +1,349 @@ +# ListVoices - 获取音色列表 + +获取音色列表 + +## 调试 + +[您可以在OpenAPI Explorer中直接运行该接口，免去您计算签名的困扰。运行成功后，OpenAPI Explorer可以自动生成SDK代码示例。](https://api.aliyun.com/api/BailianVoiceBot/2025-01-01/ListVoices) + + [![](https://img.alicdn.com/tfs/TB16JcyXHr1gK0jSZR0XXbP8XXa-24-26.png) 调试](https://api.aliyun.com/api/BailianVoiceBot/2025-01-01/ListVoices) + +## **授权信息** + +下表是API对应的授权信息，可以在RAM权限策略语句的`Action`元素中使用，用来给RAM用户或RAM角色授予调用此API的权限。具体说明如下： + +- 操作：是指具体的权限点。 + +- 访问级别：是指每个操作的访问级别，取值为写入（Write）、读取（Read）或列出（List）。 + +- 资源类型：是指操作中支持授权的资源类型。具体说明如下： + + - 对于必选的资源类型，用前面加 \* 表示。 + + - 对于不支持资源级授权的操作，用`全部资源`表示。 + +- 条件关键字：是指云产品自身定义的条件关键字。 + +- 关联操作：是指成功执行操作所需要的其他权限。操作者必须同时具备关联操作的权限，操作才能成功。 + + +**操作** + +**访问级别** + +**资源类型** + +**条件关键字** + +**关联操作** + +bailianvoicebot:ListVoices + +list + +\*全部资源 + +`*` + +无 + +无 + +## 请求参数 + +**名称** + +**类型** + +**必填** + +**描述** + +**示例值** + +BusinessUnitId + +string + +否 + +百炼业务空间 ID + +llm-c11iig67g863rih8 + +PageNumber + +integer + +否 + +页号 + +1 + +PageSize + +integer + +否 + +每页条数(固定值 1000，不可修改) + +1000 + +NlsAccessType + +string + +否 + +TTS 调用方式 + +**枚举值：** + +- PROVIDED : + + PROVIDED + +- MANAGED : + + MANAGED + + +MANAGED + +NlsEngine + +string + +否 + +TTS 引擎 + +**枚举值：** + +- IFLYTEK : + + IFLYTEK + +- VOLC : + + VOLC + +- BAILIAN : + + BAILIAN + + +BAILIAN + +## **返回参数** + +**名称** + +**类型** + +**描述** + +**示例值** + +object + +Schema of Response + +RequestId + +string + +请求 ID + +CF6D3484-19A1-5C77-863B-AC8B5754D37C + +Code + +string + +内部错误码 + +OK + +HttpStatusCode + +integer + +http 状态码 + +200 + +Message + +string + +错误信息 + +Instance llm-xdne77rxe14ziszr does not exist. + +Data + +object + +音色分页结果 + +PageNumber + +integer + +页号 + +1 + +PageSize + +integer + +每页条数 + +10 + +TotalCount + +integer + +总数量 + +100 + +Voices + +array + +音色列表 + +object + +音色 + +Category + +string + +类目 + +通义千问音色 + +Name + +string + +音色名称 + +芊悦 + +Style + +string + +风格 + +女声 + +Voice + +string + +发音人 + +Cherry + +Language + +string + +支持语言 + +中文及中英文混合 + +NlsEngine + +string + +TTS 引擎 + +BAILIAN + +Model + +string + +TTS 模型 + +Qwen + +SupportedParams + +array + +支持参数列表 + +string + +支持参数 + +speechRate + +Params + +array + +动态错误参数列表 + +string + +动态错误参数 + +llm-xdne77rxe14ziszr + +## 示例 + +正常返回示例 + +`JSON`格式 + +``` +{ + "RequestId": "CF6D3484-19A1-5C77-863B-AC8B5754D37C", + "Code": "OK", + "HttpStatusCode": 200, + "Message": "Instance llm-xdne77rxe14ziszr\n does not exist.", + "Data": { + "PageNumber": 1, + "PageSize": 10, + "TotalCount": 100, + "Voices": [ + { + "Category": "通义千问音色", + "Name": "芊悦", + "Style": "女声", + "Voice": "Cherry", + "Language": "中文及中英文混合", + "NlsEngine": "BAILIAN", + "Model": "Qwen", + "SupportedParams": [ + "speechRate" + ] + } + ] + }, + "Params": [ + "llm-xdne77rxe14ziszr" + ] +} +``` + +## 错误码 + +访问[错误中心](https://api.aliyun.com/document/BailianVoiceBot/2025-01-01/errorCode)查看更多错误码。 + +## **变更历史** + +更多信息，参考[变更详情](https://api.aliyun.com/document/BailianVoiceBot/2025-01-01/ListVoices#workbench-doc-change-demo)。 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-tongyi-translate/web-page-translation-jssdk.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-tongyi-translate/web-page-translation-jssdk.md index 90ebfa90..bcb1000c 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-tongyi-translate/web-page-translation-jssdk.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/official-application-tongyi-translate/web-page-translation-jssdk.md @@ -10,218 +10,633 @@ ## 接入JSSDK -#### **步骤一：在HTML插入脚本** +## 当前版本 -``` - - - - - - - -``` +项目 + +值 + +**版本号** + +`3.0.0` + +CDN 根路径 + +`https://g.alicdn.com/code/npm/@alife/translate-js-sdk/3.0.0/` + +下文示例中的版本号均以 `3.0.0` 为准，升级时请同步替换 CDN 链接中的版本号。 + +* * * + +SDK 提供两种翻译模式： + +- 页面翻译（`pageTranslate`）：直接替换页面原文为译文，适合“单语展示”场景。 + +- 段落对照翻译（`paragraphTranslate`）：在原文后插入译文，适合“原文+译文对照阅读”场景。 + + +同时支持： + +- 懒加载翻译（只翻译可视区域） + +- 动态内容翻译（DOM 变化后自动翻译） + +- 术语表与翻译记忆 + +- 细粒度 hooks（翻译前/后插入自定义逻辑） + -#### **步骤二：配置JSSDK** +* * * + +## 引入方式 + +通过 Script 标签从 CDN 引入，支持两种方式： + +### 方式一：全量引入（推荐上手） + +一次引入 `index.js`（已包含核心 + 全部插件），即可同时使用页面翻译与段落对照： + +文件 + +说明 + +地址 + +`index.js` + +全量包（核心 + 插件） + +[https://g.alicdn.com/code/npm/@alife/translate-js-sdk/3.0.0/index.js](https://g.alicdn.com/code/npm/@alife/translate-js-sdk/3.0.0/index.js) ``` -interface ITokenResponseData { - code: 200; - data: { - url: string, - host: string, - method: "POST", - headers: { - "host": string, - "x-acs-action": "BatchTranslateForHtml", - "x-acs-version": string, - "x-acs-date": string; - "x-acs-signature-nonce": string; - "content-type": "application/json", - "x-acs-content-sha256": string; - Authorization: string; - }; - body: string; - }; -} -interface ITokenRequestData { - sourceLanguage: string; - targetLanguage: string; - streaming: false, - text: {[index: number]: string} - scene: 'mt-turbo', - fallbackTimeoutMs: number; -} -interface ISetupConfig { - getToken: (data: ITokenRequestData) => Promise; -} -// 只用初始化一次，不需要每次翻译都初始化 -__AliTranslate.setup({ - getToken: async (data) => { - const res = await fetch('YOUR_GET_TOEKN_URL', { method: 'post', body: JSON.stringify(data) }); - return (await res.json()).data; - } -}); + ``` -#### **步骤三：调用页面翻译** +### 方式二：按需拆包引入 + +先加载 `core.js`，再按需加载对应插件脚本： + +文件 + +说明 + +地址 + +`core.js` -**仅展示译文** +核心 SDK + +[https://g.alicdn.com/code/npm/@alife/translate-js-sdk/3.0.0/core.js](https://g.alicdn.com/code/npm/@alife/translate-js-sdk/3.0.0/core.js) + +`page.js` + +页面翻译插件 + +[https://g.alicdn.com/code/npm/@alife/translate-js-sdk/3.0.0/page.js](https://g.alicdn.com/code/npm/@alife/translate-js-sdk/3.0.0/page.js) + +`paragraph.js` + +段落对照插件 + +[https://g.alicdn.com/code/npm/@alife/translate-js-sdk/3.0.0/paragraph.js](https://g.alicdn.com/code/npm/@alife/translate-js-sdk/3.0.0/paragraph.js) ``` -interface IPageTranslate { - srcLanguage?: string; // 默认语种，默认为auto - tgtLanguage?: string; // 目标语种，默认为en - lazyload?: boolean; // 是否只翻译可视区域，默认为false - lazyOffset?: number; // 可视区域的扩展 - target?: HTMLElement; // 要翻译的目标区域，默认为body - except?: string; // 要排除翻译的区域，默认为空 -} -const instance = __AliTranslate.pageTranslate({ - // 详细参数见PageTranslate参数 - lazyload: true, - lazyOffset: 500 -}); + + + + + + ``` -**双语对照翻译** +说明： + +- 全量引入只需一个文件；拆包引入必须先加载 `core.js`，再加载插件。 + +- 全局对象为 `window.AliTranslate`，同时提供别名 `window.__AliTranslate`（两者指向同一实例）。 + + +* * * + +## 快速开始 + +### 页面翻译（Pure Page） ``` -interface IParagraphTranslate { - srcLanguage?: string; // 默认语种，默认为auto - tgtLanguage?: string; // 目标语种，默认为en - target?: HTMLElement; // 要翻译的目标区域，默认为body - dynamic?: boolean; // 是否支持动态翻译，默认为false -} -const instance = __AliTranslate.paragraphTranslate({ - // 详细参数见ParagrapthTranslate参数 - srcLanguage: 'zh', - tgtLanguage: 'en' -}); + + ``` -#### **步骤四：取消翻译** +### 段落对照翻译（Paragraph Compare） ``` -// 触发取消翻译后，页面将回到原始文案 -instance.destroy(); + + ``` -### **PageTranslate参数** +* * * + +## 初始化参数（SDKConfig） + +`AliTranslate.pageTranslate` / `AliTranslate.paragraphTranslate` 中共用的 SDK 级参数如下： 参数 +类型 + +必填 + 说明 -类型 +`getToken` + +`(params) => Promise` + +是 -默认值 +鉴权签名函数，返回可直接请求翻译服务的签名请求 -srcLanguage +`terminologies` -默认的原语种 +`Array<{ src, tgt }>` -String +否 -auto（会根据页面文案自动选择） +术语表，提升领域词一致性 -tgtLanguage +`examples` -默认的目标语种 +`Array<{ src, tgt }>` -String +否 -en +翻译记忆 -lazyload +`domainHint` -是否只翻译可视区域 +`string` -Boolean +否 -false +业务领域（透传至翻译请求） -lazyOffset +`chunkSize` -当lazyload为true时，可视区域扩展 +`number` -Number +否 -\-1 (-1表示不做任何offset) +分块阈值（影响请求合并与吞吐） -target +`maxRetries` -要翻译的目标区域 +`number` -HTMLElement | HTMLElement\[\] +否 -body +翻译失败最大重试次数 -except +`afterTranslate` -不翻译的区域 +`(result) => void` -Selector(String) +否 -空 +翻译完成回调 -### **ParagraphTranslate参数** +* * * + +## 页面翻译参数（PageTranslateConfig） 参数 +类型 + +必填 + 说明 +`tgtLanguage` + +`LANGUAGE` + +是 + +目标语种 + +`srcLanguage` + +`LANGUAGE` + +否 + +源语种，不传默认自动检测 + +`targetSelectors` + +`string | string[]` + +否 + +翻译根节点，默认 `body` + +`excludeSelectors` + +`string | string[]` + +否 + +排除节点（命中后整棵子树不翻译） + +`lazyload` + +`boolean` + +否 + +是否启用懒加载翻译 + +`lazyOffset` + +`number` + +否 + +懒加载视口偏移（px） + +`lazySelectors` + +`string | string[]` + +否 + +懒翻译选择器（当前版本为预留字段，建议优先使用 `targetSelectors`） + +`dynamic` + +`boolean` + +否 + +是否监听 DOM 变化并自动翻译 + +`translateDelay` + +`number` + +否 + +初始化等待/动态翻译节流延迟（ms） + +`rules` + +`Array<{ selector; style?; className? }>` + +否 + +页面翻译规则（作用于原文容器） + +`onBeforeInsertTrans` + +`(ctx) => void | HTMLElement | Promise<...>` + +否 + +插入译文前回调 + +`onTranslatingStart` + +`(ctx) => void | Promise` + +否 + +单节点翻译开始回调 + +`onTranslatingEnd` + +`(ctx) => void | Promise` + +否 + +单节点翻译结束回调 + +* * * + +## 段落对照参数（ParagraphTranslateConfig） + +参数 + 类型 -默认值 +必填 + +说明 + +`tgtLanguage` + +`LANGUAGE` + +是 + +目标语种 + +`srcLanguage` + +`LANGUAGE` + +否 + +源语种，不传默认自动检测 + +`targetSelectors` + +`string | string[]` + +否 + +翻译根节点，默认 `body` + +`excludeSelectors` + +`string | string[]` + +否 + +排除节点（命中后不参与段落提取） + +`lazyload` + +`boolean` + +否 + +是否启用懒加载翻译 + +`lazyOffset` + +`number` + +否 + +懒加载视口偏移（px） + +`lazySelectors` + +`string | string[]` + +否 + +懒翻译选择器（当前版本为预留字段，建议优先使用 `targetSelectors`） -srcLanguage +`dynamic` -默认的原语种 +`boolean` -String +否 -auto（会根据页面文案自动选择） +是否监听 DOM 变化并自动翻译 -tgtLanguage +`translateDelay` -默认的目标语种 +`number` -String +否 -en +初始化等待/动态翻译节流延迟（ms） -target +`extraBlockSelectors` -要翻译的目标区域 +`string | string[]` -HTMLElement | HTMLElement\[\] +否 -body +强制按 block 方式插入译文（独占一行） -dynamic +`extraInlineSelectors` -是否动态翻译。如果开启，如果页面发生变化，会自动翻译。如果不开启，则只翻译一次 +`string | string[]` -Boolean +否 -false +强制按 inline 方式插入译文（同行展示） -lazyload +`rules` -是否只翻译可视区域 +`TransStyleRule[]` -Boolean +否 -false +对照翻译样式规则 -lazyOffset +`onBeforeInsertTrans` -当lazyload为true时，可视区域扩展 +`(ctx) => void | HTMLElement | Promise<...>` -Number +否 -300 +插入 `` 之前回调 + +`onTranslatingStart` + +`(ctx) => void | Promise` + +否 + +单段落翻译开始回调 + +`onTranslatingEnd` + +`(ctx) => void | Promise` + +否 + +单段落翻译结束回调 + +### `TransStyleRule` 说明 + +``` +interface TransStyleRule { + selector: string + transStyle?: Record // 作用于译文节点 + style?: Record // 作用于命中 selector 的父容器 + className?: string // 作用于命中 selector 的父容器 +} +``` + +* * * + +## Hooks 与可回滚 DOM 变更 + +SDK 提供了 `mutator`（可回滚变更操作器），支持在 hooks 中做安全 DOM 操作，并在 `removeTranslations()` 时自动回滚。 + +``` + + +``` + +`mutator` 可用能力： + +- `setStyle(el, prop, value, priority?)` + +- `addClass(el, className)` + +- `removeClass(el, className)` + +- `appendNode(parent, node)` + +- `removeNode(node)` + + +* * * + +## 语种与枚举值 + +常用语种值示例： + +- `auto`（自动检测） + +- `zh`、`en`、`ja`、`ko`、`fr`、`de`、`es`、`ru`、`ar`、`th`、`vi` + + +完整语言枚举请参考 SDK 类型定义 `LANGUAGE`。 + +* * * + +## 常用 API + +``` +// 移除翻译结果（保留 SDK 实例） +AliTranslate.removeTranslations() + +// 销毁 SDK（释放资源，插件与事件解绑） +AliTranslate.destroy() +``` + +* * * + +## 完整示例（段落对照 + rules + hooks） + +``` + + +``` + +* * * + +如需同时支持“页面翻译”和“段落对照翻译”，推荐直接使用全量包 `index.js`，并在业务 UI 中明确区分两种模式的切换与清理流程。 ## **获取Token服务** @@ -859,21 +1274,70 @@ public class NodeSignature { } ``` -## **常见问题** +## 注意事项（生产接入建议） + +- **必须配置** `**getToken**`：由业务侧完成鉴权签名后再发起翻译请求。 + +- **引入方式二选一**：全量引入 `index.js` 即可；若按需拆包，页面翻译需引入 `page.js`，段落对照需引入 `paragraph.js`。 + +- **重复翻译建议先清理**：多次切换参数/语言前建议先执行 `removeTranslations()`，避免旧状态干扰新配置验证。 + +- **目标与排除选择器要收敛**：`targetSelectors` 建议限定业务容器，`excludeSelectors` 排除导航、代码块、编辑区等。 + +- **避免翻译敏感区域**：如输入框、编辑器、业务脚本节点、模板容器。 + +- **Hook 里避免重逻辑**：`onTranslatingStart/End` 触发频率高，避免耗时操作阻塞主线程。 + +- **动态内容场景建议开启** `**dynamic**`：适配懒渲染和流式加载页面。 + + +## 常见问题 + +### 调用无效果 + +- 检查是否已正确引入脚本（全量 `index.js`，或 `core.js` + 对应插件）。 + +- 检查 `targetSelectors` 是否命中。 + +- 检查 `excludeSelectors` 是否误伤目标区域。 + +- 检查签名接口是否返回合法请求结构。 + + +### rules / hooks 未生效 -#### **1\. AK/SK在哪里获取？** +- 确认传参写在 `pageTranslate` / `paragraphTranslate` 的同一层级。 + +- 重复执行前先 `removeTranslations()` 再重新翻译。 + +- 检查 `selector` 是否可命中段落 `commonAncestor` 或其祖先。 + -在[阿里云](https://ram.console.aliyun.com/profile/access-keys)平台的AccessKey模块中获取 +### loading 看不到 -#### **2\. 获取Token服务中的workspaceId从哪里获取？** +- 翻译速度很快时会瞬时消失，建议在业务侧设置最小可见时长（如 300-500ms）。 + +- 用 `data-source-hash` 做唯一标识，避免并发段落相互覆盖。 + -登录AK/SK对应的阿里云账号后，在[百炼](https://bailian.console.aliyun.com/)左下角的业务空间详情中获取 +### **AK/SK在哪里获取？** -弹窗中的**业务空间id**即对应所需的workspaceId，单击字段右侧的复制图标可直接复制。 +- 在[阿里云](https://ram.console.aliyun.com/profile/access-keys)平台的AccessKey模块中获取 + -#### **3\. 网络接口调用报没有权限** +### **获取Token服务中的workspaceId从哪里获取？** -未开通通义多模态翻译产品：需要使用AK所属的阿里云主账号，在百炼通义多模态翻译上进行开通 +- 登录AK/SK对应的阿里云账号后，在[百炼](https://bailian.console.aliyun.com/)左下角的业务空间详情中获取 + +- 弹窗中的**业务空间id**即对应所需的workspaceId，单击字段右侧的复制图标可直接复制。 + + +### **网络接口调用报没有权限** + +- 未开通 通义多模态翻译 产品：需要使用AK所属的阿里云主账号，在百炼通义多模态翻译上进行开通 + + +* * * ## **变更记录** @@ -885,6 +1349,12 @@ public class NodeSignature { 发布内容 +3.0.0 + +2026年7月16日 + +大版本升级，增加按需引入、术语/语料/domain配置、样式干预、更多的配置参数 + 2.0.6 2026年3月5日 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-getvideoanalysistask.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-getvideoanalysistask.md index 83fdd1da..f0939963 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-getvideoanalysistask.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/quanmiao-light-application-series/development-documentation/api-reference-1/api-quanmiaolightapp-2024-08-01-dir/api-quanmiaolightapp-2024-08-01-dir-film-and-television-media-video-understanding/api-quanmiaolightapp-2024-08-01-getvideoanalysistask.md @@ -133,7 +133,7 @@ code string -状态码 +状态码：仅代表 http 请求状态 successful @@ -179,7 +179,7 @@ string 异常错误码 -success +ERR-00000000 errorMessage @@ -187,15 +187,22 @@ string 异常错误信息 -Deduct task already success,Please do not resubmit.token '369e8f2c-d283-424a-96c4-c83efe08c89e' +url download error: http://xx event string -事件类型 +事件类型：任务执行状态 + +- task-finished：任务结束（成功） + +- task-failed：任务结束（失败） + +- task-xxx：中间态，可以不关注 + -TIMEOUT\_CLOSE\_ORDER +task-finshed eventInfo @@ -203,7 +210,7 @@ string 事件描述 -xxx +url download error: http://xx sessionId @@ -1025,7 +1032,7 @@ taskStatus string -任务状态。 +任务状态：代表任务调度执行状态，如果任务状态 SUCCESSED，请查看 event 字段取真实业务状态。 - PENDING：待执行 @@ -1073,10 +1080,10 @@ Access was denied, message: No such namespace namespaces/mjp-test-default. "success": true, "data": { "header": { - "errorCode": "success", - "errorMessage": "Deduct task already success,Please do not resubmit.token '369e8f2c-d283-424a-96c4-c83efe08c89e'", - "event": "TIMEOUT_CLOSE_ORDER", - "eventInfo": "xxx", + "errorCode": "ERR-00000000", + "errorMessage": "url download error: http://xx", + "event": "task-finshed", + "eventInfo": "url download error: http://xx", "sessionId": "d5c38cf6-a4bf-4a57-a697-9f449926f0c9", "taskId": "6e223291-729b-4e84-9271-c13ada1a776b", "traceId": "215045f817272303448235204efdef" diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-overview.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-overview.md index 8f161902..d2d7873d 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-overview.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-gallery/quanmiao-solution-products/ai-quan-miao-development-document/amb-api-reference/api-aimiaobi-2023-08-01-overview.md @@ -462,24 +462,6 @@ API API概述 -[QueryAuditTask](https://help.aliyun.com/zh/model-studio/api-aimiaobi-2023-08-01-queryaudittask) - -查询审核结果 - -查询审核结果。 - -[SubmitAuditTask](https://help.aliyun.com/zh/model-studio/api-aimiaobi-2023-08-01-submitaudittask) - -提交审核任务 - -提交审核任务 - -[CancelAuditTask](https://help.aliyun.com/zh/model-studio/api-aimiaobi-2023-08-01-cancelaudittask) - -取消审核任务 - -取消审核任务 - [SubmitSmartAudit](https://help.aliyun.com/zh/model-studio/api-aimiaobi-2023-08-01-submitsmartaudit) 提交智能审校任务 @@ -1607,3 +1589,21 @@ SubmitParseDocumentLayoutTask 获取排版任务结果 获取排版任务结果 + +[CancelAuditTask](https://help.aliyun.com/zh/model-studio/api-aimiaobi-2023-08-01-cancelaudittask) + +取消审核任务 + +取消审核任务 + +[QueryAuditTask](https://help.aliyun.com/zh/model-studio/api-aimiaobi-2023-08-01-queryaudittask) + +查询审核结果 + +查询审核结果。 + +[SubmitAuditTask](https://help.aliyun.com/zh/model-studio/api-aimiaobi-2023-08-01-submitaudittask) + +提交审核任务 + +提交审核任务 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-support/application-after-sales-service-scope.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-support/application-after-sales-service-scope.md new file mode 100644 index 00000000..54844525 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-support/application-after-sales-service-scope.md @@ -0,0 +1,57 @@ +# 阿里云百炼平台售后服务范围说明 + +## **阿里云百炼平台售后服务范围说明** + +欢迎您使用阿里云百炼。本《**阿里云百炼平台售后服务范围说明》是对您使用阿里云百炼相关产品和服务时适用的售后服务范围的说明。** + +1\. 在您购买的服务期限内，我们将为您提供如下售后基础服务，即通过官网、电话及阿里云APP提供7×24的电话咨询（95187、400电话）、智能在线和标准工单支持。支持范围包括： + +（1）关于阿里云百炼模型服务与产品功能、架构的咨询； + +（2）使用、配置阿里云百炼模型服务的最佳实践； + +（3）阿里云百炼模型服务的使用咨询、技术问题及故障诊断； + +（4）阿里云百炼API及阿里云百炼官方SDK问题的故障诊断； + +（5）与阿里云百炼管理控制台相关的问题； + +（6）与阿里云相关的账号问题咨询支持； + +（7）与阿里云相关的财务、合同及计费问题的咨询支持。 + +2\. 阿里云百炼将以阿里云官网页面公布的[客户服务权益](https://www.aliyun.com/service/customer-service-benefits?spm=5176.support-home.J_3451238410.1.12d1156fPBBxO0)标准向您提供相应的售后服务支持。 + +3\. 阿里云百炼同时提供付费版的售后增值服务（包括支持计划等），该等服务需在阿里云官网订购后生效使用。您还可通过阿里云获得其他付费的售后服务，具体详见阿里云的网站相关页面的收费售后服务内容。如您的项目需要更深度的技术支持（如业务代码编写指导、定制化集成方案等），建议联系阿里云商务经理沟通定制化服务方案。 + +4\. 为了方便您的生产或使用，如您选择将阿里云百炼服务与外部（非阿里云百炼平台上的）第三方工具或产品进行对接，阿里云将尽商业上合理的努力为您提供第三方工具在接入阿里云百炼模型推理服务过程中的方向性建议，但针对非阿里云百炼服务相关的问题，我们无法提供专业意见。 + +（1）我们提供建议的范围包括： + +（i）确认阿里云百炼API接口及服务端的可用状态； + +（ii）阿里云百炼官方API调用示例及SDK使用说明参考； + +（iii）协助核查阿里云百炼服务端调用明细和计费记录； + +（iv）基本的连通性测试建议（如通过curl等标准工具测试阿里云百炼服务地址的可达性）。 + +**（2）我们提供建议的范围不包括：** + +（i）第三方工具（如Cursor、Windsurf、Cline、OpenClaw等）的安装、部署、配置、升级及日常使用指导； + +（ii）第三方工具的产品功能、交互逻辑及内部实现问题的排查； + +（iii）其他云厂商、企业或社区提供的产品或服务的配置与运维； + +（iv）用户业务代码的编写、调试与实现； + +（v）用户本地环境（含内网、代理、VPN、防火墙、操作系统等）导致的连通性或兼容性问题的排查； + +（vi）第三方工具内部显示的Token数量、费用预估值或调用统计与阿里云计费数据之间的差异解释； + +（vii）所有第三方工具的安装、补丁更新、测试、故障诊断、优化等日常运维服务； + +（viii）基于阿里云百炼模型服务原生能力之上的第三方自建业务相关支持。 + +5\. 但请您注意，除我们另有书面说明外，第三方工具不构成我们的代理、受托或联合服务主体，**我们不对外部第三方工具的任何陈述、承诺或行为承担责任。**您知悉并确认，阿里云仅负责阿里云百炼平台自身的运营维护，即百炼服务端的技术架构、API接口、计量计费系统、控制台功能等；**第三方工具的运行维护（如AI编程工具的安装配置、开源代理框架的部署调优等）由您及相应工具提供方负责。**当您的问题出现在百炼模型服务的使用过程中，但其原因、责任范围或依赖关系已超出阿里云百炼平台本身可直接提供支持和保障的范围时——通常涉及您侧系统、外部第三方服务、网络环境、账号权限、业务流程或非标集成，阿里云将协助进行初步排查。若经排查确认问题来源于非阿里云侧，阿里云将给予方向性建议并引导您联系相应的服务主体。 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-support/application-related-agreements.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-support/application-related-agreements.md index 941388e9..87bc4bba 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-support/application-related-agreements.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/application-support/application-related-agreements.md @@ -2,6 +2,6 @@ - [阿里云百炼服务协议](https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html?spm=5176.28197581.0.0.16e829a4HTC9FE) -- [阿里云百炼服务特别说明](https://help.aliyun.com/zh/model-studio/bailian-service-notes) +- [阿里云百炼体验功能特别说明](https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20260716114753386/20260716114753386.html) - [开源模型协议条款说明](https://help.aliyun.com/zh/model-studio/open-source-model-terms) diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/billing-for-knowledge-base.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/billing-for-knowledge-base.md index 29f6ac10..36597cef 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/billing-for-knowledge-base.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/billing-for-knowledge-base.md @@ -271,7 +271,45 @@ ### **2.2 模型调用费用** -在创建、更新、检索、命中测试知识库时，会调用向量模型（用于内容向量化）和排序模型（Rerank，用于重排序），这些调用会产生费用。 +在创建、更新、检索知识库以及使用知识问答服务时，会调用以下模型，这些调用会产生独立于规格费用之外的模型调用费用： + +**模型类别** + +**模型名称** + +**用途** + +**向量模型** + +[text-embedding-v4](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/text-embedding-v4)等 + +文档类知识库的文本向量化 + +[qwen3-vl-embedding](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-vl-embedding) + +图片问答类、音视频搜索类知识库的多模态向量化 + +**排序模型** + +[qwen3-rerank](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-rerank) + +文档类知识库检索结果的二次排序（可选） + +[qwen3-vl-rerank](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-vl-rerank) + +图片问答类、音视频搜索类知识库检索结果的二次排序（可选） + +**路由模型** + +[qwen-plus](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen-plus-latest) + +开启知识库路由时，系统调用 qwen-plus 判断查询应路由至哪些知识库 + +**问答模型** + +[qwen3.7-plus](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3.7-plus?serviceSite=asia-pacific-china) 等 + +知识问答服务中生成回答的大语言模型，由用户在应用中自行选择 **重要** @@ -281,20 +319,28 @@ **多个知识库计费规则：**阿里云百炼应用挂载了多个知识库时，会在多个知识库内执行检索，Token 消耗量（Query 向量化和 Rerank 排序）**按知识库数量倍数增加**（N 个知识库则消耗量 × N）。 -#### **2.2.1 创建/更新知识库** +#### **2.2.1 知识管理（创建与更新知识库）** - **调用场景：**上传新文件或增量更新时，调用向量模型对文本内容进行向量化处理。 - **计费说明：按新增内容的 Token 数量计费。**删除文件不产生模型调用费用。 +- **调用的模型：** + + - 文档搜索类知识库：[text-embedding-v4](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/text-embedding-v4) 或 [text-embedding-v3](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/text-embedding-v3)（文本向量模型）。 + + - 图片问答类、音视频搜索类知识库：[qwen3-vl-embedding](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-vl-embedding)（多模态向量模型）。 + -#### **2.2.2 检索知识库** +#### **2.2.2 知识检索** - **调用场景** - 1. 调用向量模型，对用户的查询（Query）进行向量化。 + 1. **向量化：**调用向量模型，对用户的查询（Query）进行向量化。 + + 2. **知识库路由（可选）：**若应用关联了多个知识库并开启了知识库路由功能，系统会调用 [qwen-plus](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen-plus-latest) 判断用户查询应路由至哪些知识库，该调用按 qwen-plus 的 Token 用量计费。 - 2. 调用排序模型（Rerank），对初步检索到的结果进行重新排序，以提升最终答案的精准度。 + 3. **排序（可选）：**调用排序模型对初步检索到的结果进行重新排序，以提升最终答案的精准度。文档搜索类知识库使用 [qwen3-rerank](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-rerank)，图片问答类和音视频搜索类知识库使用 [qwen3-vl-rerank](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-vl-rerank)。 - **计费说明** @@ -304,7 +350,7 @@ - **检索流程与计费关系详解** - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0647879771/CAEQaxiBgMCFrtjd3BkiIDA2ZWRiNzYxYzZiNzRkNGM5Mzg4NGQ5ZjhlODBlOWZj6139615_20260107153729.136.svg) + ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6004524871/CAEQaxiBgMCFrtjd3BkiIDA2ZWRiNzYxYzZiNzRkNGM5Mzg4NGQ5ZjhlODBlOWZj6139615_20260107153729.136.svg) 1. **初步召回** 系统根据以下参数从知识库中召回文本切片： @@ -328,7 +374,22 @@ Rerank 模型排序后，系统会根据**最终召回最大数量**参数（例如 5）返回相应数量的切片。 -#### **2.2.3 费用优化建议** +#### **2.2.3 知识问答** + +通过百炼应用（智能体应用、工作流应用）使用知识库进行问答时，除了检索阶段的模型费用外，还会产生以下模型调用费用： + +- **问答生成模型：**系统根据您在应用中选择的问答模型（如 qwen-plus 等）生成回答，按该模型的 Token 用量计费。具体价格以[模型计费标准](https://help.aliyun.com/zh/model-studio/model-pricing)为准。 + +- **预文件解析（可选）：**当用户在对话中上传文件并开启预文件解析功能时，系统会调用 [qwen3-rerank](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-rerank) 对文件内容进行排序处理，按排序模型的 Token 用量计费。 + +- **知识库路由（可选）：**若应用关联了多个知识库并开启了路由功能，系统会调用 [qwen-plus](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen-plus-latest) 进行路由判断（详见 [2.2.2 知识检索](#c868ef3a653qx)）。 + + +**重要** + +知识问答服务的完整费用 = **规格费用**（知识库运行时长）+ **检索阶段的模型费用**（向量化 + 排序 + 路由）+ **问答阶段的模型费用**（问答生成 + 预文件解析）。各模型费用按实际 Token 消耗量独立计算，请关注[模型计费标准](https://help.aliyun.com/zh/model-studio/model-pricing)了解各模型的单价。 + +#### **2.2.4 费用优化建议** 有以下两种方式： @@ -366,6 +427,19 @@ 点击知识库卡片上的**命中测试**，进入配置调试页面进行测试，会产生相应的模型（向量模型、排序模型）调用计费。 +#### **2.2.5 节省计划抵扣说明** + +知识库使用的向量模型（如 text-embedding-v4）和排序模型（如 qwen3-rerank）属于百炼平台 A 类模型，其调用费用支持通过以下节省计划抵扣： + +- **AI 通用型节省计划**（推荐）：覆盖 A 类全部模型（含文本向量、多模态向量、排序模型），按月承诺消费享阶梯折扣。详情请参见[节省计划与资源包](https://help.aliyun.com/zh/model-studio/savings-plan-and-resource-package)。 + +- **向量及排序模型节省计划**：专门针对向量和排序模型的节省计划，一次性购买固定金额。详情请参见[节省计划与资源包 > 向量及排序模型节省计划](https://help.aliyun.com/zh/model-studio/savings-plan-and-resource-package)。 + + +**说明** + +节省计划仅可抵扣模型调用费用，不可抵扣知识库的规格费用（运行时长费用）。规格费用的优化请参见[资源包](#a06c023507qq3)。 + ## 3\. 计费示例 ### 3.1 连续运行 1 天 @@ -396,7 +470,7 @@ ### 3.2 创建、更新与检索知识库 -基于 **text-embedding-v4**（向量模型）与 **qwen3-rerank**（排序模型），价格均为 **0.0005 元/千 Token**。 +以下示例基于文档搜索类知识库，使用 [**text-embedding-v4**](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/text-embedding-v4)（向量模型）与 [**qwen3-rerank**](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-rerank)（排序模型），价格均为 **0.0005 元/千 Token**。图片问答类和音视频搜索类知识库使用的多模态模型（[qwen3-vl-embedding](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-vl-embedding)、[qwen3-vl-rerank](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3-vl-rerank)）价格请参见对应模型详情页。 **计费逻辑**：费用 = Token 消耗量（以“千 Token”为单位） × 模型单价 @@ -550,7 +624,7 @@ 5. **为什么我的排序（Rerank）费用特别高？如何降低模型调用费用？** - 排序（Rerank）模型的费用与您最终返回的结果数量无关，而是由**初步召回**的文本切片总数决定的。降低模型调用费用详见本文[2.2.3 费用优化建议](#9a9e30ecc3pbe)内容。 + 排序（Rerank）模型的费用与您最终返回的结果数量无关，而是由**初步召回**的文本切片总数决定的。降低模型调用费用详见本文[2.2.4 费用优化建议](#9a9e30ecc3pbe)内容。 6. **如何彻底停止知识库的计费？删除库内文件可以吗？** diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-knowledge-base-specifications.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-knowledge-base-specifications.md index 61db382a..b85e5222 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-knowledge-base-specifications.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-knowledge-base-specifications.md @@ -117,6 +117,17 @@ aac、amr、flac、flv、m4a、mp3、mpeg、ogg、opus、wav、webm、wma、mp4 最大512MB +### **表格格式要求** + +上传 Excel 文件时，表格格式需满足以下要求，否则模型无法正确解析数据。 + +- **不支持合并单元格表头**：请使用标准单行表头格式，每列只有一个独立的字段名，不跨列合并。合并单元格会导致模型无法正确映射列关系，造成数据解析异常。 + +- **备注信息不要写在 Excel 首行**：Excel 文件的首行会被识别为表头或数据。请将备注和说明信息写在智能体应用的**系统提示词**中作为上下文，避免模型将说明文字与实际数据混淆。 + +- **格式修改后需重新上传**：修改表格格式后，需重新上传文件到知识库，并重新发布智能体应用，变更才能生效。 + + ### **上传操作** **类别** @@ -129,7 +140,7 @@ aac、amr、flac、flv、m4a、mp3、mpeg、ogg、opus、wav、webm、wma、mp4 在控制台单次操作可同时导入的文件数量。 -> 通过API进行批量导入时不受此限制，但建议单次导入不超过 10,000 个文件。 +> 通过API进行批量导入时不受此限制，但建议单次导入不超过 500 个文件。 50 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-knowledge-base.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-knowledge-base.md index c5cdc98d..73ce26f5 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-knowledge-base.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-knowledge-base.md @@ -62,7 +62,7 @@ 2. 填写**知识库名称**和**知识库描述**，其余设置保持默认，点击**下一步**。 -3. 选择**默认类目**，上传[阿里云百炼系列手机产品介绍.docx](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250603/duuuxk/%E9%98%BF%E9%87%8C%E4%BA%91%E7%99%BE%E7%82%BC%E7%B3%BB%E5%88%97%E6%89%8B%E6%9C%BA%E4%BA%A7%E5%93%81%E4%BB%8B%E7%BB%8D.docx)文件。点击**下一步**，然后点击**完成**。 +3. 选择**配置类目**，上传[阿里云百炼系列手机产品介绍.docx](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250603/duuuxk/%E9%98%BF%E9%87%8C%E4%BA%91%E7%99%BE%E7%82%BC%E7%B3%BB%E5%88%97%E6%89%8B%E6%9C%BA%E4%BA%A7%E5%93%81%E4%BB%8B%E7%BB%8D.docx)文件。点击**下一步**，然后点击**完成**。 ### 2\. 集成到业务应用 @@ -184,12 +184,10 @@ 2. **填写基础信息** - 根据应用场景选择合适的**知识库类型**（单一知识库不支持同时选择多个类型）。选择**文档搜索**类型后，还需选择**使用场景**（基础文档问答、图文并茂回复、视觉理解（富文本文档）或极速问答）： + 根据应用场景选择合适的**知识库类型**（单一知识库不支持同时选择多个类型）。选择**文档搜索**类型后，还需选择**使用场景**（基础文档问答、视觉理解（富文本文档）或极速问答）： - **基础文档问答：**适用于纯文本文档的语义检索。 - - **图文并茂回复：**适用于需要返回图文混排内容的场景。 - - **视觉理解（富文本文档）：**使用多模态向量模型对 PDF、图片等富文本文档进行视觉级理解和索引，保留原始版面信息。适合含有复杂排版、图表、公式的文档，支持文字、图片和图文组合三种命中测试模式。 - **极速问答：**针对检索速度进行优化，适合高度结构化或简单文档类型（如 FAQ、产品参数表等），提供极速低延时的问答体验。索引配置与基础文档问答一致，差异在于后端检索策略针对低延迟场景进行了专项优化。仅支持文本查询，不支持图片输入。 @@ -466,7 +464,7 @@ - **知识库类型限制：**仅适用于文档搜索类、数据查询类、音视频搜索类知识库。**图片问答类知识库不支持**。 - - **使用场景限制：**仅「基础文档问答」和「图文并茂回复」两种使用场景支持。「视觉理解（富文本文档）」和「极速问答」**不支持**。 + - **使用场景限制：**仅**基础文档问答**使用场景支持。**视觉理解（富文本文档）**和**极速问答**不支持。 ## **相似度阈值** @@ -534,7 +532,7 @@ > 文档搜索类知识库无法实现此效果。 - - 支持导入多个Excel文件，但要求各文件的**表结构完全一致**。 + - 支持导入单份xlsx、xls格式的文档，文件大小限制20MB以内。 - **选择连接器：**选择指定的数据连接器。如尚未创建数据连接器，请参阅[数据连接](https://help.aliyun.com/zh/model-studio/data-connection)。 @@ -609,7 +607,7 @@ - **知识库类型限制：**仅适用于文档搜索类、数据查询类、音视频搜索类知识库。**图片问答类知识库不支持**。 - - **使用场景限制：**仅「基础文档问答」和「图文并茂回复」两种使用场景支持。「视觉理解（富文本文档）」和「极速问答」**不支持**。 + - **使用场景限制：**仅**基础文档问答**使用场景支持。**视觉理解（富文本文档）**和**极速问答**不支持。 ### **相似度阈值** @@ -721,7 +719,7 @@ - **知识库类型限制：**仅适用于文档搜索类、数据查询类、音视频搜索类知识库。**图片问答类知识库不支持**。 - - **使用场景限制：**仅「基础文档问答」和「图文并茂回复」两种使用场景支持。「视觉理解（富文本文档）」和「极速问答」**不支持**。 + - **使用场景限制：**仅**基础文档问答**使用场景支持。**视觉理解（富文本文档）**和**极速问答**不支持。 ### **相似度阈值** @@ -865,7 +863,7 @@ - **知识库类型限制：**仅适用于文档搜索类、数据查询类、音视频搜索类知识库。**图片问答类知识库不支持**。 - - **使用场景限制：**仅「基础文档问答」和「图文并茂回复」两种使用场景支持。「视觉理解（富文本文档）」和「极速问答」**不支持**。 + - **使用场景限制：**仅**基础文档问答**使用场景支持。**视觉理解（富文本文档）**和**极速问答**不支持。 ### **相似度阈值** @@ -894,7 +892,7 @@ - **使用场景**可根据需求选择**基础文档问答**、**图文并茂回复**、**视觉理解（富文本文档）**或**极速问答**（适用于高度结构化或简单文档类型，任务明确，提供极低延迟的问答体验）。 + **使用场景**可根据需求选择**基础文档问答**、**视觉理解（富文本文档）**或**极速问答**（适用于高度结构化或简单文档类型，任务明确，提供极低延迟的问答体验）。 在请求高峰时段，创建过程可能需要数小时（取决于数据量），请耐心等待。 @@ -905,11 +903,15 @@ ## **文档搜索类知识库** -- **自动更新（推荐）** +1. **自动更新（推荐）** - 通过对象存储OSS管理文件，借助函数计算 FC 监听文件变更事件，自动同步更新至知识库，实现知识的实时更新。详见[告别手动操作，让AI知识库自动更新](https://www.aliyun.com/solution/tech-solution/auto-updated-knowledge-base)。 + 通过对象存储OSS类型连接器监听OSS内数据变动，自动同步更新至知识库，实现知识的实时更新。操作步骤如下： -- **手动更新** + 1. **创建连接器：**前往[创建连接器](https://bailian.console.aliyun.com/cn-beijing?tab=app#/connector/create)，创建**OSS**类型的连接器并选择进行监听的Bucket。 + + 2. **创建知识库：**创建一个知识库，在**选择数据**步骤的**数据来源**选项选择创建完成的OSS类型的连接器。 + +2. **手动更新** 在[知识库](https://bailian.console.aliyun.com/?tab=app#/knowledge-base)页面，找到目标知识库，单击卡片上的**查看详情**。 @@ -1230,22 +1232,12 @@ **方式一（仅适用智能体应用）** - 1. 在[构建知识库](#c0fa1080aerzp)时，知识库类型选择**文档搜索**，使用场景选择**图文并茂回复**。 - - > 选择图文并茂回复后，知识库将从文件插图中提取摘要，大模型根据摘要与问题的相关性自主决定是否插入图片。 + 1. 在[构建知识库](#c0fa1080aerzp)时，知识库类型选择**文档搜索**，使用场景选择**视觉理解（富文本文档）**。 - **重要** - - 上传文档时不能选择**电子文档解析**，否则无法获取图片内容。电子文档解析不识别文档中的图片，会导致图文并茂回复功能无法正常使用。 + > 选择视觉理解（富文本文档）回复后，知识库将从文件插图中提取摘要，大模型根据摘要与问题的相关性自主决定是否插入图片。 2. 创建或编辑智能体应用时，选择**千问-Plus**或**千问-Plus-Latest**模型（经测试，两款模型效果最佳）。点击**文档知识库**右侧的**+**按钮，添加上一步构建的知识库。 - **说明** - - 召回长度须小于文档实际长度。若召回长度超过文档实际长度，系统将直接返回完整文档内容，不执行图文并茂的逻辑判断。 - - > 注意：当前"图文并茂回复"与"展示回答来源"功能暂不支持同时开启。 - 3. 实际问答效果： ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1699676371/p903021.png) diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-optimization.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-optimization.md index 49fb108e..308e13dd 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-optimization.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/knowledge-base/rag-optimization.md @@ -107,6 +107,8 @@ RAG（Retrieval Augmented Generation，检索增强生成）是一种结合了 在创建知识库的**索引设置**步骤中，打开页面底部的**多轮对话改写**开关即可启用该功能。 + ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3562168571/p1008316.png) + 请注意，多轮对话改写功能是与知识库绑定的，开启后仅对当前知识库相关的查询生效。且如果您在创建知识库时未启用该配置，则后续无法再为该知识库开启，除非重新创建知识库。 @@ -129,7 +131,7 @@ RAG（Retrieval Augmented Generation，检索增强生成）是一种结合了 - **在数据管理页面编辑标签**：对于已上传的文件，可单击右侧的**标签**进行编辑，相关API是[UpdateFileTag](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-updatefiletag)。 - 目前阿里云百炼支持以下方式使用标签： + 目前阿里云百炼支持以下两种方式使用标签： - [通过API调用阿里云百炼应用](https://help.aliyun.com/zh/model-studio/application-calling-guide#4100253b7chc3)时，可以在请求参数`tags`中指定标签。 @@ -141,42 +143,6 @@ RAG（Retrieval Augmented Generation，检索增强生成）是一种结合了 切换至**召回策略**页签，在**标签过滤**中选择需要应用的标签（如示例中的『硬件』『智能手机』）即可。 - - 在新版智能体中，通过系统提示词引导智能体按标签进行过滤。 - - 1. **为知识库文档设置标签**：在管理知识库时，为各文件添加标签。 - - 例如，百炼电脑的标签为`bailian_pc`，百炼系列手机产品介绍的标签为`bailian_mobile`，百炼电脑服务及优惠政策的标签为`bailian_service`和`bailian_pc`。 - - 2. **在系统提示词中引导智能体**：在智能体的系统提示词中，清晰地说明各标签的含义与使用规则。例如： - - ``` - 你是阿里云百炼产品的智能客服，请回答与阿里云百炼产品相关的问题。 - - 当问题比较宽泛时，检索整个知识库。 - 当问题比较明确时，使用知识库标签进行定向检索。标签映射关系如下： - 1. 与百炼手机相关：bailian_mobile - 2. 与百炼电脑相关：bailian_pc - 3. 与百炼产品相关：bailian_mobile 或 bailian_pc - 4. 与百炼电脑服务及优惠政策相关：bailian_service 和 bailian_pc - ``` - - 3. 智能体会识别用户意图，匹配对应的标签，并仅从带有这些标签的文件中检索内容。匹配逻辑包含以下几种模式： - - - **单标签匹配**：对于问题“百炼有哪些手机产品？”，智能体会检索所有带有`bailian_mobile`标签的文件。 - - - **多标签“或”逻辑**：对于问题“百炼有哪些产品？”，智能体会检索所有带有`bailian_mobile`或`bailian_pc`标签的文件。 - - - **多标签“与”逻辑**：对于问题“百炼电脑的服务政策是什么？”，智能体仅检索同时带有`bailian_pc`和`bailian_service`标签的文件。 - - - 您可以在知识库交互卡片中查看标签匹配逻辑，并通过调整提示词来优化匹配效果。匹配逻辑的格式如下： - - - **单标签匹配**：`[{"tags":["bailian_mobile"]}]` - - - **多标签“或”逻辑**：`[{"tags":["bailian_mobile","bailian_pc"]}]` - - - **多标签“与”逻辑：**`[{"tags":["bailian_service"]},{"tags":["bailian_pc"]}]` - 2. **典型问题****：**知识库中含多个结构内容相同/相近的文件，如文件A和文件B的内容都包含“功能概述”章节，只希望在A文件的“功能概述”中检索。 @@ -290,6 +256,23 @@ RAG（Retrieval Augmented Generation，检索增强生成）是一种结合了 此处只是修改了知识库中的文本切片，并未修改数据管理（临时存储）中的源文件或数据表。因此后续将此文件/数据表再次导入知识库时，仍需人工检查和修正。 +3. **必要时新增或删除单条切片：**除修正现有切片外，您还可以直接在切片层做更细粒度的管理，应对智能切分无法覆盖的边界情况。 + + **说明：新增切片与删除切片** + + 在知识库的切片视图中，可以对单条切片做以下两类操作： + + - **新增切片：**当原文档存在关键信息但被切片策略遗漏，或希望额外补充上下文时，可手动新增一条切片。新增时需选择所属文档，并填写切片内容（最多 6000 字符）；如该文档为图片问答类知识库，还可上传关联图片（每张图片不超过 20 MB，支持 PNG、JPG、JPEG、BMP、GIF 等常见格式）。 + + - **删除切片：**当某条切片包含错误信息或重复内容，且无法通过修正解决时，可单独删除该切片，不影响同文档其他切片继续被检索。 + + + 新增和删除均仅作用于当前知识库的检索结果，不会修改数据管理中的源文件。再次将源文件导入知识库时，需要重新进行此类调整。 + + **说明** + + 不同知识库类型支持的切片操作存在差异：**编辑切片（更新内容）所有类型的知识库均支持**，在切片详情弹窗右上角单击编辑即可操作。新增和删除方面，文档搜索类、数据查询类、图片问答类知识库均支持；**音视频搜索类知识库仅支持删除切片**。 + ### **3.4 重排不佳** @@ -338,7 +321,7 @@ RAG（Retrieval Augmented Generation，检索增强生成）是一种结合了 从下方示意图可以看到，目标知识库中实际与用户提示词相关，需要返回的文本切片总共有7个（下图左侧，已用绿色标出），但由于已经超出了当前设定的最大召回片段数K，因此包含优势5（超长待机）和优势6（拍照清晰）的文本切片被舍弃，没有提供给大模型。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6954804871/CAEQURiBgMCatMinohkiIDY3YTVhOWY2MjNjMjRkYzc5NTU1ZmVhNGQ2MGQ5ODc24762899_20250109142407.621.svg) + ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3188584871/CAEQURiBgMCatMinohkiIDY3YTVhOWY2MjNjMjRkYzc5NTU1ZmVhNGQ2MGQ5ODc24762899_20250109142407.621.svg) 由于 RAG 本身无法判断需要多少个文本切片才能给出“完整”的答案，因此即使最终提供的文本切片有遗漏，随后大模型仍然会基于缺失的文本切片生成不完整的回答。 @@ -555,6 +538,47 @@ RAG（Retrieval Augmented Generation，检索增强生成）是一种结合了 **enable\_thinking**：是否开启思考模式。 +## **3.6 知识检索与知识问答服务的效果优化** + +如果您使用知识检索或知识问答服务，以下策略可帮助提升检索和回答质量。 + +### **多知识库场景的优化** + +- **合理设置知识库权重**：当绑定多个知识库时，为核心知识库设置更高的权重，使其检索结果在混排中排序更靠前。 + +- **启用知识库路由**：绑定多个知识库但查询通常只涉及部分库时，启用知识库路由可让系统自动判断需要查询哪些库，减少无关结果的干扰。 + +- **选择合适的混排模型模式**： + + - **问答模式**：适用于用户输入为问题、期望从知识库中找到答案的场景。 + + - **相似模式**：适用于需要查找与输入内容语义相似的段落或文档的场景。 + + - **自定义高级模式**：当上述两种模式效果不理想时，可自定义排序指令来干预排序行为。 + + +### **知识库独立检索参数调优** + +在知识检索或知识问答的配置页中，点击知识库旁的![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6339082871/p1080483.png)图标可展开独立检索参数，针对不同知识库进行精细化调优： + +- **相似度阈值**：提高阈值可过滤低质量结果，但阈值过高可能导致召回不足。建议从 0.2 开始调整。 + +- **最大召回数量**：增加召回数量可获得更全面的结果，但可能引入噪音。根据知识库规模和内容相关度调整。 + +- **标签过滤**：利用文档标签缩小检索范围，提高结果的精准度。 + + +### **知识问答的生成效果优化** + +- **调整提示词**：通过提示词设定模型的角色、回答风格和关注领域。例如设定"你是一位专业的技术支持工程师，请基于知识库内容简洁回答"可显著提升回答质量。 + +- **选择检索模式**：简单问题使用**极速**模式降低延时；复杂或模糊问题使用**多轮智能**模式，让模型自动规划多轮检索。 + +- **设置知识库优先级**：开启优先级后，为更权威的知识库设置"高"优先级，确保其内容在回答中占主导地位。 + +- **启用拒答和防泄漏**：如果希望模型在知识库无相关内容时明确拒绝回答（而非编造答案），可开启拒答功能并配置触发条件。如果需要保护知识库原文不被直接输出，可开启防泄漏。 + + ## **4\. 后续步骤** ### **4.1 持续迭代** diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/application-introduction.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/application-introduction.md index 0e578a2d..2a13317b 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/application-introduction.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/application-introduction.md @@ -45,12 +45,15 @@ Python 编码（专业代码） AI 自主决策、动态规划 由大模型根据提示词自主规划任务步骤 + 由预定义流程精确控制 每一步都由预设的节点定义，逻辑确定 + 完全由代码控制 所有逻辑和执行路径由代码定义 + 适合人群 @@ -74,9 +77,9 @@ AI 工程师、开发者 - **智能体应用** - - 如何创建和配置？请参考：[新版智能体应用](https://help.aliyun.com/zh/model-studio/new-single-agent-application)（推荐）、[智能体应用](https://help.aliyun.com/zh/model-studio/single-agent-application)。 + - 如何创建和配置？请参考：[新版智能体应用（Agent 2.0）](https://help.aliyun.com/zh/model-studio/new-single-agent-application)（推荐）、[智能体应用（Agent 1.0）](https://help.aliyun.com/zh/model-studio/single-agent-application)。 - - 如何快速上手实际案例？请参考：[创建智能问答 AI 电商客服助手](https://help.aliyun.com/zh/document_detail/2878136.html#0a9fbaf6a71q7)、[集成高德 MCP 的旅行规划智能体](https://help.aliyun.com/zh/document_detail/2880695.html)。 + - 如何快速上手实际案例？请参考：[创建智能问答 AI 电商客服助手](https://help.aliyun.com/zh/document_detail/2878136.html#0a9fbaf6a71q7)、[集成高德 MCP 的旅行规划智能体](https://help.aliyun.com/zh/model-studio/use-cases/integrate-amap-mcp-travel-planning-agent)。 - 如何通过 API 调用？请参考：[新版智能体应用 API](https://help.aliyun.com/zh/model-studio/new-agent-application-api-reference)、[调用智能体应用](https://help.aliyun.com/zh/model-studio/call-single-agent-application/)。 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/new-single-agent-application.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/new-single-agent-application.md index ede0fcdc..48466f2f 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/new-single-agent-application.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/new-single-agent-application.md @@ -1,4 +1,4 @@ -# 新版智能体应用（Agent 2.0） +# 新版智能体应用 新版智能体应用（Agent 2.0）将知识库、MCP 等多种能力统一为工具，并通过自主思考和规划来调用，以解决复杂任务。 @@ -53,7 +53,7 @@ - **最长回复长度**：模型生成的长度限制，不包含提示词。 - - **temperature**：控制生成随机性和多样性，数值越高随机性越强。 + - **温度系数**：控制生成随机性和多样性，数值越高随机性越强。 - **enable\_thinking**：是否开启思考模式。开启思考模式有助于提升智能体的反思效果。不支持思考模式的模型无法配置 enable\_thinking 参数。 @@ -245,3 +245,15 @@ - 意图与技能的相关性：请评估问题的表述是否清晰，其意图是否能明确指向特定技能。如果意图模糊或与技能功能不相关，模型可能选择不调用。 - 执行轮次限制：请检查是否达到了 ReAct 轮次上限。智能体可能已规划调用该技能，但在执行到该步骤前因轮次耗尽而被强制终止。 + + +### 智能体应用是否支持上下文缓存？ + +支持**隐式缓存**，暂不支持在应用中配置**显式缓存**。 + +- **隐式缓存**：智能体在调用支持隐式缓存的模型时会自动生效，无需任何配置、也无法关闭。系统会自动识别并缓存请求的公共前缀（如相同的系统提示词、多轮对话历史、知识库召回内容等），命中缓存的输入 Token 按标准输入单价的 20% 计费，可相应降低模型调用成本。 + +- **显式缓存**：需要在模型调用请求中主动为指定内容创建缓存标记。智能体应用由平台统一构造模型请求，暂不支持配置显式缓存。 + + +上下文缓存的工作模式、支持的模型及计费详情，请参见[上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)。 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/rich-code-application.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/rich-code-application.md index 764cb376..0ae6fa78 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/rich-code-application.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/llm-application/rich-code-application.md @@ -30,7 +30,7 @@ - **命令行创建**：在控制台创建空白应用后，通过 AgentScope-AI 命令行工具上传本地代码包部署。适合已有项目代码的开发者，详细流程请参考[API 开发指南](https://help.aliyun.com/zh/model-studio/rich-code-app-develop-guide)。 - 填写**应用名称**后，单击**确认**。 + 填写**应用名称**后，单击**立即创建**。 ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/2506686771/p1070627.png) @@ -49,14 +49,14 @@ **提交代码** - - **使用模板代码**（默认）：选择一个预置的应用模板快速开始。 + - **使用模版代码**（默认）：选择一个预置的应用模版快速开始。 - **上传代码包**：上传本地开发的 .whl 格式代码包。代码包制作方式请参考[API 开发指南](https://help.aliyun.com/zh/model-studio/rich-code-app-develop-guide)。 - 平台提供以下应用模板： + 平台提供以下应用模版： - **模板** + **模版** **说明** @@ -76,7 +76,7 @@ 根据需要配置**规格方案**（vCPU、内存、磁盘大小）、**最小实例数**、**单实例并发度**和**部署地域**。初次体验保持默认即可。 - > 时延敏感业务建议最小实例数 ≥ 1，可实现毫秒级热启动、保障服务不中断。详见[实例类型和规格](https://help.aliyun.com/zh/functioncompute/fc/product-overview/instance-types-and-specifications#section-mfv-5fb-ehw)。 + > 时延敏感业务建议最小实例数 ≥ 1，可实现毫秒级热启动、保障服务不中断。详见[实例类型和规格](https://help.aliyun.com/zh/functioncompute/instance-types-and-specifications#section-mfv-5fb-ehw)。 配置完成后，单击**立即部署**开始部署。 diff --git a/skills/bailian-docs-llm-wiki/raw/application-user-guide/managed-agents/managed-agents-context.md b/skills/bailian-docs-llm-wiki/raw/application-user-guide/managed-agents/managed-agents-context.md index b93e1eff..efe28a55 100644 --- a/skills/bailian-docs-llm-wiki/raw/application-user-guide/managed-agents/managed-agents-context.md +++ b/skills/bailian-docs-llm-wiki/raw/application-user-guide/managed-agents/managed-agents-context.md @@ -20,8 +20,6 @@ - **创建会话时挂载**：在 `resources` 字段中指定资源列表和挂载路径。 -- **运行时追加**：通过 **POST** `/sessions/{session_id}/resources` 追加，实时生效，无需重启会话。 - ### 路径约定 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md index 62ca0940..382676af 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md @@ -21,7 +21,9 @@ Tripo 3D模型生成支持**文生3D模型**、**单图生3D模型和多图生3D ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -35,7 +37,8 @@ Tripo 3D模型生成支持**文生3D模型**、**单图生3D模型和多图生3D ## 文生3D模型（有贴图） ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -53,7 +56,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## 单图生3D模型（有贴图） ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -71,7 +75,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## 多图生3D模型（有贴图） ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -96,7 +101,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener > 图片顺序为前、左、后、右，不需要的视角传入空对象`{}`即可。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -121,7 +127,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener > 需同时将`texture`和`pbr`设为`false`。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -324,7 +331,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -343,7 +350,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-realtime-python-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-realtime-python-sdk.md index 601bfe7e..b7ba1534 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-realtime-python-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-real-time-speech-recognition-api-reference/fun-asr-realtime-python-sdk.md @@ -39,8 +39,6 @@ 点击查看完整示例 -示例中用到的音频为：[asr\_example.wav](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250210/iwaouc/asr_example.wav)。 - ``` from http import HTTPStatus import dashscope @@ -58,7 +56,7 @@ recognition = Recognition(model='fun-asr-realtime', format='wav', sample_rate=16000, callback=None) -result = recognition.call('asr_example.wav') +result = recognition.call('{YOUR_AUDIO_FILE}') if result.status_code == HTTPStatus.OK: print('识别结果：') print(result.get_sentence()) @@ -221,8 +219,6 @@ if __name__ == '__main__': ## 识别本地语音文件 -示例中用到的音频为：[asr\_example.wav](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250210/acoict/asr_example.wav)。 - ``` import os import time @@ -270,8 +266,8 @@ recognition = Recognition(model='fun-asr-realtime', try: audio_data: bytes = None - f = open("asr_example.wav", 'rb') - if os.path.getsize("asr_example.wav"): + f = open("{YOUR_AUDIO_FILE}", 'rb') + if os.path.getsize("{YOUR_AUDIO_FILE}"): # 一次性将文件数据全部读入buffer file_buffer = f.read() f.close() diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-android-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-android-sdk.md index 10ee8202..3a99c0b6 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-android-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-android-sdk.md @@ -1,6 +1,6 @@ -# Fun-ASR录音文件识别Android SDK +# Fun-ASR非实时语音识别Android SDK -本文档提供了Fun-ASR录音文件识别Android SDK的详细使用指南，帮助您将语音转换为文本。 +本文档提供了Fun-ASR非实时语音识别Android SDK的详细使用指南，帮助您将语音转换为文本。 **用户指南：**[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide)。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 @@ -13,7 +13,7 @@ - [下载最新SDK整合包](https://help.aliyun.com/zh/isi/sdk-selection-and-download)。 - 解压 ZIP 包。在 `app/libs` 目录中获取 AAR 格式 SDK，并添加到项目依赖。 - 需要 Android CPP 接入时，使用 ZIP 包内的 `android_libs` 与 `android_include` 获取动态库和头文件。 + 需要 Android CPP 接入时，使用 ZIP 包内的 `android_libs` 与 `android_include` 获取动态库和头文件。 - 用 Android Studio 打开工程。示例代码位于`DashFunAsrFileTranscriberActivity.java`，替换 API Key 后体验功能。 @@ -99,7 +99,7 @@ 是 - 运行模式。录音文件识别固定为 `"1"`。 + 运行模式。非实时语音识别固定为 `"1"`。 `device_id` @@ -170,7 +170,7 @@ ``` { "file_urls": [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ], "async_request": false, "nls_config": { @@ -596,7 +596,7 @@ ``` { "file_urls": [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ], "async_request": false, "nls_config": { diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-http-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-http-api.md index 2dbd2ce9..2ccfc352 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-http-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-http-api.md @@ -1,12 +1,10 @@ -# Fun-ASR录音文件识别HTTP API参考 +# Fun-ASR非实时语音识别HTTP API参考 -本文介绍Fun-ASR录音文件识别HTTP API的参数和接口细节。 +本文介绍Fun-ASR非实时语音识别HTTP API的参数和接口细节。 **用户指南：**[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide)。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 -## **DashScope异步调用（Fun-ASR）** - -### **流程说明** +## **流程说明** 与DashScope同步调用（一次请求、立即返回结果）不同，异步调用专为处理长音频文件或耗时较长的任务设计，该模式采用“提交-轮询”的两步式流程，避免了因长时间等待而导致的请求超时： @@ -23,7 +21,7 @@ - 当任务处理完成后，结果查询接口将返回最终的识别结果。 -### **服务端点** +## **服务端点** ## 华北2（北京） @@ -56,7 +54,7 @@ 使用新版域名（`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`）提交任务时，请求体中必须包含`parameters`对象。即使无需设置任何参数，也必须传入空对象`{}`，否则任务可正常提交，但识别将失败。 -### **请求头** +## **请求头** **参数** @@ -90,11 +88,11 @@ string 异步任务标识。仅提交任务接口需要传入，固定为`enable`，请勿遗漏，否则无法提交任务。 -### **提交任务接口** +## **提交任务接口** 提交语音识别任务。该接口异步返回，业务侧需结合[查询任务接口](#480630e0582sb)轮询任务状态。 -#### **请求体** +### **请求体** 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 @@ -107,7 +105,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi "model": "fun-asr", "input": { "file_urls": [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ] }, "parameters": { @@ -289,7 +287,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi - en: 英文 -#### **返回体** +### **返回体** ``` { @@ -319,11 +317,11 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi 任务状态。提交成功时返回`PENDING`。 -### **查询任务接口** +## **查询任务接口** 查询语音识别任务的执行情况和结果。建议轮询调用直至任务终态。 -#### **请求体** +### **请求体** 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 @@ -340,7 +338,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks 查询任务需指定其ID，该ID为[提交任务接口](#418f2ac8ecxm4)被调用后返回的`task_id`。 -#### **返回体** +### **返回体** ## 正常示例 @@ -355,7 +353,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks "end_time": "2024-09-12 15:11:40.903", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url": "{YOUR_AUDIO_URL}", "transcription_url": "https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/pre/filetrans-16k/20240912/15%3A11/409a4b92-445b-4dd8-8c1d-f110954d82d8-1.json?Expires=1726211500&OSSAccessKeyId=yourOSSAccessKeyId&Signature=v5Owy5qoAfT7mzGmQgH0g8C****%3D", "subtask_status": "SUCCEEDED" } @@ -383,7 +381,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "FILE_DOWNLOAD_FAILED", "message": "FILE_DOWNLOAD_FAILED", "subtask_status": "FAILED" @@ -485,11 +483,11 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks 失败的子任务数。 -### **其他接口：批量查询任务状态/取消任务** +## **其他接口：批量查询任务状态/取消任务** -详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的录音文件识别任务，同时支持取消`PENDING`（排队）状态的任务。 +详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的非实时语音识别任务，同时支持取消`PENDING`（排队）状态的任务。 -### **识别结果说明** +## **识别结果说明** 识别结果保存为JSON文件。 @@ -497,7 +495,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks ``` { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "properties":{ "audio_format":"pcm_s16le", "channels":[ @@ -637,887 +635,3 @@ punctuation string 预测出的词之后的标点符号（如有）。 - -## **DashScope同步调用（Fun-ASR-Flash）** - -**重要** - -该功能不支持SDK调用。 - -### **服务端点** - -## 华北2（北京） - -`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` - -调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - -## 新加坡 - -`POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` - -调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - -**重要** - -阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，能够为推理请求提供卓越的性能和更高的稳定性，建议迁移至新域名： - -- 华北2（北京）地域：从 `dashscope.aliyuncs.com` 迁移至 `{WorkspaceId}.cn-beijing.maas.aliyuncs.com` - -- 新加坡地域：从 `dashscope-intl.aliyuncs.com` 迁移至 `{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` - - -`{WorkspaceId}`需要替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。现有域名仍可正常使用。 - -### **请求头** - -**参数** - -**类型** - -**是否必选** - -**说明** - -Authorization - -string - -是 - -鉴权令牌，格式为`Bearer `，使用时将"``"替换为实际的API Key。 - -Content-Type - -string - -是 - -请求体的媒体类型，固定为`application/json`。 - -X-DashScope-SSE - -string - -是 - -用于控制是否以SSE流式方式返回结果。设置为`enable`时开启SSE流式返回模式，服务端会分多次返回中间识别结果和最终结果；设置为`disable`或不传该参数则仅返回最终结果。 - -### **请求体** - -以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 - -## 非流式 - -``` -curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ - --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ - --header "Content-Type: application/json" \ - --header "X-DashScope-SSE: disable" \ - --data '{ - "model": "fun-asr-flash-2026-06-15", - "input": { - "messages": [ - { - "role": "user", - "content": [ - { - "type": "input_audio", - "input_audio": { - "data": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" - } - } - ] - } - ] - }, - "parameters": { - "format": "wav", - "sample_rate": "16000" - } -}' -``` - -## 流式 - -``` -curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ - --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ - --header "Content-Type: application/json" \ - --header "X-DashScope-SSE: enable" \ - --data '{ - "model": "fun-asr-flash-2026-06-15", - "input": { - "messages": [ - { - "role": "user", - "content": [ - { - "type": "input_audio", - "input_audio": { - "data": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" - } - } - ] - } - ] - }, - "parameters": { - "format": "wav", - "sample_rate": "16000" - } -}' -``` - -## 携带上下文-非流式 - -``` -curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ - --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ - --header "Content-Type: application/json" \ - --header "X-DashScope-SSE: disable" \ - --data '{ - "model": "fun-asr-flash-2026-06-15", - "input": { - "messages": [ - { - "role": "user", - "content": [ - { - "type": "input_text", - "text": "你好啊" - } - ] - }, - { - "role": "assistant", - "content": [ - { - "type": "text", - "text": "你好啊，我是通义千问，有什么可以帮助你的？" - } - ] - }, - { - "role": "user", - "content": [ - { - "type": "input_audio", - "input_audio": { - "data": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" - } - } - ] - } - ] - }, - "parameters": { - "format": "wav", - "sample_rate": "16000" - } -}' -``` - -## 携带上下文-流式 - -``` -curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ - --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ - --header "Content-Type: application/json" \ - --header "X-DashScope-SSE: enable" \ - --data '{ - "model": "fun-asr-flash-2026-06-15", - "input": { - "messages": [ - { - "role": "user", - "content": [ - { - "type": "input_text", - "text": "你好啊" - } - ] - }, - { - "role": "assistant", - "content": [ - { - "type": "text", - "text": "你好啊，我是通义千问，有什么可以帮助你的？" - } - ] - }, - { - "role": "user", - "content": [ - { - "type": "input_audio", - "input_audio": { - "data": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" - } - } - ] - } - ] - }, - "parameters": { - "format": "wav", - "sample_rate": "16000" - } -}' -``` - -## Base64 - -可输入Base64编码数据（[Data URL](https://www.rfc-editor.org/rfc/rfc2397)），格式为：`data:;base64,`。 - -- ``：MIME类型 - - 因音频格式而异，例如： - - - WAV：`audio/wav` - - - MP3：`audio/mpeg` - -- ``：音频转成的Base64编码的字符串 - - Base64编码会增大体积，请控制原文件大小，确保编码后仍符合输入音频大小限制（10MB） - -- 示例：`data:audio/wav;base64,SUQzBAAAAAAAI1RTU0UAAAAPAAADTGF2ZjU4LjI5LjEwMAAAAAAAAAAAAAAA//PAxABQ/BXRbMPe4IQAhl9` - - **点击查看示例代码** - - Python - - ``` - import base64, pathlib - - # input.mp3为待识别的本地音频文件，请替换为自己的音频文件路径，确保其符合音频要求 - file_path = pathlib.Path("input.mp3") - base64_str = base64.b64encode(file_path.read_bytes()).decode() - data_uri = f"data:audio/mpeg;base64,{base64_str}" - ``` - - Java - - ``` - import java.nio.file.*; - import java.util.Base64; - - public class Main { - /** - * filePath为待识别的本地音频文件，请替换为自己的音频文件路径，确保其符合音频要求 - */ - public static String toDataUrl(String filePath) throws Exception { - byte[] bytes = Files.readAllBytes(Paths.get(filePath)); - String encoded = Base64.getEncoder().encodeToString(bytes); - return "data:audio/mpeg;base64," + encoded; - } - - public static void main(String[] args) throws Exception { - System.out.println(toDataUrl("input.mp3")); - } - } - ``` - - -``` -import base64, pathlib -import os -import requests - -# input.wav为待识别的本地音频文件，请替换为自己的音频文件路径，确保其符合音频要求 -file_path = pathlib.Path("input.wav") -base64_str = base64.b64encode(file_path.read_bytes()).decode() -data_uri = f"data:audio/wav;base64,{base64_str}" - -url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation" - -headers = { - "Authorization": f"Bearer {os.environ['DASHSCOPE_API_KEY']}", - "Content-Type": "application/json", - "X-DashScope-SSE": "disable", -} - -payload = { - "model": "fun-asr-flash-2026-06-15", - "input": { - "messages": [ - { - "role": "user", - "content": [ - { - "type": "input_audio", - "input_audio": { - "data": data_uri, - }, - } - ], - } - ] - }, - "parameters": { - "format": "wav", - "sample_rate": "16000", - }, -} - -response = requests.post(url, headers=headers, json=payload) -print(response.status_code) -print(response.json()) -``` - -**model** `_string_` **（必选）** - -模型名称，固定为`fun-asr-flash-2026-06-15`。 - -**input** `_object_` **（必选）** - -输入信息。 - -**属性** - -**messages** `_array(object)_` **（必选）** - -消息列表。包含当前待识别的音频，以及可选的对话上下文（用于提升识别效果）。 - -**重要** - -上下文功能用于提升专有词汇的识别准确率，使用方法详见[快速开始](https://help.aliyun.com/zh/model-studio/improve-asr-accuracy#ctx-quickstart-sec)。约束：上下文消息（`input_text` 和 `text` 类型）各最多 5 条，超出时保留最近的 5 条。每轮上下文文本总长度（`user` 和 `assistant` 的 `text` 字段长度之和）不超过 400 个字符（按字符数计算，每个字符计为 1），超出部分从末尾截断。 - -**重要** - -携带上下文时，`messages` 中的消息顺序有要求：上下文消息必须按对话轮次排列，每轮中 `user`（`input_text` 类型）必须在对应的 `assistant`（`text` 类型）之前；包含 `input_audio` 的 `user` 消息必须放在 `messages` 数组的最后。 - -**属性** - -**role** `_string_` **（必选）** - -消息角色。取值范围： - -- `user`（必选）：用户消息。type为`input_audio`时表示当前待识别的音频；type为`input_text`时表示前几轮的识别结果或领域相关的词表（可选，上下文）。 - -- `assistant`（可选，上下文）：前几轮大语言模型的回复内容。 - - -**content** `_array(object)_` **（必选）** - -消息内容列表。 - -**属性** - -**type** `_string_` **（必选）** - -内容类型。每个请求至少需要一条`input_audio`类型的消息。取值范围： - -- `input_audio`（必选）：当前待识别的音频输入（role为user），需同时传入`input_audio`对象。 - -- `input_text`（可选，上下文）：前几轮用户语音的识别结果或领域相关的词表（role为user），需同时传入`text`字段。 - -- `text`（可选，上下文）：前几轮大语言模型的回复内容（role为assistant），需同时传入`text`字段。 - - -**input\_audio** `_object_` **（条件必选）** - -当`type`为`input_audio`时必填。 - -**属性** - -**data** `_string_` **（必选）** - -待识别音频数据。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。支持以下两种方式： - -- **音频文件URL**：直接传入可公开访问的音频文件地址。 - -- **Base64 Data URI**：采用Data URI格式传入Base64编码的音频数据，值由`data:{MIME_TYPE};base64,`前缀与Base64编码的音频数据拼接而成。支持的MIME类型包括`audio/wav`、`audio/mp3`等。 - - -示例（URL方式）：`https://example.com/audio/sample.wav` - -示例（Base64方式）：`data:audio/wav;base64,{BASE64_ENCODED_DATA}` - -**text** `_string_` **（条件必选）** - -当`type`为`input_text`时，填入前几轮用户语音的识别结果或领域相关的词表；当`type`为`text`时，填入前几轮大语言模型的回复内容。文本按字符数计算，每个字符计为 1。每轮上下文中所有消息的 `text` 字段长度之和不超过 400 个字符，超出部分从末尾截断。 - -**parameters** `_object_` **（必选）** - -模型参数。 - -**属性** - -**format** `_string_` **（必选）** - -音频格式。根据实际音频格式填写，支持`wav`、`mp3`、`opus`等。详情请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 - -**sample\_rate** `_string_` （可选） - -音频采样率，单位Hz。例如`16000`表示16kHz采样率。详情请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 - -### **返回体** - -## 非流式 - -``` -{ - "output": { - "sentence": { - "begin_time": 760, - "channel_id": 0, - "end_time": 3800, - "sentence_end": true, - "sentence_id": 1, - "text": "Hello World，这里是阿里巴巴语音实验室。", - "words": [ - {"begin_time": 760, "end_time": 1040, "fixed": true, "punctuation": "", "text": "Hello"}, - {"begin_time": 1040, "end_time": 1240, "fixed": true, "punctuation": "，", "text": " World"}, - {"begin_time": 1360, "end_time": 1880, "fixed": true, "punctuation": "", "text": "这里是"}, - {"begin_time": 1880, "end_time": 2520, "fixed": true, "punctuation": "", "text": "阿里巴巴"}, - {"begin_time": 2520, "end_time": 2840, "fixed": true, "punctuation": "", "text": "语音"}, - {"begin_time": 2840, "end_time": 3800, "fixed": true, "punctuation": "。", "text": "实验室"} - ] - }, - "text": "Hello World，这里是阿里巴巴语音实验室。" - }, - "usage": { - "duration": 4 - }, - "request_id": "40e0734d-096f-9ae3-86c1-a8c013287561" -} -``` - -## 流式 - -设置`X-DashScope-SSE: enable`时，服务端以Server-Sent Events协议返回识别结果。SSE事件格式如下： - -``` -id:{序列号} -event:result -:HTTP_STATUS/200 -data:{JSON数据} -``` - -返回示例： - -``` -id:1 -event:result -:HTTP_STATUS/200 -data:{"output":{"sentence":{"sentence_id":1,"sentence_end":true,"end_time":3800,"words":[{"end_time":1040,"punctuation":"","begin_time":760,"fixed":true,"text":"Hello"},{"end_time":1240,"punctuation":"，","begin_time":1040,"fixed":true,"text":" World"},{"end_time":1880,"punctuation":"","begin_time":1360,"fixed":true,"text":"这里是"},{"end_time":2520,"punctuation":"","begin_time":1880,"fixed":true,"text":"阿里巴巴"},{"end_time":2840,"punctuation":"","begin_time":2520,"fixed":true,"text":"语音"},{"end_time":3800,"punctuation":"。","begin_time":2840,"fixed":true,"text":"实验室"}],"begin_time":760,"text":"Hello World，这里是阿里巴巴语音实验室。","channel_id":0},"text":"Hello World，这里是阿里巴巴语音实验室。"},"usage":{"duration":4},"request_id":"fc1582e4-935c-9fc2-a482-a98bf43daa69"} -``` - -**request\_id** `_string_` - -本次请求的唯一标识。 - -**output** `_object_` - -输出结果。 - -**属性** - -**text** `_string_` - -当前累积的完整识别文本。 - -**sentence** `_object_` - -当前句子的详细信息。 - -**属性** - -**sentence\_id** `_integer_` - -句子编号，从1开始。 - -**sentence\_end** `_boolean_` - -是否为该句的最终结果。为`true`时表示该句识别完成。 - -**begin\_time** `_integer_` - -句子开始时间，单位毫秒。 - -**end\_time** `_integer_` - -句子结束时间，单位毫秒。仅在`sentence_end`为`true`时返回。 - -**text** `_string_` - -当前句子的识别文本。 - -**channel\_id** `_integer_` - -声道编号，从0开始。 - -**words** `_array_` - -词级别时间戳列表。 - -**属性** - -**text** `_string_` - -词文本。 - -**begin\_time** `_integer_` - -词开始时间，单位毫秒。 - -**end\_time** `_integer_` - -词结束时间，单位毫秒。 - -**punctuation** `_string_` - -词后的标点符号。无标点时为空字符串。 - -**fixed** `_boolean_` - -词是否已稳定。`false`表示后续事件中该词的时间戳可能调整。 - -**usage** `_object_` - -用量信息。仅在`sentence_end`为`true`时返回。 - -**属性** - -**duration** `_integer_` - -已处理的音频时长，单位秒。 - -### **SSE 流式结果处理逻辑** - -在流式模式下，客户端需关注以下处理要点： - -1. 每收到一个SSE事件，解析`data`字段中的JSON。 - -2. 通过`output.sentence.sentence_end`判断当前句子是否结束：当该值为`true`时，该句识别完成，词级时间戳已稳定，可作为最终结果使用；当该值为`false`时，识别仍在进行中，文本和时间戳可能在后续事件中更新。 - -3. `usage`信息仅在句子结束事件中返回，可用于计量音频处理时长。 - - -## **DashScope同步调用（Fun-ASR-R**ealtime**）** - -**重要** - -- 该功能只支持北京地域。 - -- 不支持SDK调用。 - - -### **服务端点** - -`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` - -调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - -**重要** - -阿里云百炼为华北2（北京）地域推出了业务空间专属域名，能够为推理请求提供卓越的性能和更高的稳定性，建议从 `dashscope.aliyuncs.com` 迁移至 `{WorkspaceId}.cn-beijing.maas.aliyuncs.com`。 - -`{WorkspaceId}`需要替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。现有域名仍可正常使用。 - -### **请求头** - -**参数** - -**类型** - -**是否必选** - -**说明** - -Authorization - -string - -是 - -鉴权令牌，格式为`Bearer `，使用时将“``”替换为实际的API Key。 - -Content-Type - -string - -是 - -请求体的媒体类型，固定为`application/json`。 - -X-DashScope-SSE - -string - -是 - -用于控制是否以SSE流式方式返回结果。设置为`enable`时开启SSE流式返回模式，服务端会分多次返回中间识别结果和最终结果；设置为`disable`或不传该参数则仅返回最终结果。 - -### **请求体** - -## 非流式 - -``` -curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ - --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ - --header "Content-Type: application/json" \ - --header "X-DashScope-SSE: disable" \ - --data '{ - "model": "fun-asr-realtime", - "input": { - "messages": [] - }, - "parameters": { - "audio_address": "https://example.com/audio/sample.mp3", - "format": "mp3" - }, - "resources": [] -}' -``` - -## 流式 - -``` -curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ - --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ - --header "Content-Type: application/json" \ - --header "X-DashScope-SSE: enable" \ - --data '{ - "model": "fun-asr-realtime", - "input": { - "messages": [] - }, - "parameters": { - "audio_address": "https://example.com/audio/sample.mp3", - "format": "mp3" - }, - "resources": [] -}' -``` - -**model** `_string_` **（必选）** - -模型名称。 - -取值范围： - -- `fun-asr-realtime`（稳定版模型） - -- `fun-asr-realtime-2026-02-28`（即 Fun-Realtime-ASR-preview） - - -**input** `_object_` **（条件必选）** - -输入信息。与输入音频文件URL方式（通过`parameters.audio_address`传入）二选一，使用Base64方式上传音频时需要填写。 - -**属性** - -**messages** `_array(object)_` **（必选）** - -消息列表。 - -**属性** - -**content** `_array(object)_` **（必选）** - -用户消息的内容。仅允许设置一组消息。 - -**属性** - -**audio** `_string_` **（必选）** - -待识别音频，采用Data URI格式传入Base64编码的音频数据。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 - -使用Base64方式上传音频时需要填写。 - -值由`data:{MIME_TYPE};base64,`前缀与Base64编码的音频数据拼接而成。支持的MIME类型包括`audio/wav`、`audio/mp3`等。 - -示例：`data:audio/wav;base64,{BASE64_ENCODED_DATA}` - -**role** `_string_` **（必选）** - -用户消息的角色，固定为`user`。使用Base64方式上传音频时需要填写。 - -**parameters** `_object_` **（必选）** - -模型参数。 - -**属性** - -**audio\_address** `_string_` **（条件必选）** - -音频文件URL地址。与Base64方式（通过`input.messages`传入）二选一，使用URL方式时必填。需为可公开访问的地址。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 - -**format** `_string_` **（必选）** - -音频格式。根据实际音频格式填写，支持`wav`、`mp3`、`opus`等。详情请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 - -**vad\_enabled** `_boolean_` （可选） - -是否启用端点检测（VAD）。 - -默认为`true`（启用 VAD 检测）。 - -设为 `false` 时关闭 VAD 检测；但当音频时长超过 1 分钟时，系统将自动启用 VAD 检测，忽略此参数设置。 - -**resources** `_array_` （可选） - -资源列表，预留字段，当前可传空数组`[]`。 - -### **返回体** - -## 非流式 - -``` -{ - "output": { - "sentence": { - "begin_time": 160, - "channel_id": 0, - "end_time": 1680, - "sentence_end": true, - "sentence_id": 1, - "text": "欢迎使用阿里云。", - "words": [ - {"begin_time": 160, "end_time": 520, "fixed": true, "punctuation": "", "text": "欢迎"}, - {"begin_time": 520, "end_time": 880, "fixed": true, "punctuation": "", "text": "使用"}, - {"begin_time": 880, "end_time": 1280, "fixed": true, "punctuation": "", "text": "阿里"}, - {"begin_time": 1280, "end_time": 1680, "fixed": true, "punctuation": "。", "text": "云"} - ] - }, - "text": "欢迎使用阿里云。" - }, - "usage": { - "duration": 2 - }, - "request_id": "eff4c092-2289-9b43-a4cd-80e591fa90f5" -} -``` - -## 流式 - -设置`X-DashScope-SSE: enable`时，服务端以Server-Sent Events协议分多次返回中间识别结果和最终结果。每个SSE事件格式如下： - -``` -id:{序列号} -event:result -:HTTP_STATUS/200 -data:{JSON数据} -``` - -中间结果（句子开始、词逐步增长）： - -``` -id:1 -event:result -:HTTP_STATUS/200 -data:{"output":{"sentence":{"sentence_id":1,"sentence_end":false,"sentence_begin":true,"words":[],"begin_time":0,"text":"","channel_id":0},"text":""},"request_id":"372d19b3-993f-9288-adf0-a99f7606bd30"} - -id:2 -event:result -:HTTP_STATUS/200 -data:{"output":{"sentence":{"words":[{"end_time":520,"punctuation":"","begin_time":160,"fixed":false,"text":"欢迎"}],"begin_time":160,"text":"欢迎","channel_id":0,"sentence_id":1,"sentence_end":false},"text":"欢迎"},"request_id":"372d19b3-993f-9288-adf0-a99f7606bd30"} - -id:3 -event:result -:HTTP_STATUS/200 -data:{"output":{"sentence":{"words":[{"end_time":520,"punctuation":"","begin_time":160,"fixed":false,"text":"欢迎"},{"end_time":880,"punctuation":"","begin_time":520,"fixed":false,"text":"使用"}],"begin_time":160,"text":"欢迎使用","channel_id":0,"sentence_id":1,"sentence_end":false},"text":"欢迎使用"},"request_id":"372d19b3-993f-9288-adf0-a99f7606bd30"} -``` - -最终结果（句子结束，包含`usage`）： - -``` -id:4 -event:result -:HTTP_STATUS/200 -data:{"output":{"sentence":{"sentence_id":1,"sentence_end":true,"end_time":1680,"words":[{"end_time":520,"punctuation":"","begin_time":160,"fixed":true,"text":"欢迎"},{"end_time":880,"punctuation":"","begin_time":520,"fixed":true,"text":"使用"},{"end_time":1280,"punctuation":"","begin_time":880,"fixed":true,"text":"阿里"},{"end_time":1680,"punctuation":"。","begin_time":1280,"fixed":true,"text":"云"}],"begin_time":160,"text":"欢迎使用阿里云。","channel_id":0},"text":"欢迎使用阿里云。"},"usage":{"duration":2},"request_id":"372d19b3-993f-9288-adf0-a99f7606bd30"} -``` - -**request\_id** `_string_` - -本次请求的唯一标识。 - -**output** `_object_` - -输出结果。 - -**属性** - -**text** `_string_` - -当前累积的完整识别文本。 - -**sentence** `_object_` - -当前句子的详细信息。 - -**属性** - -**sentence\_id** `_integer_` - -句子编号，从1开始。 - -**sentence\_end** `_boolean_` - -是否为该句的最终结果。为`true`时表示该句识别完成。 - -**begin\_time** `_integer_` - -句子开始时间，单位毫秒。 - -**end\_time** `_integer_` - -句子结束时间，单位毫秒。仅在`sentence_end`为`true`时返回。 - -**text** `_string_` - -当前句子的识别文本。 - -**channel\_id** `_integer_` - -声道编号，从0开始。 - -**words** `_array_` - -词级别时间戳列表。 - -**属性** - -**text** `_string_` - -词文本。 - -**begin\_time** `_integer_` - -词开始时间，单位毫秒。 - -**end\_time** `_integer_` - -词结束时间，单位毫秒。 - -**punctuation** `_string_` - -词后的标点符号。无标点时为空字符串。 - -**fixed** `_boolean_` - -词是否已稳定。`false`表示后续事件中该词的时间戳可能调整。 - -**usage** `_object_` - -用量信息。仅在`sentence_end`为`true`时返回。 - -**属性** - -**duration** `_integer_` - -已处理的音频时长，单位秒。 - -### **SSE 流式结果处理逻辑** - -在流式模式下，客户端需关注以下处理要点： - -1. 每收到一个SSE事件，解析`data`字段中的JSON。 - -2. 通过`output.sentence.sentence_end`判断当前句子是否结束：当该值为`true`时，该句识别完成，词级时间戳已稳定，可作为最终结果使用；当该值为`false`时，识别仍在进行中，文本和时间戳可能在后续事件中更新。 - -3. `usage`信息仅在句子结束事件中返回，可用于计量音频处理时长。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-ios-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-ios-sdk.md index e59f1695..928864d9 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-ios-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-ios-sdk.md @@ -1,6 +1,6 @@ -# Fun-ASR录音文件识别iOS SDK +# Fun-ASR非实时语音识别iOS SDK -本文档提供了Fun-ASR录音文件识别iOS SDK的详细使用指南，帮助您将语音转换为文本。 +本文档提供了Fun-ASR非实时语音识别iOS SDK的详细使用指南，帮助您将语音转换为文本。 **用户指南：**[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide)。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 @@ -102,7 +102,7 @@ 是 - 运行模式。录音文件识别固定为 `"1"`。 + 运行模式。非实时语音识别固定为 `"1"`。 `device_id` @@ -173,7 +173,7 @@ ``` { "file_urls": [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ], "async_request": false, "nls_config": { @@ -590,7 +590,7 @@ ``` { "file_urls": [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ], "async_request": false, "nls_config": { diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-java-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-java-sdk.md index 8008ce57..47e93c0f 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-java-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/fun-asr-recorded-speech-recognition-java-sdk.md @@ -1,6 +1,6 @@ -# Fun-ASR录音文件识别Java SDK +# Fun-ASR非实时语音识别Java SDK -本文介绍Fun-ASR录音文件识别Java SDK的参数和接口细节。 +本文介绍Fun-ASR非实时语音识别Java SDK的参数和接口细节。 **重要** @@ -32,7 +32,7 @@ ## **快速开始** -[核心类（Transcription）](#adcb5e9bddbyq)提供了异步提交任务、同步等待任务结束和异步查询任务执行结果的接口。可通过如下两种调用方式进行录音文件识别： +[核心类（Transcription）](#adcb5e9bddbyq)提供了异步提交任务、同步等待任务结束和异步查询任务执行结果的接口。可通过如下两种调用方式进行非实时语音识别： - 异步提交任务+同步等待任务结束：提交任务后，阻塞当前线程直到任务结束并获取识别结果。 @@ -41,7 +41,7 @@ ### **异步提交任务+同步等待任务结束** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/4867892871/CAEQURiBgMCO2_fRpxkiIDBlNzI4YmMyNTU3ODRlM2Y4NjUxZWU4YmUxNjliMmFl4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6278074871/CAEQURiBgMCO2_fRpxkiIDBlNzI4YmMyNTU3ODRlM2Y4NjUxZWU4YmUxNjliMmFl4709861_20241015153444.149.svg) 1. 配置[请求参数](#48ea212b1d08r)。 @@ -85,7 +85,7 @@ public class Main { .model("fun-asr") // 此处以fun-asr为例，可按需更换模型名称。模型列表：https://help.aliyun.com/zh/model-studio/models .fileUrls( Arrays.asList( - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav")) + "{YOUR_AUDIO_URL}")) .build(); try { Transcription transcription = new Transcription(); @@ -107,7 +107,7 @@ public class Main { ### **异步提交任务+异步查询任务执行结果** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/4867892871/CAEQURiBgIDnxvjRpxkiIGI1NjJjOTgyNTVhMTRiMjM4OWVjYzFmZTExNGZjYzE14709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6278074871/CAEQURiBgIDnxvjRpxkiIGI1NjJjOTgyNTVhMTRiMjM4OWVjYzFmZTExNGZjYzE14709861_20241015153444.149.svg) 1. 配置[请求参数](#48ea212b1d08r)。 @@ -152,7 +152,7 @@ public class Main { .model("fun-asr") // 此处以fun-asr为例，可按需更换模型名称。模型列表：https://help.aliyun.com/zh/model-studio/models .fileUrls( Arrays.asList( - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav")) + "{YOUR_AUDIO_URL}")) .build(); try { Transcription transcription = new Transcription(); @@ -188,7 +188,7 @@ TranscriptionParam param = TranscriptionParam.builder() .model("fun-asr") .fileUrls( Arrays.asList( - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav")) + "{YOUR_AUDIO_URL}")) .build(); ``` @@ -507,7 +507,7 @@ public JsonObject getOutput() "end_time":"2025-02-13 16:12:10.189", "results":[ { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "transcription_url":"https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/prod/paraformer-v2/20250213/16%3A12/3baafe5f-d09d-46c6-8b01-724927670edb-1.json?Expires=1739520730&OSSAccessKeyId=yourOSSAccessKeyId&Signature=BF7vPxlsJN9hkJlY%2BLReezxOwK8%3D", "subtask_status":"SUCCEEDED" } @@ -533,7 +533,7 @@ public JsonObject getOutput() "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" @@ -615,7 +615,7 @@ public String getMessage() ``` { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "properties":{ "audio_format":"pcm_s16le", "channels":[ @@ -775,7 +775,7 @@ TranscriptionParam param = .model("fun-asr") .fileUrls( Arrays.asList( - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav")) + "{YOUR_AUDIO_URL}")) .build(); try { Transcription transcription = new Transcription(); @@ -854,7 +854,7 @@ public TranscriptionResult fetch(TranscriptionQueryParam queryParam) ## **其他接口：批量查询任务状态/取消任务** -详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的录音文件识别任务，同时支持取消`PENDING`（排队）状态的任务。 +详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的非实时语音识别任务，同时支持取消`PENDING`（排队）状态的任务。 ## **错误码** @@ -873,7 +873,7 @@ public TranscriptionResult fetch(TranscriptionQueryParam queryParam) "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/funauidio-asr-recorded-speech-recognition-python-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/funauidio-asr-recorded-speech-recognition-python-sdk.md index 97a0b5b3..051845e2 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/funauidio-asr-recorded-speech-recognition-python-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/fun-asr-recorded-speech-recognition-api-reference/funauidio-asr-recorded-speech-recognition-python-sdk.md @@ -1,6 +1,6 @@ -# Fun-ASR录音文件识别Python SDK +# Fun-ASR非实时语音识别Python SDK -本文介绍Fun-ASR录音文件识别Python SDK的参数和接口细节。 +本文介绍Fun-ASR非实时语音识别Python SDK的参数和接口细节。 **重要** @@ -32,7 +32,7 @@ ## **快速开始** -[核心类（Transcription）](#adcb5e9bddbyq)提供了异步提交任务、同步等待任务结束和异步查询任务执行结果的接口。可通过如下两种调用方式进行录音文件识别： +[核心类（Transcription）](#adcb5e9bddbyq)提供了异步提交任务、同步等待任务结束和异步查询任务执行结果的接口。可通过如下两种调用方式进行非实时语音识别： - 异步提交任务+同步等待任务结束：提交任务后，阻塞当前线程直到任务结束并获取识别结果。 @@ -41,7 +41,7 @@ ### **异步提交任务+同步等待任务结束** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/5594892871/CAEQURiBgMCvo5zjpxkiIDQyNzUwZjVjMWM3MjQ5Nzg4ODBjNDRjNzE1ZGFiOGFj4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3178074871/CAEQURiBgMCvo5zjpxkiIDQyNzUwZjVjMWM3MjQ5Nzg4ODBjNDRjNzE1ZGFiOGFj4709861_20241015153444.149.svg) 1. 调用[核心类（Transcription）](#adcb5e9bddbyq)的`async_call`方法并设置[请求参数](#340f6879fci7d)。 @@ -77,7 +77,7 @@ dashscope.api_key = os.getenv("DASHSCOPE_API_KEY") task_response = Transcription.async_call( model='fun-asr', - file_urls=['https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav'] + file_urls=['{YOUR_AUDIO_URL}'] ) transcribe_response = Transcription.wait(task=task_response.output.task_id) @@ -88,7 +88,7 @@ if transcribe_response.status_code == HTTPStatus.OK: ### **异步提交任务+异步查询任务执行结果** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/5594892871/CAEQURiBgMCN3qzkpxkiIGE0YmU4YTdjMWNiNzRmYjJhMjFlMWZkZmFmOWQ1NmEx4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3178074871/CAEQURiBgMCN3qzkpxkiIGE0YmU4YTdjMWNiNzRmYjJhMjFlMWZkZmFmOWQ1NmEx4709861_20241015153444.149.svg) 1. 调用[核心类（Transcription）](#adcb5e9bddbyq)的`async_call`方法并设置[请求参数](#340f6879fci7d)。 @@ -124,7 +124,7 @@ dashscope.api_key = os.getenv("DASHSCOPE_API_KEY") transcribe_response = Transcription.async_call( model='fun-asr', - file_urls=['https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav'] + file_urls=['{YOUR_AUDIO_URL}'] ) while True: @@ -407,7 +407,7 @@ list\[str\] "end_time":"2025-02-13 17:31:21.867", "results":[ { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "transcription_url":"https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/prod/paraformer-v2/20250213/17%3A31/20ee4e4f-0404-4806-b617-c7d4c62eed19-1.json?Expires=1739525481&OSSAccessKeyId=yourOSSAccessKeyId&Signature=3q%2B1uQmRwltd7FPn5HQM2mBKw74%3D", "subtask_status":"SUCCEEDED" } @@ -440,7 +440,7 @@ list\[str\] "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" @@ -563,7 +563,7 @@ transcription\_url "end_time":"2025-02-13 17:59:28.828", "results":[ { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "transcription_url":"https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/prod/paraformer-v2/20250213/17%3A59/70e737cc-bf8c-418b-b0c8-83fab192a0fa-1.json?Expires=1739527168&OSSAccessKeyId=yourOSSAccessKeyId&Signature=AtGjIKI%2BdgbzjJIu%2BHsr1R5nSAY%3D", "subtask_status":"SUCCEEDED" } @@ -589,7 +589,7 @@ transcription\_url "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" @@ -657,7 +657,7 @@ transcription\_url ``` { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "properties":{ "audio_format":"pcm_s16le", "channels":[ @@ -857,7 +857,7 @@ def fetch(cls, ## **其他接口：批量查询任务状态/取消任务** -详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的录音文件识别任务，同时支持取消`PENDING`（排队）状态的任务。 +详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的非实时语音识别任务，同时支持取消`PENDING`（排队）状态的任务。 ## **错误码** @@ -876,7 +876,7 @@ def fetch(cls, "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-flash.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-flash.md new file mode 100644 index 00000000..ecf7e674 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-flash.md @@ -0,0 +1,561 @@ +# 非实时语音识别（Fun-ASR-Flash）API参考 + +本文介绍Fun-ASR-Flash非实时语音识别HTTP API的参数和接口细节。 + +**用户指南：**[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide)。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 + +**重要** + +该功能不支持SDK调用。 + +## **服务端点** + +## 华北2（北京） + +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` + +调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + +## 新加坡 + +`POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` + +调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + +**重要** + +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，能够为推理请求提供卓越的性能和更高的稳定性，建议迁移至新域名： + +- 华北2（北京）地域：从 `dashscope.aliyuncs.com` 迁移至 `{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `dashscope-intl.aliyuncs.com` 迁移至 `{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +`{WorkspaceId}`需要替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。现有域名仍可正常使用。 + +## **请求头** + +**参数** + +**类型** + +**是否必选** + +**说明** + +Authorization + +string + +是 + +鉴权令牌，格式为`Bearer `，使用时将"``"替换为实际的API Key。 + +Content-Type + +string + +是 + +请求体的媒体类型，固定为`application/json`。 + +X-DashScope-SSE + +string + +是 + +用于控制是否以SSE流式方式返回结果。设置为`enable`时开启SSE流式返回模式，服务端会分多次返回中间识别结果和最终结果；设置为`disable`或不传该参数则仅返回最终结果。 + +## **请求体** + +以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 + +## 非流式 + +``` +curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ + --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ + --header "Content-Type: application/json" \ + --header "X-DashScope-SSE: disable" \ + --data '{ + "model": "fun-asr-flash-2026-06-15", + "input": { + "messages": [ + { + "role": "user", + "content": [ + { + "type": "input_audio", + "input_audio": { + "data": "{YOUR_AUDIO_URL}" + } + } + ] + } + ] + }, + "parameters": { + "format": "wav", + "sample_rate": "16000" + } +}' +``` + +## 流式 + +``` +curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ + --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ + --header "Content-Type: application/json" \ + --header "X-DashScope-SSE: enable" \ + --data '{ + "model": "fun-asr-flash-2026-06-15", + "input": { + "messages": [ + { + "role": "user", + "content": [ + { + "type": "input_audio", + "input_audio": { + "data": "{YOUR_AUDIO_URL}" + } + } + ] + } + ] + }, + "parameters": { + "format": "wav", + "sample_rate": "16000" + } +}' +``` + +## 携带上下文-非流式 + +``` +curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ + --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ + --header "Content-Type: application/json" \ + --header "X-DashScope-SSE: disable" \ + --data '{ + "model": "fun-asr-flash-2026-06-15", + "input": { + "messages": [ + { + "role": "user", + "content": [ + { + "type": "input_text", + "text": "你好啊" + } + ] + }, + { + "role": "assistant", + "content": [ + { + "type": "text", + "text": "你好啊，我是通义千问，有什么可以帮助你的？" + } + ] + }, + { + "role": "user", + "content": [ + { + "type": "input_audio", + "input_audio": { + "data": "{YOUR_AUDIO_URL}" + } + } + ] + } + ] + }, + "parameters": { + "format": "wav", + "sample_rate": "16000" + } +}' +``` + +## 携带上下文-流式 + +``` +curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ + --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ + --header "Content-Type: application/json" \ + --header "X-DashScope-SSE: enable" \ + --data '{ + "model": "fun-asr-flash-2026-06-15", + "input": { + "messages": [ + { + "role": "user", + "content": [ + { + "type": "input_text", + "text": "你好啊" + } + ] + }, + { + "role": "assistant", + "content": [ + { + "type": "text", + "text": "你好啊，我是通义千问，有什么可以帮助你的？" + } + ] + }, + { + "role": "user", + "content": [ + { + "type": "input_audio", + "input_audio": { + "data": "{YOUR_AUDIO_URL}" + } + } + ] + } + ] + }, + "parameters": { + "format": "wav", + "sample_rate": "16000" + } +}' +``` + +## Base64 + +可输入Base64编码数据（[Data URL](https://www.rfc-editor.org/rfc/rfc2397)），格式为：`data:;base64,`。 + +- ``：MIME类型 + + 因音频格式而异，例如： + + - WAV：`audio/wav` + + - MP3：`audio/mpeg` + +- ``：音频转成的Base64编码的字符串 + + Base64编码会增大体积，请控制原文件大小，确保编码后仍符合输入音频大小限制（10MB） + +- 示例：`data:audio/wav;base64,SUQzBAAAAAAAI1RTU0UAAAAPAAADTGF2ZjU4LjI5LjEwMAAAAAAAAAAAAAAA//PAxABQ/BXRbMPe4IQAhl9` + + **点击查看示例代码** + + Python + + ``` + import base64, pathlib + + # 请替换为自己的音频文件路径，确保其符合音频要求 + file_path = pathlib.Path("{YOUR_AUDIO_FILE}") + base64_str = base64.b64encode(file_path.read_bytes()).decode() + data_uri = f"data:audio/mpeg;base64,{base64_str}" + ``` + + Java + + ``` + import java.nio.file.*; + import java.util.Base64; + + public class Main { + /** + * 请替换为自己的音频文件路径，确保其符合音频要求 + */ + public static String toDataUrl(String filePath) throws Exception { + byte[] bytes = Files.readAllBytes(Paths.get(filePath)); + String encoded = Base64.getEncoder().encodeToString(bytes); + return "data:audio/mpeg;base64," + encoded; + } + + public static void main(String[] args) throws Exception { + System.out.println(toDataUrl("{YOUR_AUDIO_FILE}")); + } + } + ``` + + +``` +import base64, pathlib +import os +import requests + +# 请替换为自己的音频文件路径，确保其符合音频要求 +file_path = pathlib.Path("{YOUR_AUDIO_FILE}") +base64_str = base64.b64encode(file_path.read_bytes()).decode() +data_uri = f"data:audio/wav;base64,{base64_str}" + +url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation" + +headers = { + "Authorization": f"Bearer {os.environ['DASHSCOPE_API_KEY']}", + "Content-Type": "application/json", + "X-DashScope-SSE": "disable", +} + +payload = { + "model": "fun-asr-flash-2026-06-15", + "input": { + "messages": [ + { + "role": "user", + "content": [ + { + "type": "input_audio", + "input_audio": { + "data": data_uri, + }, + } + ], + } + ] + }, + "parameters": { + "format": "wav", + "sample_rate": "16000", + }, +} + +response = requests.post(url, headers=headers, json=payload) +print(response.status_code) +print(response.json()) +``` + +**model** `_string_` **（必选）** + +模型名称，固定为`fun-asr-flash-2026-06-15`。 + +**input** `_object_` **（必选）** + +输入信息。 + +**属性** + +**messages** `_array(object)_` **（必选）** + +消息列表。包含当前待识别的音频，以及可选的对话上下文（用于提升识别效果）。 + +**重要** + +上下文功能用于提升专有词汇的识别准确率，使用方法详见[快速开始](https://help.aliyun.com/zh/model-studio/improve-asr-accuracy#ctx-quickstart-sec)。约束：上下文消息（`input_text` 和 `text` 类型）各最多 5 条，超出时保留最近的 5 条。每轮上下文文本总长度（`user` 和 `assistant` 的 `text` 字段长度之和）不超过 400 个字符（按字符数计算，每个字符计为 1），超出部分从末尾截断。 + +**重要** + +携带上下文时，`messages` 中的消息顺序有要求：上下文消息必须按对话轮次排列，每轮中 `user`（`input_text` 类型）必须在对应的 `assistant`（`text` 类型）之前；包含 `input_audio` 的 `user` 消息必须放在 `messages` 数组的最后。 + +**属性** + +**role** `_string_` **（必选）** + +消息角色。取值范围： + +- `user`（必选）：用户消息。type为`input_audio`时表示当前待识别的音频；type为`input_text`时表示前几轮的识别结果或领域相关的词表（可选，上下文）。 + +- `assistant`（可选，上下文）：前几轮大语言模型的回复内容。 + + +**content** `_array(object)_` **（必选）** + +消息内容列表。 + +**属性** + +**type** `_string_` **（必选）** + +内容类型。每个请求至少需要一条`input_audio`类型的消息。取值范围： + +- `input_audio`（必选）：当前待识别的音频输入（role为user），需同时传入`input_audio`对象。 + +- `input_text`（可选，上下文）：前几轮用户语音的识别结果或领域相关的词表（role为user），需同时传入`text`字段。 + +- `text`（可选，上下文）：前几轮大语言模型的回复内容（role为assistant），需同时传入`text`字段。 + + +**input\_audio** `_object_` **（条件必选）** + +当`type`为`input_audio`时必填。 + +**属性** + +**data** `_string_` **（必选）** + +待识别音频数据。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。支持以下两种方式： + +- **音频文件URL**：直接传入可公开访问的音频文件地址。 + +- **Base64 Data URI**：采用Data URI格式传入Base64编码的音频数据，值由`data:{MIME_TYPE};base64,`前缀与Base64编码的音频数据拼接而成。支持的MIME类型包括`audio/wav`、`audio/mp3`等。 + + +示例（URL方式）：`https://example.com/audio/sample.wav` + +示例（Base64方式）：`data:audio/wav;base64,{BASE64_ENCODED_DATA}` + +**text** `_string_` **（条件必选）** + +当`type`为`input_text`时，填入前几轮用户语音的识别结果或领域相关的词表；当`type`为`text`时，填入前几轮大语言模型的回复内容。文本按字符数计算，每个字符计为 1。每轮上下文中所有消息的 `text` 字段长度之和不超过 400 个字符，超出部分从末尾截断。 + +**parameters** `_object_` **（必选）** + +模型参数。 + +**属性** + +**format** `_string_` **（必选）** + +音频格式。根据实际音频格式填写，支持`wav`、`mp3`、`opus`等。详情请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 + +**sample\_rate** `_string_` （可选） + +音频采样率，单位Hz。例如`16000`表示16kHz采样率。详情请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 + +## **返回体** + +## 非流式 + +``` +{ + "output": { + "sentence": { + "begin_time": 760, + "channel_id": 0, + "end_time": 3800, + "sentence_end": true, + "sentence_id": 1, + "text": "Hello World，这里是阿里巴巴语音实验室。", + "words": [ + {"begin_time": 760, "end_time": 1040, "fixed": true, "punctuation": "", "text": "Hello"}, + {"begin_time": 1040, "end_time": 1240, "fixed": true, "punctuation": "，", "text": " World"}, + {"begin_time": 1360, "end_time": 1880, "fixed": true, "punctuation": "", "text": "这里是"}, + {"begin_time": 1880, "end_time": 2520, "fixed": true, "punctuation": "", "text": "阿里巴巴"}, + {"begin_time": 2520, "end_time": 2840, "fixed": true, "punctuation": "", "text": "语音"}, + {"begin_time": 2840, "end_time": 3800, "fixed": true, "punctuation": "。", "text": "实验室"} + ] + }, + "text": "Hello World，这里是阿里巴巴语音实验室。" + }, + "usage": { + "duration": 4 + }, + "request_id": "40e0734d-096f-9ae3-86c1-a8c013287561" +} +``` + +## 流式 + +设置`X-DashScope-SSE: enable`时，服务端以Server-Sent Events协议返回识别结果。SSE事件格式如下： + +``` +id:{序列号} +event:result +:HTTP_STATUS/200 +data:{JSON数据} +``` + +返回示例： + +``` +id:1 +event:result +:HTTP_STATUS/200 +data:{"output":{"sentence":{"sentence_id":1,"sentence_end":true,"end_time":3800,"words":[{"end_time":1040,"punctuation":"","begin_time":760,"fixed":true,"text":"Hello"},{"end_time":1240,"punctuation":"，","begin_time":1040,"fixed":true,"text":" World"},{"end_time":1880,"punctuation":"","begin_time":1360,"fixed":true,"text":"这里是"},{"end_time":2520,"punctuation":"","begin_time":1880,"fixed":true,"text":"阿里巴巴"},{"end_time":2840,"punctuation":"","begin_time":2520,"fixed":true,"text":"语音"},{"end_time":3800,"punctuation":"。","begin_time":2840,"fixed":true,"text":"实验室"}],"begin_time":760,"text":"Hello World，这里是阿里巴巴语音实验室。","channel_id":0},"text":"Hello World，这里是阿里巴巴语音实验室。"},"usage":{"duration":4},"request_id":"fc1582e4-935c-9fc2-a482-a98bf43daa69"} +``` + +**request\_id** `_string_` + +本次请求的唯一标识。 + +**output** `_object_` + +输出结果。 + +**属性** + +**text** `_string_` + +当前累积的完整识别文本。 + +**sentence** `_object_` + +当前句子的详细信息。 + +**属性** + +**sentence\_id** `_integer_` + +句子编号，从1开始。 + +**sentence\_end** `_boolean_` + +是否为该句的最终结果。为`true`时表示该句识别完成。 + +**begin\_time** `_integer_` + +句子开始时间，单位毫秒。 + +**end\_time** `_integer_` + +句子结束时间，单位毫秒。仅在`sentence_end`为`true`时返回。 + +**text** `_string_` + +当前句子的识别文本。 + +**channel\_id** `_integer_` + +声道编号，从0开始。 + +**words** `_array_` + +词级别时间戳列表。 + +**属性** + +**text** `_string_` + +词文本。 + +**begin\_time** `_integer_` + +词开始时间，单位毫秒。 + +**end\_time** `_integer_` + +词结束时间，单位毫秒。 + +**punctuation** `_string_` + +词后的标点符号。无标点时为空字符串。 + +**fixed** `_boolean_` + +词是否已稳定。`false`表示后续事件中该词的时间戳可能调整。 + +**usage** `_object_` + +用量信息。仅在`sentence_end`为`true`时返回。 + +**属性** + +**duration** `_integer_` + +已处理的音频时长，单位秒。 + +## **SSE 流式结果处理逻辑** + +在流式模式下，客户端需关注以下处理要点： + +1. 每收到一个SSE事件，解析`data`字段中的JSON。 + +2. 通过`output.sentence.sentence_end`判断当前句子是否结束：当该值为`true`时，该句识别完成，词级时间戳已稳定，可作为最终结果使用；当该值为`false`时，识别仍在进行中，文本和时间戳可能在后续事件中更新。 + +3. `usage`信息仅在句子结束事件中返回，可用于计量音频处理时长。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-realtime.md new file mode 100644 index 00000000..3c19542d --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/non-real-time-speech-recognition-for-fun-asr-realtime.md @@ -0,0 +1,328 @@ +# 非实时语音识别（Fun-ASR-Realtime）API参考 + +本文介绍Fun-ASR-Realtime非实时语音识别HTTP API的参数和接口细节。 + +**用户指南：**[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide)。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 + +**重要** + +- 该功能只支持北京地域。 + +- 不支持SDK调用。 + + +## **服务端点** + +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` + +调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + +**重要** + +阿里云百炼为华北2（北京）地域推出了业务空间专属域名，能够为推理请求提供卓越的性能和更高的稳定性，建议从 `dashscope.aliyuncs.com` 迁移至 `{WorkspaceId}.cn-beijing.maas.aliyuncs.com`。 + +`{WorkspaceId}`需要替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。现有域名仍可正常使用。 + +## **请求头** + +**参数** + +**类型** + +**是否必选** + +**说明** + +Authorization + +string + +是 + +鉴权令牌，格式为`Bearer `，使用时将“``”替换为实际的API Key。 + +Content-Type + +string + +是 + +请求体的媒体类型，固定为`application/json`。 + +X-DashScope-SSE + +string + +是 + +用于控制是否以SSE流式方式返回结果。设置为`enable`时开启SSE流式返回模式，服务端会分多次返回中间识别结果和最终结果；设置为`disable`或不传该参数则仅返回最终结果。 + +## **请求体** + +## 非流式 + +``` +curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ + --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ + --header "Content-Type: application/json" \ + --header "X-DashScope-SSE: disable" \ + --data '{ + "model": "fun-asr-realtime", + "input": { + "messages": [] + }, + "parameters": { + "audio_address": "https://example.com/audio/sample.mp3", + "format": "mp3" + }, + "resources": [] +}' +``` + +## 流式 + +``` +curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ + --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ + --header "Content-Type: application/json" \ + --header "X-DashScope-SSE: enable" \ + --data '{ + "model": "fun-asr-realtime", + "input": { + "messages": [] + }, + "parameters": { + "audio_address": "https://example.com/audio/sample.mp3", + "format": "mp3" + }, + "resources": [] +}' +``` + +**model** `_string_` **（必选）** + +模型名称。 + +取值范围： + +- `fun-asr-realtime`（稳定版模型） + +- `fun-asr-realtime-2026-02-28`（即 Fun-Realtime-ASR-preview） + + +**input** `_object_` **（条件必选）** + +输入信息。与输入音频文件URL方式（通过`parameters.audio_address`传入）二选一，使用Base64方式上传音频时需要填写。 + +**属性** + +**messages** `_array(object)_` **（必选）** + +消息列表。 + +**属性** + +**content** `_array(object)_` **（必选）** + +用户消息的内容。仅允许设置一组消息。 + +**属性** + +**audio** `_string_` **（必选）** + +待识别音频，采用Data URI格式传入Base64编码的音频数据。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 + +使用Base64方式上传音频时需要填写。 + +值由`data:{MIME_TYPE};base64,`前缀与Base64编码的音频数据拼接而成。支持的MIME类型包括`audio/wav`、`audio/mp3`等。 + +示例：`data:audio/wav;base64,{BASE64_ENCODED_DATA}` + +**role** `_string_` **（必选）** + +用户消息的角色，固定为`user`。使用Base64方式上传音频时需要填写。 + +**parameters** `_object_` **（必选）** + +模型参数。 + +**属性** + +**audio\_address** `_string_` **（条件必选）** + +音频文件URL地址。与Base64方式（通过`input.messages`传入）二选一，使用URL方式时必填。需为可公开访问的地址。关于支持的音频格式、文件大小限制、时长限制等输入要求，请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 + +**format** `_string_` **（必选）** + +音频格式。根据实际音频格式填写，支持`wav`、`mp3`、`opus`等。详情请参见[音频规格](https://help.aliyun.com/zh/model-studio/asr-model/#asr-audio-spec02)。 + +**vad\_enabled** `_boolean_` （可选） + +是否启用端点检测（VAD）。 + +默认为`true`（启用 VAD 检测）。 + +设为 `false` 时关闭 VAD 检测；但当音频时长超过 1 分钟时，系统将自动启用 VAD 检测，忽略此参数设置。 + +**resources** `_array_` （可选） + +资源列表，预留字段，当前可传空数组`[]`。 + +## **返回体** + +## 非流式 + +``` +{ + "output": { + "sentence": { + "begin_time": 160, + "channel_id": 0, + "end_time": 1680, + "sentence_end": true, + "sentence_id": 1, + "text": "欢迎使用阿里云。", + "words": [ + {"begin_time": 160, "end_time": 520, "fixed": true, "punctuation": "", "text": "欢迎"}, + {"begin_time": 520, "end_time": 880, "fixed": true, "punctuation": "", "text": "使用"}, + {"begin_time": 880, "end_time": 1280, "fixed": true, "punctuation": "", "text": "阿里"}, + {"begin_time": 1280, "end_time": 1680, "fixed": true, "punctuation": "。", "text": "云"} + ] + }, + "text": "欢迎使用阿里云。" + }, + "usage": { + "duration": 2 + }, + "request_id": "eff4c092-2289-9b43-a4cd-80e591fa90f5" +} +``` + +## 流式 + +设置`X-DashScope-SSE: enable`时，服务端以Server-Sent Events协议分多次返回中间识别结果和最终结果。每个SSE事件格式如下： + +``` +id:{序列号} +event:result +:HTTP_STATUS/200 +data:{JSON数据} +``` + +中间结果（句子开始、词逐步增长）： + +``` +id:1 +event:result +:HTTP_STATUS/200 +data:{"output":{"sentence":{"sentence_id":1,"sentence_end":false,"sentence_begin":true,"words":[],"begin_time":0,"text":"","channel_id":0},"text":""},"request_id":"372d19b3-993f-9288-adf0-a99f7606bd30"} + +id:2 +event:result +:HTTP_STATUS/200 +data:{"output":{"sentence":{"words":[{"end_time":520,"punctuation":"","begin_time":160,"fixed":false,"text":"欢迎"}],"begin_time":160,"text":"欢迎","channel_id":0,"sentence_id":1,"sentence_end":false},"text":"欢迎"},"request_id":"372d19b3-993f-9288-adf0-a99f7606bd30"} + +id:3 +event:result +:HTTP_STATUS/200 +data:{"output":{"sentence":{"words":[{"end_time":520,"punctuation":"","begin_time":160,"fixed":false,"text":"欢迎"},{"end_time":880,"punctuation":"","begin_time":520,"fixed":false,"text":"使用"}],"begin_time":160,"text":"欢迎使用","channel_id":0,"sentence_id":1,"sentence_end":false},"text":"欢迎使用"},"request_id":"372d19b3-993f-9288-adf0-a99f7606bd30"} +``` + +最终结果（句子结束，包含`usage`）： + +``` +id:4 +event:result +:HTTP_STATUS/200 +data:{"output":{"sentence":{"sentence_id":1,"sentence_end":true,"end_time":1680,"words":[{"end_time":520,"punctuation":"","begin_time":160,"fixed":true,"text":"欢迎"},{"end_time":880,"punctuation":"","begin_time":520,"fixed":true,"text":"使用"},{"end_time":1280,"punctuation":"","begin_time":880,"fixed":true,"text":"阿里"},{"end_time":1680,"punctuation":"。","begin_time":1280,"fixed":true,"text":"云"}],"begin_time":160,"text":"欢迎使用阿里云。","channel_id":0},"text":"欢迎使用阿里云。"},"usage":{"duration":2},"request_id":"372d19b3-993f-9288-adf0-a99f7606bd30"} +``` + +**request\_id** `_string_` + +本次请求的唯一标识。 + +**output** `_object_` + +输出结果。 + +**属性** + +**text** `_string_` + +当前累积的完整识别文本。 + +**sentence** `_object_` + +当前句子的详细信息。 + +**属性** + +**sentence\_id** `_integer_` + +句子编号，从1开始。 + +**sentence\_end** `_boolean_` + +是否为该句的最终结果。为`true`时表示该句识别完成。 + +**begin\_time** `_integer_` + +句子开始时间，单位毫秒。 + +**end\_time** `_integer_` + +句子结束时间，单位毫秒。仅在`sentence_end`为`true`时返回。 + +**text** `_string_` + +当前句子的识别文本。 + +**channel\_id** `_integer_` + +声道编号，从0开始。 + +**words** `_array_` + +词级别时间戳列表。 + +**属性** + +**text** `_string_` + +词文本。 + +**begin\_time** `_integer_` + +词开始时间，单位毫秒。 + +**end\_time** `_integer_` + +词结束时间，单位毫秒。 + +**punctuation** `_string_` + +词后的标点符号。无标点时为空字符串。 + +**fixed** `_boolean_` + +词是否已稳定。`false`表示后续事件中该词的时间戳可能调整。 + +**usage** `_object_` + +用量信息。仅在`sentence_end`为`true`时返回。 + +**属性** + +**duration** `_integer_` + +已处理的音频时长，单位秒。 + +## **SSE 流式结果处理逻辑** + +在流式模式下，客户端需关注以下处理要点： + +1. 每收到一个SSE事件，解析`data`字段中的JSON。 + +2. 通过`output.sentence.sentence_end`判断当前句子是否结束：当该值为`true`时，该句识别完成，词级时间戳已稳定，可作为最终结果使用；当该值为`false`时，识别仍在进行中，文本和时间戳可能在后续事件中更新。 + +3. `usage`信息仅在句子结束事件中返回，可用于计量音频处理时长。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-real-time-speech-recognition-java-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-real-time-speech-recognition-java-sdk.md index 2872f1a4..b9a5a871 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-real-time-speech-recognition-java-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-real-time-speech-recognition-java-sdk.md @@ -143,14 +143,12 @@ 提交单个语音实时转写任务，通过传入本地文件的方式同步阻塞地拿到转写结果。 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9365892871/CAEQURiBgMDS0c2RpxkiIDNmYjBlMTE3ODQxYTQ3Nzk4MGMxNTc5MjY3OWVjZjlj4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3558074871/CAEQURiBgMDS0c2RpxkiIDNmYjBlMTE3ODQxYTQ3Nzk4MGMxNTc5MjY3OWVjZjlj4709861_20241015153444.149.svg) 实例化[Recognition类](#adcb5e9bddbyq)，调用`call`方法绑定[请求参数](#d72d661a1brzp)和待识别文件，进行识别并最终获取识别结果。 点击查看完整示例 -示例中用到的音频为：[asr\_example.wav](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250210/elouas/asr_example.wav)。 - ``` import com.alibaba.dashscope.audio.asr.recognition.Recognition; import com.alibaba.dashscope.audio.asr.recognition.RecognitionParam; @@ -161,7 +159,7 @@ import java.io.File; public class Main { public static void main(String[] args) { // 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 - Constants.baseWebsocketApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; + Constants.baseWebsocketApiUrl = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/inference"; // 创建Recognition实例 Recognition recognizer = new Recognition(); // 创建RecognitionParam @@ -177,7 +175,7 @@ public class Main { .build(); try { - System.out.println("识别结果：" + recognizer.call(param, new File("asr_example.wav"))); + System.out.println("识别结果：" + recognizer.call(param, new File("{YOUR_AUDIO_FILE}"))); } catch (Exception e) { e.printStackTrace(); } finally { @@ -200,7 +198,7 @@ public class Main { 提交单个语音实时转写任务，通过实现回调接口的方式流式输出实时识别结果。 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9365892871/CAEQURiBgID1ooWUpxkiIDcyOTEyYjZiZmUxNzRkZjVhMTNhYmNkYjI2NzYzYTMy4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3558074871/CAEQURiBgID1ooWUpxkiIDcyOTEyYjZiZmUxNzRkZjVhMTNhYmNkYjI2NzYzYTMy4709861_20241015153444.149.svg) 1. 启动流式语音识别 @@ -244,7 +242,7 @@ import java.util.concurrent.TimeUnit; public class Main { public static void main(String[] args) throws InterruptedException { // 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 - Constants.baseWebsocketApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; + Constants.baseWebsocketApiUrl = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/inference"; ExecutorService executorService = Executors.newSingleThreadExecutor(); executorService.submit(new RealtimeRecognitionTask()); executorService.shutdown(); @@ -332,8 +330,6 @@ class RealtimeRecognitionTask implements Runnable { ## 识别本地语音文件 -示例中用到的音频为：[asr\_example.wav](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250210/oiydrd/asr_example.wav)。 - ``` import com.alibaba.dashscope.audio.asr.recognition.Recognition; import com.alibaba.dashscope.audio.asr.recognition.RecognitionParam; @@ -364,9 +360,9 @@ class TimeUtils { public class Main { public static void main(String[] args) throws InterruptedException { // 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 - Constants.baseWebsocketApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; + Constants.baseWebsocketApiUrl = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/inference"; ExecutorService executorService = Executors.newSingleThreadExecutor(); - executorService.submit(new RealtimeRecognitionTask(Paths.get(System.getProperty("user.dir"), "asr_example.wav"))); + executorService.submit(new RealtimeRecognitionTask(Paths.get(System.getProperty("user.dir"), "{YOUR_AUDIO_FILE}"))); executorService.shutdown(); // wait for all tasks to complete @@ -503,7 +499,7 @@ import java.nio.ByteBuffer; public class Main { public static void main(String[] args) throws NoApiKeyException { // 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 - Constants.baseWebsocketApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; + Constants.baseWebsocketApiUrl = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/inference"; // 创建一个Flowable Flowable audioSource = Flowable.create( diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-real-time-speech-recognition-python-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-real-time-speech-recognition-python-sdk.md index bba41596..38973852 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-real-time-speech-recognition-python-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-real-time-speech-recognition-api-reference/paraformer-real-time-speech-recognition-python-sdk.md @@ -143,19 +143,17 @@ 提交单个语音实时转写任务，通过传入本地文件的方式同步阻塞地拿到转写结果。 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6455892871/CAEQURiBgMDS0c2RpxkiIDNmYjBlMTE3ODQxYTQ3Nzk4MGMxNTc5MjY3OWVjZjlj4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/4858074871/CAEQURiBgMDS0c2RpxkiIDNmYjBlMTE3ODQxYTQ3Nzk4MGMxNTc5MjY3OWVjZjlj4709861_20241015153444.149.svg) 实例化[Recognition类](#d6bc1f133f871)绑定[请求参数](#555007db2033f)，调用`call`进行识别/翻译并最终获取[识别结果（RecognitionResult）](#bc3e1a43d6hhy)。 点击查看完整示例 -示例中用到的音频为：[asr\_example.wav](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250210/iwaouc/asr_example.wav)。 - ``` from http import HTTPStatus from dashscope.audio.asr import Recognition # 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 -dashscope.base_websocket_api_url = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" +dashscope.base_websocket_api_url = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/inference" # 若没有将API Key配置到环境变量中，需将下面这行代码注释放开，并将apiKey替换为自己的API Key # import dashscope @@ -167,13 +165,15 @@ recognition = Recognition(model='paraformer-realtime-v2', # “language_hints”只支持paraformer-realtime-v2模型 language_hints=['zh', 'en'], callback=None) -result = recognition.call('asr_example.wav') +result = recognition.call('{YOUR_AUDIO_FILE}') if result.status_code == HTTPStatus.OK: print('识别结果：') - print(result.get_sentence()) + sentences = result.get_sentence() + for sentence in sentences: + print(sentence['text']) else: print('Error: ', result.message) - + print( '[Metric] requestId: {}, first package delay ms: {}, last package delay ms: {}' .format( @@ -183,11 +183,15 @@ print( )) ``` +`result.get_sentence()`在非流式调用（`call`）中返回**句子列表**（`List[Dict]`），每个元素为`Dict[str, Any]`，包含`text`（识别文本）、`begin_time` / `end_time`（时间戳）、`words`（字时间戳）等字段。如需获取识别文本，需遍历列表并通过`sentence['text']`提取。 + +在流式回调（`on_event`）中，`result.get_sentence()`返回**单句信息**（`Dict[str, Any]`），可直接使用`sentence['text']`获取识别文本。两种调用模式的返回值类型不同，详见[识别结果（RecognitionResult）](#bc3e1a43d6hhy)中的`get_sentence`方法说明。 + ### **双向流式调用** 提交单个语音实时转写任务，通过实现回调接口的方式流式输出实时识别结果。 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6455892871/CAEQURiBgIDvi..2pxkiIGE4NTc3Njg4ZGM2YzQ2NzVhZGI3MzE2YWUwYTA3OGEy4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/4858074871/CAEQURiBgIDvi..2pxkiIGE4NTc3Njg4ZGM2YzQ2NzVhZGI3MzE2YWUwYTA3OGEy4709861_20241015153444.149.svg) 1. 启动流式语音识别 @@ -221,7 +225,7 @@ import dashscope import pyaudio from dashscope.audio.asr import * # 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 -dashscope.base_websocket_api_url = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" +dashscope.base_websocket_api_url = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/inference" mic = None stream = None @@ -342,14 +346,12 @@ if __name__ == '__main__': ## 识别本地语音文件 -示例中用到的音频为：[asr\_example.wav](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250210/acoict/asr_example.wav)。 - ``` import os import time from dashscope.audio.asr import * # 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 -dashscope.base_websocket_api_url = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" +dashscope.base_websocket_api_url = "wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/inference" # 若没有将API Key配置到环境变量中，需将下面这行代码注释放开，并将apiKey替换为自己的API Key # import dashscope @@ -393,8 +395,8 @@ recognition.start() try: audio_data: bytes = None - f = open("asr_example.wav", 'rb') - if os.path.getsize("asr_example.wav"): + f = open("{YOUR_AUDIO_FILE}", 'rb') + if os.path.getsize("{YOUR_AUDIO_FILE}"): while True: audio_data = f.read(3200) if not audio_data: diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-android-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-android-sdk.md index ff800b01..697261be 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-android-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-android-sdk.md @@ -1,6 +1,6 @@ -# Paraformer录音文件识别Android SDK +# Paraformer非实时语音识别Android SDK -本文档提供了Paraformer录音文件识别Android SDK的详细使用指南，帮助您将语音转换为文本。 +本文档提供了Paraformer非实时语音识别Android SDK的详细使用指南，帮助您将语音转换为文本。 **用户指南：**[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide) @@ -17,7 +17,10 @@ - [下载最新SDK整合包](https://help.aliyun.com/zh/isi/sdk-selection-and-download)。 - 解压 ZIP 包。在 `app/libs` 目录中获取 AAR 格式 SDK，并添加到项目依赖。 - 需要 Android CPP 接入时，使用 ZIP 包内的 `android_libs` 与 `android_include` 获取动态库和头文件。 + 需要 Android CPP 接入时，使用 ZIP 包内的 `android_libs` 与 `android_include` 获取动态库和头文件。 + + + - 用 Android Studio 打开工程。示例代码位于`DashParaformerFileTranscriberActivity.java`，替换 API Key 后体验功能。 @@ -103,7 +106,7 @@ 是 - 运行模式。录音文件识别固定为 `"1"`。 + 运行模式。非实时语音识别固定为 `"1"`。 `device_id` @@ -174,7 +177,7 @@ ``` { "file_urls": [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ], "async_request": false, "nls_config": { @@ -617,7 +620,7 @@ ``` { "file_urls": [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ], "async_request": false, "nls_config": { diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-ios-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-ios-sdk.md index 2147dda6..0ecf2ea8 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-ios-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-ios-sdk.md @@ -1,6 +1,6 @@ -# Paraformer录音文件识别iOS SDK +# Paraformer非实时语音识别iOS SDK -本文档提供了Paraformer录音文件识别iOS SDK的详细使用指南，帮助您将语音转换为文本。 +本文档提供了Paraformer非实时语音识别iOS SDK的详细使用指南，帮助您将语音转换为文本。 **用户指南：**[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide) @@ -106,7 +106,7 @@ 是 - 运行模式。录音文件识别固定为 `"1"`。 + 运行模式。非实时语音识别固定为 `"1"`。 `device_id` @@ -177,7 +177,7 @@ ``` { "file_urls": [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ], "async_request": false, "nls_config": { @@ -611,7 +611,7 @@ ``` { "file_urls": [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ], "async_request": false, "nls_config": { diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-java-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-java-sdk.md index b3d42809..65269d08 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-java-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-java-sdk.md @@ -1,6 +1,6 @@ -# Paraformer录音文件识别Java SDK +# Paraformer非实时语音识别Java SDK -本文介绍Paraformer录音文件识别Java SDK的参数和接口细节。 +本文介绍Paraformer非实时语音识别Java SDK的参数和接口细节。 **重要** @@ -27,7 +27,7 @@ ## **快速开始** -[核心类（Transcription）](#adcb5e9bddbyq)提供了异步提交任务、同步等待任务结束和异步查询任务执行结果的接口。可通过如下两种调用方式进行录音文件识别： +[核心类（Transcription）](#adcb5e9bddbyq)提供了异步提交任务、同步等待任务结束和异步查询任务执行结果的接口。可通过如下两种调用方式进行非实时语音识别： - 异步提交任务+同步等待任务结束：提交任务后，阻塞当前线程直到任务结束并获取识别结果。 @@ -36,7 +36,7 @@ ### **异步提交任务+同步等待任务结束** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7695892871/CAEQURiBgMCO2_fRpxkiIDBlNzI4YmMyNTU3ODRlM2Y4NjUxZWU4YmUxNjliMmFl4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0509074871/CAEQURiBgMCO2_fRpxkiIDBlNzI4YmMyNTU3ODRlM2Y4NjUxZWU4YmUxNjliMmFl4709861_20241015153444.149.svg) 1. 配置[请求参数](#48ea212b1d08r)。 @@ -80,7 +80,7 @@ public class Main { .parameter("language_hints", new String[]{"zh", "en"}) .fileUrls( Arrays.asList( - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav")) + "{YOUR_AUDIO_URL}")) .build(); try { Transcription transcription = new Transcription(); @@ -102,7 +102,7 @@ public class Main { ### **异步提交任务+异步查询任务执行结果** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7695892871/CAEQURiBgIDnxvjRpxkiIGI1NjJjOTgyNTVhMTRiMjM4OWVjYzFmZTExNGZjYzE14709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0509074871/CAEQURiBgIDnxvjRpxkiIGI1NjJjOTgyNTVhMTRiMjM4OWVjYzFmZTExNGZjYzE14709861_20241015153444.149.svg) 1. 配置[请求参数](#48ea212b1d08r)。 @@ -147,7 +147,7 @@ public class Main { .parameter("language_hints", new String[]{"zh", "en"}) .fileUrls( Arrays.asList( - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav")) + "{YOUR_AUDIO_URL}")) .build(); try { Transcription transcription = new Transcription(); @@ -185,7 +185,7 @@ TranscriptionParam param = TranscriptionParam.builder() .parameter("language_hints", new String[]{"zh", "en"}) .fileUrls( Arrays.asList( - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav")) + "{YOUR_AUDIO_URL}")) .build(); ``` @@ -537,7 +537,7 @@ public JsonObject getOutput() "end_time":"2025-02-13 16:12:10.189", "results":[ { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "transcription_url":"https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/prod/paraformer-v2/20250213/16%3A12/3baafe5f-d09d-46c6-8b01-724927670edb-1.json?Expires=1739520730&OSSAccessKeyId=yourOSSAccessKeyId&Signature=BF7vPxlsJN9hkJlY%2BLReezxOwK8%3D", "subtask_status":"SUCCEEDED" } @@ -563,7 +563,7 @@ public JsonObject getOutput() "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" @@ -645,7 +645,7 @@ public String getMessage() ``` { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "properties":{ "audio_format":"pcm_s16le", "channels":[ @@ -807,7 +807,7 @@ TranscriptionParam param = .parameter("language_hints", new String[]{"zh", "en"}) .fileUrls( Arrays.asList( - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav")) + "{YOUR_AUDIO_URL}")) .build(); try { Transcription transcription = new Transcription(); @@ -886,7 +886,7 @@ public TranscriptionResult fetch(TranscriptionQueryParam queryParam) ## **其他接口：批量查询任务状态/取消任务** -详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的录音文件识别任务，同时支持取消`PENDING`（排队）状态的任务。 +详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的非实时语音识别任务，同时支持取消`PENDING`（排队）状态的任务。 ## **错误码** @@ -907,7 +907,7 @@ public TranscriptionResult fetch(TranscriptionQueryParam queryParam) "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-python-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-python-sdk.md index e3ffe0b6..175befce 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-python-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-python-sdk.md @@ -1,6 +1,6 @@ -# Paraformer录音文件识别Python SDK +# Paraformer非实时语音识别Python SDK -本文介绍Paraformer录音文件识别Python SDK的参数和接口细节。 +本文介绍Paraformer非实时语音识别Python SDK的参数和接口细节。 **重要** @@ -27,7 +27,7 @@ ## **快速开始** -[核心类（Transcription）](#adcb5e9bddbyq)提供了异步提交任务、同步等待任务结束和异步查询任务执行结果的接口。可通过如下两种调用方式进行录音文件识别： +[核心类（Transcription）](#adcb5e9bddbyq)提供了异步提交任务、同步等待任务结束和异步查询任务执行结果的接口。可通过如下两种调用方式进行非实时语音识别： - 异步提交任务+同步等待任务结束：提交任务后，阻塞当前线程直到任务结束并获取识别结果。 @@ -36,7 +36,7 @@ ### **异步提交任务+同步等待任务结束** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/2057892871/CAEQURiBgMCvo5zjpxkiIDQyNzUwZjVjMWM3MjQ5Nzg4ODBjNDRjNzE1ZGFiOGFj4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1609074871/CAEQURiBgMCvo5zjpxkiIDQyNzUwZjVjMWM3MjQ5Nzg4ODBjNDRjNzE1ZGFiOGFj4709861_20241015153444.149.svg) 1. 调用[核心类（Transcription）](#adcb5e9bddbyq)的`async_call`方法并设置[请求参数](#340f6879fci7d)。 @@ -69,7 +69,7 @@ dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.co task_response = Transcription.async_call( model='paraformer-v2', - file_urls=['https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav'], + file_urls=['{YOUR_AUDIO_URL}'], language_hints=['zh', 'en'] # “language_hints”只支持paraformer-v2模型 ) @@ -81,7 +81,7 @@ if transcribe_response.status_code == HTTPStatus.OK: ### **异步提交任务+异步查询任务执行结果** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/2057892871/CAEQURiBgMCN3qzkpxkiIGE0YmU4YTdjMWNiNzRmYjJhMjFlMWZkZmFmOWQ1NmEx4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/2609074871/CAEQURiBgMCN3qzkpxkiIGE0YmU4YTdjMWNiNzRmYjJhMjFlMWZkZmFmOWQ1NmEx4709861_20241015153444.149.svg) 1. 调用[核心类（Transcription）](#adcb5e9bddbyq)的`async_call`方法并设置[请求参数](#340f6879fci7d)。 @@ -114,7 +114,7 @@ dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.co transcribe_response = Transcription.async_call( model='paraformer-v2', - file_urls=['https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav'], + file_urls=['{YOUR_AUDIO_URL}'], language_hints=['zh', 'en'] # “language_hints”只支持paraformer-v2模型 ) @@ -451,7 +451,7 @@ int "end_time":"2025-02-13 17:31:21.867", "results":[ { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "transcription_url":"https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/prod/paraformer-v2/20250213/17%3A31/20ee4e4f-0404-4806-b617-c7d4c62eed19-1.json?Expires=1739525481&OSSAccessKeyId=yourOSSAccessKeyId&Signature=3q%2B1uQmRwltd7FPn5HQM2mBKw74%3D", "subtask_status":"SUCCEEDED" } @@ -484,7 +484,7 @@ int "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" @@ -607,7 +607,7 @@ transcription\_url "end_time":"2025-02-13 17:59:28.828", "results":[ { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "transcription_url":"https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/prod/paraformer-v2/20250213/17%3A59/70e737cc-bf8c-418b-b0c8-83fab192a0fa-1.json?Expires=1739527168&OSSAccessKeyId=yourOSSAccessKeyId&Signature=AtGjIKI%2BdgbzjJIu%2BHsr1R5nSAY%3D", "subtask_status":"SUCCEEDED" } @@ -633,7 +633,7 @@ transcription\_url "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" @@ -701,7 +701,7 @@ transcription\_url ``` { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "properties":{ "audio_format":"pcm_s16le", "channels":[ @@ -903,7 +903,7 @@ def fetch(cls, ## **其他接口：批量查询任务状态/取消任务** -详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的录音文件识别任务，同时支持取消`PENDING`（排队）状态的任务。 +详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的非实时语音识别任务，同时支持取消`PENDING`（排队）状态的任务。 ## **错误码** @@ -924,7 +924,7 @@ def fetch(cls, "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-restful-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-restful-api.md index 971b8184..2fb481c4 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-restful-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/paraformer-recorded-speech-recognition-api-reference/paraformer-recorded-speech-recognition-restful-api.md @@ -1,6 +1,6 @@ -# Paraformer录音文件识别RESTful API +# Paraformer非实时语音识别HTTP API -本文介绍Paraformer录音文件识别RESTful API的参数和接口细节。 +本文介绍Paraformer非实时语音识别HTTP API的参数和接口细节。 **重要** @@ -59,7 +59,7 @@ X-DashScope-Async: enable // 请勿遗漏该请求头，否则无法提交任务 "model":"paraformer-v2", //模型名，必选 "input":{ "file_urls":[ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav" + "{YOUR_AUDIO_URL}" ] //待识别文件，必选 } "parameters":{ @@ -90,7 +90,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ --header "Content-Type: application/json" \ --header "X-DashScope-Async: enable" \ - --data '{"model":"paraformer-v2","input":{"file_urls":["https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav"]},"parameters":{"channel_id":[0]}}' + --data '{"model":"paraformer-v2","input":{"file_urls":["{YOUR_AUDIO_URL}"]},"parameters":{"channel_id":[0]}}' ``` **参数** @@ -450,7 +450,7 @@ string "end_time": "2024-09-12 15:11:40.903", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url": "{YOUR_AUDIO_URL}", "transcription_url": "https://dashscope-result-bj.oss-cn-beijing.aliyuncs.com/pre/filetrans-16k/20240912/15%3A11/409a4b92-445b-4dd8-8c1d-f110954d82d8-1.json?Expires=1726211500&OSSAccessKeyId=yourOSSAccessKeyId&Signature=v5Owy5qoAfT7mzGmQgH0g8C****%3D", "subtask_status": "SUCCEEDED" } @@ -480,7 +480,7 @@ string "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" @@ -546,7 +546,7 @@ JSON数据中各字段含义请参见[识别结果说明](#a9021178ccl7s)。 ``` { - "file_url":"https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "file_url":"{YOUR_AUDIO_URL}", "properties":{ "audio_format":"pcm_s16le", "channels":[ @@ -689,7 +689,7 @@ string ## **其他接口：批量查询任务状态/取消任务** -详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的录音文件识别任务，同时支持取消`PENDING`（排队）状态的任务。 +详情请参见[管理异步任务](https://help.aliyun.com/zh/model-studio/manage-asynchronous-tasks)：支持批量查询24小时内提交的非实时语音识别任务，同时支持取消`PENDING`（排队）状态的任务。 ## **完整示例** @@ -704,7 +704,7 @@ import time api_key = "your-dashscope-api-key" # 在此处替换为您的API Key file_urls = [ - "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/paraformer/hello_world_female2.wav", + "{YOUR_AUDIO_URL}", ] language_hints = ["zh", "en"] @@ -797,7 +797,7 @@ print("transcription result: ", result) "end_time": "2024-12-16 16:31:02.375", "results": [ { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/samples/audio/sensevoice/rich_text_exaple_1.wav", + "file_url": "{YOUR_AUDIO_URL}", "code": "InvalidFile.DownloadFailed", "message": "The audio file cannot be downloaded.", "subtask_status": "FAILED" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-api-reference.md index 4a80c542..e847488f 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-api-reference.md @@ -1,4 +1,4 @@ -# 录音文件识别（Qwen-ASR）API参考 +# 非实时语音识别（Qwen-ASR）API参考 本文介绍 Qwen-ASR 模型的输入与输出参数。可通过OpenAI 兼容或DashScope协议调用 API。 @@ -70,7 +70,7 @@ try: # 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key # 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：api_key = "sk-xxx", api_key=os.getenv("DASHSCOPE_API_KEY"), - # 以下为华北2（北京）地域的URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 + # 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) @@ -84,7 +84,7 @@ try: { "type": "input_audio", "input_audio": { - "data": "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3" + "data": "{YOUR_AUDIO_URL}" } } ], @@ -130,7 +130,7 @@ const client = new OpenAI({ // 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key // 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：apiKey: "sk-xxx", apiKey: process.env.DASHSCOPE_API_KEY, - // 以下为华北2（北京）地域的URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 + // 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", }); @@ -146,7 +146,7 @@ async function main() { { type: "input_audio", input_audio: { - data: "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3" + data: "{YOUR_AUDIO_URL}" } } ] @@ -157,11 +157,9 @@ async function main() { // stream_options: { // "include_usage": true // }, - extra_body: { - asr_options: { - // language: "zh", - enable_itn: false - } + asr_options: { + // language: "zh", + enable_itn: false } }); @@ -191,7 +189,7 @@ main(); #### cURL -以下为华北2（北京）地域的URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 +以下为华北2（北京）地域的配置，调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)，各地域的配置不同。 ``` curl -X POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions' \ @@ -205,7 +203,7 @@ curl -X POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode { "type": "input_audio", "input_audio": { - "data": "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3" + "data": "{YOUR_AUDIO_URL}" } } ], @@ -245,7 +243,7 @@ curl -X POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode import base64, pathlib # input.mp3为用于声音复刻的本地音频文件，请替换为自己的音频文件路径，确保其符合音频要求 - file_path = pathlib.Path("input.mp3") + file_path = pathlib.Path("{YOUR_AUDIO_FILE}") base64_str = base64.b64encode(file_path.read_bytes()).decode() data_uri = f"data:audio/mpeg;base64,{base64_str}" ``` @@ -268,7 +266,7 @@ curl -X POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode // 使用示例 public static void main(String[] args) throws Exception { - System.out.println(toDataUrl("input.mp3")); + System.out.println(toDataUrl("{YOUR_AUDIO_FILE}")); } } ``` @@ -276,8 +274,6 @@ curl -X POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode #### Python SDK -示例中用到的音频文件为：[welcome.mp3](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260105/wotsae/welcome.mp3)。 - ``` import base64 from openai import OpenAI @@ -286,7 +282,7 @@ import pathlib try: # 请替换为实际的音频文件路径 - file_path = "welcome.mp3" + file_path = "{YOUR_AUDIO_FILE}" # 请替换为实际的音频文件MIME类型 audio_mime_type = "audio/mpeg" @@ -301,7 +297,7 @@ try: # 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key # 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：api_key = "sk-xxx", api_key=os.getenv("DASHSCOPE_API_KEY"), - # 以下为华北2（北京）地域的URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 + # 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) @@ -349,8 +345,6 @@ except Exception as e: #### Node.js SDK -示例中用到的音频文件为：[welcome.mp3](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260105/wotsae/welcome.mp3)。 - ``` // 运行前的准备工作: // Windows/Mac/Linux 通用: @@ -364,8 +358,8 @@ const client = new OpenAI({ // 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key // 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：apiKey: "sk-xxx", apiKey: process.env.DASHSCOPE_API_KEY, - // 以下为华北2（北京）地域的URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 - baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + // 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 + baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", }); const encodeAudioFile = (audioFilePath) => { @@ -374,7 +368,7 @@ const encodeAudioFile = (audioFilePath) => { }; // 请替换为实际的音频文件路径 -const dataUri = `data:audio/mpeg;base64,${encodeAudioFile("welcome.mp3")}`; +const dataUri = `data:audio/mpeg;base64,${encodeAudioFile("{YOUR_AUDIO_FILE}")}`; async function main() { try { @@ -399,11 +393,9 @@ async function main() { // stream_options: { // "include_usage": true // }, - extra_body: { - asr_options: { - // language: "zh", - enable_itn: false - } + asr_options: { + // language: "zh", + enable_itn: false } }); @@ -443,7 +435,7 @@ main(); System Message `_object_`（可选） -模型的目标或角色。如果设置系统消息，请放在messages列表的第一位。 +用于为语音识别提供上下文（Context），如背景文本和实体词表等参考信息，不支持设置模型角色等传统系统提示词。如果设置系统消息，请放在messages列表的第一位。 **属性** @@ -912,15 +904,7 @@ curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services { "content": [ { - "text": "" - } - ], - "role": "system" - }, - { - "content": [ - { - "audio": "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3" + "audio": "{YOUR_AUDIO_URL}" } ], "role": "user" @@ -961,7 +945,7 @@ public class Main { MultiModalMessage userMessage = MultiModalMessage.builder() .role(Role.USER.getValue()) .content(Arrays.asList( - Collections.singletonMap("audio", "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3"))) + Collections.singletonMap("audio", "{YOUR_AUDIO_URL}"))) .build(); Map asrOptions = new HashMap<>(); @@ -981,7 +965,7 @@ public class Main { } public static void main(String[] args) { try { - // 以下为华北2（北京）地域的URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 + // 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; simpleMultiModalConversationCall(); } catch (ApiException | NoApiKeyException | UploadFileException e) { @@ -998,11 +982,11 @@ public class Main { import os import dashscope -# 以下为华北2（北京）地域的URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 +# 以下为华北2（北京）地域的配置，调用时请将"{WorkspaceId}"替换为真实的业务空间ID，各地域的配置不同。 dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' messages = [ - {"role": "user", "content": [{"audio": "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3"}]} + {"role": "user", "content": [{"audio": "{YOUR_AUDIO_URL}"}]} ] response = dashscope.MultiModalConversation.call( @@ -1035,7 +1019,7 @@ print(response) System Message `_object_`（可选） -模型的目标或角色。如果设置系统消息，请放在messages列表的第一位。 +用于为语音识别提供上下文（Context），如背景文本和实体词表等参考信息，不支持设置模型角色等传统系统提示词。如果设置系统消息，请放在messages列表的第一位。 仅千问3-ASR-Flash支持该参数。 @@ -1460,7 +1444,7 @@ curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.c --data '{ "model": "qwen3-asr-flash-filetrans", "input": { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3" + "file_url": "{YOUR_AUDIO_URL}" }, "parameters": { "channel_id":[ @@ -1498,7 +1482,7 @@ public class Main { { "model": "qwen3-asr-flash-filetrans", "input": { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3" + "file_url": "{YOUR_AUDIO_URL}" }, "parameters": { "channel_id": [0], @@ -1514,7 +1498,7 @@ public class Main { { "model": "qwen3-asr-flash-filetrans", "input": { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3" + "file_url": "{YOUR_AUDIO_URL}" }, "parameters": { "channel_id": [0], @@ -1595,7 +1579,7 @@ headers = { payload = { "model": "qwen3-asr-flash-filetrans", "input": { - "file_url": "https://dashscope.oss-cn-beijing.aliyuncs.com/audios/welcome.mp3" + "file_url": "{YOUR_AUDIO_URL}" }, "parameters": { "channel_id": [0], @@ -2010,6 +1994,14 @@ print(response.json()) + + + + + + + + 详情参见[异步调用识别结果说明](#2c27ad3e80p4y)。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-interaction-process.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-interaction-process.md index 9839c136..d5bc3a81 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-interaction-process.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-recognition-api-reference/qwen-asr-realtime-api/qwen-asr-realtime-interaction-process.md @@ -100,7 +100,7 @@ Authorization 鉴权在 WebSocket 握手阶段验证。如果 API Key 无效或 **启用方式：**配置客户端`[session.update](https://help.aliyun.com/zh/model-studio/qwen-asr-realtime-client-events#af43722339yva)`事件的`session.turn_detection`参数。 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1592892871/CAEQaxiBgICW9b6H3RkiIGM5MDgwMTNkMjBjMDRlNTNiOGZlODNjZGJhNDQ3NGJm5812623_20251022102739.334.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7151064871/CAEQaxiBgICW9b6H3RkiIGM5MDgwMTNkMjBjMDRlNTNiOGZlODNjZGJhNDQ3NGJm5812623_20251022102739.334.svg) - 客户端通过发送`[input_audio_buffer.append](https://help.aliyun.com/zh/model-studio/qwen-asr-realtime-client-events#a42f8e9111n72)`事件将音频追加到缓冲区。 @@ -112,6 +112,10 @@ Authorization 鉴权在 WebSocket 握手阶段验证。如果 API Key 无效或 - 客户端在音频提交完后，发送`[session.finish](https://help.aliyun.com/zh/model-studio/qwen-asr-realtime-client-events#147ce70052d4z)`事件通知服务端结束当前会话。 + **警告** + + 在 VAD 模式下，推完音频后必须先发送`[session.finish](https://help.aliyun.com/zh/model-studio/qwen-asr-realtime-client-events#147ce70052d4z)`事件再关闭连接。如果客户端直接关闭 WebSocket 连接而未发送该事件，服务端将丢弃当前 in\_progress item，`conversation.item.input_audio_transcription.completed` 等事件将不会到达。建议在调用 `ws.Close()` 之前，先发送 `{"type":"session.finish"}`，并等待收到`[session.finished](https://help.aliyun.com/zh/model-studio/qwen-asr-realtime-server-events#6eaa77339djdv)`事件后再关闭连接。 + - 服务端在检测到语音结束时返回`[input_audio_buffer.speech_stopped](https://help.aliyun.com/zh/model-studio/qwen-asr-realtime-server-events#3d73b074cak7k)`事件。 - 服务端返回`[input_audio_buffer.committed](https://help.aliyun.com/zh/model-studio/qwen-asr-realtime-server-events#1108a3764an0e)`事件。 @@ -131,7 +135,7 @@ Authorization 鉴权在 WebSocket 握手阶段验证。如果 API Key 无效或 **启用方式：**将客户端`[session.update](https://help.aliyun.com/zh/model-studio/qwen-asr-realtime-client-events#af43722339yva)`事件的`session.turn_detection`设为null。 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1592892871/CAEQaxiBgMDUp8qH3RkiIGEyYTc0NTI1ZmQ1OTQ5NjliNWE0OTYwYTAwMDBlMjBm5812623_20251022102739.334.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/8151064871/CAEQaxiBgMDUp8qH3RkiIGEyYTc0NTI1ZmQ1OTQ5NjliNWE0OTYwYTAwMDBlMjBm5812623_20251022102739.334.svg) - 客户端通过发送`[input_audio_buffer.append](https://help.aliyun.com/zh/model-studio/qwen-asr-realtime-client-events#a42f8e9111n72)`事件将音频追加到缓冲区。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-android-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-android-sdk.md index 3d4ec241..2c798bb3 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-android-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-android-sdk.md @@ -277,7 +277,7 @@ API Key。建议使用时效性短、安全性更高的[临时API Key](https://h 语音合成所使用的音色。 -- **系统音色**：参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) +- **系统音色**：参见[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) - **复刻音色**：通过声音复刻功能定制 @@ -394,7 +394,7 @@ SSML 的使用限制（支持的模型、音色和接口），请参见[使用 默认值：false。 -仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 +仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 > 时间戳结果在[INativeStreamInputTtsCallback](#secstreamcallback)的all\_response中。 @@ -440,7 +440,7 @@ cosyvoice-v1不支持该参数。 - zh：中文 -- en：英文 +- en：英语 - fr：法语 @@ -460,9 +460,15 @@ cosyvoice-v1不支持该参数。 - vi：越南语 +- es：西班牙语 + - it：意大利语 -- ms：马来语 +- ms：马来西亚语 + +- fil：菲律宾语 + +- ar：阿拉伯语 `instruction` @@ -834,6 +840,7 @@ STREAM\_INPUT\_TTS\_EVENT\_TASK\_FAILED - 解压 ZIP 包。在 `app/libs` 目录中获取 AAR 格式 SDK，并添加到项目依赖。 需要 Android CPP 接入时，使用 ZIP 包内的 `android_libs` 与 `android_include` 获取动态库和头文件。 + - 用 Android Studio 打开工程。示例代码位于`DashCosyVoiceStreamTtsActivity.java`，替换 API Key 后体验功能。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-client-events.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-client-events.md index dabe4c85..69c32225 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-client-events.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-client-events.md @@ -91,7 +91,7 @@ 语音合成所使用的音色。 -- **系统音色**：参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) +- **系统音色**：参见[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) - **复刻音色**：通过声音复刻功能定制 @@ -173,7 +173,7 @@ SSML 的使用限制（支持的模型、音色和接口），请参见[使用 默认值：false。 -仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 +仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 **seed** `_integer_` （可选） @@ -209,7 +209,7 @@ cosyvoice-v1不支持该参数。 - zh：中文 -- en：英文 +- en：英语 - fr：法语 @@ -229,9 +229,15 @@ cosyvoice-v1不支持该参数。 - vi：越南语 +- es：西班牙语 + - it：意大利语 -- ms：马来语 +- ms：马来西亚语 + +- fil：菲律宾语 + +- ar：阿拉伯语 **instruction** `_string_` （可选） @@ -365,7 +371,7 @@ qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v2、cosyvoice-v1 ## **finish-task** -**说明**：通知服务端文本发送完毕，请求结束任务。 +**说明**：通知服务端文本发送完毕，请求结束任务。如需取消当前轮次的语音合成任务，可在 `input` 中设置 `directive` 为 `cancel`。 **发送时机**：所有文本发送完毕后立即发送。 @@ -400,10 +406,41 @@ qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v2、cosyvoice-v1 } ``` +**取消任务示例**： + +``` +{ + "header": { + "action": "finish-task", + "task_id": "2bf83b9a-baeb-4fda-8d9a-xxxxxxxxxxxx", + "streaming": "duplex" + }, + "payload": { + "input": { + "directive": "cancel" + } + } +} +``` + **payload** `_object_` **（必选）** **属性** **input** `_object_` **（必选）** -固定为 `{}`。 +任务输入。为空对象 `{}` 时表示正常结束任务；包含 `directive` 时可用于取消当前轮次的语音合成任务。 + +**directive** `_string_` （可选） + +控制任务结束行为。当前仅支持取值为 `cancel`，表示取消当前轮次的语音合成任务，服务端会立即返回 `task-finished` 事件，且不会输出后续音频。 + +取消后，可在当前 WebSocket 连接上重新发起语音合成任务（发送新的 `run-task` 事件），无需重新建立连接。 + +**重要** + +**模型限制**： + +- 华北2（北京）地域：Qwen-Audio-TTS 系列模型的所有模型都支持该功能；CosyVoice 系列模型仅 v2 及以上版本支持该功能。 + +- 新加坡地域：Qwen-Audio-TTS 系列模型的所有模型都支持该功能；CosyVoice 系列模型不支持该功能。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-ios-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-ios-sdk.md index 5a534022..6631da6e 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-ios-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-ios-sdk.md @@ -243,7 +243,7 @@ Qwen-Audio-TTS/CosyVoice 支持一次性输入和流式输入两种调用方式 语音合成所使用的音色。 - - **系统音色**：参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) + - **系统音色**：参见[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) - **复刻音色**：通过声音复刻功能定制 @@ -360,7 +360,7 @@ Qwen-Audio-TTS/CosyVoice 支持一次性输入和流式输入两种调用方式 默认值：false。 - 仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 + 仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 > 时间戳结果在[onStreamInputTtsEventCallback](#bea29bbafcosq)的all\_response中。 @@ -406,7 +406,7 @@ Qwen-Audio-TTS/CosyVoice 支持一次性输入和流式输入两种调用方式 - zh：中文 - - en：英文 + - en：英语 - fr：法语 @@ -426,9 +426,15 @@ Qwen-Audio-TTS/CosyVoice 支持一次性输入和流式输入两种调用方式 - vi：越南语 + - es：西班牙语 + - it：意大利语 - - ms：马来语 + - ms：马来西亚语 + + - fil：菲律宾语 + + - ar：阿拉伯语 `instruction` diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-java-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-java-sdk.md index f3e4feda..d8dec276 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-java-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-java-sdk.md @@ -121,6 +121,29 @@ public void streamingComplete() 结束双向流式调用，通知服务端所有文本已发送完毕。 +### **streamingCancel() - 取消双向流式调用** + +**方法签名**： + +``` +public void streamingCancel() +``` + +**说明**：取消当前轮次的双向流式语音合成任务。调用后，SDK 会立即结束当前任务。取消后可在当前连接上继续发起新的合成任务，无需重新初始化 `SpeechSynthesizer` 实例。 + +**重要** + +**版本要求**：使用该功能需要 Java SDK 版本不低于 2.22.26。 + +**重要** + +**模型限制**： + +- 华北2（北京）地域：Qwen-Audio-TTS 系列模型的所有模型都支持该功能；CosyVoice 系列模型仅 v2 及以上版本支持该功能。 + +- 新加坡地域：Qwen-Audio-TTS 系列模型的所有模型都支持该功能；CosyVoice 系列模型不支持该功能。 + + ### **callAsFlowable() - 单向流式合成（响应式）** **方法签名**： @@ -242,7 +265,7 @@ public long getFirstPackageDelay() ``` SpeechSynthesisParam param = SpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") // 模型 - .voice("longanlingxi") // 音色 + .voice("longanhuan_v3.6") // 音色 .format(SpeechSynthesisAudioFormat.WAV_8000HZ_MONO_16BIT) // 音频编码格式、采样率 .volume(50) // 音量，取值范围：[0, 100] .speechRate(1.0f) // 语速，取值范围：[0.5, 2] @@ -278,7 +301,7 @@ String 语音合成所使用的音色。 -- **系统音色**：参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) +- **系统音色**：参见[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) - **复刻音色**：通过声音复刻功能定制 @@ -345,7 +368,7 @@ boolean 默认值：false。 -仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 +仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 `seed(int)` @@ -391,7 +414,7 @@ List - zh：中文 -- en：英文 +- en：英语 - fr：法语 @@ -411,9 +434,15 @@ List - vi：越南语 +- es：西班牙语 + - it：意大利语 -- ms：马来语 +- ms：马来西亚语 + +- fil：菲律宾语 + +- ar：阿拉伯语 `instruction(String)` @@ -458,7 +487,7 @@ paramHotFix.setReplace(replaceItems); SpeechSynthesisParam param = SpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") // 模型 - .voice("your_voice") // 替换成qwen-audio-3.0-tts-flash复刻音色 + .voice("longanhuan_v3.6") // 音色 .hotFix(paramHotFix) .build(); ``` @@ -488,7 +517,7 @@ Map ``` SpeechSynthesisParam param = SpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("enable_markdown_filter", true) .build(); ``` @@ -842,7 +871,7 @@ SDK提供了语音合成的关键接口，支持以下几种调用方式： ### **非流式调用** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1094204871/CAEQURiBgIDHpsn4phkiIDQ0ZGE2OTk3NmY5NTRhNDVhZDQwNWE3ZGZiMzk4Yjk54709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6538354871/CAEQURiBgIDHpsn4phkiIDQ0ZGE2OTk3NmY5NTRhNDVhZDQwNWE3ZGZiMzk4Yjk54709861_20241015153444.149.svg) 发送的文本长度不得超过20000字符。 @@ -864,7 +893,7 @@ public class Main { // 模型 private static String model = "qwen-audio-3.0-tts-flash"; // 音色 - private static String voice = "longanlingxi"; + private static String voice = "longanhuan_v3.6"; public static void streamAudioDataToSpeaker() { // 请求参数 @@ -917,7 +946,7 @@ public class Main { ### **单向流式调用** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1094204871/CAEQVRiBgMCfo..hrBkiIGEyMjNkZjVlMWZiYzRhZDU4ZjEyZjdjMmMzYjM1YzMz4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7538354871/CAEQVRiBgMCfo..hrBkiIGEyMjNkZjVlMWZiYzRhZDU4ZjEyZjdjMmMzYjM1YzMz4709861_20241015153444.149.svg) 发送的文本长度不得超过20000字符。 @@ -949,7 +978,7 @@ public class Main { // 模型 private static String model = "qwen-audio-3.0-tts-flash"; // 音色 - private static String voice = "longanlingxi"; + private static String voice = "longanhuan_v3.6"; public static void streamAudioDataToSpeaker() { CountDownLatch latch = new CountDownLatch(1); @@ -1024,7 +1053,7 @@ public class Main { ### 双向流式调用 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1094204871/CAEQVRiBgICHxPGhrBkiIGE3ZTVmMzY0YzI3NzQxYTFiYWE2MmU2NTBhMDgzZGM14709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7538354871/CAEQVRiBgICHxPGhrBkiIGE3ZTVmMzY0YzI3NzQxYTFiYWE2MmU2NTBhMDgzZGM14709861_20241015153444.149.svg) 单次发送文本长度不得超过 20000 字符，且累计发送文本总长度不得超过 20 万字符。 @@ -1076,7 +1105,7 @@ public class Main { "减少了用户等待时间。", "适用于调用大规模", "语言模型（LLM），以", "流式输入文本的方式", "进行语音合成的场景。"}; private static String model = "qwen-audio-3.0-tts-flash"; // 模型 - private static String voice = "longanlingxi"; // 音色 + private static String voice = "longanhuan_v3.6"; // 音色 public static void streamAudioDataToSpeaker() { // 配置回调函数 @@ -1177,7 +1206,7 @@ class TimeUtils { public class Main { private static String model = "qwen-audio-3.0-tts-flash"; // 模型 - private static String voice = "longanlingxi"; // 音色 + private static String voice = "longanhuan_v3.6"; // 音色 public static void streamAudioDataToSpeaker() throws NoApiKeyException { // 请求参数 @@ -1252,7 +1281,7 @@ public class Main { "减少了用户等待时间。", "适用于调用大规模", "语言模型（LLM），以", "流式输入文本的方式", "进行语音合成的场景。"}; private static String model = "qwen-audio-3.0-tts-flash"; - private static String voice = "longanlingxi"; + private static String voice = "longanhuan_v3.6"; public static void streamAudioDataToSpeaker() throws NoApiKeyException { // 模拟流式输入 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-python-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-python-sdk.md index 8e94f9c5..4d00d326 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-python-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-python-sdk.md @@ -125,6 +125,47 @@ def streaming_complete(self) -> None **说明**：通知服务端所有文本已发送完毕，阻塞当前线程直到剩余文本合成完成并返回所有音频数据。未调用此方法可能导致尾部文本无法转换为语音。 +### **streaming\_cancel() - 取消流式合成** + +**方法签名**： + +``` +def streaming_cancel(self, complete_timeout_millis: int = 10000) -> None +``` + +**参数说明**： + +**参数** + +**类型** + +**必填** + +**说明** + +complete\_timeout\_millis + +int + +否 + +等待服务端返回 task-finished 事件的超时时间，单位毫秒。默认值：10000。 + +**说明**：取消当前轮次的流式语音合成任务。调用后，SDK 会立即结束当前任务。取消后可在当前连接上继续发起新的合成任务，无需重新初始化 `SpeechSynthesizer` 实例。 + +**重要** + +**版本要求**：使用该功能需要 Python SDK 版本不低于 1.26.4。 + +**重要** + +**模型限制**： + +- 华北2（北京）地域：Qwen-Audio-TTS 系列模型的所有模型都支持该功能；CosyVoice 系列模型仅 v2 及以上版本支持该功能。 + +- 新加坡地域：Qwen-Audio-TTS 系列模型的所有模型都支持该功能；CosyVoice 系列模型不支持该功能。 + + ### **get\_last\_request\_id() - 获取请求ID** **方法签名**： @@ -185,7 +226,7 @@ str 语音合成所使用的音色。 -- **系统音色**：参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) +- **系统音色**：参见[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) - **复刻音色**：通过声音复刻功能定制 @@ -263,7 +304,7 @@ int ``` synthesizer = SpeechSynthesizer( model="qwen-audio-3.0-tts-flash", - voice="longanlingxi", + voice="longanhuan_v3.6", additional_params={"bit_rate": 128000} ) ``` @@ -278,7 +319,7 @@ bool 默认值：false。 -仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 +仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 **说明** @@ -287,7 +328,7 @@ bool ``` synthesizer = SpeechSynthesizer( model="qwen-audio-3.0-tts-flash", - voice="your_voice", + voice="longanhuan_v3.6", additional_params={"word_timestamp_enabled": True} ) ``` @@ -334,7 +375,7 @@ list\[str\] - zh：中文 -- en：英文 +- en：英语 - fr：法语 @@ -354,9 +395,15 @@ list\[str\] - vi：越南语 +- es：西班牙语 + - it：意大利语 -- ms：马来语 +- ms：马来西亚语 + +- fil：菲律宾语 + +- ar：阿拉伯语 instruction @@ -388,7 +435,7 @@ bool ``` synthesizer = SpeechSynthesizer( model="qwen-audio-3.0-tts-flash", - voice="longanlingxi", + voice="longanhuan_v3.6", additional_params={ "enable_aigc_tag": True, "aigc_propagator": "your_propagator", @@ -447,7 +494,7 @@ qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v2、cosyvoice-v1 ``` synthesizer = SpeechSynthesizer( model="qwen-audio-3.0-tts-flash", - voice="your_voice", # 替换成qwen-audio-3.0-tts-flash复刻音色 + voice="longanhuan_v3.6", # 音色 hot_fix={ "pronunciation": [{"天气": "tian1 qi4"}], "replace": [{"今天": "金天"}] @@ -483,7 +530,7 @@ bool ``` synthesizer = SpeechSynthesizer( model="qwen-audio-3.0-tts-flash", - voice="your_voice", # 替换成qwen-audio-3.0-tts-flash复刻音色 + voice="longanhuan_v3.6", # 音色 additional_params={"enable_markdown_filter": True} ) ``` @@ -693,7 +740,7 @@ SDK提供了语音合成的关键接口，支持以下几种调用方式： ### **非流式调用** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1194204871/CAEQURiBgMDRr9T4phkiIGNmYzBiZjFkZjQ4MDQzZGU4NDIyZDU2NWJjYjkyZTQ04709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0658354871/CAEQURiBgMDRr9T4phkiIGNmYzBiZjFkZjQ4MDQzZGU4NDIyZDU2NWJjYjkyZTQ04709861_20241015153444.149.svg) 单次调用发送的文本长度不得超过20000字符，超出限制将返回错误。 @@ -718,7 +765,7 @@ dashscope.base_websocket_api_url='wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.c # 模型 model = "qwen-audio-3.0-tts-flash" # 音色 -voice = "longanlingxi" +voice = "longanhuan_v3.6" # 实例化SpeechSynthesizer，并在构造方法中传入模型（model）、音色（voice）等请求参数 synthesizer = SpeechSynthesizer(model=model, voice=voice) @@ -736,7 +783,7 @@ with open('output.mp3', 'wb') as f: ### **单向流式调用** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1194204871/CAEQVRiBgIDv9fShrBkiIDhmNTk5YmQ1ZDgwNzRjZjRiN2VlMTU5YzI1ZGMwMTlm4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0658354871/CAEQVRiBgIDv9fShrBkiIDhmNTk5YmQ1ZDgwNzRjZjRiN2VlMTU5YzI1ZGMwMTlm4709861_20241015153444.149.svg) 单次调用发送的文本长度不得超过20000字符，超出限制将返回错误。 @@ -769,7 +816,7 @@ dashscope.base_websocket_api_url='wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.c # 模型 model = "qwen-audio-3.0-tts-flash" # 音色 -voice = "longanlingxi" +voice = "longanhuan_v3.6" # 定义回调接口 class Callback(ResultCallback): @@ -823,7 +870,7 @@ synthesizer.call("今天天气怎么样？") ### **双向流式调用** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1194204871/CAEQVRiBgMDb7PahrBkiIDVkNjEwOTMxYjEwOTRmOWFhMmI1OTRiY2Q3ZDgzZmE54709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0658354871/CAEQVRiBgMDb7PahrBkiIDVkNjEwOTMxYjEwOTRmOWFhMmI1OTRiY2Q3ZDgzZmE54709861_20241015153444.149.svg) 单次发送文本长度不得超过 20000 字符，且累计发送文本总长度不得超过 20 万字符。 @@ -888,7 +935,7 @@ dashscope.base_websocket_api_url='wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.c # 模型 model = "qwen-audio-3.0-tts-flash" # 音色 -voice = "longanlingxi" +voice = "longanhuan_v3.6" # 定义回调接口 class Callback(ResultCallback): diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-websocket-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-websocket-api.md index 90389819..56124c1b 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-websocket-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/cosyvoice-large-model-for-speech-synthesis/cosyvoice-websocket-api.md @@ -87,7 +87,7 @@ Authorization 鉴权在 WebSocket 握手阶段验证。如果 API Key 无效或 ## 交互流程 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/8684204871/CAEQaxiBgID50pCW3hkiIDVlOWNkODdhOGYyYjQ2ZDFiMzgyYjNmMmUzOGZkNGVh4709861_20241015153444.149.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9336814871/CAEQaxiBgID50pCW3hkiIDVlOWNkODdhOGYyYjQ2ZDFiMzgyYjNmMmUzOGZkNGVh4709861_20241015153444.149.svg) 客户端事件和服务端事件的详细说明，请参见[客户端事件](https://help.aliyun.com/zh/model-studio/cosyvoice-client-events)和[服务端事件](https://help.aliyun.com/zh/model-studio/cosyvoice-server-events)。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-http-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-http-api.md index e78e4fb1..a25c2588 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-http-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-http-api.md @@ -66,7 +66,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ "model": "qwen-audio-3.0-tts-flash", "input": { "text": "我家的后面有一个很大的花园。", - "voice": "longanlingxi", + "voice": "longanhuan_v3.6", "format": "wav", "sample_rate": 24000 } @@ -84,7 +84,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ "model": "qwen-audio-3.0-tts-flash", "input": { "text": "我家的后面有一个很大的花园。", - "voice": "longanlingxi", + "voice": "longanhuan_v3.6", "format": "wav", "sample_rate": 24000 } @@ -135,7 +135,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ 取值范围： -- 系统音色：参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) +- 系统音色：参见[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) - 声音复刻音色：如何创建音色请参见[CosyVoice声音复刻/设计API](https://help.aliyun.com/zh/model-studio/cosyvoice-clone-design-api) @@ -211,7 +211,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ 默认值：false。 -仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 +仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 **seed** `_integer_` （可选） @@ -245,7 +245,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ - zh：中文 -- en：英文 +- en：英语 - fr：法语 @@ -265,6 +265,16 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ - vi：越南语 +- es：西班牙语 + +- it：意大利语 + +- ms：马来西亚语 + +- fil：菲律宾语 + +- ar：阿拉伯语 + **instruction** `_string_` （可选） diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-java-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-java-sdk.md index 7e77c178..1c261549 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-java-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-java-sdk.md @@ -207,7 +207,7 @@ String 取值范围： -- 系统音色：参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) +- 系统音色：参见[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) - 声音复刻音色：如何创建音色请参见[CosyVoice声音复刻/设计API](https://help.aliyun.com/zh/model-studio/cosyvoice-clone-design-api) @@ -301,7 +301,7 @@ boolean HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("你好") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("enable_ssml", true) .build(); ``` @@ -312,7 +312,7 @@ HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("你好") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(Collections.singletonMap("enable_ssml", true)) .build(); ``` @@ -327,7 +327,7 @@ boolean 默认值：false。 -仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 +仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 **说明** @@ -339,7 +339,7 @@ boolean HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("word_timestamp_enabled", true) .build(); ``` @@ -350,7 +350,7 @@ HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(Collections.singletonMap("word_timestamp_enabled", true)) .build(); ``` @@ -377,7 +377,7 @@ int HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("seed", 1234) .build(); ``` @@ -388,7 +388,7 @@ HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(Collections.singletonMap("seed", 1234)) .build(); ``` @@ -421,7 +421,7 @@ List - zh：中文 -- en：英文 +- en：英语 - fr：法语 @@ -441,6 +441,16 @@ List - vi：越南语 +- es：西班牙语 + +- it：意大利语 + +- ms：马来西亚语 + +- fil：菲律宾语 + +- ar：阿拉伯语 + **说明** @@ -452,7 +462,7 @@ List HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("language_hints", Arrays.asList("zh")) .build(); ``` @@ -463,7 +473,7 @@ HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(Collections.singletonMap("language_hints", Arrays.asList("zh"))) .build(); ``` @@ -488,7 +498,7 @@ String HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("instruction", "请用非常开心的语气说话。") .build(); ``` @@ -499,7 +509,7 @@ HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(Collections.singletonMap("instruction", "请用非常开心的语气说话。")) .build(); ``` @@ -530,7 +540,7 @@ int HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .format("opus") .parameter("bit_rate", 32) .build(); @@ -542,7 +552,7 @@ HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .format("opus") .parameters(Collections.singletonMap("bit_rate", 32)) .build(); @@ -570,7 +580,7 @@ boolean HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("enable_aigc_tag", true) .build(); ``` @@ -581,7 +591,7 @@ HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(Collections.singletonMap("enable_aigc_tag", true)) .build(); ``` @@ -608,7 +618,7 @@ String HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("enable_aigc_tag", true) .parameter("aigc_propagator", "xxxx") .build(); @@ -624,7 +634,7 @@ map.put("aigc_propagator", "xxxx"); HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(map) .build(); ``` @@ -651,7 +661,7 @@ String HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("enable_aigc_tag", true) .parameter("aigc_propagate_id", "xxxx") .build(); @@ -667,7 +677,7 @@ map.put("aigc_propagate_id", "xxxx"); HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(map) .build(); ``` @@ -726,7 +736,7 @@ hotFix.put("replace", replace); HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("今天天气真好。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("hot_fix", hotFix) .build(); ``` @@ -738,7 +748,7 @@ HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("今天天气真好。") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(Collections.singletonMap("hot_fix", hotFix)) .build(); ``` @@ -774,7 +784,7 @@ boolean HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("# 标题\n正文内容") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameter("enable_markdown_filter", true) .build(); ``` @@ -785,7 +795,7 @@ HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() HttpSpeechSynthesisParam param = HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") .text("# 标题\n正文内容") - .voice("longanlingxi") + .voice("longanhuan_v3.6") .parameters(Collections.singletonMap("enable_markdown_filter", true)) .build(); ``` @@ -837,7 +847,7 @@ public class CosyVoiceSyncExample { HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") // 更换模型时，需同步更换为对应版本的音色 .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") // 该音色适用于qwen-audio-3.0-tts系列，cosyvoice-v3请使用longanyang等v3音色，cosyvoice-v2请使用longxiaochun_v2等v2音色 + .voice("longanhuan_v3.6") // 该音色适用于qwen-audio-3.0-tts系列，cosyvoice-v3请使用longanyang等v3音色，cosyvoice-v2请使用longxiaochun_v2等v2音色 .format("wav") .sampleRate(24000) // 未配置环境变量时，将下行替换为：apiKey("sk-xxx")，即替换为实际的API Key @@ -877,7 +887,7 @@ public class CosyVoiceSyncExample { HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") // 更换模型时，需同步更换为对应版本的音色 .text("我家的后面有一个很大的花园。") - .voice("longanlingxi") // 该音色适用于qwen-audio-3.0-tts系列，cosyvoice-v3请使用longanyang等v3音色，cosyvoice-v2请使用longxiaochun_v2等v2音色 + .voice("longanhuan_v3.6") // 该音色适用于qwen-audio-3.0-tts系列，cosyvoice-v3请使用longanyang等v3音色，cosyvoice-v2请使用longxiaochun_v2等v2音色 .format("wav") .sampleRate(24000) // 未配置环境变量时，将下行替换为：apiKey("sk-xxx")，即替换为实际的API Key @@ -938,7 +948,7 @@ public class CosyVoiceStreamExample { HttpSpeechSynthesisParam.builder() .model("qwen-audio-3.0-tts-flash") // 更换模型时，需同步更换为对应版本的音色 .text("今天天气真好，适合出去玩。") - .voice("longanlingxi") // 该音色适用于qwen-audio-3.0-tts系列，cosyvoice-v3请使用longanyang等v3音色，cosyvoice-v2请使用longxiaochun_v2等v2音色 + .voice("longanhuan_v3.6") // 该音色适用于qwen-audio-3.0-tts系列，cosyvoice-v3请使用longanyang等v3音色，cosyvoice-v2请使用longxiaochun_v2等v2音色 .format("wav") .sampleRate(24000) // 未配置环境变量时，将下行替换为：apiKey("sk-xxx")，即替换为实际的API Key diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-python-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-python-sdk.md index 7143911a..a3f50903 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-python-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/non-realtime-cosyvoice-api/cosyvoice-tts-python-sdk.md @@ -109,7 +109,7 @@ str 取值范围： -- 系统音色：参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) +- 系统音色：参见[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list) - 声音复刻音色：如何创建音色请参见[CosyVoice声音复刻/设计API](https://help.aliyun.com/zh/model-studio/cosyvoice-clone-design-api) @@ -222,7 +222,7 @@ bool - False：关闭。 -仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 +仅在流式输出模式下可用。支持的音色范围：cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash、cosyvoice-v3-plus和cosyvoice-v2模型的复刻音色，以及[Qwen-Audio-TTS音色列表](https://help.aliyun.com/zh/model-studio/qwen-audio-tts-voice-list)、[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)中标记为支持的系统音色。qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash及其他模型的复刻音色不支持此功能。 seed @@ -264,7 +264,7 @@ list - zh：中文 -- en：英文 +- en：英语 - fr：法语 @@ -284,6 +284,16 @@ list - vi：越南语 +- es：西班牙语 + +- it：意大利语 + +- ms：马来西亚语 + +- fil：菲律宾语 + +- ar：阿拉伯语 + instruction @@ -443,7 +453,7 @@ dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.co result = HttpSpeechSynthesizer.call( model="qwen-audio-3.0-tts-flash", # 更换模型时，需同步更换为对应版本的音色 text="今天是个好日子，适合构建人们喜爱的产品！", - voice="longanhuan", # 该音色适用于cosyvoice-v3系列，cosyvoice-v2请使用longxiaochun_v2等v2音色 + voice="longanhuan_v3.6", # 该音色适用于cosyvoice-v3系列，cosyvoice-v2请使用longxiaochun_v2等v2音色 format="wav", sample_rate=24000, stream=False, @@ -478,7 +488,7 @@ dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.co stream_result = HttpSpeechSynthesizer.call( model="qwen-audio-3.0-tts-flash", # 更换模型时，需同步更换为对应版本的音色 text="今天是个好日子，适合构建人们喜爱的产品！", - voice="longanhuan", # 该音色适用于cosyvoice-v3系列，cosyvoice-v2请使用longxiaochun_v2等v2音色 + voice="longanhuan_v3.6", # 该音色适用于cosyvoice-v3系列，cosyvoice-v2请使用longxiaochun_v2等v2音色 format="wav", sample_rate=24000, stream=True, diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-api.md index 18e9d458..37905ab0 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/qwen-tts-api.md @@ -63,7 +63,7 @@ public class Main { .apiKey(System.getenv("DASHSCOPE_API_KEY")) .text("Today is a wonderful day to build something people love!") .voice(AudioParameters.Voice.CHERRY) - .languageType("English") + .parameter("language_type", "English") // 如需使用指令控制功能，请取消下方注释，并将model替换为qwen3-tts-instruct-flash // .parameter("instructions","语速较快，带有明显的上扬语调，适合介绍时尚产品。") // .parameter("optimize_instructions",true) @@ -166,7 +166,7 @@ public class Main { .apiKey(System.getenv("DASHSCOPE_API_KEY")) .text("Today is a wonderful day to build something people love!") .voice(AudioParameters.Voice.CHERRY) - .languageType("English") + .parameter("language_type", "English") // 如需使用指令控制功能，请取消下方注释，并将model替换为qwen3-tts-instruct-flash // .parameter("instructions","语速较快，带有明显的上扬语调，适合介绍时尚产品。") // .parameter("optimize_instructions",true) @@ -270,7 +270,7 @@ curl -X POST 'https://dashscope.aliyuncs.com/api/v1/services/aigc/multimodal-gen 支持语言：仅支持中文和英文。 -适用范围：该功能仅适用于千问3-TTS-Instruct-Flash-Realtime系列模型。 +适用范围：该功能仅适用于千问3-TTS-Instruct-Flash系列模型。 **optimize\_instructions** `_boolean_` （可选） @@ -363,140 +363,6 @@ HTTP状态码。遵循 [RFC 9110](https://www.rfc-editor.org/rfc/rfc9110.html#n - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - **request\_id** `_string_` diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-design-http-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-design-http-api.md index 2673fe86..a2402160 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-design-http-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-design-http-api.md @@ -181,7 +181,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ **重要** -仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash、v3-plus和v3-flash模型支持。 +仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash、v3-plus和v3-flash模型支持。 辅助模型识别样本音频的语种，从而更准确地提取音色特征，提升复刻效果。若设置的语种与实际音频语种不符（例如为中文音频设置 `en`），系统将忽略该设置并自动检测语种。 @@ -189,11 +189,11 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ 取值范围（因模型而异）： -- qwen-audio-3.0-tts-flash： +- qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash： - zh：中文 - - en：英文 + - en：英语 - fr：法语 @@ -215,7 +215,13 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ - it：意大利语 - - ms：马来语 + - es：西班牙语 + + - ms：马来西亚语 + + - fil：菲律宾语 + + - ar：阿拉伯语 - cosyvoice-v3-plus： @@ -297,7 +303,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ **重要** -仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 +仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 音频预处理后用于声音复刻的参考音频最大时长（秒）。取值范围：\[3.0, 30.0\]。时间越长效果越好。 @@ -307,7 +313,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/ **重要** -仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 +仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 是否开启音频预处理（降噪、音频增强、音量规整）。有背景噪音时建议开启；安静环境建议关闭以最大程度还原音色。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-java-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-java-sdk.md index be92e944..36aa6d87 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-java-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-java-sdk.md @@ -121,7 +121,7 @@ customParam 否 -自定义参数，可指定languageHints、maxPromptAudioLength等。 +自定义参数，可通过 parameter() 方法指定 language\_hints、max\_prompt\_audio\_length 等参数。 **返回值**：`Voice` 对象，通过 `getVoiceId()` 方法获取音色ID。 @@ -283,108 +283,11 @@ String 声音复刻模型，固定为"voice-enrollment"。 -languageHints(List) - -List - -**重要** - -仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash、v3-plus和v3-flash模型支持。 - -辅助模型识别样本音频的语种，从而更准确地提取音色特征，提升复刻效果。若设置的语种与实际音频语种不符（例如为中文音频设置 `en`），系统将忽略该设置并自动检测语种。 - -此参数为数组，但当前版本仅处理第一个元素。 - -取值范围（因模型而异）： - -- qwen-audio-3.0-tts-flash： - - - zh：中文 - - - en：英文 - - - fr：法语 - - - de：德语 - - - ja：日语 - - - ko：韩语 - - - ru：俄语 - - - pt：葡萄牙语 - - - th：泰语 - - - id：印尼语 - - - vi：越南语 - - - it：意大利语 - - - ms：马来语 - -- cosyvoice-v3-plus： - - - zh：中文 - - - en：英文 - - - fr：法语 - - - de：德语 - - - ja：日语 - - - ko：韩语 - - - ru：俄语 - -- cosyvoice-v3.5-plus、cosyvoice-v3.5-flash、cosyvoice-v3-flash： - - - zh：中文 - - - en：英文 - - - fr：法语 - - - de：德语 - - - ja：日语 - - - ko：韩语 - - - ru：俄语 - - - pt：葡萄牙语 - - - th：泰语 - - - id：印尼语 - - - vi：越南语 - - -默认值：\["zh"\]。 - -maxPromptAudioLength(Float) - -Float - -**重要** - -仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 - -音频预处理后用于声音复刻的参考音频最大时长（秒）。取值范围：\[3.0, 30.0\]。时间越长效果越好。 - -默认值：10.0。 - parameter(String, Object) Object -设置[扩展参数](#a18b66cba924j)，如 parameter("enable\_preprocess", false)。 +设置自定义参数，如 parameter("language\_hints", Arrays.asList("zh"))、parameter("max\_prompt\_audio\_length", 10.0f)、parameter("enable\_preprocess", false)。 ### **扩展参数** @@ -404,7 +307,7 @@ boolean **重要** -仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 +仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 是否开启音频预处理（降噪、音频增强、音量规整）。有背景噪音时建议开启；安静环境建议关闭以最大程度还原音色。 @@ -422,7 +325,7 @@ import com.alibaba.dashscope.utils.Constants; import org.slf4j.Logger; import org.slf4j.LoggerFactory; -import java.util.Collections; +import java.util.Arrays; public class Main { private static final Logger logger = LoggerFactory.getLogger(Main.class); @@ -444,8 +347,8 @@ public class Main { fileUrl, VoiceEnrollmentParam.builder() .model(cloneModelName) - .languageHints(Collections.singletonList("zh")) - // .maxPromptAudioLength(10.0f) + .parameter("language_hints", Arrays.asList("zh")) + // .parameter("max_prompt_audio_length", 10.0f) // .parameter("enable_preprocess", false) .build()); diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-python-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-python-sdk.md index 14b73cd2..42410c27 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-python-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/speech-synthesis-api-reference/sound-reengraving/voice-clone-python-sdk.md @@ -112,7 +112,7 @@ List\[str\] **重要** -仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash、v3-plus和v3-flash模型支持。 +仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash、v3-plus和v3-flash模型支持。 辅助模型识别样本音频的语种，从而更准确地提取音色特征，提升复刻效果。若设置的语种与实际音频语种不符（例如为中文音频设置 `en`），系统将忽略该设置并自动检测语种。 @@ -120,11 +120,11 @@ List\[str\] 取值范围（因模型而异）： -- qwen-audio-3.0-tts-flash： +- qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash： - zh：中文 - - en：英文 + - en：英语 - fr：法语 @@ -146,7 +146,13 @@ List\[str\] - it：意大利语 - - ms：马来语 + - es：西班牙语 + + - ms：马来西亚语 + + - fil：菲律宾语 + + - ar：阿拉伯语 - cosyvoice-v3-plus： @@ -199,7 +205,7 @@ float **重要** -仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 +仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 音频预处理后用于声音复刻的参考音频最大时长（秒）。取值范围：\[3.0, 30.0\]。时间越长效果越好。 @@ -213,7 +219,7 @@ bool **重要** -仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 +仅适用于Qwen-Audio-TTS/CosyVoice声音复刻（model为`voice-enrollment`时），且仅qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash、cosyvoice-v3.5-plus、v3.5-flash和v3-flash模型支持。 是否开启音频预处理（降噪、音频增强、音量规整）。有背景噪音时建议开启；安静环境建议关闭以最大程度还原音色。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/qwen-audio-realtime-websocket-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/fun-audiochat-realtime-websocket-api.md similarity index 67% rename from skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/qwen-audio-realtime-websocket-api.md rename to skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/fun-audiochat-realtime-websocket-api.md index 25294168..75868574 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/qwen-audio-realtime-websocket-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/audio-api-references/voice-conversation-api-references/real-time-voice-conversation-api-references/fun-audiochat-realtime-websocket-api.md @@ -127,7 +127,7 @@ Qwen-Audio Realtime API 支持三种交互模式，通过 `session.update` 事 下图展示了 server\_vad 模式下的典型交互时序： -服务端 客户端 服务端 客户端 会话初始化 语音输入 loop \[用户说话中\] loop \[用户说话中\] 响应生成 loop \[流式输出\] loop \[多轮交互\] connect session.created session.update session.updated input\_audio\_buffer.append input\_audio\_buffer.speech\_started input\_audio\_buffer.append conversation.item.input\_audio\_transcription.delta input\_audio\_buffer.speech\_stopped conversation.item.input\_audio\_transcription.completed commit audio buffer input\_audio\_buffer.committed conversation.item.created response.created response.output\_item.added conversation.item.created response.content\_part.added response.audio\_transcript.delta response.audio.delta response.audio\_transcript.done response.audio.done response.content\_part.done response.output\_item.done response.done +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7268354871/p1088432.svg) 按时间顺序，客户端与服务端的交互流程如下： @@ -148,7 +148,7 @@ Qwen-Audio Realtime API 支持三种交互模式，通过 `session.update` 事 模型播报期间，若 VAD 检测到用户开始说话，服务端会取消当前响应（返回 `response.done`，状态为 `cancelled`），随后开始新一轮语音输入和响应。下图展示了用户打断的交互时序： -服务端 客户端 服务端 客户端 服务端正在流式输出 loop \[流式输出中\] 用户打断 新一轮语音输入 loop \[用户说话中\] 新一轮推理 response.audio.delta input\_audio\_buffer.append（用户开始说话） response.done（status=cancelled） input\_audio\_buffer.speech\_started input\_audio\_buffer.append conversation.item.input\_audio\_transcription.delta input\_audio\_buffer.speech\_stopped conversation.item.input\_audio\_transcription.completed commit audio buffer input\_audio\_buffer.committed conversation.item.created response.created +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7268354871/p1088435.svg) ### **smart\_turn 模式** @@ -160,7 +160,7 @@ Qwen-Audio Realtime API 支持三种交互模式，通过 `session.update` 事 下图展示了 smart\_turn 模式下的典型交互时序： -服务端 客户端 服务端 客户端 会话初始化 loop \[用户说话中\] 无效语音（可能出现 0~N 次） loop \[用户说话中\] loop \[用户说话中\] 响应生成 loop \[流式输出\] loop \[多轮交互\] connect session.created session.update session.updated input\_audio\_buffer.append conversation.item.ambient\_audio\_transcription.delta conversation.item.ambient\_audio\_transcription.completed input\_audio\_buffer.append input\_audio\_buffer.speech\_started input\_audio\_buffer.append conversation.item.input\_audio\_transcription.delta input\_audio\_buffer.speech\_stopped conversation.item.input\_audio\_transcription.completed commit audio buffer input\_audio\_buffer.committed conversation.item.created response.created response.output\_item.added conversation.item.created response.content\_part.added response.audio\_transcript.delta response.audio.delta response.audio\_transcript.done response.audio.done response.content\_part.done response.output\_item.done response.done +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7268354871/p1088441.svg) 与 server\_vad 模式的主要区别： @@ -175,13 +175,13 @@ Qwen-Audio Realtime API 支持三种交互模式，通过 `session.update` 事 与 server\_vad 模式的打断处理基本一致。下图展示了用户打断的交互时序： -服务端 客户端 服务端 客户端 服务端正在流式输出 loop \[流式输出中\] 用户打断 新一轮语音输入 loop \[用户说话中\] 新一轮推理 response.audio.delta input\_audio\_buffer.append（用户开始说话） response.done（status=cancelled） input\_audio\_buffer.speech\_started input\_audio\_buffer.append conversation.item.input\_audio\_transcription.delta conversation.item.input\_audio\_transcription.completed input\_audio\_buffer.speech\_stopped commit audio buffer input\_audio\_buffer.committed conversation.item.created response.created +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7268354871/p1088443.svg) ## **无效轮次** 已判定有效的语音可能被撤回（`input_audio_buffer.speech_stopped` 返回 `reason=turn_invalid`），此时不触发推理，客户端应继续发送音频等待下一轮有效语音。下图展示了无效轮次的交互时序： -服务端 客户端 服务端 客户端 loop \[用户说话中\] loop \[用户说话中\] 轮次无效 继续发送音频，等待下一轮 input\_audio\_buffer.append input\_audio\_buffer.speech\_started input\_audio\_buffer.append conversation.item.input\_audio\_transcription.delta input\_audio\_buffer.speech\_stopped（reason=turn\_invalid） +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7268354871/p1088444.svg) ### **说话人增强配置流程** @@ -216,7 +216,7 @@ Qwen-Audio Realtime API 支持三种交互模式，通过 `session.update` 事 下图展示了 push-to-talk 模式下的典型交互时序： -服务端 客户端 服务端 客户端 会话初始化 语音输入 loop \[用户说话中\] 响应生成 loop \[流式输出\] loop \[多轮交互\] connect session.created session.update session.updated input\_audio\_buffer.append conversation.item.input\_audio\_transcription.delta input\_audio\_buffer.commit（用户松开按键） conversation.item.input\_audio\_transcription.completed input\_audio\_buffer.committed conversation.item.created response.create（手动触发推理） response.created response.output\_item.added conversation.item.created response.content\_part.added response.audio\_transcript.delta response.audio.delta response.audio\_transcript.done response.audio.done response.content\_part.done response.output\_item.done response.done +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7268354871/p1088447.svg) 按时间顺序，客户端与服务端的交互流程如下： @@ -233,7 +233,7 @@ Qwen-Audio Realtime API 支持三种交互模式，通过 `session.update` 事 客户端发送 `response.cancel` 取消当前响应，服务端返回 `response.done`（状态为 `cancelled`，原因为 `client_cancelled`）。下图展示了用户打断的交互时序： -服务端 客户端 服务端 客户端 服务端正在流式输出 loop \[流式输出中\] 用户打断 新一轮语音输入 loop \[用户说话中\] 新一轮推理 response.audio\_transcript.delta response.audio.delta response.cancel response.done（status=cancelled, reason=client\_cancelled） input\_audio\_buffer.append conversation.item.input\_audio\_transcription.delta input\_audio\_buffer.commit conversation.item.input\_audio\_transcription.completed input\_audio\_buffer.committed conversation.item.created response.create response.created +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7268354871/p1088449.svg) ## **各模式操作约束** diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/image-creative-tools-api-reference/creative-poster-generation-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/image-creative-tools-api-reference/creative-poster-generation-api.md index 0a687e48..96d7ae74 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/image-creative-tools-api-reference/creative-poster-generation-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/image-creative-tools-api-reference/creative-poster-generation-api.md @@ -80,7 +80,6 @@ curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.c "generate_mode":"generate", "generate_num":1 }, - "auxiliary_parameters": "WMq4SC4......", "parameters":{} }' ``` diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/image-creative-tools-api-reference/wanx-background-generation-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/image-creative-tools-api-reference/wanx-background-generation-api-reference.md index 720eeeec..825a4dde 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/image-creative-tools-api-reference/wanx-background-generation-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/image-creative-tools-api-reference/wanx-background-generation-api-reference.md @@ -71,14 +71,16 @@ wanx-background-generation-v2 ### **步骤1：创建任务获取任务ID** -`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/background-generation/generation/` +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/background-generation/generation` #### **请求头（Headers）** ## 图像背景生成 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/background-generation/generation/' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/background-generation/generation' \ --header 'X-DashScope-Async: enable' \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ --header 'Content-Type: application/json' \ @@ -396,7 +398,7 @@ foreground\_edge图像列表和background\_edge图像列表之和不得超过10 ### **步骤2：根据任务ID查询结果** -`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` #### **请求头（Headers）** @@ -404,10 +406,10 @@ foreground\_edge图像列表和background\_edge图像列表之和不得超过10 请将`86ecf553-d340-4e21-xxxxxxxxx`替换为真实的task\_id。 -> 若使用新加坡地域的模型，需将base\_url替换为https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx，其中WorkspaceId需替换为真实的业务空间ID。 +> 若使用新加坡地域的模型，需将base\_url替换为https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx，其中{WorkspaceId}需替换为真实的业务空间ID。 ``` -curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx \ +curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" ``` diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-generation-and-editing-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-generation-and-editing-api-reference.md new file mode 100644 index 00000000..1537ef67 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-generation-and-editing-api-reference.md @@ -0,0 +1,459 @@ +# 千问-图像生成与编辑3.0 API参考 + +千问-图像生成与编辑3.0模型同时支持文生图（T2I）和图生图/图像编辑（I2I），可根据文本提示词直接生成图像，也可基于1-3张参考图结合编辑指令进行精确编辑。 + +**重要** + +该模型目前处于邀测阶段，您需要前往模型广场申请开通后方可使用。 + +## **模型概览** + +**模型名称** + +**模型简介** + +**输出图像规格** + +qwen-image-3.0-pro + +千问图像生成与编辑3.0模型，同时支持文生图（T2I）和图生图/图像编辑（I2I）。 + +图像分辨率： + +- **文生图（T2I）**：总像素需在512\*512至2048\*2048之间。 + +- **图生图（I2I）**：总像素需在512\*512至2048\*2048之间。 + +- **默认**：不指定`size`时，模型根据提示词自动推荐分辨率。 + + +图像格式：png + +## **前提条件** + +在调用前，您需要[获取API Key](https://help.aliyun.com/zh/model-studio/get-api-key)，再[配置API Key到环境变量](https://help.aliyun.com/zh/model-studio/configure-api-key-through-environment-variables)。 + +如需通过SDK进行调用，请[安装DashScope SDK](https://help.aliyun.com/zh/model-studio/install-sdk)。目前，该SDK已支持Python和Java。 + +**重要** + +华北2（北京）和新加坡地域拥有独立的 **API Key** 与**请求地址**，不可混用，跨地域调用将导致鉴权失败或服务报错。 + +**重要** + +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： + +- 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 + +## HTTP调用 + +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` + +**新加坡地域**：`POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` + +调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + +#### 请求参数 + +## 文生图（T2I） + +``` +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ +--header 'Content-Type: application/json' \ +--header "Authorization: Bearer $DASHSCOPE_API_KEY" \ +--data '{ + "model": "qwen-image-3.0-pro", + "input": { + "messages": [ + { + "role": "user", + "content": [ + { + "text": "画面是一张竖幅户外人像摄影，整体从上到下呈现温暖的午后街景氛围。顶部左侧到上方大面积被深绿色藤蔓和橙色小花覆盖，花叶从建筑檐口自然垂落，受阳光照射的叶片呈黄绿色高光，阴影处则偏深绿，形成浓密而柔和的背景层次。左上至中上区域是一块深蓝色横向招牌，招牌表面较暗、略带磨砂质感，上面以白色哥特体大字写着 Il Messaggero，文字位于画面左侧偏上，部分被前景花叶轻微遮挡，字体高对比、带装饰性尖角和粗细变化。招牌下方是报刊亭或书报摊的玻璃展示窗，黑色金属框架将橱窗分隔成多个矩形区域，内部陈列着许多报纸、杂志和书刊封面，但大多因景深虚化和光线反射而难以辨读，形成浅色纸张与深色边框交错的背景纹理。画面右上方是强烈的逆光区域，阳光从街道尽头照入，背景建筑被虚化成米灰色块面，边缘柔和，呈现明显的浅景深效果。画面中部偏右是一名年轻成年女性的半身至膝上人像，她回头面向镜头微笑，身体略向右转，肩背朝向观者，姿态自然放松。她有长而浓密的黑色波浪卷发，发丝被逆光勾勒出金色轮廓光，发梢在右侧向外散开，显得轻盈蓬松。她肤色白皙，脸型柔和偏鹅蛋形，眉形细致，眼睛明亮，眼妆清透，睫毛明显，面部带有自然高光，唇部为柔和珊瑚红色，笑容露齿，表情亲切明朗。她佩戴小巧耳饰，身穿黑色细肩带露背连衣裙，面料颜色深黑、轮廓简洁，细肩带从肩部向背部延伸，背部线条清晰。画面下部偏左到中部，她双手抱着一束玫瑰花，花束体积较大，主要由橙色、杏色、粉色和浅桃色玫瑰组成，花瓣层层卷曲，边缘被阳光照亮，绿色叶片和长花茎从花束下方垂出，花束与黑色裙装形成鲜明色彩对比。右侧背景是一条被阳光照亮的城市街道，地面呈暖灰与金黄色调，远处建筑、街边设施和一个模糊的红色圆形交通标志位于右下远景，均因焦外虚化而只保留色块和轮廓。整张照片采用暖色胶片感处理，带有细腻颗粒、柔和对比和明显逆光边缘光，人物位于视觉焦点，背景报刊亭、花藤、街道和阳光共同营造出浪漫、明亮、都市漫步式的氛围。" + } + ] + } + ] + }, + "parameters": { + "prompt_extend": true + } +}' +``` + +## 图生图/图像编辑（I2I） + +``` +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \ +--header 'Content-Type: application/json' \ +--header "Authorization: Bearer $DASHSCOPE_API_KEY" \ +--data '{ + "model": "qwen-image-3.0-pro", + "input": { + "messages": [ + { + "role": "user", + "content": [ + { + "image": "https://alidocs.oss-cn-zhangjiakou.aliyuncs.com/res/yBRq1ZPYEaXdyOdv/img/33a80a19-7ac7-4c64-b0fa-7d685b7046a0.png" + }, + { + "text": "帮我生成一张充满高级感的都市风格女性写真，画面中人物完美保留输入图片中这位年轻女性的面部特征与一头柔顺的黑色长发。人物脱下原本的米色针织上衣，换上一套彰显高雅气质的都市职场穿搭，身穿一件质感垂顺的香槟色真丝衬衫，外搭一件剪裁利落的深灰色休闲西装外套，下身搭配同色系的高腰阔腿裤，整体造型既干练又富有女人味。场景设定在一家装修现代简约的高端咖啡店内，背景是通透的落地玻璃窗，窗外隐约可见繁华的城市街景，室内摆放着深色实木长桌和舒适的皮质座椅，桌面上放置着一台打开的银色笔记本电脑、一份文件和一杯热气腾腾的美式咖啡。人物呈现出慵懒而放松的办公姿态，身体微微后仰倚靠在椅背上，一只手臂自然搭在扶手上，另一只手轻轻握着咖啡杯置于桌边，头部微侧，眼神清澈从容且带有一丝慵懒地直视镜头，嘴角挂着一抹优雅自信的微笑。人物化着精致得体的正式场合妆容，底妆清透干净，眉眼线条清晰利落，唇部涂抹着显气色的豆沙色口红，展现出成熟知性的魅力。光线采用午后柔和的自然光，从侧面透过落地窗洒入，在人物的面部轮廓和衣物褶皱上留下细腻的光影过渡，背景呈现自然的景深虚化效果，色彩以大地色、灰色和暖白色为主调，营造出宁静、高级且充满故事感的都市办公氛围，构图采用经典的竖幅七分身人像视角，人物位于画面视觉中心略偏右，比例协调，画质清晰细腻。" + } + ] + } + ] + }, + "parameters": { + "prompt_extend": true + } +}' +``` + +##### 请求头（Headers） + +**Content-Type** `_string_` **（必选）** + +请求内容类型。此参数必须设置为`application/json`。 + +**Authorization** `_string_`**（必选）** + +请求身份认证。接口使用阿里云百炼API Key进行身份认证。示例值：Bearer sk-xxxx。 + +##### 请求体（Request Body） + +**model** `_string_` **（必选）** + +模型名称，当前可用模型为`qwen-image-3.0-pro`。 + +**input** `_object_` **（必选）** + +输入参数对象，包含以下字段： + +**属性** + +**messages** `_array_` **（必选）** + +请求内容数组。**当前仅支持单轮对话**，因此数组内**有且只有一个对象**，该对象包含`role`和`content`两个属性。 + +**属性** + +**role** `_string_` **（必选）** + +消息发送者角色，必须设置为`user`。 + +**content** `_array_` **（必选）** + +消息内容数组，根据使用场景有不同的组合方式： + +- **文生图（T2I）**：仅包含一个`{"text": "..."}`对象。 + +- **图生图（I2I）**：包含1-3个`{"image": "..."}`对象和1个`{"text": "..."}`对象。 + + +**属性** + +**image** `_string_` （可选） + +输入图像的 URL 或 Base64 编码数据。I2I场景下支持传入1-3张图像。多图输入时，按照数组顺序定义图像顺序。 + +**图像要求：** + +- 图像格式：JPG、JPEG、PNG、BMP、TIFF、WEBP和GIF。 + +- 图像分辨率：建议图像的宽和高均在384像素至2048像素之间。 + +- 图像大小：不超过10MB。 + + +**支持的输入格式** + +1. 公网URL：支持 HTTP 和 HTTPS 协议。您也可在此[获取临时公网URL](https://help.aliyun.com/zh/model-studio/get-temporary-file-url)。 + +2. Base64 编码：格式为`data:{MIME_type};base64,{base64_data}`。 + + +**text** `_string_` **（必选）** + +正向提示词，用于描述您期望生成或编辑的图像内容、风格和构图。支持中英文。 + +**注意**：仅支持传入一个text，不传或传入多个将报错。 + +**parameters** `_object_` （可选） + +控制图像生成的附加参数。 + +**属性** + +**prompt\_extend** `_boolean_` （可选） + +是否开启提示词智能改写，默认值为 `true`（建议开启）。开启后，模型会优化正向提示词，对描述较简单的提示词效果提升明显。 + +**n** `_integer_` （可选） + +输出图像的数量，支持输出1-6张图片，默认值为1。 + +**size** `_string_` （可选） + +设置输出图像的分辨率，格式为`宽*高`，例如`"1024*1024"`。未指定时由模型根据提示词自动推荐分辨率。 + +- **文生图（T2I）**：像素范围512\*512至2048\*2048。 + +- **图生图（I2I）**：像素范围512\*512至2048\*2048。 + + +**negative\_prompt** `_string_` （可选） + +反向提示词，用来描述不希望在画面中看到的内容，可以对画面进行限制。 + +**seed** `_integer_` （可选） + +随机数种子，取值范围`[0, 2147483647]`。固定种子可使生成结果相对稳定。 + +**watermark** `_boolean_` （可选） + +是否添加水印，默认值为 `false`。 + +#### 响应参数 + +## 任务执行成功 + +任务数据（如任务状态、图像URL等）仅保留24小时，超时后会被自动清除。请您务必及时保存生成的图像。 + +``` +{ + "output": { + "choices": [ + { + "finish_reason": "stop", + "message": { + "content": [ + { + "image": "https://dashscope-result-sz.oss-cn-shenzhen.aliyuncs.com/xxx.png?Expires=xxx" + } + ], + "role": "assistant" + } + } + ] + }, + "usage": { + "width": 1024, + "height": 1024, + "image_count": 1 + }, + "request_id": "571ae02f-5c9d-436c-83c2-f221e6df0xxx" +} +``` + +## 任务执行异常 + +如果因为某种原因导致任务执行失败，将返回相关信息，可以通过code和message字段明确指示错误原因。请参见[错误码](https://help.aliyun.com/zh/model-studio/error-code)进行解决。 + +``` +{ + "request_id": "31f808fd-8eef-9004-xxxxx", + "code": "InvalidApiKey", + "message": "Invalid API-key provided." +} +``` + +**output** `_object_` + +包含模型生成结果。 + +**属性** + +**choices** `_array_` + +结果选项列表。 + +**属性** + +**finish\_reason** `_string_` + +任务停止原因，自然停止时为`stop`。 + +**message** `_object_` + +模型返回的消息。 + +**属性** + +**role** `_string_` + +消息的角色，固定为`assistant`。 + +**content** `_array_` + +消息内容，包含生成的图像信息。 + +**属性** + +**image** `_string_` + +生成图像的 URL，格式为PNG。**链接有效期为24小时**，请及时下载并保存图像。 + +**usage** `_object_` + +本次调用的资源使用情况，仅调用成功时返回。 + +**属性** + +**width** `_integer_` + +生成图像的宽度（像素）。 + +**height** `_integer_` + +生成图像的高度（像素）。 + +**image\_count** `_integer_` + +生成图像的张数。 + +**request\_id** `_string_` + +请求唯一标识。可用于请求明细溯源和问题排查。 + +**code** `_string_` + +请求失败的错误码。请求成功时不会返回此参数，详情请参见[错误码](https://help.aliyun.com/zh/model-studio/error-code)。 + +**message** `_string_` + +请求失败的详细信息。请求成功时不会返回此参数，详情请参见[错误码](https://help.aliyun.com/zh/model-studio/error-code)。 + +## SDK调用 + +以下以图生图/图像编辑（I2I）为示例，展示Python和Java SDK的调用方式。 + +## Python + +``` +import os +import base64 +import mimetypes +import dashscope +from dashscope import MultiModalConversation + +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' + +def encode_file(file_path): + mime_type, _ = mimetypes.guess_type(file_path) + if not mime_type or not mime_type.startswith("image/"): + raise ValueError("Unsupported or unrecognized image format") + with open(file_path, "rb") as image_file: + encoded_string = base64.b64encode(image_file.read()).decode('utf-8') + return f"data:{mime_type};base64,{encoded_string}" + +# [方法一] 使用公网图像URL +image_url = "https://alidocs.oss-cn-zhangjiakou.aliyuncs.com/res/yBRq1ZPYEaXdyOdv/img/33a80a19-7ac7-4c64-b0fa-7d685b7046a0.png" + +# [方法二] 使用Base64编码图像 +# image_url = encode_file("./your_image.png") + +response = MultiModalConversation.call( + api_key=os.getenv("DASHSCOPE_API_KEY"), + model="qwen-image-3.0-pro", + messages=[{ + "role": "user", + "content": [ + {"image": image_url}, + {"text": "帮我生成一张充满高级感的都市风格女性写真，画面中人物完美保留输入图片中这位年轻女性的面部特征与一头柔顺的黑色长发。人物换上一套彰显高雅气质的都市职场穿搭，场景设定在一家装修现代简约的高端咖啡店内。"} + ] + }], + prompt_extend=True +) + +print(response) +if response.status_code == 200: + url = response.output.choices[0].message.content[0]["image"] + print(f"Generated image URL: {url}") +else: + print(f"Error: {response.code} - {response.message}") +``` + +## Java + +``` +import java.util.Arrays; +import java.util.Base64; +import java.util.Collections; +import java.io.IOException; +import java.nio.file.Files; +import java.nio.file.Path; +import java.nio.file.Paths; +import com.alibaba.dashscope.aigc.multimodalconversation.MultiModalConversation; +import com.alibaba.dashscope.aigc.multimodalconversation.MultiModalConversationParam; +import com.alibaba.dashscope.aigc.multimodalconversation.MultiModalConversationResult; +import com.alibaba.dashscope.common.MultiModalMessage; +import com.alibaba.dashscope.common.Role; +import com.alibaba.dashscope.utils.Constants; + +public class ImageEditExample { + public static void main(String[] args) { + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; + + // [方法一] 使用公网图像URL + String imageUrl = "https://alidocs.oss-cn-zhangjiakou.aliyuncs.com/res/yBRq1ZPYEaXdyOdv/img/33a80a19-7ac7-4c64-b0fa-7d685b7046a0.png"; + + // [方法二] 使用Base64编码图像 + // String imageUrl = encodeFile("/path/to/your/image.png"); + + MultiModalConversation conv = new MultiModalConversation(); + MultiModalMessage userMessage = MultiModalMessage.builder() + .role(Role.USER.getValue()) + .content(Arrays.asList( + Collections.singletonMap("image", imageUrl), + Collections.singletonMap("text", "帮我生成一张充满高级感的都市风格女性写真，画面中人物完美保留输入图片中这位年轻女性的面部特征与一头柔顺的黑色长发。人物换上一套彰显高雅气质的都市职场穿搭，场景设定在一家装修现代简约的高端咖啡店内。") + )) + .build(); + MultiModalConversationParam param = MultiModalConversationParam.builder() + .apiKey(System.getenv("DASHSCOPE_API_KEY")) + .model("qwen-image-3.0-pro") + .messages(Arrays.asList(userMessage)) + .parameter("prompt_extend", true) + .build(); + try { + MultiModalConversationResult result = conv.call(param); + System.out.println(result); + } catch (Exception e) { + e.printStackTrace(); + } + } + + public static String encodeFile(String filePath) { + Path path = Paths.get(filePath); + if (!Files.exists(path)) { + throw new IllegalArgumentException("File does not exist: " + filePath); + } + String mimeType = null; + try { + mimeType = Files.probeContentType(path); + } catch (IOException e) { + throw new IllegalArgumentException("Cannot detect file type: " + filePath); + } + if (mimeType == null || !mimeType.startsWith("image/")) { + throw new IllegalArgumentException("Unsupported or unrecognized image format"); + } + byte[] fileBytes = null; + try { + fileBytes = Files.readAllBytes(path); + } catch (IOException e) { + throw new IllegalArgumentException("Cannot read file content: " + filePath); + } + String encodedString = Base64.getEncoder().encodeToString(fileBytes); + return "data:" + mimeType + ";base64," + encodedString; + } +} +``` + +## **错误码** + +如果模型调用失败并返回报错信息，请参见[错误码](https://help.aliyun.com/zh/model-studio/error-code)进行解决。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/client-events.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/client-events.md index bb4d9f20..0441479a 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/client-events.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/client-events.md @@ -89,7 +89,7 @@ Qwen-Omni-Realtime API的客户端事件参考。 默认音色： -- Qwen3.5-Omni-Realtime 系列：`Tina` +- Qwen3.5-Omni-Realtime 系列模型：`Tina` - Qwen3-Omni-Flash-Realtime：`Cherry` @@ -135,7 +135,7 @@ VAD 类型，可选值： - `server_vad`（默认值）：基于声学特征检测语音结束。 -- `semantic_vad`：基于语义有效性检测语音结束，可过滤回应语、背景音等无意义声音。仅 `qwen3.5-omni-realtime` 模型支持。 +- `semantic_vad`：基于语义有效性检测语音结束，可过滤回应语、背景音等无意义声音。仅 `qwen3.5-omni-realtime` 系列模型支持。 **threshold** `_float_`（可选） @@ -160,7 +160,7 @@ VAD 灵敏度。值越低，VAD 越灵敏，越容易将微弱声音（包括背 **enable\_search** `_boolean_`（可选） -**仅在使用 Qwen3.5-Omni-Realtime 模型时生效。** +**仅在使用 Qwen3.5-Omni-Realtime 系列模型时生效。** 是否启用联网搜索。设为 `true` 启用，默认为 `false`。启用后，模型可自主判断是否需要联网搜索来回答用户问题。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/omni-realtime-java-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/omni-realtime-java-sdk.md index 0aece341..5d4f89d4 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/omni-realtime-java-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/omni-realtime-java-sdk.md @@ -176,7 +176,7 @@ VAD类型，取值如下： - `server_vad`（默认值）：基于声学特征检测用户语音结束。 -- `semantic_vad`：基于语义有效性检测用户语音结束，可过滤无意义语音（如回应语、背景音）。仅`qwen3.5-omni-realtime`模型支持。 +- `semantic_vad`：基于语义有效性检测用户语音结束，可过滤无意义语音（如回应语、背景音）。仅`qwen3.5-omni-realtime`系列模型支持。 turnDetectionThreshold @@ -210,7 +210,7 @@ enable\_search Boolean -**仅在使用 Qwen3.5-Omni-Realtime 模型时生效。** +**仅在使用 Qwen3.5-Omni-Realtime 系列模型时生效。** 是否启用联网搜索功能。设置为 `true` 启用，默认为 `false`。启用后，模型可自主判断是否需要搜索来回应用户的即时问题。 @@ -230,7 +230,7 @@ tools List> -**仅在使用 Qwen3.5-Omni-Realtime 模型时生效。** +**仅在使用 Qwen3.5-Omni-Realtime 系列模型时生效。** 工具定义列表。启用后，模型可自主判断是否需要调用外部工具来回应用户的问题。命中工具调用时，模型不生成音频，仅返回工具调用参数。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/omni-realtime-python-sdk.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/omni-realtime-python-sdk.md index 869f563c..cd0f1593 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/omni-realtime-python-sdk.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/omni-realtime-python-sdk.md @@ -95,7 +95,7 @@ str 默认音色： -- `Qwen3.5-Omni-Realtime`: `"Tina"` +- `Qwen3.5-Omni-Realtime`系列模型: `"Tina"` - `Qwen3-Omni-Flash-Realtime`: `"Cherry"` @@ -157,7 +157,7 @@ VAD类型，取值如下： - `server_vad`（默认值）：基于声学特征检测用户语音结束。 -- `semantic_vad`：基于语义有效性检测用户语音结束，可过滤无意义语音（如回应语、背景音）。仅`Qwen3.5-Omni-Realtime`模型支持。 +- `semantic_vad`：基于语义有效性检测用户语音结束，可过滤无意义语音（如回应语、背景音）。仅`Qwen3.5-Omni-Realtime`系列模型支持。 turn\_detection\_threshold @@ -191,7 +191,7 @@ enable\_search bool -**仅在使用 Qwen3.5-Omni-Realtime 模型时生效。** +**仅在使用 Qwen3.5-Omni-Realtime 系列模型时生效。** 是否启用联网搜索功能。设置为 `true` 启用，默认为 `false`。启用后，模型可自主判断是否需要搜索来回应用户的即时问题。 @@ -209,7 +209,7 @@ tools list\[dict\] -**仅在使用 Qwen3.5-Omni-Realtime 模型时生效。** +**仅在使用 Qwen3.5-Omni-Realtime 系列模型时生效。** 工具定义列表。启用后，模型可自主判断是否需要调用外部工具来回应用户的问题。命中工具调用时，模型不生成音频，仅返回工具调用参数。 @@ -244,7 +244,7 @@ temperature越高，生成的内容更多样，反之，生成的内容更确定 由于temperature与top\_p均可以控制生成内容的多样性，因此建议只设置其中一个值。 -- `Qwen3.5-Omni-Realtime`系列：0.7 +- `Qwen3.5-Omni-Realtime`系列模型：0.7 - `Qwen3-Omni-Flash-Realtime`系列：0.9 @@ -267,7 +267,7 @@ top\_p越高，生成的内容更多样。反之，生成的内容更确定。 top\_p默认值： -- `Qwen3.5-Omni-Realtime`系列：0.8 +- `Qwen3.5-Omni-Realtime`系列模型：0.8 - `Qwen3-Omni-Flash-Realtime`系列：1.0 @@ -286,7 +286,7 @@ integer top\_k默认值： -- `Qwen3.5-Omni-Realtime`系列：20 +- `Qwen3.5-Omni-Realtime`系列模型：20 - `Qwen3-Omni-Flash-Realtime`系列：50 @@ -317,7 +317,7 @@ float repetition\_penalty默认值： -- `Qwen3.5-Omni-Realtime`系列：1.0 +- `Qwen3.5-Omni-Realtime`系列模型：1.0 - 其他模型：1.05 @@ -334,7 +334,7 @@ float presence\_penalty默认值： -- `Qwen3.5-Omni-Realtime`系列：1.5 +- `Qwen3.5-Omni-Realtime`系列模型：1.5 - 其他模型：0.0 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/qwen-omni-voice-cloning.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/qwen-omni-voice-cloning.md index 7fe83582..f35b6a89 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/qwen-omni-voice-cloning.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/qwen-omni-voice-cloning.md @@ -10,9 +10,14 @@ **重要** -百炼为新加坡地域推出了业务空间专属域名 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`，**能够为推理请求提供卓越的性能和更高的稳定性**，建议从 `https://dashscope-intl.aliyuncs.com` 迁移至新域名。 +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +- 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## **音频要求** @@ -54,7 +59,7 @@ WAV (16bit)、MP3、M4A ## 快速开始：复刻与使用音色 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1397661871/CAEQbxiBgICd6_Do8BkiIDM3NjYwZDQxMGIyMTQzMDdhOGMyY2YwNWFhMmM2NjVi5899512_20251120114927.389.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/2365984871/CAEQbxiBgICd6_Do8BkiIDM3NjYwZDQxMGIyMTQzMDdhOGMyY2YwNWFhMmM2NjVi5899512_20251120114927.389.svg) ### 1\. 工作流程 @@ -134,6 +139,7 @@ DEFAULT_AUDIO_MIME_TYPE = "audio/mpeg" VOICE_FILE_PATH = "voice.mp3" # 用于声音复刻的本地音频文件的相对路径 def create_voice(file_path: str, + url: str, target_model: str = DEFAULT_TARGET_MODEL, preferred_name: str = DEFAULT_PREFERRED_NAME, audio_mime_type: str = DEFAULT_AUDIO_MIME_TYPE) -> str: @@ -150,9 +156,6 @@ def create_voice(file_path: str, base64_str = base64.b64encode(file_path_obj.read_bytes()).decode() data_uri = f"data:{audio_mime_type};base64,{base64_str}" - - # 以下为华北2（北京）地域的URL，各地域的URL不同。 - url = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization" payload = { "model": "qwen-voice-enrollment", "input": { @@ -198,17 +201,18 @@ class SimpleCallback(OmniRealtimeCallback): if __name__ == '__main__': # 若没有配置环境变量，请用百炼API Key将下行替换为：dashscope.api_key = "sk-xxx" dashscope.api_key = os.getenv("DASHSCOPE_API_KEY") - # 以下为华北2（北京）地域的URL，各地域的URL不同。 - url = "wss://dashscope.aliyuncs.com/api-ws/v1/realtime" + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com" + ws_url = base_url.replace("https://", "wss://") + "/api-ws/v1/realtime" # 1. 声音复刻：创建专属音色 - voice = create_voice(VOICE_FILE_PATH) + voice = create_voice(VOICE_FILE_PATH, url=f"{base_url}/api/v1/services/audio/tts/customization") print(f"声音复刻完成，音色: {voice}") # 2. 使用复刻音色进行实时对话 pya = pyaudio.PyAudio() callback = SimpleCallback(pya) - conv = OmniRealtimeConversation(model=DEFAULT_TARGET_MODEL, callback=callback, url=url) + conv = OmniRealtimeConversation(model=DEFAULT_TARGET_MODEL, callback=callback, url=ws_url) conv.connect() conv.update_session( output_modalities=[MultiModality.AUDIO, MultiModality.TEXT], @@ -234,6 +238,7 @@ if __name__ == '__main__': ``` import com.alibaba.dashscope.audio.omni.*; import com.alibaba.dashscope.exception.NoApiKeyException; +import com.alibaba.dashscope.utils.Constants; import com.google.gson.Gson; import com.google.gson.JsonObject; @@ -258,6 +263,8 @@ public class Main { // 用于声音复刻的本地音频文件的相对路径 private static final String AUDIO_FILE = "voice.mp3"; private static final String AUDIO_MIME_TYPE = "audio/mpeg"; + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + private static final String BASE_URL = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com"; // 生成 data URI public static String toDataUrl(String filePath) throws IOException { @@ -285,8 +292,7 @@ public class Main { + "}" + "}"; - // 以下为华北2（北京）地域的URL，各地域的URL不同。 - String url = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization"; + String url = BASE_URL + "/api/v1/services/audio/tts/customization"; HttpURLConnection con = (HttpURLConnection) new URL(url).openConnection(); con.setRequestMethod("POST"); con.setRequestProperty("Authorization", "Bearer " + apiKey); @@ -366,8 +372,7 @@ public class Main { // 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key // 若没有配置环境变量，请用百炼API Key将下行替换为：.apikey("sk-xxx") .apikey(System.getenv("DASHSCOPE_API_KEY")) - // 以下为华北2（北京）地域的URL，各地域的URL不同。 - .url("wss://dashscope.aliyuncs.com/api-ws/v1/realtime") + .url(BASE_URL.replace("https://", "wss://") + "/api-ws/v1/realtime") .build(); OmniRealtimeConversation conversation = new OmniRealtimeConversation(param, new OmniRealtimeCallback() { @@ -466,8 +471,7 @@ def create_voice(file_path: str, base64_str = base64.b64encode(file_path_obj.read_bytes()).decode() data_uri = f"data:{audio_mime_type};base64,{base64_str}" - # 以下为华北2（北京）地域的URL，各地域的URL不同。 - url = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization" + url = dashscope.base_http_api_url + "/services/audio/tts/customization" payload = { "model": "qwen-voice-enrollment", "input": { @@ -492,8 +496,8 @@ def create_voice(file_path: str, raise RuntimeError(f"解析 voice 响应失败: {e}") if __name__ == '__main__': - # 以下为华北2（北京）地域的URL，各地域的URL不同。 - dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 1. 声音复刻：创建专属音色 voice = create_voice(VOICE_FILE_PATH) @@ -556,6 +560,8 @@ public class Main { // 用于声音复刻的本地音频文件的相对路径 private static final String AUDIO_FILE = "voice.mp3"; private static final String AUDIO_MIME_TYPE = "audio/mpeg"; + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + private static final String BASE_URL = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com"; // 将 PCM 数据写入标准 WAV 文件 public static void writeWav(String path, byte[] pcmData, int sampleRate) throws IOException { @@ -601,8 +607,7 @@ public class Main { + "}" + "}"; - // 以下为华北2（北京）地域的URL，各地域的URL不同。 - String url = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization"; + String url = BASE_URL + "/api/v1/services/audio/tts/customization"; HttpURLConnection con = (HttpURLConnection) new URL(url).openConnection(); con.setRequestMethod("POST"); con.setRequestProperty("Authorization", "Bearer " + apiKey); @@ -650,8 +655,7 @@ public class Main { + "\"stream_options\": {\"include_usage\": true}" + "}"; - // 以下为华北2（北京）地域的URL，各地域的URL不同。 - String url = "https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions"; + String url = BASE_URL + "/compatible-mode/v1/chat/completions"; HttpURLConnection con = (HttpURLConnection) new URL(url).openConnection(); con.setRequestMethod("POST"); con.setRequestProperty("Authorization", "Bearer " + apiKey); @@ -712,13 +716,13 @@ public class Main { - **URL** - 中国内地： + 华北2（北京）： ``` - POST https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization + POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization ``` - 国际： + 新加坡： ``` POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/audio/tts/customization @@ -1024,11 +1028,11 @@ public class Main { ``` # ======= 重要提示 ======= - # 以下为华北2（北京）地域的URL，各地域的URL不同。 + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 # 新加坡地域和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key # === 执行时请删除该注释 === - curl -X POST https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization \ + curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -1062,8 +1066,8 @@ public class Main { # 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key = "sk-xxx" api_key = os.getenv("DASHSCOPE_API_KEY") - # 以下为华北2（北京）地域的URL，各地域的URL不同。 - url = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization" + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization" payload = { "model": "qwen-voice-enrollment", # 不要修改这个值 @@ -1121,8 +1125,8 @@ public class Main { // 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key // 若没有配置环境变量，请用百炼API Key将下行替换为：String apiKey = "sk-xxx" String apiKey = System.getenv("DASHSCOPE_API_KEY"); - // 以下为华北2（北京）地域的URL，各地域的URL不同。 - String apiUrl = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization"; + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + String apiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization"; try { // 构造 JSON 请求体（注意内部的引号需转义） @@ -1188,13 +1192,13 @@ public class Main { - **URL** - 中国内地： + 华北2（北京）： ``` - POST https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization + POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization ``` - 国际： + 新加坡： ``` POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/audio/tts/customization @@ -1387,11 +1391,11 @@ public class Main { ``` # ======= 重要提示 ======= - # 以下为华北2（北京）地域的URL，各地域的URL不同。 + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 # 新加坡地域和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key # === 执行时请删除该注释 === - curl --location --request POST 'https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization' \ + curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization' \ --header 'Authorization: Bearer $DASHSCOPE_API_KEY' \ --header 'Content-Type: application/json' \ --data '{ @@ -1413,8 +1417,8 @@ public class Main { # 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key = "sk-xxx" api_key = os.getenv("DASHSCOPE_API_KEY") - # 以下为华北2（北京）地域的URL，各地域的URL不同。 - url = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization" + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization" payload = { "model": "qwen-voice-enrollment", # 不要修改该值 @@ -1463,8 +1467,8 @@ public class Main { // 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key // 若没有配置环境变量，请用百炼API Key将下行替换为：String apiKey = "sk-xxx" String apiKey = System.getenv("DASHSCOPE_API_KEY"); - // 以下为华北2（北京）地域的URL，各地域的URL不同。 - String apiUrl = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization"; + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + String apiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization"; // JSON 请求体（旧版本 Java 无 """ 多行字符串） String jsonPayload = @@ -1533,13 +1537,13 @@ public class Main { - **URL** - 中国内地： + 华北2（北京）： ``` - POST https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization + POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization ``` - 国际： + 新加坡： ``` POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/audio/tts/customization @@ -1678,11 +1682,11 @@ public class Main { ``` # ======= 重要提示 ======= - # 以下为华北2（北京）地域的URL，各地域的URL不同。 + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 # 新加坡地域和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key # === 执行时请删除该注释 === - curl --location --request POST 'https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization' \ + curl --location --request POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization' \ --header 'Authorization: Bearer $DASHSCOPE_API_KEY' \ --header 'Content-Type: application/json' \ --data '{ @@ -1703,8 +1707,8 @@ public class Main { # 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key = "sk-xxx" api_key = os.getenv("DASHSCOPE_API_KEY") - # 以下为华北2（北京）地域的URL，各地域的URL不同。 - url = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization" + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization" voice_to_delete = "yourVoice" # 要删除的音色（替换为真实值） @@ -1752,8 +1756,8 @@ public class Main { // 新加坡和北京地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key // 若没有配置环境变量，请用百炼API Key将下行替换为：String apiKey = "sk-xxx" String apiKey = System.getenv("DASHSCOPE_API_KEY"); - // 以下为华北2（北京）地域的URL，各地域的URL不同。 - String apiUrl = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization"; + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + String apiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/audio/tts/customization"; String voiceToDelete = "yourVoice"; // 要删除的音色（替换为真实值） // 构造 JSON 请求体（字符串拼接，兼容 Java 8） diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/server-events.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/server-events.md index 1b9a731f..98d52374 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/server-events.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/omni-realtime-api/server-events.md @@ -162,7 +162,7 @@ VAD检测阈值。 **enable\_search** `_boolean_` -是否启用联网搜索功能。仅 Qwen3.5-Omni-Realtime 模型支持。 +是否启用联网搜索功能。仅 Qwen3.5-Omni-Realtime 系列模型支持。 **search\_options** `_object_` @@ -308,7 +308,7 @@ VAD检测阈值。 **enable\_search** `_boolean_`（可选） -是否启用联网搜索功能。仅 Qwen3.5-Omni-Realtime 模型支持。 +是否启用联网搜索功能。仅 Qwen3.5-Omni-Realtime 系列模型支持。 **search\_options** `_object_`（可选） diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/preparations/error-code.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/preparations/error-code.md index 59f0dda7..1ef9f9e0 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/preparations/error-code.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/preparations/error-code.md @@ -213,6 +213,17 @@ AI 助理准确分析出原因，并给出解决方案： - **使用正确的模型名称：**请对照模型列表中的模型名称，检查输入的`model`是否正确。请勿混用开源社区的模型名与百炼模型ID，如应该使用`qwen3-235b-a22b-instruct-2507`，而非`Qwen/Qwen3-235B-A22B-Instruct-2507`。 +### **The product is not activated, please confirm that you have activated products and try again after activation.** + +**原因：** 所调用的模型服务未开通（未激活）。通过 OpenAI 兼容接口调用时，若目标模型未在模型市场开通，网关会返回该错误，错误码为 `invalid_parameter_error`。 + +**解决方案：** + +- **开通模型服务：**请前往阿里云百炼控制台[模型市场](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market)，确认所调用的模型（如 `kimi-k3`）处于已开通状态后再行调用。 + +- **开通百炼服务：**若尚未开通阿里云百炼服务，请先开通百炼服务后再调用。 + + ### The result\_format parameter must be \\"message\\" when enable\_thinking is tru**e** **原因：** 调用思考模式模型，`result_format`参数未设置为`"message"`。 @@ -767,13 +778,13 @@ AI 助理准确分析出原因，并给出解决方案： ### **request timeout after 23 seconds.** -**原因：** 超过23秒未向服务发送数据。该报错信息在使用[实时语音合成（Sambert）](https://help.aliyun.com/zh/model-studio/sambert-speech-synthesis/)、[语音识别（Paraformer）](https://help.aliyun.com/zh/model-studio/paraformer-speech-recognition)和[实时语音合成（CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)时产生。 +**原因：** 超过23秒未向服务发送数据。该报错信息在使用[实时语音合成（Sambert）](https://help.aliyun.com/zh/model-studio/sambert-speech-synthesis/)、[语音识别（Paraformer）](https://help.aliyun.com/zh/model-studio/paraformer-speech-recognition)和[实时语音合成（Qwen-Audio-TTS/CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)时产生。 **解决方案：** 请检查为什么长时间未向服务器发送数据。如果长时间（超过23秒）不向服务端发送消息，请及时结束任务。 ### **Please ensure input text is valid.** -**原因：** 若您使用[实时语音合成（CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)，此错误通常是由于未发送待合成文本引起的。可能原因包括：参数遗漏（未为 `text` 参数赋值）或代码异常（导致对 `text` 参数的赋值失败）。 +**原因：** 若您使用[实时语音合成（Qwen-Audio-TTS/CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)，此错误通常是由于未发送待合成文本引起的。可能原因包括：参数遗漏（未为 `text` 参数赋值）或代码异常（导致对 `text` 参数的赋值失败）。 **解决方案：** 请排查代码，确保 `text` 参数被正确赋值并发送。 @@ -816,7 +827,7 @@ AI 助理准确分析出原因，并给出解决方案： ### **\[tts:\]Engine return error code: 418** -**原因：** 使用[实时语音合成（CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)，请求参数 `voice`（音色）不正确，或 `model`（模型）与 `voice`（音色）版本不匹配。 +**原因：** 使用[实时语音合成（Qwen-Audio-TTS/CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)，请求参数 `voice`（音色）不正确，或 `model`（模型）与 `voice`（音色）版本不匹配。 **解决方案：** @@ -831,7 +842,7 @@ AI 助理准确分析出原因，并给出解决方案： ### **Request voice is invalid!** -**原因：** 若您使用[实时语音合成（CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)，此错误通常是因为未设置音色。 +**原因：** 若您使用[实时语音合成（Qwen-Audio-TTS/CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)，此错误通常是因为未设置音色。 **解决方案：** 请检查是否对`voice`参数赋值。若您使用[WebSocket API参考](https://help.aliyun.com/zh/model-studio/cosyvoice-websocket-api)，请参照API文档按照正确JSON格式配置参数。 @@ -1015,7 +1026,7 @@ AI 助理准确分析出原因，并给出解决方案： **解决方案：**前往[费用与成本](https://usercenter2.aliyun.com/home)查看是否欠费： -- 未欠费：请确认该 API Key 是否属于当前账号； +- 未欠费：请确认该 API Key 是否属于当前账号。如果账号不存在欠费的情况，可能账户出现异常，详情请联系客服进一步排查。 - 欠费：请及时充值。充值后，系统余额可能存在延迟，请稍等后重试。 @@ -1745,7 +1756,7 @@ A：请核对资源包的可抵扣范围。以qwen-plus/qwen-plus-latest系列 - **填写错误**：阿里云百炼的 API Key 以 `sk-` 开头，请确认未误填其他模型提供商的密钥，且复制时未包含多余空格或换行符。 -- **套餐专属 API Key（Coding Plan / Token Plan 团队版）**：Coding Plan 与 Token Plan 团队版均提供以 `sk-sp-` 开头的专属 API Key，**必须配合各自的专属 Base URL 使用**，不可与通用 API Key/Base URL 混用（混用会返回本鉴权错误）。其中 Coding Plan 的专属地址为 https://coding.dashscope.aliyuncs.com/v1；Token Plan 团队版的专属 Base URL 可在控制台**我的订阅**的 API Key 区域查看。请确认同时更新了 API Key 和 Base URL，具体配置方法请分别参见[接入AI工具](https://help.aliyun.com/zh/model-studio/use-coding-plan-in-ai-tools/)与[快速开始](https://help.aliyun.com/zh/model-studio/token-plan-quickstart)。 +- **套餐专属 API Key（Coding Plan / Token Plan 团队版）**：Coding Plan 与 Token Plan 团队版均提供以 `sk-sp-` 开头的专属 API Key，**必须配合各自的专属 Base URL 使用**，不可与通用 API Key/Base URL 混用（混用会返回本鉴权错误）。其中 Coding Plan 的专属地址为 https://coding.dashscope.aliyuncs.com/v1；Token Plan 团队版的专属 Base URL 可在控制台**我的订阅**的 API Key 区域查看。请确认同时更新了 API Key 和 Base URL，具体配置方法请分别参见[接入AI工具](https://help.aliyun.com/zh/model-studio/use-coding-plan-in-ai-tools/)与[快速开始](https://help.aliyun.com/zh/model-studio/token-plan-team-quickstart)。 - **地域不匹配**：API Key 和 Base URL 属于不同的地域，例如使用了华北2（北京）地域的 API Key 和新加坡地域的 Base URL。请确认您使用的 API Key 位于[北京](https://bailian.console.aliyun.com/?tab=globalset#/efm/api_key)地域页面或[新加坡](https://modelstudio.console.aliyun.com/?tab=globalset#/efm/api_key)地域页面，或[美国](https://modelstudio.console.aliyun.com/us-east-1)地域页面。各地域对应的 Base URL 如下： @@ -1806,6 +1817,8 @@ A：请核对资源包的可抵扣范围。以qwen-plus/qwen-plus-latest系列 - 中国站用户请使用**华北2（北京）地域的接入地址；国际站用户请使用新加坡**地域的接入地址。使用[在线调试](https://api.aliyun.com/api/bailian/2023-12-29/CreateIndex)时，确认服务地址正确。 + ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7513215871/p1090718.png) + ## **401-**invalid access token or token expired @@ -1958,6 +1971,8 @@ A：请核对资源包的可抵扣范围。以qwen-plus/qwen-plus-latest系列 - 请前往模型广场开通模型服务。 +- 如果您通过国际站 API 端点（如 `dashscope-us.aliyuncs.com`）发起调用，请注意不同地域可用的模型列表不同。调用前请确认目标模型是否在该地域可用，部分模型在美国地域需使用带 `-us` 后缀的模型名称（如 `qwen-max-us`）。 + ## **404-**model\_not\_supported @@ -2038,7 +2053,7 @@ A：请核对资源包的可抵扣范围。以qwen-plus/qwen-plus-latest系列 - 如需更高调用频率，可参考[限流](https://help.aliyun.com/zh/model-studio/rate-limit)申请提额。 -## **429-**Throttling.RateQuota/LimitRequests/limit\_requests +## **429-**Throttling.RateQuota/LimitRequests/limit\_requests/ResourceExhausted/Too many requests ### **You have exceeded your request limit./Requests rate limit exceeded, please try again later.** /You exceeded your current requests list. @@ -2186,7 +2201,7 @@ A：请核对资源包的可抵扣范围。以qwen-plus/qwen-plus-latest系列 **原因：** 语音合成中使用的音色不存在。 -**解决方案：** 请检查`voice`参数，确保指定了正确的音色名称。可用音色请参见[实时语音合成（CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)。 +**解决方案：** 请检查`voice`参数，确保指定了正确的音色名称。可用音色请参见[实时语音合成（Qwen-Audio-TTS/CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)。 ## **500-**InternalError.FileUpload @@ -2438,7 +2453,7 @@ A：请核对资源包的可抵扣范围。以qwen-plus/qwen-plus-latest系列 ### **Cannot resolve symbol 'ttsv2'** -**原因：** 若您使用[实时语音合成（CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)，出现该问题的原因是DashScope SDK版本过低。 +**原因：** 若您使用[实时语音合成（Qwen-Audio-TTS/CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)，出现该问题的原因是DashScope SDK版本过低。 **解决方案：** 请[安装最新版 DashScope SDK](https://help.aliyun.com/zh/model-studio/install-sdk#f80a232bb24v7)。 @@ -2458,7 +2473,7 @@ A：请核对资源包的可抵扣范围。以qwen-plus/qwen-plus-latest系列 ### **InputRequiredException: Parameter invalid: text is null** -**原因**：使用[实时语音合成（CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)时未发送待合成文本。 +**原因**：使用[实时语音合成（Qwen-Audio-TTS/CosyVoice）](https://help.aliyun.com/zh/model-studio/cosyvoice-large-model-for-speech-synthesis/)时未发送待合成文本。 **解决方案：**调用语音合成接口时为 `text` 参数赋值。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/batch-interfaces-compatible-with-openai.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/batch-interfaces-compatible-with-openai.md index 26d8278b..0a419a8e 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/batch-interfaces-compatible-with-openai.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/batch-interfaces-compatible-with-openai.md @@ -6,7 +6,7 @@ ## **工作流程** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/2515653871/CAEQaxiBgIDB5qWk4BkiIDViYzQ0MWUwNTYyNDQ3NDM5NzM0ZTc4N2Y3NTU2NjA56318723_20260129171731.699.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3199194871/CAEQaxiBgIDB5qWk4BkiIDViYzQ0MWUwNTYyNDQ3NDM5NzM0ZTc4N2Y3NTU2NjA56318723_20260129171731.699.svg) ## **前提条件** @@ -18,9 +18,9 @@ - **服务端点** - - **中国内地：**`https://dashscope.aliyuncs.com/compatible-mode/v1` + - **华北2（北京）：**`https://dashscope.aliyuncs.com/compatible-mode/v1` - - **国际：**`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` + - **新加坡：**`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` **重要** @@ -39,7 +39,7 @@ - 千问 Plus：qwen3.7-plus、qwen3.6-plus、qwen3.5-plus、qwen-plus、qwen-plus-latest - - 千问 Flash：qwen3.6-flash、qwen3.5-flash、qwen-flash + - 千问 Flash：qwen3.7-flash、qwen3.7-flash-2026-07-15、qwen3.6-flash、qwen3.5-flash、qwen-flash - 千问 Long：qwen-long、qwen-long-latest @@ -47,7 +47,7 @@ - **多模态模型** - - [图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)：qwen3.7-plus、qwen3.6-plus、qwen3.6-flash、qwen3.5-plus、qwen3.5-flash、qwen3-vl-plus、qwen3-vl-flash + - [图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)：qwen3.7-plus、qwen3.6-plus、qwen3.7-flash、qwen3.7-flash-2026-07-15、qwen3.6-flash、qwen3.5-plus、qwen3.5-flash、qwen3-vl-plus、qwen3-vl-flash - [文字提取](https://help.aliyun.com/zh/model-studio/qwen-vl-ocr)：qwen-vl-ocr、qwen-vl-ocr-latest @@ -58,7 +58,7 @@ **重要** -- 在Batch 场景下，`qwen3.7-max`、`qwen3.7-plus`、`qwen3.6-plus`、`qwen3.6-flash`、`qwen3.5-plus`、`qwen3.5-flash`和`qwen3.5-omni-plus`单次请求的上下文 Token 数最大支持 256K，`qwen3.5-omni-plus`不支持语音输出。 +- 在Batch 场景下，`qwen3.7-max`、`qwen3.7-plus`、`qwen3.6-plus`、`qwen3.7-flash`、`qwen3.7-flash-2026-07-15`、`qwen3.6-flash`、`qwen3.5-plus`、`qwen3.5-flash`和`qwen3.5-omni-plus`单次请求的上下文 Token 数最大支持 256K，`qwen3.5-omni-plus`不支持语音输出。 - 部分模型支持思考模式，开启后会产生思考`tokens`导致成本增加。 @@ -771,9 +771,9 @@ fi JSONL 批量生成工具 -**请选择模式：** +**请选择地域：** -中国内地 国际 +华北2（北京） 新加坡 **选择模型系列:** 文本生成模型 多模态模型 通用文本向量模型 @@ -785,9 +785,9 @@ fi 生成 -**请选择模式：** +**请选择地域：** -中国内地 国际 +华北2（北京） 新加坡 **选择模型系列:** 文本生成模型 @@ -878,7 +878,7 @@ client = OpenAI( # 若没有配置环境变量，可用阿里云百炼API Key将下行替换为：api_key="sk-xxx"。但不建议在生产环境中直接将API Key硬编码到代码中，以减少API Key泄露风险。 # 新加坡和北京地域的API Key不同。 api_key=os.getenv("DASHSCOPE_API_KEY"), - # 以下是北京地域base_url，如果使用新加坡地域的模型，需要将base_url替换为：https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1 + # 以下是北京地域base_url，如果使用新加坡地域的模型，需要将base_url替换为：https://dashscope-intl.aliyuncs.com/compatible-mode/v1 # 注意：切换地域时，API Key也需要对应更换 base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", ) @@ -909,7 +909,7 @@ const fs = require('fs'); // 北京地域配置（默认） const BASE_URL = 'https://dashscope.aliyuncs.com/compatible-mode/v1'; // 如果使用新加坡地域，请将上面的 BASE_URL 替换为： -// const BASE_URL = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1'; +// const BASE_URL = 'https://dashscope-intl.aliyuncs.com/compatible-mode/v1'; // 注意：切换地域时，API Key也需要对应更换 const apiKey = process.env.DASHSCOPE_API_KEY; @@ -955,7 +955,7 @@ import java.util.regex.Matcher; * * 地域配置： * - 北京地域：https://dashscope.aliyuncs.com/compatible-mode/v1 - * - 新加坡地域：https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1 + * - 新加坡地域：https://dashscope-intl.aliyuncs.com/compatible-mode/v1 * 注意：切换地域时，API Key也需要对应更换 */ public class BatchAPIUploadFile { @@ -963,7 +963,7 @@ public class BatchAPIUploadFile { // 北京地域配置（默认） private static final String BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1"; // 如果使用新加坡地域，请将上面的 BASE_URL 替换为： - // private static final String BASE_URL = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1"; + // private static final String BASE_URL = "https://dashscope-intl.aliyuncs.com/compatible-mode/v1"; // 注意：切换地域时，API Key也需要对应更换 private static String API_KEY; @@ -1035,7 +1035,7 @@ String fileId = uploadFile("test.jsonl"); ``` # ======= 重要提示 ======= # 新加坡和北京地域的API Key不同。 -# 以下是北京地域base_url，如果使用新加坡地域的模型，需要将base_url替换为：https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1/files +# 以下是北京地域base_url，如果使用新加坡地域的模型，需要将base_url替换为：https://dashscope-intl.aliyuncs.com/compatible-mode/v1/files # === 执行时请删除该注释 === curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/files \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ @@ -2469,7 +2469,7 @@ System.out.println(content); ``` # ======= 重要提示 ======= # 新加坡和北京地域的API Key不同。 -# 以下是北京地域base_url，如果使用新加坡地域的模型，需要将base_url替换为：https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1/files/file-batch_output-xxx/content +# 以下是北京地域base_url，如果使用新加坡地域的模型，需要将base_url替换为：https://dashscope-intl.aliyuncs.com/compatible-mode/v1/files/file-batch_output-xxx/content # === 执行时请删除该注释 === curl -X GET https://dashscope.aliyuncs.com/compatible-mode/v1/files/file-batch_output-xxx/content \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" > result.jsonl diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/compatibility-with-openai-responses-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/compatibility-with-openai-responses-api.md index 53fb9ef1..823f8b6c 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/compatibility-with-openai-responses-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/compatibility-with-openai-responses-api.md @@ -19,7 +19,7 @@ ## 支持的模型 -`qwen3-max`、`qwen3-max-2026-01-23`、`qwen3.7-max`、`qwen3.7-max-2026-05-20`、`qwen3.7-max-2026-06-08`、`qwen3.7-plus`、`qwen3.7-plus-2026-05-26`、`qwen3.6-plus`、`qwen3.6-plus-2026-04-02`、`qwen3.5-plus`、`qwen3.5-plus-2026-02-15`、`qwen3.5-plus-2026-04-20`、`qwen3.6-flash`、`qwen3.6-flash-2026-04-16`、`qwen3.5-flash`、`qwen3.5-flash-2026-02-23`、`qwen3.6-35b-a3b`、`qwen3.5-397b-a17b`、`qwen3.5-122b-a10b`、`qwen3.5-27b`、`qwen3.5-35b-a3b`、`qwen-plus`、`qwen-flash`、`qwen3-coder-plus`、`qwen3-coder-flash`、`qwen3-coder-next`。 +`qwen3-max`、`qwen3-max-2026-01-23`、`qwen3.7-max`、`qwen3.7-max-2026-05-20`、`qwen3.7-max-2026-06-08`、`qwen3.7-plus`、`qwen3.7-plus-2026-05-26`、`qwen3.6-plus`、`qwen3.6-plus-2026-04-02`、`qwen3.5-plus`、`qwen3.5-plus-2026-02-15`、`qwen3.5-plus-2026-04-20`、`qwen3.7-flash`、`qwen3.7-flash-2026-07-15`、`qwen3.6-flash`、`qwen3.6-flash-2026-04-16`、`qwen3.5-flash`、`qwen3.5-flash-2026-02-23`、`qwen3.6-35b-a3b`、`qwen3.5-397b-a17b`、`qwen3.5-122b-a10b`、`qwen3.5-27b`、`qwen3.5-35b-a3b`、`qwen-plus`、`qwen-flash`、`qwen3-coder-plus`、`qwen3-coder-flash`、`qwen3-coder-next`。 ## 服务地址 @@ -771,7 +771,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/ **使用方式**：在请求 Header 中添加 `x-dashscope-session-cache: enable` 开启，或设置为 `disable` 关闭。默认值为 `disable`。 -**支持的模型：**`qwen3-max`、`qwen3.7-max`、`qwen3.7-max-2026-05-20`、`qwen3.7-max-2026-06-08`、`qwen3.7-plus`、`qwen3.7-plus-2026-05-26`、`qwen3.6-plus`、`qwen3.5-plus`、`qwen3.6-flash`、`qwen3.5-flash`、`qwen-plus`、`qwen-flash`、`qwen3-coder-plus`、`qwen3-coder-flash` +**支持的模型：**`qwen3-max`、`qwen3.7-max`、`qwen3.7-max-2026-05-20`、`qwen3.7-max-2026-06-08`、`qwen3.7-plus`、`qwen3.7-plus-2026-05-26`、`qwen3.6-plus`、`qwen3.5-plus`、`qwen3.7-flash`、`qwen3.6-flash`、`qwen3.5-flash`、`qwen-plus`、`qwen-flash`、`qwen3-coder-plus`、`qwen3-coder-flash` > Session 缓存 最小可缓存提示词长度为 1024 Token，缓存有效期为 5 分钟。相关约束限制与[显式缓存](https://help.aliyun.com/zh/model-studio/context-cache)一致。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/completions.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/completions.md index a88b645e..849e2704 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/completions.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/completions.md @@ -4,7 +4,7 @@ Completions 接口专为文本补全场景设计，适合代码补全、内容 **说明** -本文档仅适用于中国内地（北京地域），需使用中国（北京）地域的[API Key](https://bailian.console.aliyun.com/?tab=model#/api-key)。 +本文档仅适用于华北2（北京）地域，需使用华北2（北京）地域的[API Key](https://bailian.console.aliyun.com/?tab=model#/api-key)。 ## **支持的模型** @@ -46,7 +46,7 @@ import os from openai import OpenAI client = OpenAI( - base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", api_key=os.getenv("DASHSCOPE_API_KEY") ) @@ -67,7 +67,7 @@ const openai = new OpenAI( { // 若没有配置环境变量，请用阿里云百炼API Key将下行替换为：apiKey: "sk-xxx", apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" + baseURL: "https://dashscope.aliyuncs.com/compatible-mode/v1" } ); @@ -85,7 +85,7 @@ main(); curl ``` -curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/completions \ +curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -113,7 +113,7 @@ import os from openai import OpenAI client = OpenAI( - base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", api_key=os.getenv("DASHSCOPE_API_KEY") ) @@ -145,7 +145,7 @@ Node.js import OpenAI from 'openai'; const client = new OpenAI({ - baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + baseURL: "https://dashscope.aliyuncs.com/compatible-mode/v1", apiKey: process.env.DASHSCOPE_API_KEY }); @@ -178,7 +178,7 @@ main(); curl ``` -curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/completions \ +curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/openai-compatible-batch-chat.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/openai-compatible-batch-chat.md index 3b3cff7e..0cdb9d23 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/openai-compatible-batch-chat.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/openai-compatible-batch-chat.md @@ -19,14 +19,14 @@ ### **华北2（北京）** -- **文本生成模型：**qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash、qwen3.5-plus、qwen3.5-flash、qwen3-max、qwen-plus、qwen-flash、deepseek-v3.2 +- **文本生成模型：**qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.7-flash、qwen3.7-flash-2026-07-15、qwen3.6-flash、qwen3.5-plus、qwen3.5-flash、qwen3-max、qwen-plus、qwen-flash、deepseek-v3.2 -- **图像与视频理解模型：**qwen3.7-plus、qwen3.6-plus、qwen3.6-flash、qwen3.5-plus、qwen3.5-flash、qwen3.5-omni-plus、qwen3-vl-plus、qwen3-vl-flash +- **图像与视频理解模型：**qwen3.7-plus、qwen3.6-plus、qwen3.7-flash、qwen3.7-flash-2026-07-15、qwen3.6-flash、qwen3.5-plus、qwen3.5-flash、qwen3.5-omni-plus、qwen3-vl-plus、qwen3-vl-flash **重要** -- 在Batch 场景下，`qwen3.7-max`、`qwen3.7-plus`、`qwen3.6-plus`、`qwen3.6-flash`、`qwen3.5-plus`、`qwen3.5-flash`和`qwen3.5-omni-plus`单次请求的上下文 Token 数最大支持 256K，`qwen3.5-omni-plus`不支持语音输出。 +- 在Batch 场景下，`qwen3.7-max`、`qwen3.7-plus`、`qwen3.6-plus`、`qwen3.7-flash`、`qwen3.7-flash-2026-07-15`、`qwen3.6-flash`、`qwen3.5-plus`、`qwen3.5-flash`和`qwen3.5-omni-plus`单次请求的上下文 Token 数最大支持 256K，`qwen3.5-omni-plus`不支持语音输出。 - 部分模型支持思考模式，开启后会产生思考`tokens`导致成本增加。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/openai-compatible-conversations.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/openai-compatible-conversations.md index b7487808..37eae351 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/openai-compatible-conversations.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/toolkits-and-frameworks/openai-compatible-conversations.md @@ -16,14 +16,14 @@ **重要** -百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： - 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` - 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 **items** `_array_`（可选） @@ -948,3 +948,5 @@ console.log("第二轮响应:", response2.output_text); - 创建会话或添加消息项时，`items` 最多包含20条。 - `metadata` 最多16对键值对，key最大长度64字符，value最大长度512字符。 + +- 会话信息最多保存7天。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-batch-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-batch-api.md index d94c7a68..480aa12c 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-batch-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-batch-api.md @@ -64,9 +64,7 @@ text-embedding-async-v1 通用文本向量批处理接口API支持通过HTTP和DashScope SDK进行调用。 -在调用前，您需要[获取API Key](https://help.aliyun.com/zh/model-studio/get-api-key)，再[配置API Key到环境变量](https://help.aliyun.com/zh/model-studio/configure-api-key-through-environment-variables)。 - -如需通过SDK进行调用，请[安装DashScope SDK](https://help.aliyun.com/zh/model-studio/install-sdk)。目前，该SDK已支持Python和Java。 +在调用前，先[获取API Key](https://help.aliyun.com/zh/model-studio/get-api-key)，再[配置API Key到环境变量](https://help.aliyun.com/zh/model-studio/configure-api-key-through-environment-variables)。如需通过SDK进行调用，请[安装DashScope SDK](https://help.aliyun.com/zh/model-studio/install-sdk)。 ## HTTP调用 @@ -79,14 +77,16 @@ HTTP调用仅支持异步模式，需通过两步完成： **通过HTTP调用时需配置的endpoint：** -`POST https://dashscope.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding` +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ### 创建任务 ##### **请求参数** ``` -curl -X POST 'https://dashscope.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \ +curl -X POST 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ -H 'X-DashScope-Async: enable' \ @@ -232,7 +232,7 @@ url `_string_` **（必选）** ### **根据任务ID查询结果** -`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` ##### **请求参数** @@ -240,10 +240,10 @@ url `_string_` **（必选）** 请将`86ecf553-d340-4e21-xxxxxxxxx`替换为真实的task\_id。 -> 若使用新加坡地域的模型，需将base\_url替换为https://dashscope-intl.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx +> 若使用新加坡地域的模型，需将base\_url替换为https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx，其中{WorkspaceId}需替换为真实的业务空间ID。 ``` -curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx \ +curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" ``` @@ -398,6 +398,9 @@ SDK与HTTP接口的参数名基本一致，参数结构根据不同语言的SDK ``` from dashscope import BatchTextEmbedding +import dashscope +# 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 +dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" result = BatchTextEmbedding.call(BatchTextEmbedding.Models.text_embedding_async_v1, url="https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241016/nigwvr/text_embedding_file.txt", @@ -416,8 +419,11 @@ import com.alibaba.dashscope.exception.NoApiKeyException; import com.alibaba.dashscope.task.AsyncTaskListParam; import com.alibaba.dashscope.task.AsyncTaskListResult; import com.alibaba.dashscope.utils.JsonUtils; +import com.alibaba.dashscope.utils.Constants; public class Main { + // 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; public static void basicCall() throws ApiException, NoApiKeyException { BatchTextEmbeddingParam param = BatchTextEmbeddingParam.builder() .model(BatchTextEmbedding.Models.TEXT_EMBEDDING_ASYNC_V1) @@ -430,6 +436,8 @@ public class Main { } public static void main(String[] args) { + // 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; try { basicCall(); } catch (ApiException | NoApiKeyException e) { @@ -446,7 +454,10 @@ public class Main { ``` from dashscope import BatchTextEmbedding +import dashscope from http import HTTPStatus +# 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 +dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" # 创建异步任务 def create_async_task(): @@ -508,8 +519,11 @@ import com.alibaba.dashscope.exception.NoApiKeyException; import com.alibaba.dashscope.task.AsyncTaskListParam; import com.alibaba.dashscope.task.AsyncTaskListResult; import com.alibaba.dashscope.utils.JsonUtils; +import com.alibaba.dashscope.utils.Constants; public class Main { + // 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; /**创建批处理任务*/ public static BatchTextEmbeddingResult createTask() throws ApiException, NoApiKeyException { @@ -738,7 +752,7 @@ url `_string_` **（必选）** 任务状态 -- SUCCESSED: 任务执行成功 +- SUCCEEDED: 任务执行成功 - FAILED: 任务执行失败 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/vector-and-sort/multimodal-vector/multimodal-embedding-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/vector-and-sort/multimodal-vector/multimodal-embedding-api-reference.md index f65767c7..0d7a583d 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/vector-and-sort/multimodal-vector/multimodal-embedding-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/vector-and-sort/multimodal-vector/multimodal-embedding-api-reference.md @@ -68,7 +68,7 @@ qwen3-vl-embedding 100万Token -有效期：百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天（以较晚者为准） qwen2.5-vl-embedding @@ -178,16 +178,24 @@ qwen3-vl-embedding 支持中、英、日、韩、法、德等33种主流语言 +**所有支持语言** + +中文、日语、韩语、印尼语、越南语、泰语、英语、法语、德语、俄语、葡萄牙语、西班牙语、意大利语、瑞典语、丹麦语、捷克语、挪威语、荷兰语、芬兰语、土耳其语、波兰语、斯瓦希里语、罗马尼亚语、塞尔维亚语、希腊语、哈萨克语、乌兹别克语、宿务语、阿拉伯语、乌尔都语、波斯语、印地语 / 天城语、希伯来语。 + JPEG, PNG, WEBP, BMP, TIFF, ICO, DIB, ICNS, SGI（支持URL或Base64） MP4, AVI, MOV（仅支持URL） -一次请求中传入内容元素总数不超过 20。图片数量不超过5，视频数量不超过1。 +一次请求中传入内容元素总数不超过 20。图片数量不超过10，视频数量不超过1。 qwen2.5-vl-embedding 支持中、英、日、韩、法、德等11种主流语言 +**所有支持语言** + +中文、英语、日语、韩语、法语、德语、俄语、葡萄牙语、西班牙语、意大利语、印尼语 + 一次请求内，图片、文本、视频、融合对象每种类型最多出现 1 次。 **多模态向量模型** @@ -206,6 +214,10 @@ tongyi-embedding-vision-plus-2026-03-06 支持中、英、日、韩等超30种主流语言 +**所有支持语言** + +中文、日语、韩语、印尼语、越南语、泰语、英语、法语、德语、俄语、葡萄牙语、西班牙语、意大利语、瑞典语、丹麦语、捷克语、挪威语、荷兰语、芬兰语、土耳其语、波兰语、斯瓦希里语、罗马尼亚语、塞尔维亚语、希腊语、哈萨克语、乌兹别克语、宿务语、阿拉伯语、乌尔都语、波斯语、印地语 / 天城语、希伯来语。 + JPEG, PNG, WEBP, BMP, TIFF, ICO, DIB, ICNS, SGI（支持URL或Base64） MP4, MPEG, MOV, MPG, WEBM, AVI, FLV, MKV（仅支持URL） @@ -542,7 +554,7 @@ contents `_array_`**（必选）** ] }, "usage": { - "input_tokens": 10, + "input_tokens": 903, "input_tokens_details": { "image_tokens": 896, "text_tokens": 7 @@ -644,11 +656,11 @@ contents `_array_`**（必选）** **image\_tokens** `_int_` -输入的图片或视频的 Token 数量。 +输入内容中图片或视频等**视觉部分**消耗的 Token 数量，**不包含文本**（文本部分见 `text_tokens`）。图片消耗的 Token 数量与输入图片的分辨率有关，分辨率越高消耗的 Token 越多；若输入为视频，系统会先对视频抽帧，再基于抽帧结果计算 Token。 **text\_tokens** `_int_` -输入的文本的 Token 数量。 +输入内容中**文本部分**消耗的 Token 数量（不包含图片或视频等视觉部分）。 **output\_tokens** `_int_` @@ -660,7 +672,7 @@ contents `_array_`**（必选）** **image\_tokens** `_int_` -本次请求输入的图片或视频的 Token 数量。系统会对输入视频进行抽帧处理，帧数上限受系统配置控制，随后基于处理结果计算 Token。仅 `qwen3-vl-embedding`、`qwen2.5-vl-embedding` 和 `multimodal-embedding-v1` 返回此字段（作为顶层字段），`tongyi-embedding-vision-*` 系列模型的图片 Token 包含在 `input_tokens_details.image_tokens` 中。 +本次请求输入的图片或视频等**视觉部分**消耗的 Token 数量（**不包含文本**）。图片消耗的 Token 数量与输入图片的分辨率有关；系统会对输入视频进行抽帧处理，帧数上限受系统配置控制，随后基于处理结果计算 Token。仅 `qwen3-vl-embedding`、`qwen2.5-vl-embedding` 和 `multimodal-embedding-v1` 返回此字段（作为顶层字段），`tongyi-embedding-vision-*` 系列模型的图片 Token 包含在 `input_tokens_details.image_tokens` 中。 **image\_count** `_int_` diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-image-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-image-to-video-api-reference.md index 79c7fb66..07f2c9b9 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-image-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-image-to-video-api-reference.md @@ -4,13 +4,13 @@ HappyHorse图生视频模型，以首帧图片为基础，支持通过文本描 ## 适用范围 -为确保调用成功，请务必保证模型、endpoint URL 和 API Key 均属于**同一地域**。跨地域调用将会失败。 +为确保调用成功，请务必保证模型、endpoint URL和API Key 均属于**同一地域**。跨地域调用将会失败。 - [**选择模型**](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)：确认模型所属的地域。 - **选择 URL**：选择对应的地域 Endpoint URL，支持HTTP URL。 -- **配置 API Key**：获取该地域的[API Key](https://help.aliyun.com/zh/model-studio/get-api-key)，再[配置API Key到环境变量](https://help.aliyun.com/zh/model-studio/configure-api-key-through-environment-variables)。 +- **配置API Key**：获取该地域的[API Key](https://help.aliyun.com/zh/model-studio/get-api-key)，再[配置API Key到环境变量](https://help.aliyun.com/zh/model-studio/configure-api-key-through-environment-variables)。 **说明** @@ -19,14 +19,14 @@ HappyHorse图生视频模型，以首帧图片为基础，支持通过文本描 **重要** -百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： - 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` - 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -38,14 +38,10 @@ HappyHorse图生视频模型，以首帧图片为基础，支持通过文本描 `POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **美国（弗吉尼亚）** `POST https://dashscope-us.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` @@ -54,7 +50,11 @@ HappyHorse图生视频模型，以首帧图片为基础，支持通过文本描 `POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +## **日本（东京）** + +`POST https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -68,6 +68,7 @@ HappyHorse图生视频模型，以首帧图片为基础，支持通过文本描 ## 图生视频-基于首帧 ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ @@ -169,7 +170,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi 1. 公网URL： - - 支持 HTTP 或 HTTPS 协议。 + - 支持HTTP或HTTPS协议。 - 示例值：https://xxx/xxx.png。 @@ -326,14 +327,10 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **美国（弗吉尼亚）** `GET https://dashscope-us.aliyuncs.com/api/v1/tasks/{task_id}` @@ -342,7 +339,9 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +## **日本（东京）** + +`GET https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -363,7 +362,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-reference-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-reference-to-video-api-reference.md index b93ad55b..57e4536b 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-reference-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-reference-to-video-api-reference.md @@ -19,14 +19,14 @@ HappyHorse-参考生视频模型支持传入**多张参考图像**，通过**文 **重要** -百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： - 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` - 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -38,14 +38,10 @@ HappyHorse-参考生视频模型支持传入**多张参考图像**，通过**文 `POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **美国（弗吉尼亚）** `POST https://dashscope-us.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` @@ -54,7 +50,11 @@ HappyHorse-参考生视频模型支持传入**多张参考图像**，通过**文 `POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +## **日本（东京）** + +`POST https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -68,6 +68,7 @@ HappyHorse-参考生视频模型支持传入**多张参考图像**，通过**文 ## 参考生视频（多图像） ``` +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ @@ -368,14 +369,10 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **美国（弗吉尼亚）** `GET https://dashscope-us.aliyuncs.com/api/v1/tasks/{task_id}` @@ -384,7 +381,9 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +## **日本（东京）** + +`GET https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -403,7 +402,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-text-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-text-to-video-api-reference.md index 385ff8f8..373f79e0 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-text-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-text-to-video-api-reference.md @@ -19,14 +19,14 @@ HappyHorse文生视频模型，输入文本提示词生成物理真实、运动 **重要** -百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： - 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` - 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -38,14 +38,10 @@ HappyHorse文生视频模型，输入文本提示词生成物理真实、运动 `POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **美国（弗吉尼亚）** `POST https://dashscope-us.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` @@ -54,7 +50,11 @@ HappyHorse文生视频模型，输入文本提示词生成物理真实、运动 `POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +## **日本（东京）** + +`POST https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -68,6 +68,7 @@ HappyHorse文生视频模型，输入文本提示词生成物理真实、运动 ## 文生视频 ``` +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ @@ -266,14 +267,10 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **美国（弗吉尼亚）** `GET https://dashscope-us.aliyuncs.com/api/v1/tasks/{task_id}` @@ -282,7 +279,9 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +## **日本（东京）** + +`GET https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -303,7 +302,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-video-edit-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-video-edit-api-reference.md index edfc94c8..58de1d49 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-video-edit-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-video-edit-api-reference.md @@ -19,14 +19,14 @@ HappyHorse 视频编辑模型支持输入视频与参考图，结合文本指令 **重要** -百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： - 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` - 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -38,14 +38,10 @@ HappyHorse 视频编辑模型支持输入视频与参考图，结合文本指令 `POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **美国（弗吉尼亚）** `POST https://dashscope-us.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` @@ -54,7 +50,11 @@ HappyHorse 视频编辑模型支持输入视频与参考图，结合文本指令 `POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +## **日本（东京）** + +`POST https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -68,6 +68,7 @@ HappyHorse 视频编辑模型支持输入视频与参考图，结合文本指令 ## 视频编辑（指令+参考图） ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ @@ -369,14 +370,10 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **美国（弗吉尼亚）** `GET https://dashscope-us.aliyuncs.com/api/v1/tasks/{task_id}` @@ -385,7 +382,9 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +## **日本（东京）** + +`GET https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -404,7 +403,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ @@ -561,7 +560,7 @@ task\_id查询有效期为 24 小时，超时后将无法查询，返回以下 **属性** -**duration** `_float_` +**duration** `_float_` 生成视频的总视频时长，用于计费。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/kling-api-reference/kling-video-generation-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/kling-api-reference/kling-video-generation-api-reference.md index 4a7b656c..f2b1777b 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/kling-api-reference/kling-video-generation-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/kling-api-reference/kling-video-generation-api-reference.md @@ -27,7 +27,9 @@ ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -43,7 +45,8 @@ 支持模型：`kling/kling-v3-omni-video-generation`、`kling/kling-v3-video-generation`。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -67,7 +70,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 支持模型：`kling/kling-v3-omni-video-generation`、`kling/kling-v3-video-generation` 。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -107,7 +111,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 支持模型：`kling/kling-v3-omni-video-generation`、`kling/kling-v3-video-generation` 。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -136,7 +141,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 支持模型：`kling/kling-v3-omni-video-generation`、`kling/kling-v3-video-generation` 。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -169,7 +175,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 支持模型：`kling/kling-v3-omni-video-generation`。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -640,7 +647,7 @@ audio直接影响费用，请前往[百炼控制台](https://bailian.console.ali ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -659,7 +666,7 @@ audio直接影响费用，请前往[百炼控制台](https://bailian.console.ali ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-image-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-image-to-video-api-reference.md index 7fea4312..7e98f28e 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-image-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-image-to-video-api-reference.md @@ -17,20 +17,26 @@ 为确保调用成功，请务必保证模型、endpoint URL 和 API Key 均属于**同一地域**。跨地域调用将会失败。 -- [**选择模型**](https://help.aliyun.com/zh/model-studio/use-video-generation#183f1b6fa0lox)：确认模型所属的地域。 +- [**选择模型**](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/all)：确认模型所属的地域。 - **选择 URL**：选择对应的地域 Endpoint URL，支持HTTP URL或 DashScope SDK URL。 - **配置 API Key**：获取该地域的[API Key](https://help.aliyun.com/zh/model-studio/get-api-key)，再[配置API Key到环境变量](https://help.aliyun.com/zh/model-studio/configure-api-key-through-environment-variables)。 +**说明** + +本文的示例代码适用于**华北2（北京）地域**。 + ## HTTP调用 图生视频任务耗时较长（通常为1-5分钟），API采用异步调用的方式。整个流程包含 **"创建任务 -> 轮询获取"** 两个核心步骤，具体如下： ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -46,7 +52,8 @@ 支持模型：pixverse/pixverse-c1-it2v、pixverse/pixverse-v6-it2v、pixverse/pixverse-v5.6-it2v。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -79,7 +86,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 在`prompt`中描述多镜头场景即可，不支持设置 `shot_type`参数。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -110,7 +118,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 在`prompt`中描述多镜头场景，并设置 `shot_type` 为`multi`， 即可生成有声多镜头视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -155,7 +164,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener **model** `_string_` **（必选）** -模型名称。模型输出规格请参见[模型列表](https://help.aliyun.com/zh/model-studio/use-video-generation#183f1b6fa0lox)。 +模型名称。 可选值： @@ -383,7 +392,7 @@ audio直接影响费用，请在调用前确认[模型价格](https://help.aliyu ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -402,7 +411,7 @@ audio直接影响费用，请在调用前确认[模型价格](https://help.aliyu ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-keyframe-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-keyframe-to-video-api-reference.md index 99e60eb8..557607a9 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-keyframe-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-keyframe-to-video-api-reference.md @@ -17,20 +17,26 @@ 为确保调用成功，请务必保证**模型、Endpoint URL 和 API Key 均属于同一地域**。跨地域调用将会失败。 -- [**选择模型**](https://help.aliyun.com/zh/model-studio/use-video-generation#f32f686472enw)：确认模型所属的地域。 +- [**选择模型**](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/all)：确认模型所属的地域。 - **选择 URL**：选择对应的地域 Endpoint URL，支持HTTP URL或 DashScope SDK URL。 - **配置 API Key**：选择地域并[获取API Key](https://help.aliyun.com/zh/model-studio/get-api-key)，再[配置API Key到环境变量](https://help.aliyun.com/zh/model-studio/configure-api-key-through-environment-variables)。 +**说明** + +本文的示例代码适用于**华北2（北京）地域**。 + ## HTTP调用 由于首尾帧生视频任务耗时较长（通常为1-5分钟），API采用异步调用。整个流程包含 **"创建任务 -> 轮询获取"** 两个核心步骤，具体如下： ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -46,7 +52,8 @@ 支持模型：pixverse/pixverse-c1-kf2v、pixverse/pixverse-v6-kf2v、pixverse/pixverse-v5.6-kf2v。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -95,7 +102,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener **model** `_string_` **（必选）** -模型名称。模型输出规格请参见[模型列表](https://help.aliyun.com/zh/model-studio/use-video-generation#f32f686472enw)。 +模型名称。 可选值： @@ -310,7 +317,7 @@ audio直接影响费用，请在调用前确认[模型价格](https://help.aliyu ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -329,7 +336,7 @@ audio直接影响费用，请在调用前确认[模型价格](https://help.aliyu ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-lipsync-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-lipsync-api-reference.md index 7a9a606c..61775d7e 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-lipsync-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-lipsync-api-reference.md @@ -34,7 +34,9 @@ ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -50,7 +52,8 @@ 传入视频和音频文件，生成对口型视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -79,7 +82,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 传入视频和 TTS 文本，由模型合成语音并生成对口型视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -273,7 +277,7 @@ auto ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-motioncontrol-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-motioncontrol-api-reference.md index f6766e71..44892823 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-motioncontrol-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-motioncontrol-api-reference.md @@ -34,7 +34,9 @@ ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -50,7 +52,8 @@ 传入角色图片和动作视频，生成角色模仿动作的视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -162,7 +165,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-reference-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-reference-to-video-api-reference.md index c90d87c5..1e14eac3 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-reference-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-reference-to-video-api-reference.md @@ -30,7 +30,9 @@ ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -46,7 +48,8 @@ 通过 `media` 传入参考图像，设置 `size` 和 `duration` 控制视频分辨率和时长。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -83,7 +86,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 在 `media` 传入参考图像，并传入`ref_name`用于设置参考图像的主体名称。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -759,7 +763,7 @@ audio直接影响费用，请在调用前确认[爱诗-参考生视频](https:// ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -778,7 +782,7 @@ audio直接影响费用，请在调用前确认[爱诗-参考生视频](https:// ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-text-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-text-to-video-api-reference.md index 4775d1a3..320e8241 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-text-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-text-to-video-api-reference.md @@ -17,7 +17,7 @@ 为确保调用成功，请务必保证**模型、Endpoint URL 和 API Key 均属于同一地域**。跨地域调用将会失败。 -- [**选择模型**](https://help.aliyun.com/zh/model-studio/use-video-generation#3ad2d09509ldb)：确认模型所属的地域。 +- [**选择模型**](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/all)：确认模型所属的地域。 - **选择 URL**：选择对应的地域 Endpoint URL，支持HTTP URL或 DashScope SDK URL。 @@ -26,13 +26,19 @@ - **安装 SDK**：如需通过SDK进行调用，请[安装DashScope SDK](https://help.aliyun.com/zh/model-studio/install-sdk)。 +**说明** + +本文的示例代码适用于**华北2（北京）地域**。 + ## HTTP调用 由于文生视频任务耗时较长（通常为1-5分钟），API采用异步调用。整个流程包含 **"创建任务 -> 轮询获取"** 两个核心步骤，具体如下： ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -48,7 +54,8 @@ 支持模型：pixverse/pixverse-c1-t2v、pixverse/pixverse-v6-t2v、pixverse/pixverse-v5.6-t2v。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -74,7 +81,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 在`prompt`中描述多镜头场景即可，不支持设置 `shot_type`参数。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -99,7 +107,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 在`prompt`中描述多镜头场景，并设置 `shot_type` 为`multi`， 即可生成有声多镜头视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -140,7 +149,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener **model** `_string_` **（必选）** -模型名称。模型输出规格请参见[模型列表](https://help.aliyun.com/zh/model-studio/use-video-generation#3ad2d09509ldb)。 +模型名称。 可选值： @@ -574,7 +583,7 @@ audio直接影响费用，请在调用前确认[模型价格](https://help.aliyu ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -593,7 +602,7 @@ audio直接影响费用，请在调用前确认[模型价格](https://help.aliyu ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ @@ -798,7 +807,9 @@ SDK 的参数命名与[HTTP接口](#pv101a0h2http)基本一致，参数结构根 若版本过低，可能会触发 "url error, please check url!" 等错误。请参考[安装SDK](https://help.aliyun.com/zh/model-studio/install-sdk)进行更新。 -**北京地域**：`dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1'` +**北京地域**：`dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1'` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 @@ -811,7 +822,7 @@ import dashscope import os # 以下为北京地域URL -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -881,7 +892,7 @@ import dashscope import os # 以下为北京地域URL -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -984,13 +995,16 @@ if __name__ == '__main__': 若版本过低，可能会触发 “url error, please check url!” 等错误。请参考[安装SDK](https://help.aliyun.com/zh/model-studio/install-sdk)进行更新。 -**北京地域**：`Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1";` +**北京地域**：`Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1";` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 ##### 请求示例 ``` +// 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 // Copyright (c) Alibaba, Inc. and its affiliates. import com.alibaba.dashscope.aigc.videosynthesis.VideoSynthesis; @@ -1006,7 +1020,7 @@ public class Text2Video { static { // 以下为北京地域url - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" @@ -1080,6 +1094,7 @@ public class Text2Video { ##### 请求示例 ``` +// 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 // Copyright (c) Alibaba, Inc. and its affiliates. import com.alibaba.dashscope.aigc.videosynthesis.VideoSynthesis; @@ -1097,7 +1112,7 @@ public class Text2Video { static { // 以下为北京地域url - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-upscale-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-upscale-api-reference.md index 43f148d0..ac899560 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-upscale-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-upscale-api-reference.md @@ -34,7 +34,9 @@ ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -50,7 +52,8 @@ 传入视频文件，输出超分辨率视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 调用时请将 {WorkspaceId} 替换为真实的业务空间ID。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -159,7 +162,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-image-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-image-to-video-api-reference.md index c12815fa..fcb9c26c 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-image-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-image-to-video-api-reference.md @@ -27,7 +27,9 @@ Vidu-图生视频模型根据**输入图像**和**文本提示词**，生成一 ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -40,8 +42,10 @@ Vidu-图生视频模型根据**输入图像**和**文本提示词**，生成一 ## 图生视频 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -299,7 +303,7 @@ duration直接影响费用，按秒计费，时间越长费用越高，请在调 ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -318,7 +322,7 @@ duration直接影响费用，按秒计费，时间越长费用越高，请在调 ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-keyframe-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-keyframe-to-video-api-reference.md index 5547a29f..425a4bc4 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-keyframe-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-keyframe-to-video-api-reference.md @@ -27,7 +27,9 @@ Vidu-首尾帧生视频模型基于**首帧图像**、**尾帧图像和文本提 ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -40,8 +42,10 @@ Vidu-首尾帧生视频模型基于**首帧图像**、**尾帧图像和文本提 ## 首尾帧生视频 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -292,7 +296,7 @@ duration直接影响费用，按秒计费，时间越长费用越高，请在调 ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -311,7 +315,7 @@ duration直接影响费用，按秒计费，时间越长费用越高，请在调 ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-reference-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-reference-to-video-api-reference.md index 833a58c2..84700aa6 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-reference-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-reference-to-video-api-reference.md @@ -27,7 +27,9 @@ Vidu-参考生视频模型支持传入**参考图片**和**文本提示词**， ### **步骤 1：创建任务获取任务 ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -42,8 +44,10 @@ Vidu-参考生视频模型支持传入**参考图片**和**文本提示词**， 支持模型：vidu/viduq3-ad\_reference2video。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -79,8 +83,10 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 支持模型：vidu/viduq3-drama\_reference2video。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -118,8 +124,10 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## 参考生视频（仅参考图像） +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -155,8 +163,10 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 支持模型：vidu/viduq2-pro\_reference2video。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -595,7 +605,7 @@ vidu/viduq3-drama\_reference2video 不支持该参数，默认输出有声视频 ### **步骤 2：根据任务 ID 查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -614,7 +624,7 @@ vidu/viduq3-drama\_reference2video 不支持该参数，默认输出有声视频 ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-text-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-text-to-video-api-reference.md index bde2d5df..1254f4e7 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-text-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-text-to-video-api-reference.md @@ -29,7 +29,9 @@ Vidu-文生视频模型基于**文本提示词**，生成一段流畅的视频 ### **步骤1：创建任务获取任务ID** -**北京地域**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +**北京地域**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -42,8 +44,10 @@ Vidu-文生视频模型基于**文本提示词**，生成一段流畅的视频 ## 文生视频 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -322,7 +326,7 @@ duration直接影响费用，按秒计费，时间越长费用越高，请在调 ### **步骤2：根据任务ID查询结果** -**北京地域**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +**北京地域**：`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` **说明** @@ -341,7 +345,7 @@ duration直接影响费用，按秒计费，时间越长费用越高，请在调 ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ @@ -546,7 +550,7 @@ SDK 的参数命名与[HTTP接口](#vd101a0h2http)基本一致，参数结构根 请确保 DashScope Python SDK 版本**不低于** `**1.25.8**`，再运行以下代码。 -**北京地域**：`dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1'` +**北京地域**：`dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1'` ## 同步调用 @@ -558,8 +562,8 @@ from dashscope import VideoSynthesis import dashscope import os -# 以下为北京地域URL -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为北京地域URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" api_key = os.getenv("DASHSCOPE_API_KEY") @@ -628,8 +632,8 @@ from dashscope import VideoSynthesis import dashscope import os -# 以下为北京地域URL -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为北京地域URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" api_key = os.getenv("DASHSCOPE_API_KEY") @@ -728,7 +732,7 @@ if __name__ == '__main__': 请确保 DashScope Java SDK 版本**不低于** `**2.22.6**`，再运行以下代码。 -**北京地域**：`Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1";` +**北京地域**：`Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1";` ## 同步调用 @@ -749,8 +753,8 @@ import com.alibaba.dashscope.utils.Constants; public class Text2Video { static { - // 以下为北京地域url - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + // 以下为北京地域URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" @@ -836,8 +840,8 @@ import com.alibaba.dashscope.utils.Constants; public class Text2Video { static { - // 以下为北京地域url - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + // 以下为北京地域URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/image-to-video-general-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/image-to-video-general-api-reference.md index 08695f0a..083f04cc 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/image-to-video-general-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/image-to-video-general-api-reference.md @@ -29,9 +29,14 @@ **重要** -百炼为新加坡地域推出了业务空间专属域名 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`，**能够为推理请求提供卓越的性能和更高的稳定性**，建议从 `https://dashscope-intl.aliyuncs.com` 迁移至新域名。 +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +- 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -45,13 +50,13 @@ ## **北京** -`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -67,7 +72,8 @@ 基于首帧图像和音频生成视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -101,7 +107,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 传入首帧和尾帧生成视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -134,7 +141,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 基于首段视频片段，让模型生成后续内容。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -526,13 +534,13 @@ duration直接影响费用，按秒计费，请在调用前确认[模型价格]( ## **北京** -`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -553,10 +561,10 @@ duration直接影响费用，按秒计费，请在调用前确认[模型价格]( ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` -curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id} \ +curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" ``` @@ -754,13 +762,13 @@ SDK 的参数命名与[HTTP接口](#9c71bffa84zm6)基本一致，参数结构根 ## **北京** -`dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1'` +`dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1'` ## **新加坡** `dashscope.base_http_api_url = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1'` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 @@ -773,8 +781,8 @@ from dashscope import VideoSynthesis import dashscope import os -# 以下为北京地域URL，各地域的URL不同，获取URL：https://help.aliyun.com/zh/model-studio/image-to-video-general-api-reference -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -854,8 +862,8 @@ from dashscope import VideoSynthesis import dashscope import os -# 以下为北京地域URL，各地域的URL不同，获取URL：https://help.aliyun.com/zh/model-studio/image-to-video-general-api-reference -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -973,13 +981,13 @@ if __name__ == '__main__': ## **北京** -`Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"` +`Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"` ## **新加坡** `Constants.baseHttpApiUrl = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1"` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 @@ -1005,29 +1013,12 @@ import java.util.List; public class Image2Video { static { - // 以下为华北2（北京）地域的URL，各地域的URL不同。 - -import com.alibaba.dashscope.aigc.videosynthesis.VideoSynthesis; -import com.alibaba.dashscope.aigc.videosynthesis.VideoSynthesisParam; -import com.alibaba.dashscope.aigc.videosynthesis.VideoSynthesisResult; -import com.alibaba.dashscope.exception.ApiException; -import com.alibaba.dashscope.exception.InputRequiredException; -import com.alibaba.dashscope.exception.NoApiKeyException; -import com.alibaba.dashscope.utils.Constants; -import com.alibaba.dashscope.utils.JsonUtils; - -import java.util.ArrayList; -import java.util.List; - -public class Image2Video { - - static { - // 以下为新加坡地域URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 - Constants.baseHttpApiUrl = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1"; + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：apiKey="sk-xxx" - // 各地域的API Key不同。获取API Key：https://www.alibabacloud.com/help/en/model-studio/get-api-key + // 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key static String apiKey = System.getenv("DASHSCOPE_API_KEY"); public static void syncCall() { @@ -1093,29 +1084,12 @@ import java.util.List; public class Image2Video { static { - // 以下为华北2（北京）地域的URL，各地域的URL不同。 - -import com.alibaba.dashscope.aigc.videosynthesis.VideoSynthesis; -import com.alibaba.dashscope.aigc.videosynthesis.VideoSynthesisParam; -import com.alibaba.dashscope.aigc.videosynthesis.VideoSynthesisResult; -import com.alibaba.dashscope.exception.ApiException; -import com.alibaba.dashscope.exception.InputRequiredException; -import com.alibaba.dashscope.exception.NoApiKeyException; -import com.alibaba.dashscope.utils.Constants; -import com.alibaba.dashscope.utils.JsonUtils; - -import java.util.ArrayList; -import java.util.List; - -public class Image2Video { - - static { - // 以下为新加坡地域URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 - Constants.baseHttpApiUrl = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1"; + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：apiKey="sk-xxx" - // 各地域的API Key不同。获取API Key：https://www.alibabacloud.com/help/en/model-studio/get-api-key + // 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key static String apiKey = System.getenv("DASHSCOPE_API_KEY"); public static void asyncCall() { diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-api-reference.md index bf2dd8df..352341c9 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-api-reference.md @@ -29,9 +29,14 @@ **重要** -百炼为新加坡地域推出了业务空间专属域名 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`，**能够为推理请求提供卓越的性能和更高的稳定性**，建议从 `https://dashscope-intl.aliyuncs.com` 迁移至新域名。 +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +- 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -41,14 +46,12 @@ ## **北京** -`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **弗吉尼亚** `POST https://dashscope-us.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` @@ -57,7 +60,7 @@ `POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -75,7 +78,9 @@ 可通过设置`"prompt_extend": true`和`"shot_type":"multi"`启用。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -102,7 +107,9 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 若不提供 `input.audio_url` ，模型将根据视频内容自动生成匹配的背景音乐或音效。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -127,7 +134,9 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 如需为视频指定背景音乐或配音，可通过 `input.audio_url` 参数传入自定义音频的 URL。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -156,7 +165,9 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -182,7 +193,9 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 示例：下载[img\_base64](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20250722/pmcjis/img_base64.txt)文件，并将完整内容粘贴至`img_url`参数中。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -203,11 +216,13 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener - prompt 字段将被忽略，建议留空。 -- 特效的可用性与模型相关。调用前请查阅[万相-图生视频-视频特效](https://help.aliyun.com/zh/model-studio/wanx-video-effects)，以免调用失败。 +- 特效的可用性与模型相关。调用前请查阅[万相-图生视频-视频特效列表](https://help.aliyun.com/zh/model-studio/wanx-video-effects)，以免调用失败。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -228,7 +243,9 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 通过 negative\_prompt 指定生成的视频避免出现“花朵”元素。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -322,7 +339,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 支持输入的格式： -1. 公网URL: +1. 公网URL： - 支持 HTTP 或 HTTPS 协议。 @@ -381,7 +398,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 视频特效模板的名称。若未填写，表示不使用任何视频特效。 -不同模型支持不同的特效模板。调用前请查阅[万相-图生视频-视频特效](https://help.aliyun.com/zh/model-studio/wanx-video-effects)，以免调用失败。 +不同模型支持不同的特效模板。调用前请查阅[万相-图生视频-视频特效列表](https://help.aliyun.com/zh/model-studio/wanx-video-effects)，以免调用失败。 示例值：flying，表示使用“魔法悬浮”特效。 @@ -596,14 +613,12 @@ audio直接影响费用，有声视频与无声视频价格不同，请前往百 ## **北京** -`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **弗吉尼亚** `GET https://dashscope-us.aliyuncs.com/api/v1/tasks/{task_id}` @@ -612,7 +627,7 @@ audio直接影响费用，有声视频与无声视频价格不同，请前往百 `GET https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -633,10 +648,10 @@ audio直接影响费用，有声视频与无声视频价格不同，请前往百 ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` -curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id} \ +curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" ``` @@ -881,18 +896,16 @@ SDK 的参数命名与[HTTP接口](#42703589880ts)基本一致，参数结构根 若版本过低，可能会触发 “url error, please check url!” 等错误。请参考[安装SDK](https://help.aliyun.com/zh/model-studio/install-sdk)进行更新。 -根据模型所在地域设置 `**base_http_api_url**`: +根据模型所在地域设置 `**base_http_api_url**`： ## **北京** -`dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1'` +`dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1'` ## **新加坡** `dashscope.base_http_api_url = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1'` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **弗吉尼亚** `dashscope.base_http_api_url = 'https://dashscope-us.aliyuncs.com/api/v1'` @@ -901,7 +914,7 @@ SDK 的参数命名与[HTTP接口](#42703589880ts)基本一致，参数结构根 `dashscope.base_http_api_url = 'https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1'` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 #### **示例代码** @@ -919,8 +932,9 @@ from dashscope import VideoSynthesis import mimetypes import dashscope -# 以下为北京地域url，获取url：https://help.aliyun.com/zh/model-studio/image-to-video-api-reference -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +# 获取URL：https://help.aliyun.com/zh/model-studio/image-to-video-api-reference +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -1032,8 +1046,9 @@ from http import HTTPStatus from dashscope import VideoSynthesis import dashscope -# 以下为北京地域url，获取url：https://help.aliyun.com/zh/model-studio/image-to-video-api-reference -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +# 获取URL：https://help.aliyun.com/zh/model-studio/image-to-video-api-reference +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -1145,18 +1160,16 @@ if __name__ == '__main__': 若版本过低，可能会触发 “url error, please check url!” 等错误。请参考[安装SDK](https://help.aliyun.com/zh/model-studio/install-sdk)进行更新。 -根据模型所在地域设置 `**baseHttpApiUrl**`: +根据模型所在地域设置 `**baseHttpApiUrl**`： ## **北京** -`Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1";` +`Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1";` ## **新加坡** `Constants.baseHttpApiUrl = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1";` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## **弗吉尼亚** `Constants.baseHttpApiUrl = "https://dashscope-us.aliyuncs.com/api/v1";` @@ -1165,7 +1178,7 @@ if __name__ == '__main__': `Constants.baseHttpApiUrl = "https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1";` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 #### **示例代码** @@ -1198,8 +1211,9 @@ import java.util.Map; public class Image2Video { static { - // 以下为北京地域url，获取url：https://help.aliyun.com/zh/model-studio/image-to-video-api-reference - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + // 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + // 获取URL：https://help.aliyun.com/zh/model-studio/image-to-video-api-reference + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：apiKey="sk-xxx" @@ -1351,8 +1365,9 @@ import java.util.Map; public class Image2Video { static { - // 以下为北京地域url，获取url：https://help.aliyun.com/zh/model-studio/image-to-video-api-reference - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + // 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + // 获取URL：https://help.aliyun.com/zh/model-studio/image-to-video-api-reference + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：apiKey="sk-xxx" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-by-first-and-last-frame-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-by-first-and-last-frame-api-reference.md index 7dc1d88e..dd4f30e4 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-by-first-and-last-frame-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-by-first-and-last-frame-api-reference.md @@ -23,9 +23,14 @@ **重要** -百炼为新加坡地域推出了业务空间专属域名 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`，**能够为推理请求提供卓越的性能和更高的稳定性**，建议从 `https://dashscope-intl.aliyuncs.com` 迁移至新域名。 +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +- 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -35,13 +40,15 @@ ## **北京** -`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis` +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -57,7 +64,8 @@ 根据首帧、尾帧和prompt生成视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -82,7 +90,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video 格式参见[如何输入图像](https://help.aliyun.com/zh/model-studio/image-to-video-first-and-last-frames-guide#69308edfbebnx)。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -104,10 +113,11 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video 必须传入`first_frame_url`和`template`，无需传入prompt和last\_frame\_url。 -不同模型支持不同的特效模板。调用前请查阅[万相-图生视频-视频特效](https://help.aliyun.com/zh/model-studio/wanx-video-effects)，以免调用失败。 +不同模型支持不同的特效模板。调用前请查阅[万相-图生视频-视频特效列表](https://help.aliyun.com/zh/model-studio/wanx-video-effects)，以免调用失败。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -129,7 +139,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video 通过 negative\_prompt 指定生成的视频避免出现“人物”元素。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -274,7 +285,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/image2video 视频特效模板的名称。使用此参数时，仅需传入 `first_frame_url`。 -不同模型支持不同的特效模板。调用前请查阅[万相-图生视频-视频特效](https://help.aliyun.com/zh/model-studio/wanx-video-effects)，以免调用失败。 +不同模型支持不同的特效模板。调用前请查阅[万相-图生视频-视频特效列表](https://help.aliyun.com/zh/model-studio/wanx-video-effects)，以免调用失败。 示例值：hufu-1，表示使用“唐韵翩然”特效。 @@ -412,13 +423,15 @@ duration直接影响费用，按秒计费，调用前请确认百炼控制台。 ## **北京** -`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -441,10 +454,10 @@ duration直接影响费用，按秒计费，调用前请确认百炼控制台。 请将`86ecf553-d340-4e21-xxxxxxxxx`替换为真实的task\_id。 -> 若使用新加坡地域的模型，需将base\_url替换为https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx，其中WorkspaceId需替换为真实的业务空间ID。 +> 若使用新加坡地域的模型，需将base\_url替换为https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx，其中{WorkspaceId}需替换为真实的业务空间ID。 ``` -curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx \ +curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/86ecf553-d340-4e21-xxxxxxxxx \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" ``` @@ -641,13 +654,15 @@ SDK 的参数命名与[HTTP接口](https://help.aliyun.com/zh/model-studio/text- ## **北京** -`dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1'` +`dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1'` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `dashscope.base_http_api_url = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1'` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 #### **示例代码** @@ -666,7 +681,7 @@ import mimetypes import dashscope # 以下为北京地域URL，各地域的URL不同，获取URL：https://help.aliyun.com/zh/model-studio/text-to-video-api-reference -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -768,7 +783,7 @@ from http import HTTPStatus from dashscope import VideoSynthesis import dashscope -# 以下为华北2（北京）地域的URL，各地域的URL不同。 +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 ``` ### Java SDK调用 @@ -816,7 +831,7 @@ import java.util.Map; public class Kf2vSync { static { - // 以下为华北2（北京）地域的URL，各地域的URL不同。 + // 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 ``` ## 异步调用 @@ -851,7 +866,7 @@ import java.util.Map; public class Kf2vAsync { static { - // 以下为华北2（北京）地域的URL，各地域的URL不同。 + // 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 ``` ## **使用限制** diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wan-reference-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wan-reference-to-video-api-reference.md index 8477b7b4..b5562bbf 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wan-reference-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wan-reference-to-video-api-reference.md @@ -21,9 +21,14 @@ **重要** -百炼为新加坡地域推出了业务空间专属域名 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`，**能够为推理请求提供卓越的性能和更高的稳定性**，建议从 `https://dashscope-intl.aliyuncs.com` 迁移至新域名。 +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +- 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -37,13 +42,15 @@ ## **北京** -`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **弗吉尼亚** @@ -53,7 +60,7 @@ `POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -69,7 +76,8 @@ 通过`reference_urls`传入图像和视频URL。同时设置`shot_type`为`multi`，生成多镜头视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -99,7 +107,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 通过`reference_urls`传入多个视频URL。同时设置`shot_type`为`multi`，生成多镜头视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -125,7 +134,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 通过`reference_urls`传入单个视频URL。同时设置`shot_type`为`multi`，生成多镜头视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -150,7 +160,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 当生成无声视频时，**必须显式设置** `parameters.audio = false`。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -505,13 +516,15 @@ audio直接影响费用，有声视频与无声视频价格不同，请在调用 ## **北京** -`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **弗吉尼亚** @@ -521,7 +534,7 @@ audio直接影响费用，有声视频与无声视频价格不同，请在调用 `GET https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -542,10 +555,10 @@ audio直接影响费用，有声视频与无声视频价格不同，请在调用 ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` -curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id} \ +curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" ``` @@ -735,13 +748,15 @@ SDK 的参数命名与[HTTP接口](#42703589880ts)基本一致，参数结构根 ## **北京** -`dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1'` +`dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1'` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `dashscope.base_http_api_url = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1'` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 美国 @@ -751,7 +766,7 @@ SDK 的参数命名与[HTTP接口](#42703589880ts)基本一致，参数结构根 `dashscope.base_http_api_url = 'https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1'` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 @@ -766,7 +781,7 @@ import dashscope import os # 以下为北京地域URL，各地域的URL不同 -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -814,7 +829,7 @@ from dashscope import VideoSynthesis import dashscope # 以下为北京地域URL，各地域的URL不同 -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -877,19 +892,21 @@ if __name__ == '__main__': ## **北京** -`Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1";` +`Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1";` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `Constants.baseHttpApiUrl = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1";` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **法兰克福** `Constants.baseHttpApiUrl = "https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1";` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 @@ -916,7 +933,7 @@ public class Ref2Video26 { static { // 以下为北京地域url，各地域的url不同 - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：apiKey="sk-xxx" @@ -986,7 +1003,7 @@ public class Ref2Video26Async { static { // 以下为北京地域url，各地域的url不同 - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：apiKey="sk-xxx" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wan-text-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wan-text-to-video-api-reference.md index 03862462..b5ad73b5 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wan-text-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wan-text-to-video-api-reference.md @@ -23,9 +23,14 @@ **重要** -百炼为新加坡地域推出了业务空间专属域名 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`，**能够为推理请求提供卓越的性能和更高的稳定性**，建议从 `https://dashscope-intl.aliyuncs.com` 迁移至新域名。 +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +- 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -39,13 +44,15 @@ ## **北京** -`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **弗吉尼亚** @@ -55,7 +62,7 @@ `POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -74,7 +81,8 @@ ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -100,7 +108,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 可通过 `input.audio_url` 参数传入自定义音频的 URL。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -125,7 +134,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 若不提供 `input.audio_url` ，模型将根据视频内容自动生成匹配的背景音乐或音效。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -149,7 +159,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener > wan2.6 及wan2.5系列模型默认生成有声视频。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -172,7 +183,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener 通过 negative\_prompt 排除“花朵”元素，避免其出现在视频画面中。 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -477,13 +489,15 @@ duration直接影响费用。费用 = 单价（基于分辨率）× 时长（秒 ## **北京** -`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **弗吉尼亚** @@ -493,7 +507,7 @@ duration直接影响费用。费用 = 单价（基于分辨率）× 时长（秒 `GET https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -514,10 +528,10 @@ duration直接影响费用。费用 = 单价（基于分辨率）× 时长（秒 ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` -curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id} \ +curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" ``` @@ -743,13 +757,15 @@ SDK 的参数命名与HTTP接口基本一致，参数结构根据语言特性进 ## **北京** -`dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1'` +`dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1'` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `dashscope.base_http_api_url = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1'` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **弗吉尼亚** @@ -759,7 +775,7 @@ SDK 的参数命名与HTTP接口基本一致，参数结构根据语言特性进 `dashscope.base_http_api_url = 'https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1'` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 @@ -771,8 +787,8 @@ from dashscope import VideoSynthesis import dashscope import os -# 以下为北京地域URL，各地域的URL不同，获取URL：https://help.aliyun.com/zh/model-studio/text-to-video-api-reference -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。获取URL：https://help.aliyun.com/zh/model-studio/text-to-video-api-reference +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -844,8 +860,8 @@ from dashscope import VideoSynthesis import dashscope import os -# 以下为北京地域URL，各地域的URL不同，获取URL：https://help.aliyun.com/zh/model-studio/text-to-video-api-reference -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。获取URL：https://help.aliyun.com/zh/model-studio/text-to-video-api-reference +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -955,13 +971,15 @@ if __name__ == '__main__': ## **北京** -`Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1";` +`Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1";` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `Constants.baseHttpApiUrl = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1";` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **弗吉尼亚** @@ -971,7 +989,7 @@ if __name__ == '__main__': `Constants.baseHttpApiUrl = "https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1";` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 @@ -994,8 +1012,8 @@ import java.util.Map; public class Text2Video { static { - // 以下为北京地域url，各地域的url不同，获取url：https://help.aliyun.com/zh/model-studio/text-to-video-api-reference - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + // 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。获取URL：https://help.aliyun.com/zh/model-studio/text-to-video-api-reference + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" @@ -1093,8 +1111,8 @@ public class Text2Video { static { - // 以下为北京地域url，各地域的url不同，获取url：https://help.aliyun.com/zh/model-studio/text-to-video-api-reference - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + // 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。获取URL：https://help.aliyun.com/zh/model-studio/text-to-video-api-reference + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wanx-vace-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wanx-vace-api-reference.md index 0fea1012..eb4e31d9 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wanx-vace-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wanx-vace-api-reference.md @@ -21,9 +21,14 @@ **重要** -百炼为新加坡地域推出了业务空间专属域名 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`，**能够为推理请求提供卓越的性能和更高的稳定性**，建议从 `https://dashscope-intl.aliyuncs.com` 迁移至新域名。 +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +- 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -33,20 +38,23 @@ ## **北京** -`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 #### 请求参数 ## **多图参考** ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ --header 'X-DashScope-Async: enable' \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ --header 'Content-Type: application/json' \ @@ -71,7 +79,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## 视频重绘 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ --header 'X-DashScope-Async: enable' \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ --header 'Content-Type: application/json' \ @@ -92,8 +101,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## **局部编辑** ``` -# 如果使用华北2（北京）地域的模型，需要将url替换为：https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ --header 'X-DashScope-Async: enable' \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ --header 'Content-Type: application/json' \ @@ -117,7 +126,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## 视频延展 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ --header 'X-DashScope-Async: enable' \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ --header 'Content-Type: application/json' \ @@ -137,7 +147,8 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## 视频画面扩展 ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ --header 'X-DashScope-Async: enable' \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ --header 'Content-Type: application/json' \ @@ -1168,22 +1179,24 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## **北京** -`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` + +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 #### 请求参数 ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` -curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id} \ +curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" ``` diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/text-to-video-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/text-to-video-api-reference.md index 9fa72fb6..4c0b78dd 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/text-to-video-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/text-to-video-api-reference.md @@ -465,7 +465,7 @@ duration直接影响费用，按秒计费，请在调用前确认[模型价格]( ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-mix-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-mix-api.md index a7b65d67..e6707b3c 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-mix-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-mix-api.md @@ -53,6 +53,8 @@ **新加坡地域**：`POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis` +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 + **说明** - 创建成功后，使用接口返回的 `task_id` 查询结果，task\_id 有效期为 24 小时。**请勿重复创建任务**，轮询获取即可。 @@ -64,7 +66,7 @@ ## 视频换人 -以下为华北2（北京）地域的URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 +以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 ``` curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/image2video/video-synthesis' \ @@ -256,13 +258,13 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 #### 新加坡 `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -285,7 +287,7 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi 请将`0385dc79-5ff8-4d82-bcb6-xxxxxx`替换为真实的task\_id。 -> 以下为华北2（北京）地域的URL，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的URL不同。 +> 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 ``` curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/0385dc79-5ff8-4d82-bcb6-xxxxxx \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-move-api.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-move-api.md index dbdb443a..3808f5dd 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-move-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-move-api.md @@ -253,13 +253,13 @@ curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/servi `GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 #### 新加坡 `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** diff --git a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-video-editing-api-reference.md b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-video-editing-api-reference.md index 24b5e582..d6728f38 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-video-editing-api-reference.md +++ b/skills/bailian-docs-llm-wiki/raw/model-api-reference/video-generation-api/wan-api-reference/wan-video-editing-api-reference.md @@ -21,9 +21,14 @@ **重要** -百炼为新加坡地域推出了业务空间专属域名 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`，**能够为推理请求提供卓越的性能和更高的稳定性**，建议从 `https://dashscope-intl.aliyuncs.com` 迁移至新域名。 +阿里云百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，**能够为推理请求提供卓越的性能和更高的稳定性**，建议迁移至新域名： -其中 `{WorkspaceId}` 为您的业务空间 ID，可在百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 +- 华北2（北京）地域：从 `https://dashscope.aliyuncs.com` 迁移至 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` + +- 新加坡地域：从 `https://dashscope-intl.aliyuncs.com` 迁移至 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + + +其中 `{WorkspaceId}` 为您的业务空间 ID，可在阿里云百炼控制台的**业务空间详情**页面查看。现有域名仍可正常使用。 ## HTTP调用 @@ -33,13 +38,13 @@ ## **北京** -`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` +`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` ## **新加坡** `POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -53,7 +58,7 @@ ## 纯指令编辑（修改视频风格） ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -79,7 +84,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## **指令+参考图编辑（局部替换）** ``` -curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis' \ -H 'X-DashScope-Async: enable' \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -499,13 +504,13 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## **北京** -`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +`GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id}` ## **新加坡** `GET https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/tasks/{task_id}` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 **说明** @@ -526,10 +531,10 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/services/aigc/video-gener ## 查询任务结果 -将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时。 +将`{task_id}`完整替换为上一步接口返回的`task_id`的值。`task_id`查询有效期为24小时，并请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ``` -curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id} \ +curl -X GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" ``` @@ -729,13 +734,13 @@ SDK 的参数命名与[HTTP接口](#e9e21dd3a6945)基本一致，参数结构根 ## **北京** -`dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1'` +`dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1'` ## **新加坡** `dashscope.base_http_api_url = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1'` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 @@ -749,8 +754,8 @@ from dashscope import VideoSynthesis import dashscope import os -# 以下为北京地域URL，各地域的URL不同 -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -855,8 +860,8 @@ from dashscope import VideoSynthesis import dashscope import os -# 以下为北京地域URL，各地域的URL不同 -dashscope.base_http_api_url = 'https://dashscope.aliyuncs.com/api/v1' +# 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx" # 各地域的API Key不同。获取API Key：https://help.aliyun.com/zh/model-studio/get-api-key @@ -997,13 +1002,13 @@ if __name__ == '__main__': ## **北京** -`Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1";` +`Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1";` ## **新加坡** `Constants.baseHttpApiUrl = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1";` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## 同步调用 @@ -1034,8 +1039,8 @@ import java.util.Base64; public class VideoEdit { static { - // 以下为北京地域url，各地域的url不同 - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + // 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：apiKey="sk-xxx" @@ -1184,8 +1189,8 @@ import java.util.Base64; public class VideoEdit { static { - // 以下为北京地域url，各地域的url不同 - Constants.baseHttpApiUrl = "https://dashscope.aliyuncs.com/api/v1"; + // 以下为华北2（北京）地域的URL，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的URL不同。 + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; } // 若没有配置环境变量，请用百炼API Key将下行替换为：apiKey="sk-xxx" diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/fine-tuning-api-guide.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/fine-tuning-api-guide.md index d4752f4b..1114b8b8 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/fine-tuning-api-guide.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/fine-tuning-api-guide.md @@ -565,6 +565,10 @@ Body ###### 文本生成 +**说明** + +Qwen3.7-Plus-2026-05-26 调优后部署请联系商务经理。 + **模型服务** **模型代码** @@ -1602,6 +1606,10 @@ FAILED 导出的参数快照保存在云存储中，暂不支持访问或下载。 +**说明** + +URL 中的 `` 取值来自上一步查询快照列表接口返回的 `checkpoint` 字段（如 `checkpoint-20`），而非 `full_name` 或其他含 job\_id 前缀的字段值。 + ``` curl --request GET 'https://dashscope.aliyuncs.com/api/v1/fine-tunes//export/?model_name=' \ --header 'Authorization: Bearer '${DASHSCOPE_API_KEY} \ @@ -1638,7 +1646,7 @@ Path Parameter 是 -要发布的 Checkpoint ID。 +要发布的 Checkpoint ID。取值为查询快照列表接口返回的 `checkpoint` 字段值（如 `checkpoint-20`），不含 job\_id 前缀。 model\_name diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-on-console.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-on-console.md index 732930c9..268db778 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-on-console.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-on-console.md @@ -4,7 +4,7 @@ ## **模型调优流程** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/8010204871/CAEQZhiBgMDg9PGS2hkiIDNlZDFiMGRlMTJhOTQ1YzJhMmNjNDM3NzQ1ZjNiOGZk4608430_20240830103738.564.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3743415871/CAEQZhiBgMDg9PGS2hkiIDNlZDFiMGRlMTJhOTQ1YzJhMmNjNDM3NzQ1ZjNiOGZk4608430_20240830103738.564.svg) ## **步骤一：选择调优方式** @@ -776,3 +776,21 @@ Checkpoint 有保存时长限制，超过保存时长后将被自动清理，届 **说明** 您可以在模型调优任务列表中，点击训练失败任务右侧的**日志**，查看具体的训练失败原因。 + +### **微调后模型体验仍回答基座模型身份怎么办？** + +**原因说明** + +控制台**模型体验**页面不支持设置 system prompt，导致微调后的身份设定无法生效。 + +**解决方案** + +可以通过以下方式验证微调模型的身份设定效果： + +- **方案一：通过 API 调用传入 system prompt** + + 调用 API 时，在请求的 `system` 字段中填写身份设定内容，即可让模型按照微调后的设定进行回答。 + +- **方案二：在百炼控制台模型调试页面测试** + + 前往百炼控制台**模型调试**页面，在系统提示词输入框中填写身份设定内容，再进行测试。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md index ae2396d5..4a8c5430 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md @@ -25,7 +25,7 @@ ### **模型调优流程** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/2700204871/CAEQZhiBgMDg9PGS2hkiIDNlZDFiMGRlMTJhOTQ1YzJhMmNjNDM3NzQ1ZjNiOGZk4608430_20240830103738.564.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3753415871/CAEQZhiBgMDg9PGS2hkiIDNlZDFiMGRlMTJhOTQ1YzJhMmNjNDM3NzQ1ZjNiOGZk4608430_20240830103738.564.svg) 详情参见： @@ -35,6 +35,10 @@ #### 文本生成 +**说明** + +Qwen3.7-Plus-2026-05-26 调优后部署请联系商务经理。 + **模型服务** **模型代码** @@ -679,6 +683,540 @@ qwen-plus-character-2025-11-06 ¥0.15/千Token +**点击此处查看更多千问 VL 模型使用限制** + +##### **使用限制** + +###### **输入文件限制** + +**图像限制** + +- **图像分辨率：** + + - 最小尺寸：图像的宽度和高度均须大于`10`像素。 + + - 宽高比：原图及缩放后的图像，长边与短边的比值不得超过 `200:1`。 + + > 图像缩放逻辑请参见 [计算图像的Token](https://help.aliyun.com/zh/model-studio/vision#ea146aaca47ng) 中 `smart_resize` 函数。 + + - 像素上限： + + - 推荐将图像分辨率控制在`8K(7680x4320)`以内。超过此分辨率的图像可能因文件过大、网络传输耗时过长而导致API调用超时。 + + - 自动缩放机制：模型可通过`max_pixels`和`min_pixels`调整图像大小；因此，提供超高分辨率的图像并不会提升识别精度，反而会增加调用失败的风险，建议在客户端提前将图像缩放至合理大小。 + +- **支持的图像格式** + + - 分辨率在4K `(3840x2160)`以下，支持的图像格式如下： + + **图像格式** + + **常见扩展名** + + **MIME Type** + + BMP + + .bmp + + image/bmp + + JPEG + + .jpe, .jpeg, .jpg + + image/jpeg + + PNG + + .png + + image/png + + TIFF + + .tif, .tiff + + image/tiff + + WEBP + + .webp + + image/webp + + HEIC + + .heic + + image/heic + + - 分辨率处于`4K(3840x2160)`到`8K(7680x4320)`范围，仅支持 JPEG、JPG 、PNG 格式 + +- **图像大小：** + + - 以公网URL传入时：Qwen3.7系列、Qwen3.6系列、Qwen3.5系列单个图像不超过 `20MB`，其他模型单个图像不超过`10MB` + + - 以本地路径传入时：单个图像不超过`10MB` + + - 以 Base64 编码传入时：编码后的字符串不超过`10MB` + + + > 如需压缩文件体积请参见 [如何将图像或视频压缩到满足要求的大小](https://help.aliyun.com/zh/model-studio/vision#ec8e0a8e03moe) 。 + +- **图片数量限制：**多图输入时根据传入方式不同，支持的图片数量上限有所区别： + + - 以公网URL或本地路径传入时： + + - Qwen3.7-Plus：最多 2048 张 + + - Qwen3.7-Flash、Qwen3.6-Plus、Qwen3.6-Flash、Qwen3.5-Plus、Qwen3.5-Flash、Qwen3-VL、Qwen-VL、QVQ系列：最多 256 张 + + - 以 Base64 编码传入时：最多 250 张 + + + Qwen-Omni系列模型请参考[全模态](https://help.aliyun.com/zh/model-studio/omni/)。 + + +> 同时受模型图文总 Token 上限（即最大输入）的限制，所有图片的总 Token 数必须小于模型的最大输入。 + +**视频限制** + +- **以图像列表传入，图像列表的数量有如下限制：** + + - `qwen3.7`系列、`qwen3.6`系列、`qwen3.5` 系列：最少传入 4 张图片，最多 8000 张图片 + + - `qwen3-vl-plus` 系列、`qwen3-vl-flash` 系列、`qwen3-vl-235b-a22b-thinking`、`qwen3-vl-235b-a22b-instruct`：最少传入 4 张图片，最多 2000 张图片 + + - 其他`Qwen3-VL`开源、`Qwen2.5-VL`（包括商业版和开源版）和`QVQ`系列模型：最少传入 4 张图片，最多 512 张图片 + + - 其他模型：最少传入 4 张图片，最多 80 张图片 + +- **以视频文件传入时：** + + - **视频大小：** + + - 以公网URL传入时： + + - `qwen3.7`系列、`qwen3.6`系列、`qwen3.5` 系列、`Qwen3-VL`系列、`qwen-vl-max` ：不超过 2GB； + + - `qwen-vl-plus` 系列、其他`qwen-vl-max`模型、`Qwen2.5-VL`开源系列及`QVQ`系列模型：不超过 1GB； + + - 其他模型不超过 150MB + + - 以 Base64 编码传入时：编码后的字符串小于 10MB； + + - 以本地文件路径传入时：视频本身不超过 100MB。 + + + > 如需压缩文件体积请参见 [如何将图像或视频压缩到满足要求的大小](https://help.aliyun.com/zh/model-studio/vision#ec8e0a8e03moe) 。 + + - **视频时长：** + + - `qwen3.7`系列、`qwen3.6`系列、`qwen3.5` 系列：2秒至2小时； + + - `qwen3-vl-plus`系列、`qwen3-vl-flash`系列、`qwen3-vl-235b-a22b-thinking`、`qwen3-vl-235b-a22b-instruct`：2 秒至 1 小时； + + - 其他`Qwen3-VL`开源系列、`qwen-vl-max` ：2 秒至 20 分钟； + + - `qwen-vl-plus`系列、 其他`qwen-vl-max`模型、`Qwen2.5-VL`开源系列及`QVQ`系列模型：2 秒至 10 分钟； + + - 其他模型：2 秒至 40 秒。 + + - **视频格式：** MP4、AVI、MKV、MOV、FLV、WMV 等。 + + - **视频尺寸：**无特定限制，模型可通过`max_pixels`和`min_pixels`自动调整视频尺寸，更大尺寸的视频文件不会有更好的理解效果。 + + - **视频数量限制：**最多可传入 64 个视频。 + + - **音频理解：**不支持对视频文件的音频进行理解。 + + +###### **文件传入方式** + +- **公网URL**：提供一个公网可访问的文件地址，支持HTTP或HTTPS协议。为获得最佳稳定性和性能，可将文件[上传至OSS](https://help.aliyun.com/zh/oss/user-guide/console-quick-start)或[上传文件获取临时URL](https://help.aliyun.com/zh/model-studio/get-temporary-file-url)，获取公网 URL。 + + **重要** + + 为确保模型能成功下载文件，提供的公网URL的响应头中**必须**包含 Content-Length（文件大小）和 Content-Type（媒体类型，如 image/jpeg）。任一字段缺失或者错误将会导致文件下载失败。 + +- **Base64编码传入：**将文件转换为 Base64 编码字符串再传入。 + +- **本地文件路径传入（仅限 DashScope SDK）：**传入本地文件的路径。 + + +> 关于文件传入方式的建议，请参见 [如何选择文件上传方式？](https://help.aliyun.com/zh/model-studio/vision#dc4e7260aauuo) + +**计算图像与视频的Token** + +## **图像** + +计算公式：`图像 Token = h_bar * w_bar / token_pixels + 2` + +- `h_bar、w_bar`：缩放后的图像长宽，模型在处理图像前会进行预处理，会将图像缩小至特定像素上限内，像素上限与`max_pixels`和`vl_high_resolution_images`参数的取值有关。 + +- `token_pixels`：每视觉`Token`对应的像素值，不同模型情况不同： + + - `qwen3.7系列`、`qwen3.6系列`、`qwen3.5系列`、`Qwen3-VL`、`qwen-vl-max`、`qwen-vl-plus`**：**每个`Token`对应 `32x32`像素 + + - `QVQ`及其他`Qwen2.5-VL`模型**：**每个Token对应`28x28`像素 + + +以下代码演示了模型内部对图像的大致缩放逻辑，可用于估算一张图像的Token，实际计费请以API响应为准。 + +**点此查看图像 token 估算示例** + +``` +import math +from PIL import Image # pip install Pillow + +def smart_size(image_path, max_pixels, vl_high_resolution_images): + """根据模型参数，计算图像缩放后的尺寸，用于估算图像 Token。""" + image = Image.open(image_path) + height, width = image.height, image.width + + # Qwen3.6、Qwen3.5、Qwen3-VL 等模型的缩放因子为 32；其他模型为 28 + factor = 32 + h_bar = round(height / factor) * factor + w_bar = round(width / factor) * factor + + # Token 下限：4 个 Token + min_pixels = 4 * factor * factor + + # vl_high_resolution_images=True 时，Token 上限固定为 16384，忽略 max_pixels + if vl_high_resolution_images: + max_pixels = 16384 * factor * factor + + # 将总像素数约束在 [min_pixels, max_pixels] 范围内 + if h_bar * w_bar > max_pixels: + beta = math.sqrt((height * width) / max_pixels) + h_bar = math.floor(height / beta / factor) * factor + w_bar = math.floor(width / beta / factor) * factor + elif h_bar * w_bar < min_pixels: + beta = math.sqrt(min_pixels / (height * width)) + h_bar = math.ceil(height * beta / factor) * factor + w_bar = math.ceil(width * beta / factor) * factor + + return h_bar, w_bar + +if __name__ == "__main__": + # 注意：max_pixels 和 vl_high_resolution_images 的值需要与调用模型时传入的参数保持一致 + h_bar, w_bar = smart_size("xxx/test.jpg", max_pixels=2560 * 32 * 32, vl_high_resolution_images=False) + print(f"缩放后的图像尺寸：高度 {h_bar}，宽度 {w_bar}") + + # 每张图像额外包含 和 各 1 个 Token + token = int(h_bar * w_bar / (32 * 32)) + 2 + print(f"图像的 Token 数：{token}") +``` + +## **视频** + +- **视频文件：** + + 模型处理视频文件时，会先进行抽帧，然后计算所有视频帧的总 Token 数。由于该计算过程较为复杂，可使用以下代码，通过传入视频路径来估算视频消耗的总 Token 数： + + ``` + # 使用前安装：pip install opencv-python + import math + import os + import logging + import cv2 + + logger = logging.getLogger(__name__) + + FRAME_FACTOR = 2 + + # Qwen3.6、Qwen3.5、Qwen3-VL、qwen-vl-max-0813、qwen-vl-plus-0815、qwen-vl-plus-0710等模型，图像缩放因子为32 + IMAGE_FACTOR = 32 + + # 其他模型，图像缩放因子为28 + # IMAGE_FACTOR = 28 + + # 视频帧的最大长宽比 + MAX_RATIO = 200 + # 视频帧的像素下限 + VIDEO_MIN_PIXELS = 4 * 32 * 32 + # 视频帧的像素上限，使用Qwen3-VL-Plus模型，VIDEO_MAX_PIXELS为640 * 32 * 32，其他模型为768 * 32 * 32 + VIDEO_MAX_PIXELS = 640 * 32 * 32 + + # 用户未传入FPS参数，则fps使用默认值 + FPS = 2.0 + # 最少抽取帧数 + FPS_MIN_FRAMES = 4 + # 最大抽取帧数（根据模型选择设置值） + FPS_MAX_FRAMES = 2000 + + # 视频输入的最大像素值，使用Qwen3-VL-Plus模型，请将VIDEO_TOTAL_PIXELS设置为131072 * 32 * 32，其他模型设置为65536 * 32 * 32 + VIDEO_TOTAL_PIXELS = int(float(os.environ.get('VIDEO_MAX_PIXELS', 131072 * 32 * 32))) + + def round_by_factor(number: int, factor: int) -> int: + """返回与”number“最接近的整数，该整数可被”factor“整除。""" + return round(number / factor) * factor + + def ceil_by_factor(number: int, factor: int) -> int: + """返回大于或等于“number”且可被“factor”整除的最小整数。""" + return math.ceil(number / factor) * factor + + def floor_by_factor(number: int, factor: int) -> int: + """返回小于或等于“number”且可被“factor”整除的最大整数。""" + return math.floor(number / factor) * factor + + def extract_vision_info(conversations): + vision_infos = [] + if isinstance(conversations[0], dict): + conversations = [conversations] + for conversation in conversations: + for message in conversation: + if isinstance(message["content"], list): + for ele in message["content"]: + if ( + "image" in ele + or "image_url" in ele + or "video" in ele + or ele.get("type","") in ("image", "image_url", "video") + ): + vision_infos.append(ele) + return vision_infos + + def smart_nframes(ele,total_frames,video_fps): + """用于计算抽取的视频帧数。 + + Args: + ele (dict): 包含视频配置的字典格式 + - fps: fps用于控制提取模型输入帧的数量。 + total_frames (int): 视频的原始总帧数。 + video_fps (int | float): 视频的原始帧率 + + Raises: + nframes应该在[FRAME_FACTOR，total_frames]间隔内，否则会报错 + + Returns: + 用于模型输入的视频帧数。 + """ + assert not ("fps" in ele and "nframes" in ele), "Only accept either `fps` or `nframes`" + fps = ele.get("fps", FPS) + min_frames = ceil_by_factor(ele.get("min_frames", FPS_MIN_FRAMES), FRAME_FACTOR) + max_frames = floor_by_factor(ele.get("max_frames", min(FPS_MAX_FRAMES, total_frames)), FRAME_FACTOR) + duration = total_frames / video_fps if video_fps != 0 else 0 + if duration-int(duration)>(1/fps): + total_frames = math.ceil(duration * video_fps) + else: + total_frames = math.ceil(int(duration)*video_fps) + nframes = total_frames / video_fps * fps + if nframes > total_frames: + logger.warning(f"smart_nframes: nframes[{nframes}] > total_frames[{total_frames}]") + nframes = int(min(min(max(nframes, min_frames), max_frames), total_frames)) + if not (FRAME_FACTOR <= nframes and nframes <= total_frames): + raise ValueError(f"nframes should in interval [{FRAME_FACTOR}, {total_frames}], but got {nframes}.") + + return nframes + + def get_video(video_path): + # 获取视频信息 + cap = cv2.VideoCapture(video_path) + + frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) + # 获取视频高度 + frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) + total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) + + video_fps = cap.get(cv2.CAP_PROP_FPS) + return frame_height, frame_width, total_frames, video_fps + + def smart_resize(ele, path, factor=IMAGE_FACTOR): + # 获取原视频的宽和高 + height, width, total_frames, video_fps = get_video(path) + # 视频帧的Token下限 + min_pixels = VIDEO_MIN_PIXELS + total_pixels = VIDEO_TOTAL_PIXELS + # 抽取的视频帧数 + nframes = smart_nframes(ele, total_frames, video_fps) + max_pixels = max(min(VIDEO_MAX_PIXELS, total_pixels / nframes * FRAME_FACTOR),int(min_pixels * 1.05)) + + # 视频的长宽比不应超过200:1或1:200 + if max(height, width) / min(height, width) > MAX_RATIO: + raise ValueError( + f"absolute aspect ratio must be smaller than {MAX_RATIO}, got {max(height, width) / min(height, width)}" + ) + + h_bar = max(factor, round_by_factor(height, factor)) + w_bar = max(factor, round_by_factor(width, factor)) + if h_bar * w_bar > max_pixels: + beta = math.sqrt((height * width) / max_pixels) + h_bar = floor_by_factor(height / beta, factor) + w_bar = floor_by_factor(width / beta, factor) + elif h_bar * w_bar < min_pixels: + beta = math.sqrt(min_pixels / (height * width)) + h_bar = ceil_by_factor(height * beta, factor) + w_bar = ceil_by_factor(width * beta, factor) + return h_bar, w_bar + + def token_calculate(video_path, fps): + # 传入视频路径和fps抽帧参数 + messages = [{"content": [{"video": video_path, "fps": fps}]}] + vision_infos = extract_vision_info(messages)[0] + + resized_height, resized_width = smart_resize(vision_infos, video_path) + + height, width, total_frames, video_fps = get_video(video_path) + num_frames = smart_nframes(vision_infos, total_frames, video_fps) + print(f"原视频尺寸：{height}*{width}， 输入模型的尺寸：{resized_height}*{resized_width}，视频总帧数:{total_frames}，fps等于{fps}时，抽取的总帧数：{num_frames}", end="，") + video_token = int(math.ceil(num_frames / 2) * resized_height / 32 * resized_width / 32) + video_token += 2 # 系统会自动添加<|vision_bos|>和<|vision_eos|>视觉标记（各计1个Token） + return video_token + + video_token = token_calculate("xxx/test.mp4", 1) + print("视频tokens:", video_token) + ``` + +- **图像列表：** + + 当以图像列表形式传入视频时，表示已预先完成视频抽帧，可使用以下代码，通过传入图像的路径和数量来计算传入图像列表时消耗的Token数： + + ``` + # 使用前安装：pip install Pillow + import math + import os + import logging + from typing import Tuple + from PIL import Image + + logger = logging.getLogger(__name__) + + # ==================== 常量定义 ==================== + FRAME_FACTOR = 2 + # Qwen3-VL、qwen-vl-max-0813、qwen-vl-plus-0815、qwen-vl-plus-0710模型，缩放因子为32 + IMAGE_FACTOR = 32 + + # 其他模型，缩放因子为28 + # IMAGE_FACTOR = 28 + + # Token计算相关常量 + TOKEN_DIVISOR = 32 # token计算时的除数 + VISION_SPECIAL_TOKENS = 2 # <|vision_bos|>和<|vision_eos|>标记 + + # 视频帧的最大长宽比 + MAX_RATIO = 200 + # 视频帧的像素下限 + VIDEO_MIN_PIXELS = 4 * 32 * 32 + # 视频帧的像素上限，使用Qwen3-VL-Plus模型，VIDEO_MAX_PIXELS为640 * 32 * 32，其他模型为768 * 32 * 32 + VIDEO_MAX_PIXELS = 640 * 32 * 32 + + # 视频输入的最大像素值，使用Qwen3-VL-Plus模型，请将VIDEO_TOTAL_PIXELS设置为131072 * 32 * 32，其他模型设置为65536 * 32 * 32 + VIDEO_TOTAL_PIXELS = int(float(os.environ.get('VIDEO_MAX_PIXELS', 131072 * 32 * 32))) + + def round_by_factor(number: int, factor: int) -> int: + """返回与”number“最接近的整数，该整数可被”factor“整除。""" + return round(number / factor) * factor + + def ceil_by_factor(number: int, factor: int) -> int: + """返回大于或等于“number”且可被“factor”整除的最小整数。""" + return math.ceil(number / factor) * factor + + def floor_by_factor(number: int, factor: int) -> int: + """返回小于或等于“number”且可被“factor”整除的最大整数。""" + return math.floor(number / factor) * factor + + def get_image_size(image_path: str) -> Tuple[int, int]: + if not os.path.exists(image_path): + raise FileNotFoundError(f"图像文件不存在: {image_path}") + + try: + image = Image.open(image_path) + height = image.height + width = image.width + image.close() # 及时关闭文件 + return height, width + except Exception as e: + raise ValueError(f"无法读取图像文件 {image_path}: {str(e)}") + + def smart_resize(height: int, width: int, nframes: int, factor: int = IMAGE_FACTOR) -> Tuple[int, int]: + """ + 计算图像缩放后的尺寸 + + Args: + height: 原始图像高度 + width: 原始图像宽度 + nframes: 视频帧数 + factor: 缩放因子，默认为IMAGE_FACTOR + + Returns: + (resized_height, resized_width) 缩放后的高度和宽度 + + Raises: + ValueError: 长宽比超过限制 + """ + # 视频帧的Token下限 + min_pixels = VIDEO_MIN_PIXELS + total_pixels = VIDEO_TOTAL_PIXELS + # 抽取的视频帧数 + max_pixels = max(min(VIDEO_MAX_PIXELS, total_pixels / nframes * FRAME_FACTOR), int(min_pixels * 1.05)) + + # 视频的长宽比不应超过200:1或1:200 + aspect_ratio = max(height, width) / min(height, width) + if aspect_ratio > MAX_RATIO: + raise ValueError( + f"图像长宽比必须小于 {MAX_RATIO}:1，当前为 {aspect_ratio:.2f}:1" + ) + + h_bar = max(factor, round_by_factor(height, factor)) + w_bar = max(factor, round_by_factor(width, factor)) + if h_bar * w_bar > max_pixels: + beta = math.sqrt((height * width) / max_pixels) + h_bar = floor_by_factor(height / beta, factor) + w_bar = floor_by_factor(width / beta, factor) + elif h_bar * w_bar < min_pixels: + beta = math.sqrt(min_pixels / (height * width)) + h_bar = ceil_by_factor(height * beta, factor) + w_bar = ceil_by_factor(width * beta, factor) + return h_bar, w_bar + + def calculate_video_tokens(image_path: str, nframes: int = 1, factor: int = IMAGE_FACTOR, verbose: bool = True) -> int: + """ + + Args: + image_path: 视频帧文件路径 + nframes: 视频帧数， + factor: 缩放因子，默认为IMAGE_FACTOR + verbose: 是否打印详细信息 + + Returns: + 所消耗的token数量 + + Raises: + FileNotFoundError: 文件不存在 + ValueError: 文件格式无效或长宽比超限 + """ + # 获取原始图像尺寸（只读取一次） + height, width = get_image_size(image_path) + + # 计算缩放后的尺寸 + resized_height, resized_width = smart_resize(height, width, nframes, factor) + + # 计算token数量 + # 公式：⌈帧数/2⌉ × (高度/TOKEN_DIVISOR) × (宽度/TOKEN_DIVISOR) + VISION_SPECIAL_TOKENS + video_token = int( + math.ceil(nframes / 2) * + (resized_height / TOKEN_DIVISOR) * + (resized_width / TOKEN_DIVISOR) + ) + # 添加视觉标记token（<|vision_bos|>和<|vision_eos|>） + video_token += VISION_SPECIAL_TOKENS + + if verbose: + print(f"原视频帧尺寸：{height}×{width}，输入模型的尺寸：{resized_height}×{resized_width}，", end="") + + return video_token + + if __name__ == "__main__": + try: + video_token = calculate_video_tokens("xxx/test.jpg", nframes=30) + print(f"视频tokens: {video_token}\n") + except Exception as e: + print(f"错误: {str(e)}\n") + ``` + + ## **模型调优前必读** - 文本生成模型调优虽然能在特定业务/场景取得非常好的效果，但有以下限制： diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/rl-training-overview.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/rl-training-overview.md new file mode 100644 index 00000000..d747c187 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/rl-training-overview.md @@ -0,0 +1,448 @@ +# 强化学习训练概述 + +本文通过端到端示例帮助您快速完成第一个强化学习（RL） 训练任务的提交与监控。 + +**如需了解 RL 训练的基本概念和适用场景，请点击这里。** + +百炼提供的多种调优方式并不互斥，而是递进的、相辅相成的。 + +`CPT（可选）→ SFT → DPO（可选）→ RL（可选）` + +1. CPT (持续预训练）- 补知识 （通用模型知识的“广度”和“浅度”，无法满足专业领域的“深度”和“精度”要求） + + - 金融模型： `学金融术语` + + - 医疗模型： `记药品病理` + + - 法律模型： `懂法条判例` + +2. SFT (监督微调）- 学做事 + + - 客服机器人： `学客服流程` + + - 代码助手： `学编程范式` + + - 工具调用 (Agent)： `学使用 MCP` + +3. DPO (直接偏好优化）- 做得更好 + + - 安全与责任感： `拒有害建议` + + - 简洁与有效性： `答干脆利落` + + - 客观与中立： `评公正客观` + +4. RL（强化学习）- 学推理（通过 Reward 信号驱动模型自主探索最优策略，无需提供标准答案） + + - 数学推理： `解题更准` + + - 工具调用 (Agent)： `调得更稳` + + - 代码生成： `写得能跑` + + +## RL 训练全流程概览 + +RL 训练从环境准备到模型部署，共 5 步： + +1. **环境准备** — 安装离线 SDK、获取 API Key、完成 RL 授权 + +2. **准备数据与代码** — 下载 Demo 包，了解数据格式与项目结构 + +3. **提交训练任务** — 运行 Demo 一步完成函数部署、数据上传、任务提交 + +4. **训练状态观测** — 查看训练状态、日志和 Reward 趋势 + +5. **发布与部署模型** — 发布 Checkpoint 至"我的模型"，部署后通过 API 调用 + + +### RL 训练原理 + +RL 训练通过"生成-评分-优化"循环不断提升模型能力。每轮训练的数据流如下： + +``` +训练数据（用户问题 + 参考答案） + │ + ▼ +策略采样（Rollout）：模型按当前策略生成候选结果 + │ + ▼ +奖励评分（Reward）：对比参考答案，给出 0~1 分（可以合并到 Rollout） + │ + ▼ +策略更新：根据奖励信号，强化高分策略、抑制低分策略 + │ + ▼ +循环迭代：用更新后的模型重新采样，逐轮提升质量 +``` + +## 支持的模型 + +**说明** + +请联系**商务经理**开启强化学习训练功能。 + +**模型名称** + +**模型 code** + +**是否为 MoE（混合专家）模型** + +**推荐的模型训练单元数量** + +千问3.5-9B + +qwen3.5-9b + +否 + +IV型模型单元 \* 24 + +千问3.6-flash-2026-04-16 + +qwen3.6-flash-2026-04-16 + +是 + +IV型模型单元 \* 72 + +千问3.5-flash-2026-02-23 + +qwen3.5-flash-2026-02-23 + +是 + +IV型模型单元 \* 72 + +## 训练单元价格 + +**计费方式** + +**说明** + +**计费公式** + +训练单元·预付费 + +按月购买模型训练单元，使用专属训练资源，**训练速度更快，无需排队等待**。 最小计费粒度：**小时**。提前退订按小时单价的 **1.2** 倍计费。 如需购买或管理预付费训练单元，可在[模型调优控制台](https://bailian.console.aliyun.com/cn-beijing/efm/model_manager?tab=model#/efm/model_manager)点击**管理训练资源**。 + +`单价 × 实例数 × 购买时长` + +训练单元·后付费 + +使用专属训练资源，**训练速度更快，无需排队等待**。按实际使用时长计费，创建任务时直接开通，无需预先购买。 + +`小时单价 × 实例数 × 使用小时数` + +**训练单元类型** + +**计费方式** + +**单价** + +**最小计费粒度** + +IV 型（MTU4） + +预付费（包月） + +19,914.00 元/月/实例 + +1 小时 + +后付费（按分钟） + +41.00 元/小时/实例 + +1 分钟 + +**说明** + +预付费扩缩容、续费、退订等运营规则与具体 MTU 容量测算请联系商务经理。 + +## 环境准备 + +**警告** + +RL 训练**仅支持**通过模型训练单元（MTU）计费，不支持按 Token 计费方式。 + +按照以下步骤完成环境配置： + +1. **获取 API Key**：在百炼控制台获取 DashScope API Key。 + +2. **完成 RL 服务授权**：在百炼控制台的**模型调优**页面完成一键授权。首次使用 RL 训练时，控制台会提示授权**阿里云 OpenTelemetry**、**函数计算（FC）**和**日志服务（SLS）**三项云服务。 + + ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7125558771/p1073237.png) + +3. **设置环境变量**：以下变量由本地 SDK 读取，控制打包、上传和部署行为： + + - **DASHSCOPE\_API\_KEY**（必填）：API 密钥。 + + - **FC\_PYPI\_LIB**（必填）：指定 FC 容器启动时安装的 dashscope 离线包。SDK 打包部署时会自动将该变量注入为 `FC_SDK_PACKAGE`，无需在 FC 侧手动安装。whl 文件名需与此变量**完全一致**，且放在项目根目录下。 + + - **LOG\_LEVEL**（可选，默认 `info`）：设为 `debug` 可输出完整请求/响应信息，便于排查提交失败等问题。 + + + ``` + export DASHSCOPE_API_KEY="sk-your-api-key" + export LOG_LEVEL="info" + ``` + + +## 准备数据与代码 + +1. 下载 Demo 包[agentic-rl-example.zip](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260721/jfslkz/agentic-rl-example-for-customer-9bae5b71d53759bc7d051cda8f4671349d6164a5.zip)。（该 Demo 运行最低要求 24 个 IV 型模型训练单元） + +2. **下载完成后需要安装 DashScope SDK**（要求 Python >= 3.10）：请按照 Demo 包中附带的指引安装相关依赖： + + ``` + # 使用PyPI源： + pip install dashscope + ``` + + +您将获得以下项目结构： + +``` +agentic-rl-example/ +├── submit_job.py ← 主入口，提交训练任务 +├── test_functions.py ← 函数注册与远程测试 +├── config.yaml ← 可选的 YAML 配置文件 +├── job.yaml ← YAML 提交配置（CLI run / SDK init） +├── cli.sh ← 本地测试与查询脚本集合 +├── dashscope-1.25.23-py3-none-any.whl ← 离线 SDK 安装包 +├── requirements.txt ← 依赖清单 +├── data/ +│ ├── calc_train.jsonl ← 训练数据（完整版） +│ ├── calc_train_min.jsonl ← 训练数据（Demo 默认使用精简版） +│ ├── calc_validation.jsonl ← 验证数据（完整版） +│ └── calc_validation_min.jsonl ← 验证数据（Demo 默认使用精简版） +├── functions/ +│ ├── __init__.py +│ ├── rollout/ +│ │ ├── rollout.py ← Rollout 函数（LangGraph + MCP 工具） +│ │ └── rollout_only.py ← Rollout 函数（简单版，含追踪示例） +│ └── reward/ +│ ├── reward.py ← Reward 函数（基础评分） +│ ├── reward_decorator.py ← Reward 函数（装饰器风格） +│ └── group_reward.py ← Reward 函数（分组评分） +├── resources/ +│ ├── rollout_input.json ← Rollout 测试输入 +│ ├── reward_input.json ← Reward 测试输入 +│ ├── reward_decorator_input.json ← 装饰器 Reward 测试输入 +│ └── group_reward_input.json ← 分组 Reward 测试输入 +└── scripts/ ← 本地启动与查询辅助脚本（build/install/start_local_*/query_local_*/query_online_tuning/convert_parquet_to_jsonl） +``` + +### Demo 简介 + +本 Demo 实现了一个数学计算 Agent： + +- **Rollout 函数**：调用训练中的模型，配合计算器工具回答数学问题 + +- **Reward 函数**：将模型输出与参考答案对比，判定回答是否正确并评分 + + +**说明** + +Demo 中的 Rollout 函数使用 LangGraph 编排 Agent 流程，通过 **MCP**（Model Context Protocol，一种 LLM 工具调用协议）调用计算器工具。 + +如需编写自己的 Rollout 和 Reward 函数，请参见[强化学习开发指南](https://help.aliyun.com/zh/model-studio/rl-function-development-guide)。 + +### 数据格式 + +训练数据采用 **JSONL** 格式，每行一个 JSON 对象，包含 `messages` 和 `rollout_extra` 两个字段： + +``` +{"messages": [{"content": "Output the answer when you are ready. The answer should be surrounded by three sharps (`###`), in the form of ### ANSWER: ###. 6.6 minus x (3/2) times equals 5.6.", "role": "user"}], "rollout_extra": {"solution": "2/3"}} +``` + +- `messages`：用户问题，格式与 ChatML 一致 + +- `rollout_extra`：存放参考答案或其他业务数据，会传递给 Reward 函数用于评分（rollout\_extra 支持自定义字段） + + +`rollout_extra` 中的数据会透传到 Rollout 函数的 `RolloutInput.rollout_extra` 和 Reward 函数的 `RewardInput.agent_output.rollout_extra`，方便在评分逻辑中使用。该字段支持自定义 key，不限于示例中的 `solution`。 + +### 数据量建议 + +- **起步**：几十到几百条即可验证方案有效性（数据量至少大于 `Batch_size`）。 + +- **正式训练**：数据量越大效果越优，大规模数据集还支持更大的 batch\_size，有助于提升训练稳定性 + + +## 提交训练任务 + +Demo 中已包含 Rollout 和 Reward 函数，运行以下命令即可一步完成函数部署、数据上传、训练任务提交： + +``` +python submit_job.py +``` + +**完整代码及参数说明** + +以下为训练脚本的完整代码，注释中说明了各参数的含义和取值建议： + +``` +import asyncio +from dashscope.finetune.reinforcement import ( + RolloutFunctionComponent, RewardFunctionComponent, + FunctionComponentModel, FunctionComponentRuntime, + TrainingDataset, ValidationDataset, DataSourceType, +) +from dashscope.finetune.agentic_rl import AgenticRL + +async def main(): + client = AgenticRL() + + # client.run() 一步完成：注册函数到 FC → 上传训练数据 → 提交训练任务 + result = await client.run( + model="qwen3.5-9b", # 基座模型 + training_datasets=[TrainingDataset( # 训练集 + data_source_type=DataSourceType.FILE_ID, + file_name="./data/calc_train_min.jsonl", + )], + validation_datasets=[ValidationDataset( # 验证集（可选） + data_source_type=DataSourceType.FILE_ID, + file_name="./data/calc_validation_min.jsonl", + )], + functions=[ + # Rollout 函数：定义模型如何执行策略采样 + RolloutFunctionComponent( + name="rollout-1", + timeout=600, + fcmodel=FunctionComponentModel( + classpath="functions.rollout.rollout.CalcXRolloutProcessor"), + runtime=FunctionComponentRuntime( + cpu=2, memory_size=4096, disk_size=512, + concurrency=30, capacity=30, + min_capacity=30, max_capacity=60, + memory_scale_threshold=0.6, + concurrency_scale_threshold=0.6, + env={})), # FC 函数环境变量 + RewardFunctionComponent( + name="reward-1", + weight=1.0, + timeout=120, + reward_metric_weight={"reward_metric_weightA": 0.3, "reward_metric_weightB": 0.7}, + fcmodel=FunctionComponentModel( + classpath="functions.reward.reward.DemoRewardProcessor"), + runtime=FunctionComponentRuntime( + cpu=2, memory_size=4096, disk_size=512, + concurrency=30, capacity=30, + min_capacity=30, max_capacity=60, + memory_scale_threshold=0.6, + concurrency_scale_threshold=0.6, + env={})), + ], + # MTU 计费配置 + resources={ + "charge_type": "mtu_postpaid", # 后付费 + "mtu_spec_code": "MTU4", # 训练单元规格 + "mtu_capacity": 24, # MTU 数量 + }, + # GSPO 算法超参数（以下为 qwen3.5-9b 非 MoE 模型的全部必选超参） + hyper_parameters={ + "algorithm": "gspo", # 训练算法 + "batch_size": 64, # 批大小 + "eval_steps": 1, # 每 N 步执行一次验证 + "kl_loss_coef": 0.002, # KL 散度损失系数 + "learning_rate": 2e-6, # 学习率 + "lr_scheduler_type": "linear", # 学习率调度策略 + "max_length": 8192, # 最大序列长度（prompt + response） + "n_epochs": 1, # 训练轮数 + "n_rollouts": 8, # 每个样本的 Rollout 次数 + "ppo_mini_batch_size": 8, # PPO 小批量大小 + "save_strategy": "epoch", # 保存策略（steps 或 epoch） + }) + + if result.status_code == 200: + print(f"训练任务已提交，Job ID: {result.output.job_id}") + else: + print(f"提交失败: {result}") + +asyncio.run(main()) +``` + +**说明** + +也可以通过配置文件或 CLI 提交训练任务（`dashscope rl run`），三种提交方式与字段逐项说明详见 [《训练配置 — 提交与配置》](https://help.aliyun.com/zh/model-studio/rl-training-config-monitoring)。 + +### 起步技巧 + +第一次跑通后，建议参考以下 4 项实践： + +- **先用精简数据**：Demo 默认 `calc_train_min.jsonl` + `n_epochs=1` 跑一次，确认链路通了再换完整数据集，避免拿 24 个 MTU 烧大数据集才发现配置错。 + +- **超参先不动**：上面 11 项超参是 qwen3.5-9b 的推荐起点，第一次训练别调。要调时一次只动一个变量（lr 或 batch\_size），跑满 `eval_steps × 3` 步再判断趋势。 + +- **盯一个核心指标**：在**指标**页签看 `critic/rewards/mean`——稳步上升说明在学；停滞或下降就停下来排查，别空跑。 + +- **FAILED 先看日志**：任务失败 → **日志**页签看末尾报错，或 SDK 拉 `AgenticRL.logs(job_id="ft-xxx", lines=100)`。配置/依赖类报错通常一眼能定位。 + + +**说明** + +算法选型、超参起点表、调参决策详见《强化学习训练配置 — 提交与配置》；Reward 函数设计哲学与 Hacking 防御详见《强化学习开发指南》；看到指标异常如何归因详见《强化学习的可观测配置与指标参考》。 + +## 训练过程观测与分析 + +任务提交后，`run()` 返回的 `job_id` 是后续查询训练状态的唯一标识。训练通常需要数十分钟到数小时，取决于数据量和模型大小。 + +在百炼控制台的**模型调优**页面点击任务进入详情页，包含以下 5 个页签： + +**页签** + +**定位** + +**主要内容** + +**详情** + +任务元数据 + +任务 ID、状态、基座模型、训练方法、起止时间、数据配置 + +**轨迹** + +Agent 行为回放 + +完整对话过程、Reward 分析（Step / Sample / Trajectory 三维度）、工具调用详情 + +**指标** + +训练健康度 + +13 组指标图表（actor / critic / trajectory / trace / timing / perf / fully\_async 等），含用户自定义指标 + +**日志** + +运行排查 + +训练过程 stdout / stderr，任务失败时用于定位报错原因 + +**产出** + +模型资产 + +Checkpoint 列表、发布状态，可将 Checkpoint 发布至"我的模型"后部署调用 + +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/5224658771/p1073270.png) + +各页签的详细说明请参见[强化学习的可观测配置](https://help.aliyun.com/zh/model-studio/observable-configuration-for-reinforcement-learning)。 + +## 发布与部署模型 + +训练完成后，最后一个 Checkpoint 会自动发布至**我的模型**。如需发布中间 Checkpoint，前往百炼控制台**模型调优** > **产出**页面手动操作，详见_模型调优控制台（百炼文档，需在线查看）_。 + +发布后的模型可在**我的模型**页面进行部署。部署完成后即可通过 API 调用模型。详见_模型部署（百炼文档，需在线查看）_。 + +## 后续步骤 + +- 了解函数开发的完整细节 → [强化学习开发指南](https://help.aliyun.com/zh/model-studio/rl-function-development-guide) + +- 了解训练配置和监控 → [强化学习训练配置 — 提交与配置](https://help.aliyun.com/zh/model-studio/rl-training-config-monitoring) + +- 训练指标参考字典 → [强化学习的可观测配置与指标参考](https://help.aliyun.com/zh/model-studio/observable-configuration-for-reinforcement-learning) + +- 了解可观测性配置与轨迹查看 → [强化学习的可观测配置与指标参考](https://help.aliyun.com/zh/model-studio/observable-configuration-for-reinforcement-learning) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/wan-image-generation-finetune-guide.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/wan-image-generation-finetune-guide.md index e688c82e..5c1b36a4 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/wan-image-generation-finetune-guide.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/wan-image-generation-finetune-guide.md @@ -115,6 +115,144 @@ curl --location --request POST 'https://dashscope.aliyuncs.com/api/v1/files' \ 请将`<替换为训练数据集的文件id>`完整替换为上一步获取的`file_id`。完整参数说明与格式约束请参见[超参数](https://help.aliyun.com/zh/model-studio/wan-generation-finetune-api-reference#5f391e4b3cezf)。 +**超参数** + +**字段** + +**类型** + +**必选** + +**描述** + +**推荐值** + +max\_steps + +int + +是 + +**训练总步数**。控制训练时长的核心参数。max\_steps 决定训练迭代次数，max\_token\_length 决定每步处理的数据量。建议不少于 500 步以确保模型充分收敛；大数据集可适当增加步数。 + +800 + +eval\_steps + +int + +是 + +**验证间隔**。取值需≥0。训练期间每隔多少个 steps 进行一次验证评估，用于阶段性评估模型训练效果。同时保存当前 step 的模型文件。 + +200 + +learning\_rate + +float + +是 + +**学习率**。控制模型权重更新的幅度。过高可能导致模型变差，过低则变化不明显。推荐使用默认值。 + +3e-5 + +generation\_type + +string + +是 + +**生成模式**。`"t2i"`：文生图模式；`"i2i"`：图生图模式。决定训练数据格式和推理方式。 + +t2i + +max\_pixels + +string + +是 + +**训练图片的最大分辨率**。例如 "1k"、"2k"（1K 即 1024×1024，2K 即 2048×2048）。设置训练集中图片分辨率的像素总数（宽×高）上限，系统仅对超过该值的图片进行缩放处理，未超限的图片保持原样。建议三个分辨率参数（max\_pixels、max\_token\_length、val\_img\_size）保持一致。 + +文生图："2k" +图生图："1k" + + +val\_img\_size + +string + +是 + +**验证图生成分辨率**。例如 "1k"、"2k"（1K 即 1024×1024，2K 即 2048×2048）。训练过程中验证评估时生成图片的目标分辨率。 + +文生图："2k" +图生图："1k" + + +max\_token\_length + +string + +是 + +**每步训练的最大 Token 长度**。例如 "1k"、"2k"。与 max\_steps 共同控制训练过程：max\_steps 决定迭代次数，max\_token\_length 决定每步处理的数据量。 + +文生图："2k" +图生图："1k" + + +gradient\_clip + +float + +是 + +**梯度裁剪**。对所有可训练参数做全局梯度范数裁剪的阈值，防止梯度爆炸。设为 -1 表示不裁剪。 + +0.5 + +weight\_decay + +float + +是 + +**权重衰减**。AdamW 解耦式权重衰减系数，对所有可训练参数生效，用于正则化防止过拟合。 + +0.02 + +lora\_rank + +int + +是 + +**LoRA 低秩矩阵的维数**。该值决定了微调参数量的大小。数值越大，模型拟合能力越强，但训练速度会变慢。取值必须为 2n（如 16、32、64）。 + +32 + +save\_total\_limit + +int + +否 + +**Checkpoint 保存数量上限**。限制最多保存的模型数量。系统将始终只保存训练生成的最后 N 个 Checkpoint（N 为该参数值）。 + +10 + +split + +float + +否 + +**训练集划分比例**。取值范围为 (0, 1)。仅在未指定 `validation_datasets` 时生效。此参数用于从训练集中自动按比例拆分出验证集。例如，0.9 表示 90% 训练集，10% 验证集。 + +0.9 + ``` curl --location 'https://dashscope.aliyuncs.com/api/v1/fine-tunes' \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/wan-video-generation-finetune-guide.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/wan-video-generation-finetune-guide.md index 34eb3507..955a59f1 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/wan-video-generation-finetune-guide.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/fine-tuning/wan-video-generation-finetune-guide.md @@ -109,10 +109,6 @@ curl --location --request POST 'https://dashscope.aliyuncs.com/api/v1/files' \ 使用步骤1中的文件ID启动训练任务。 -**说明** - -不同模型的微调参数的值有所差异，超参数设置请参见[超参数](https://help.aliyun.com/zh/model-studio/wan-generation-finetune-api-reference#5f391e4b3cezf)，更多调用示例请参见[请求示例](https://help.aliyun.com/zh/model-studio/wan-generation-finetune-api-reference#1a9196bd16o9h)。 - **请求示例** 请将`<替换为训练数据集的文件id>`完整替换为上一步获取的`file_id`。 @@ -130,12 +126,12 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/fine-tunes' \ ], "training_type": "efficient_sft", "hyper_parameters": { - "n_epochs": 400, + "n_epochs": 50, "batch_size": 1, "learning_rate": 2e-5, "split": 0.9, "max_split_val_dataset_sample": 5, - "eval_epochs": 50, + "eval_epochs": 20, "max_pixels": 102400, "save_total_limit": 10, "lora_rank": 32, @@ -157,12 +153,12 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/fine-tunes' \ ], "training_type": "efficient_sft", "hyper_parameters": { - "n_epochs": 400, + "n_epochs": 50, "batch_size": 4, "learning_rate": 2e-5, "split": 0.9, "max_split_val_dataset_sample": 5, - "eval_epochs": 50, + "eval_epochs": 20, "max_pixels": 262144, "save_total_limit": 10, "lora_rank": 32, @@ -171,6 +167,140 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/fine-tunes' \ }' ``` +**超参数（hyper\_parameters）** + +**字段** + +**类型** + +**必选** + +**描述** + +**推荐值** + +batch\_size + +int + +是 + +**批次大小**。一次性送入模型进行训练的数据条数。 + +- wan2.7-i2v：推荐为 1。 + +- wan2.5-i2v-preview：推荐为 4。 + +- wan2.2-i2v-flash：推荐为 4。 + +- wan2.2-kf2v-flash：推荐为 4。 + + +以模型为准 + +n\_epochs + +int + +是 + +**训练循环次数**。steps = n\_epochs × ⌈数据集大小 / batch\_size⌉。建议总步数 ≥ 800。 + +> 例如：数据集 5 条，batch\_size=2，每轮步数=⌈5/2⌉=3，最小 n\_epochs = 800/3 ≈ 267。 + +> 推荐训练轮数会根据数据量自动调整。数据越少，需要更多轮数来充分学习；数据越多，每轮包含的样本越多，因此所需轮数会减少。50 epochs 主要适用于 2 条左右的小数据集；当数据量达到 50-60 条视频时，通常建议训练约 3000-5000 steps 即可。 + +50 + +learning\_rate + +float + +是 + +**学习率**。控制模型权重更新幅度。过高可能导致模型变差，过低则变化不明显。 + +2e-5 + +eval\_epochs + +int + +是 + +**验证间隔**。取值需 ≥ `n_epochs/10`。每隔多少个 epoch 进行一次验证评估并保存 Checkpoint。 + +20 + +max\_pixels + +int + +是 + +**训练视频的最大分辨率**（像素总数 = 宽×高）。系统仅对超过该值的视频进行缩放处理。 + +- wan2.7-i2v：推荐 102400。范围 36864～123904。 + +- wan2.5-i2v-preview：推荐 36864。范围 16384～36864。 + +- wan2.2-i2v-flash：推荐 262144。范围 65536～262144。 + +- wan2.2-kf2v-flash：推荐 262144。范围 65536～262144。 + + +以模型为准 + +split + +float + +否 + +**训练集划分比例**。取值 (0,1)，仅在未指定 validation\_datasets 时生效。 + +0.9 + +max\_split\_val\_dataset\_sample + +int + +否 + +**自动划分验证集的最大样本数**。验证集数量 = min(总数×(1−split), 此值)。 + +5 + +save\_total\_limit + +int + +否 + +**Checkpoint 保存数量上限**。系统只保存最后 N 个 Checkpoint。 + +10 + +lora\_rank + +int + +否 + +**LoRA 低秩矩阵维数**。取值须为 2n（16/32/64）。 + +32 + +lora\_alpha + +int + +否 + +**LoRA 权重缩放系数**。取值须为 2n（16/32/64）。 + +32 + **响应示例** 关注 `output` 中的三个关键参数： @@ -1276,7 +1406,7 @@ curl --location 'https://dashscope.aliyuncs.com/api/v1/fine-tunes/<替换为微 - **n\_epochs (训练轮数)** - - 默认值：**400**，推荐使用默认值。若需调整，请遵循 **“总训练步数 (Steps) ≥ 800”** 的原则。 + - 默认值：**50**，推荐使用默认值。若需调整，请遵循 **”总训练步数 (Steps) ≥ 800”** 的原则。 - 总步数计算公式： `steps = n_epochs × 向上取整（训练集大小 / batch_size）。` diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/models.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/models.md index bb37de23..49deecfe 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/models.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/models.md @@ -12,6 +12,16 @@ [ +qwen3.8-max-preview + +](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview) + +仅 Token Plan 可用 + +qwen3.8-max-preview 目前仅面向 Token Plan 订阅用户提供，[前往开通 Token Plan →](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview) + +[ + qwen3.7-max ](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.7-max) @@ -228,27 +238,27 @@ API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/a [ -qwen3.6-flash +qwen3.7-flash -](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.6-flash) +](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.7-flash) 华北2（北京）新加坡日本（东京）德国（法兰克福）美国（弗吉尼亚） OpenAI 兼容Anthropic 兼容DashScope -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/compatible-mode/v1` API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/apps/anthropic` API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api/v1` @@ -256,19 +266,19 @@ API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api- OpenAI 兼容Anthropic 兼容DashScope -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/apps/anthropic` API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api/v1` @@ -276,19 +286,19 @@ API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=mod OpenAI 兼容Anthropic 兼容DashScope -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-northeast-1?tab=globalset#/efm/business_management).ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1` API Key[获取↗](https://modelstudio.console.aliyun.com/ap-northeast-1?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-northeast-1?tab=globalset#/efm/business_management).ap-northeast-1.maas.aliyuncs.com/apps/anthropic` API Key[获取↗](https://modelstudio.console.aliyun.com/ap-northeast-1?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-northeast-1?tab=globalset#/efm/business_management).ap-northeast-1.maas.aliyuncs.com/api/v1` @@ -296,19 +306,19 @@ API Key[获取↗](https://modelstudio.console.aliyun.com/ap-northeast-1?tab=mod OpenAI 兼容Anthropic 兼容DashScope -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management).eu-central-1.maas.aliyuncs.com/compatible-mode/v1` API Key[获取↗](https://modelstudio.console.aliyun.com/eu-central-1?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management).eu-central-1.maas.aliyuncs.com/apps/anthropic` API Key[获取↗](https://modelstudio.console.aliyun.com/eu-central-1?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management).eu-central-1.maas.aliyuncs.com/api/v1` @@ -316,19 +326,19 @@ API Key[获取↗](https://modelstudio.console.aliyun.com/eu-central-1?tab=model OpenAI 兼容Anthropic 兼容DashScope -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://dashscope-us.aliyuncs.com/compatible-mode/v1` API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://dashscope-us.aliyuncs.com/apps/anthropic` API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/api-key) -模型 ID`qwen3.6-flash` +模型 ID`qwen3.7-flash` Base URL`https://dashscope-us.aliyuncs.com/api/v1` @@ -558,92 +568,20 @@ API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/a [ -kimi-k2.7-code +kimi/kimi-k3 -](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/kimi-k2.7-code) +](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/kimi%2Fkimi-k3?serviceSite=asia-pacific-china) -华北2（北京）新加坡德国（法兰克福）美国（弗吉尼亚） +华北2（北京） -OpenAI 兼容Anthropic 兼容DashScope +OpenAI 兼容 -模型 ID`kimi-k2.7-code` +模型 ID`kimi/kimi-k3` Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/compatible-mode/v1` API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api/v1` - -API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) - -OpenAI 兼容Anthropic 兼容DashScope - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) - -OpenAI 兼容Anthropic 兼容DashScope - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management).eu-central-1.maas.aliyuncs.com/compatible-mode/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/eu-central-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management).eu-central-1.maas.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://modelstudio.console.aliyun.com/eu-central-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management).eu-central-1.maas.aliyuncs.com/api/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/eu-central-1?tab=model#/api-key) - -OpenAI 兼容Anthropic 兼容DashScope - -模型 ID`kimi-k2.7-code` - -Base URL`https://dashscope-us.aliyuncs.com/compatible-mode/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://dashscope-us.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://dashscope-us.aliyuncs.com/api/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/api-key) - [ glm-5.2 @@ -888,7 +826,7 @@ qwen3.5-omni-plus 华北2（北京）新加坡 -OpenAI 兼容Anthropic 兼容DashScope +OpenAI 兼容 模型 ID`qwen3.5-omni-plus` @@ -896,19 +834,7 @@ Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?t API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) -模型 ID`qwen3.5-omni-plus` - -Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) - -模型 ID`qwen3.5-omni-plus` - -Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api/v1` - -API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) - -OpenAI 兼容Anthropic 兼容DashScope +OpenAI 兼容 模型 ID`qwen3.5-omni-plus` @@ -916,106 +842,22 @@ Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-south API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) -模型 ID`qwen3.5-omni-plus` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) - -模型 ID`qwen3.5-omni-plus` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) - [ -kimi-k2.7-code +kimi/kimi-k3 -](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/kimi-k2.7-code) +](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/kimi%2Fkimi-k3?serviceSite=asia-pacific-china) -华北2（北京）新加坡德国（法兰克福）美国（弗吉尼亚） +华北2（北京） -OpenAI 兼容Anthropic 兼容DashScope +OpenAI 兼容 -模型 ID`kimi-k2.7-code` +模型 ID`kimi/kimi-k3` Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/compatible-mode/v1` API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api/v1` - -API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) - -OpenAI 兼容Anthropic 兼容DashScope - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) - -OpenAI 兼容Anthropic 兼容DashScope - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management).eu-central-1.maas.aliyuncs.com/compatible-mode/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/eu-central-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management).eu-central-1.maas.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://modelstudio.console.aliyun.com/eu-central-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management).eu-central-1.maas.aliyuncs.com/api/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/eu-central-1?tab=model#/api-key) - -OpenAI 兼容Anthropic 兼容DashScope - -模型 ID`kimi-k2.7-code` - -Base URL`https://dashscope-us.aliyuncs.com/compatible-mode/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://dashscope-us.aliyuncs.com/apps/anthropic` - -API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/api-key) - -模型 ID`kimi-k2.7-code` - -Base URL`https://dashscope-us.aliyuncs.com/api/v1` - -API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/api-key) - ### 生成 通过文本或图片生成图像与视频，支持编辑、参考与高分辨率输出 @@ -1024,41 +866,41 @@ API Key[获取↗](https://modelstudio.console.aliyun.com/us-east-1?tab=model#/a [ -wan2.7-image-pro +qwen-image-3.0-pro -](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/wan2.7-image-pro) +](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen-image-3.0-pro) 华北2（北京）新加坡 -模型 ID`wan2.7-image-pro` +模型 ID`qwen-image-3.0-pro` -Request URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/image-generation/generation` +Request URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) -模型 ID`wan2.7-image-pro` +模型 ID`qwen-image-3.0-pro` -Request URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/image-generation/generation` +Request URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) [ -qwen-image-2.0-pro +wan2.7-image-pro -](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen-image-2.0-pro) +](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/wan2.7-image-pro) 华北2（北京）新加坡 -模型 ID`qwen-image-2.0-pro` +模型 ID`wan2.7-image-pro` -Request URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` +Request URL`https://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/image-generation/generation` API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) -模型 ID`qwen-image-2.0-pro` +模型 ID`wan2.7-image-pro` -Request URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation` +Request URL`https://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/image-generation/generation` API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) @@ -1258,18 +1100,24 @@ API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api- [ -cosyvoice-v3.5-plus +qwen-audio-3.0-tts-plus -](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/cosyvoice-v3.5-plus) +](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen-audio-3.0-tts-plus) -华北2（北京） +华北2（北京）新加坡 -模型 ID`cosyvoice-v3.5-plus` +模型 ID`qwen-audio-3.0-tts-plus` Request URL`wss://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api-ws/v1/inference` API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) +模型 ID`qwen-audio-3.0-tts-plus` + +Request URL`wss://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api-ws/v1/inference` + +API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) + [ MiniMax/speech-2.8-hd @@ -1426,24 +1274,18 @@ API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=mod [ -qwen3.5-omni-plus-realtime +qwen-audio-3.0-realtime-plus -](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.5-omni-plus-realtime) +](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen-audio-3.0-realtime-plus) -华北2（北京）新加坡 +华北2（北京） -模型 ID`qwen3.5-omni-plus-realtime` +模型 ID`qwen-audio-3.0-realtime-plus` Request URL`wss://[{WorkspaceId}](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management).cn-beijing.maas.aliyuncs.com/api-ws/v1/realtime` API Key[获取↗](https://bailian.console.aliyun.com/cn-beijing?tab=model#/api-key) -模型 ID`qwen3.5-omni-plus-realtime` - -Request URL`wss://[{WorkspaceId}](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=globalset#/efm/business_management).ap-southeast-1.maas.aliyuncs.com/api-ws/v1/realtime` - -API Key[获取↗](https://modelstudio.console.aliyun.com/ap-southeast-1?tab=model#/api-key) - [ qwen3.5-omni-plus diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/rate-limit.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/rate-limit.md index 8e375ee1..cbad05cc 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/rate-limit.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/rate-limit.md @@ -171,8 +171,6 @@ **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -187,48 +185,36 @@ qwen3.7-max > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 30,000 5,000,000 qwen3.7-max-2026-06-08 -中国内地 - 600 1,000,000 qwen3.7-max-2026-05-20 -中国内地 - 600 1,000,000 qwen3.7-max-preview -中国内地 - 60 500,000 qwen3.7-max-2026-05-17 -中国内地 - 60 500,000 qwen3.6-max-preview -中国内地 - 600 1,000,000 @@ -237,32 +223,24 @@ qwen3-max > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 30,000 5,000,000 qwen3-max-2026-01-23 -中国内地 - 600 1,000,000 qwen3-max-2025-09-23 -中国内地 - 60 100,000 qwen3-max-preview -中国内地 - 600 1,000,000 @@ -271,24 +249,18 @@ qwen-max > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,200 1,000,000 qwen3.7-plus -中国内地 - 30,000 5,000,000 qwen3.7-plus-2026-05-26 -中国内地 - 600 1,000,000 @@ -297,15 +269,25 @@ qwen3.6-plus > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 30,000 5,000,000 qwen3.6-plus-2026-04-02 -中国内地 +600 + +1,000,000 + +qwen3.7-flash + +> 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 + +30,000 + +5,000,000 + +qwen3.7-flash-2026-07-15 600 @@ -315,16 +297,12 @@ qwen3.6-flash > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 30,000 10,000,000 qwen3.6-flash-2026-04-16 -中国内地 - 600 1,000,000 @@ -333,24 +311,18 @@ qwen3.5-plus > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 30,000 5,000,000 qwen3.5-plus-2026-04-20 -中国内地 - 600 1,000,000 qwen3.5-plus-2026-02-15 -中国内地 - 600 1,000,000 @@ -359,8 +331,6 @@ qwen-plus > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 30,000 5,000,000 @@ -369,24 +339,18 @@ qwen-plus-latest > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 15,000 1,200,000 qwen-plus-2025-12-01 -中国内地 - 120 1,000,000 qwen-plus-2025-09-11 -中国内地 - 60 1,000,000 @@ -395,8 +359,6 @@ qwen-plus-2025-07-28 （qwen-plus-0728） -中国内地 - 60 1,000,000 @@ -405,8 +367,6 @@ qwen-plus-2025-07-14 （qwen-plus-0714） -中国内地 - 60 100,000 @@ -415,8 +375,6 @@ qwen-plus-2025-04-28 （qwen-plus-0428） -中国内地 - 60 1,000,000 @@ -425,8 +383,6 @@ qwen-plus-2025-01-25 （qwen-plus-0125） -中国内地 - 60 150,000 @@ -435,8 +391,6 @@ qwen-plus-2025-01-12 （qwen-plus-0112） -中国内地 - 60 150,000 @@ -445,8 +399,6 @@ qwen-plus-2024-12-20 （qwen-plus-1220） -中国内地 - 60 150,000 @@ -455,16 +407,12 @@ qwen3.5-flash > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 30,000 10,000,000 qwen3.5-flash-2026-02-23 -中国内地 - 600 1,000,000 @@ -473,16 +421,12 @@ qwen-flash > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 30,000 10,000,000 qwen-flash-2025-07-28 -中国内地 - 60 1,000,000 @@ -491,8 +435,6 @@ qwen-turbo > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,200 5,000,000 @@ -501,8 +443,6 @@ qwq-plus > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 600 1,000,000 @@ -511,8 +451,6 @@ qwen-long > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,200 3,000,000 @@ -521,8 +459,6 @@ qwen-long-latest > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,200 60,000 @@ -531,8 +467,6 @@ qwen-long-2025-01-25 (qwen-long-0125) -中国内地 - 3 7,500 @@ -665,6 +599,14 @@ qwen3.6-flash-2026-04-16 1,000,000 +qwen3.6-flash-us + +美国 + +15,000 + +5,000,000 + qwen3.5-plus 全球 @@ -899,6 +841,22 @@ qwen3.6-plus-2026-04-02 1,000,000 +qwen3.7-flash + +国际 + +15,000 + +5,000,000 + +qwen3.7-flash-2026-07-15 + +国际 + +60 + +1,000,000 + qwen3.6-flash 国际 @@ -1393,8 +1351,6 @@ qwen3.6-flash-2026-04-16 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -1409,24 +1365,18 @@ qwen3-vl-plus > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 3,000 5,000,000 qwen3-vl-plus-2025-12-19 -中国内地 - 60 100,000 qwen3-vl-plus-2025-09-23 -中国内地 - 60 100,000 @@ -1435,24 +1385,18 @@ qwen3-vl-flash > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 3,000 5,000,000 qwen3-vl-flash-2026-01-22 -中国内地 - 60 100,000 qwen3-vl-flash-2025-10-15 -中国内地 - 60 100,000 @@ -1461,8 +1405,6 @@ qwen-vl-max > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,200 1,000,000 @@ -1471,24 +1413,18 @@ qwen-vl-plus > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,200 1,000,000 qvq-max -中国内地 - 60 100,000 qvq-plus -中国内地 - 60 100,000 @@ -1731,8 +1667,6 @@ qwen3-vl-flash-2025-10-15 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -1745,56 +1679,42 @@ qwen3-vl-flash-2025-10-15 qwen3.5-omni-plus -中国内地 - 60 100,000 qwen3.5-omni-plus-2026-03-15 -中国内地 - 60 100,000 qwen3.5-omni-flash -中国内地 - 60 100,000 qwen3.5-omni-flash-2026-03-15 -中国内地 - 60 100,000 qwen3-omni-flash -中国内地 - 60 100,000 qwen3-omni-flash-2025-12-01 -中国内地 - 60 100,000 qwen3-omni-flash-2025-09-15 -中国内地 - 60 100,000 @@ -1803,16 +1723,12 @@ qwen-omni-turbo > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 60 100,000 qwen-omni-turbo-latest -中国内地 - 60 100,000 @@ -1821,8 +1737,6 @@ qwen-omni-turbo-2025-03-26 （qwen-omni-turbo-0326） -中国内地 - 60 100,000 @@ -1831,8 +1745,6 @@ qwen-omni-turbo-2025-01-19 （qwen-omni-turbo-0119） -中国内地 - 60 100,000 @@ -1939,8 +1851,6 @@ qwen-omni-turbo-2025-03-26 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -1953,72 +1863,54 @@ qwen-omni-turbo-2025-03-26 qwen3.5-omni-plus-realtime -中国内地 - 60 100,000 qwen3.5-omni-plus-realtime-2026-03-15 -中国内地 - 60 100,000 qwen3.5-omni-flash-realtime -中国内地 - 60 100,000 qwen3.5-omni-flash-realtime-2026-03-15 -中国内地 - 60 100,000 qwen3-omni-flash-realtime -中国内地 - 60 100,000 qwen3-omni-flash-realtime-2025-12-01 -中国内地 - 60 100,000 qwen3-omni-flash-realtime-2025-09-15 -中国内地 - 60 100,000 qwen-omni-turbo-realtime-latest -中国内地 - 60 100,000 qwen-omni-turbo-realtime-2025-05-08 -中国内地 - 60 100,000 @@ -2125,8 +2017,6 @@ qwen-omni-turbo-realtime**\-**2025-05-08 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -2139,8 +2029,6 @@ qwen-omni-turbo-realtime**\-**2025-05-08 qwen3.5-ocr -中国内地 - 6,000 30,000,000 @@ -2149,8 +2037,6 @@ qwen-vl-ocr > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 600 6,000,000 @@ -2159,40 +2045,30 @@ qwen-vl-ocr-latest > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - -1,200 +6,000 -6,000,000 +30,000,000 qwen-vl-ocr-2025-11-20 -中国内地 - -1,200 +6,000 -6,000,000 +30,000,000 qwen-vl-ocr-2025-08-28 -中国内地 - 600 6,000,000 qwen-vl-ocr-2025-04-13 -中国内地 - 600 6,000,000 qwen-vl-ocr-2024-10-28 -中国内地 - 600 6,000,000 @@ -2299,8 +2175,6 @@ qwen-vl-ocr-2025-11-20 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -2313,16 +2187,12 @@ qwen-vl-ocr-2025-11-20 qwen-audio-turbo -中国内地 - 120 100,000 qwen-audio-turbo-latest -中国内地 - 60 100,000 @@ -2333,8 +2203,6 @@ qwen-audio-turbo-latest **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -2347,16 +2215,12 @@ qwen-audio-turbo-latest qwen-math-plus -中国内地 - 1,200 1,000,000 qwen-math-plus-latest -中国内地 - 1,200 1,000,000 @@ -2365,8 +2229,6 @@ qwen-math-plus-2024-09-19 （qwen-math-plus-0919） -中国内地 - 60 100,000 @@ -2375,16 +2237,12 @@ qwen-math-plus-2024-08-16 （qwen-math-plus-0816） -中国内地 - 10 20,000 qwen-math-turbo -中国内地 - 1200 1,000,000 @@ -2395,8 +2253,6 @@ qwen-math-turbo **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -2409,56 +2265,42 @@ qwen-math-turbo qwen3-coder-plus -中国内地 - 5,000 5,000,000 qwen3-coder-plus-2025-09-23 -中国内地 - 60 1,000,000 qwen3-coder-plus-2025-07-22 -中国内地 - 60 1,000,000 qwen3-coder-flash -中国内地 - 5,000 5,000,000 qwen3-coder-flash-2025-07-28 -中国内地 - 60 1,000,000 qwen-coder-plus -中国内地 - 1,200 1,000,000 qwen-coder-turbo -中国内地 - 1,200 1,000,000 @@ -2637,8 +2479,6 @@ qwen3-coder-flash-2025-07-28 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -2651,32 +2491,24 @@ qwen3-coder-flash-2025-07-28 qwen-mt-plus -中国内地 - 60 25,000 qwen-mt-flash -中国内地 - 60 35,000 qwen-mt-lite -中国内地 - 60 100,000 qwen-mt-turbo -中国内地 - 60 35,000 @@ -2823,8 +2655,6 @@ qwen-mt-lite **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -2837,8 +2667,6 @@ qwen-mt-lite qwen-doc-turbo -中国内地 - 600 3,000,000 @@ -2849,8 +2677,6 @@ qwen-doc-turbo **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -2863,8 +2689,6 @@ qwen-doc-turbo qwen-deep-research -中国内地 - 120 1,200,000 @@ -2875,8 +2699,6 @@ qwen-deep-research **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -2889,16 +2711,12 @@ qwen-deep-research tongyi-xiaomi-analysis-flash -中国内地 - 600 1,000,000 tongyi-xiaomi-analysis-pro -中国内地 - 600 1,000,000 @@ -2911,8 +2729,6 @@ tongyi-xiaomi-analysis-pro **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -2925,136 +2741,102 @@ tongyi-xiaomi-analysis-pro qwen3.6-35b-a3b -中国内地 - 600 1,000,000 qwen3.6-27b -中国内地 - 600 1,000,000 qwen3.5-397b-a17b -中国内地 - 600 1,000,000 qwen3.5-122b-a10b -中国内地 - 600 1,000,000 qwen3.5-27b -中国内地 - 600 1,000,000 qwen3.5-35b-a3b -中国内地 - 600 1,000,000 qwen3-next-80b-a3b-thinking -中国内地 - 600 1,000,000 qwen3-next-80b-a3b-instruct -中国内地 - 600 1,000,000 qwen3-235b-a22b-thinking-2507 -中国内地 - 600 1,000,000 qwen3-235b-a22b-instruct-2507 -中国内地 - 600 1,000,000 qwen3-30b-a3b-thinking-2507 -中国内地 - 600 1,000,000 qwen3-30b-a3b-instruct-2507 -中国内地 - 600 1,000,000 qwen3-235b-a22b -中国内地 - 600 1,000,000 qwen3-30b-a3b -中国内地 - 600 1,000,000 qwen3-32b -中国内地 - 2400 1,000,000 qwen3-14b -中国内地 - 600 1,000,000 qwen3-8b -中国内地 - 600 1,000,000 @@ -3505,8 +3287,6 @@ qwen3-8b **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -3519,64 +3299,48 @@ qwen3-8b qwen3-vl-32b-thinking -中国内地 - 600 1,000,000 qwen3-vl-32b-instruct -中国内地 - 600 1,000,000 qwen3-vl-30b-a3b-thinking -中国内地 - 600 1,000,000 qwen3-vl-30b-a3b-instruct -中国内地 - 600 1,000,000 qwen3-vl-8b-thinking -中国内地 - 600 1,000,000 qwen3-vl-8b-instruct -中国内地 - 600 1,000,000 qwen3-vl-235b-a22b-thinking -中国内地 - 60 100,000 qwen3-vl-235b-a22b-instruct -中国内地 - 60 100,000 @@ -3827,8 +3591,6 @@ qwen3-vl-8b-instruct **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -3841,8 +3603,6 @@ qwen3-vl-8b-instruct qwen2.5-omni-7b -中国内地 - 60 100,000 @@ -3877,8 +3637,6 @@ qwen2.5-omni-7b **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -3891,8 +3649,6 @@ qwen2.5-omni-7b qwen3-omni-30b-a3b-captioner -中国内地 - 60 100,000 @@ -3927,8 +3683,6 @@ qwen3-omni-30b-a3b-captioner **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -3945,8 +3699,6 @@ qwen3-omni-30b-a3b-captioner **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -3959,24 +3711,18 @@ qwen3-omni-30b-a3b-captioner qwen3-coder-next -中国内地 - 600 1,000,000 qwen3-coder-480b-a35b-instruct -中国内地 - 600 1,000,000 qwen3-coder-30b-a3b-instruct -中国内地 - 600 1,000,000 @@ -4101,8 +3847,6 @@ qwen3-coder-next **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -4115,16 +3859,12 @@ qwen3-coder-next deepseek-v4-pro -中国内地 - 15,000 1,200,000 deepseek-v4-flash -中国内地 - 15,000 1,200,000 @@ -4133,32 +3873,24 @@ deepseek-v3.2 > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 15,000 1,200,000 deepseek-v3.2-exp -中国内地 - 15,000 1,200,000 deepseek-v3.1 -中国内地 - 15,000 1,200,000 deepseek-r1-0528 -中国内地 - 60 100,000 @@ -4167,8 +3899,6 @@ deepseek-r1 > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 15,000 1,200,000 @@ -4177,56 +3907,42 @@ deepseek-v3 > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 15,000 1,200,000 deepseek-r1-distill-qwen-7b -中国内地 - 15,000 1,200,000 deepseek-r1-distill-qwen-14b -中国内地 - 15,000 1,200,000 deepseek-r1-distill-qwen-32b -中国内地 - 15,000 1,200,000 deepseek-r1-distill-qwen-1.5b -中国内地 - 60 100,000 deepseek-r1-distill-llama-8b -中国内地 - 60 100,000 deepseek-r1-distill-llama-70b -中国内地 - 60 100,000 @@ -4405,8 +4121,6 @@ deepseek-v4-flash **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -4419,32 +4133,24 @@ deepseek-v4-flash siliconflow/deepseek-v3.2 -中国内地 - 500 500,000 siliconflow/deepseek-v3.1-terminus -中国内地 - 500 500,000 siliconflow/deepseek-r1-0528 -中国内地 - 500 500,000 siliconflow/deepseek-v3-0324 -中国内地 - 500 500,000 @@ -4455,8 +4161,6 @@ siliconflow/deepseek-v3-0324 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -4469,40 +4173,30 @@ siliconflow/deepseek-v3-0324 vanchin/deepseek-v3.2-think -中国内地 - 30 600,000 vanchin/deepseek-v3.1-terminus -中国内地 - 500 1,000,000 vanchin/deepseek-r1 -中国内地 - 500 1,000,000 vanchin/deepseek-v3 -中国内地 - 500 1,000,000 vanchin/deepseek-ocr -中国内地 - 500 1,000,000 @@ -4513,8 +4207,6 @@ vanchin/deepseek-ocr **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -4527,40 +4219,30 @@ vanchin/deepseek-ocr kimi-k2.7-code -中国内地 - 500 1,000,000 kimi-k2.6 -中国内地 - 500 1,000,000 kimi-k2.5 -中国内地 - 500 1,000,000 kimi-k2-thinking -中国内地 - 500 1,000,000 Moonshot-Kimi-K2-Instruct -中国内地 - 500 1,000,000 @@ -4683,8 +4365,6 @@ kimi-k2.7-code **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -4695,38 +4375,30 @@ kimi-k2.7-code > **含输入与输出Token** -kimi/kimi-k2.7-code-highspeed - -中国内地 +kimi/kimi-k3 500 -> 同一个阿里云百炼API Key 下，在 4 个模型中共享 500 RPM 限流配额。即这 4 个模型的每分钟请求总数加起来不能超过 500。 +> 同一个阿里云百炼API Key 下，在 5 个模型中共享 500 RPM 限流配额。即这 5 个模型的每分钟请求总数加起来不能超过 500。 3,000,000 -> 同一个阿里云百炼API Key 下，在 4 个模型中共享 3000000 TPM 限流配额。即这 4 个模型的每分钟 Token 消耗总数加起来不能超过 3000000。 +> 同一个阿里云百炼API Key 下，在 5 个模型中共享 3000000 TPM 限流配额。即这 5 个模型的每分钟 Token 消耗总数加起来不能超过 3000000。 -kimi/kimi-k2.7-code +kimi/kimi-k2.7-code-highspeed -中国内地 +kimi/kimi-k2.7-code kimi/kimi-k2.6 -中国内地 - kimi/kimi-k2.5 -中国内地 - ### **GLM** ## **华北2（北京）** **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -4739,56 +4411,42 @@ kimi/kimi-k2.5 glm-5.2 -中国内地 - 500 2,000,000 glm-5.1 -中国内地 - 500 1,000,000 glm-5 -中国内地 - 500 1,000,000 glm-4.7 -中国内地 - 500 1,000,000 glm-4.6 -中国内地 - 60 1,000,000 glm-4.5 -中国内地 - 60 1,000,000 glm-4.5-air -中国内地 - 60 1,000,000 @@ -4907,7 +4565,15 @@ glm-5.1 glm-5.2 -全球 +国际 + +500 + +1,000,000 + +glm-5.1 + +国际 500 @@ -4919,8 +4585,6 @@ glm-5.2 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -4933,24 +4597,18 @@ glm-5.2 ZHIPU/GLM-5.2 -中国内地 - 200 3,000,000 ZHIPU/GLM-5.1 -中国内地 - 200 -10,000,000 +3,000,000 ZHIPU/GLM-5 -中国内地 - 200 3,000,000 @@ -4961,8 +4619,6 @@ ZHIPU/GLM-5 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -4975,16 +4631,12 @@ ZHIPU/GLM-5 MiniMax-M2.5 -中国内地 - 500 1,000,000 MiniMax-M2.1 -中国内地 - 500 1,000,000 @@ -4995,8 +4647,6 @@ MiniMax-M2.1 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -5009,32 +4659,24 @@ MiniMax-M2.1 MiniMax/MiniMax-M3 -中国内地 - 500 20,000,000 MiniMax/MiniMax-M2.7 -中国内地 - 500 20,000,000 MiniMax/MiniMax-M2.5 -中国内地 - 500 20,000,000 MiniMax/MiniMax-M2.1 -中国内地 - 500 20,000,000 @@ -5045,8 +4687,6 @@ MiniMax/MiniMax-M2.1 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -5059,8 +4699,6 @@ MiniMax/MiniMax-M2.1 xiaomi/mimo-v2.5-pro -中国内地 - 100 10,000,000 @@ -5071,8 +4709,6 @@ xiaomi/mimo-v2.5-pro **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -5085,8 +4721,6 @@ xiaomi/mimo-v2.5-pro stepfun/step-3.7-flash -中国内地 - 500 20,000,000 @@ -5099,17 +4733,19 @@ stepfun/step-3.7-flash **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **任务下发接口调用限制** **同时处理中任务数量（并发数）** -qwen-image-2.0-pro +qwen-image-3.0-pro + +1 次/分钟 + +同步接口无限制 -中国内地 +qwen-image-2.0-pro 2 次/分钟 @@ -5117,136 +4753,102 @@ qwen-image-2.0-pro qwen-image-2.0-pro-2026-06-22 -中国内地 - 2 次/分钟 同步接口无限制 qwen-image-2.0-pro-2026-04-22 -中国内地 - 2 次/分钟 同步接口无限制 qwen-image-2.0-pro-2026-03-03 -中国内地 - 2 次/分钟 同步接口无限制 qwen-image-2.0 -中国内地 - 2 次/秒 同步接口无限制 qwen-image-2.0-2026-03-03 -中国内地 - 2 次/秒 同步接口无限制 qwen-image-max -中国内地 - 2 次/分钟 同步接口无限制 qwen-image-max-2025-12-30 -中国内地 - 2 次/分钟 同步接口无限制 qwen-image-plus -中国内地 - 2 次/秒 同步接口无限制 / 异步接口 2 qwen-image-plus-2026-01-09 -中国内地 - 2 次/秒 同步接口无限制 qwen-image -中国内地 - 2 次/秒 同步接口无限制 / 异步接口 2 qwen-image-edit-max -中国内地 - 2 次/分钟 同步接口无限制 qwen-image-edit-max-2026-01-16 -中国内地 - 2 次/分钟 同步接口无限制 qwen-image-edit-plus -中国内地 - 2 次/秒 同步接口无限制 qwen-image-edit-plus-2025-12-15 -中国内地 - 2 次/秒 同步接口无限制 qwen-image-edit-plus-2025-10-30 -中国内地 - 2 次/秒 同步接口无限制 qwen-image-edit -中国内地 - 2 次/秒 同步接口无限制 qwen-mt-image -中国内地 - 1 次/秒 2 @@ -5263,6 +4865,14 @@ qwen-mt-image **同时处理中任务数量（并发数）** +qwen-image-3.0-pro + +国际 + +1 次/分钟 + +同步接口无限制 + qwen-image-2.0-pro 国际 @@ -5273,7 +4883,7 @@ qwen-image-2.0-pro qwen-image-2.0-pro-2026-06-22 -中国内地 +国际 2 次/分钟 @@ -5405,8 +5015,6 @@ qwen-image-edit **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每秒钟任务下发接口RPS限制** @@ -5415,8 +5023,6 @@ qwen-image-edit z-image-turbo -中国内地 - 2 同步接口无限制 @@ -5447,8 +5053,6 @@ z-image-turbo **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每秒钟任务下发接口RPS限制** @@ -5457,120 +5061,90 @@ z-image-turbo wan2.7-image-pro -中国内地 - 5 5 wan2.7-image -中国内地 - 5 5 wan2.6-image -中国内地 - 5 5 wan2.6-t2i -中国内地 - 1 5 wan2.5-t2i-preview -中国内地 - 5 5 wan2.2-t2i-plus -中国内地 - 2 2 wan2.2-t2i-flash -中国内地 - 2 2 wanx2.1-t2i-plus -中国内地 - 2 2 wanx2.1-t2i-turbo -中国内地 - 2 2 wanx2.0-t2i-turbo -中国内地 - 2 2 wan2.5-i2i-preview -中国内地 - 5 5 wanx2.1-imageedit -中国内地 - 2 2 wanx-v1 -中国内地 - 2 1 wanx-x-painting -中国内地 - 2 1 wanx-sketch-to-image-lite -中国内地 - 2 1 @@ -5729,8 +5303,6 @@ wan2.6-image **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每秒钟任务下发接口RPS限制** @@ -5739,72 +5311,54 @@ wan2.6-image shoemodel-v1 -中国内地 - 2 1 wanx-virtualmodel -中国内地 - 2 1 wanx-style-repaint-v1 -中国内地 - 2 2 wanx-poster-generation-v1 -中国内地 - 2 1 virtualmodel-v2 -中国内地 - 2 1 wanx-background-generation-v2 -中国内地 - 2 1 image-instance-segmentation -中国内地 - 2 1 image-erase-completion -中国内地 - 2 1 image-out-painting -中国内地 - 2 10 @@ -5815,8 +5369,6 @@ image-out-painting **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **作业提交接口RPS限制** @@ -5825,24 +5377,18 @@ image-out-painting facechain-facedetect -中国内地 - 5 同步接口无限制 facechain-finetune -中国内地 - 1 1 facechain-generation -中国内地 - 2 1 @@ -5853,8 +5399,6 @@ facechain-generation **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **作业提交接口RPS限制** @@ -5863,16 +5407,12 @@ facechain-generation wordart-texture -中国内地 - 2 1 wordart-semantic -中国内地 - 2 1 @@ -5883,8 +5423,6 @@ wordart-semantic **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **作业提交接口RPS限制** @@ -5893,32 +5431,24 @@ wordart-semantic aitryon -中国内地 - 10 5 aitryon-plus -中国内地 - 10 5 aitryon-parsing-v1 -中国内地 - 10 同步接口无限制 aitryon-refiner -中国内地 - 10 5 @@ -5931,8 +5461,6 @@ aitryon-refiner **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每秒钟任务下发接口RPS限制** @@ -5941,8 +5469,6 @@ aitryon-refiner kling/kling-v3-omni-image-generation -中国内地 - 5 10 @@ -5951,16 +5477,12 @@ kling/kling-v3-omni-image-generation kling/kling-v3-image-generation -中国内地 - ### **Vidu系列** ## **华北2（北京）** **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每分钟请求数RPM限制** @@ -5969,8 +5491,6 @@ kling/kling-v3-image-generation vidu/vidu-image\_reference2image -中国内地 - 300 5 @@ -5979,36 +5499,24 @@ vidu/vidu-image\_reference2image vidu/viduq3-fast\_reference2image -中国内地 - vidu/viduq2-pro\_reference2image -中国内地 - vidu/viduq2-fast\_reference2image -中国内地 - ## **音乐生成** ## **华北2（北京）** **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** fun-music-preview -中国内地 - 180 fun-music-v1 -中国内地 - 180 ## **语音对话** @@ -6019,8 +5527,6 @@ fun-music-v1 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -6033,16 +5539,12 @@ fun-music-v1 qwen-audio-3.0-realtime-plus -中国内地 - 60 100,000 qwen-audio-3.0-realtime-flash -中国内地 - 60 100,000 @@ -6055,20 +5557,14 @@ qwen-audio-3.0-realtime-flash **模型名称** -**服务部署范围** - **提交作业接口RPS限制** qwen-audio-3.0-tts-plus -中国内地 - 3 qwen-audio-3.0-tts-flash -中国内地 - 3 #### 新加坡 @@ -6099,82 +5595,58 @@ qwen-audio-3.0-tts-flash **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-tts-instruct-flash -中国内地 - 180 qwen3-tts-instruct-flash-2026-01-26 -中国内地 - 180 ##### **千问3-TTS-VD** **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-tts-vd-2026-01-26 -中国内地 - 180 ##### **千问3-TTS-VC** **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-tts-vc-2026-01-22 -中国内地 - 180 ##### 千问3-TTS-Flash **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-tts-flash -中国内地 - 180 qwen3-tts-flash-2025-11-27 -中国内地 - 180 qwen3-tts-flash-2025-09-18 -中国内地 - 10 ##### 千问-TTS **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -6187,24 +5659,16 @@ qwen3-tts-flash-2025-09-18 qwen-tts -中国内地 - 10 100,000 qwen-tts-latest -中国内地 - qwen-tts-2025-05-22 -中国内地 - qwen-tts-2025-04-10 -中国内地 - #### 新加坡 ##### **千问3-TTS-Instruct-Flash** @@ -6289,90 +5753,62 @@ qwen3-tts-flash-2025-09-18 **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-tts-instruct-flash-realtime -中国内地 - 180 qwen3-tts-instruct-flash-realtime-2026-01-22 -中国内地 - 180 ##### 千问3-TTS-VD-Realtime **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-tts-vd-realtime-2026-01-15 -中国内地 - 180 qwen3-tts-vd-realtime-2025-12-16 -中国内地 - ##### 千问3-TTS-VC-Realtime **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-tts-vc-realtime-2026-01-15 -中国内地 - 180 qwen3-tts-vc-realtime-2025-11-27 -中国内地 - ##### 千问3-TTS-Flash-Realtime **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-tts-flash-realtime -中国内地 - 180 qwen3-tts-flash-realtime-2025-11-27 -中国内地 - 180 qwen3-tts-flash-realtime-2025-09-18 -中国内地 - 10 ##### 千问-TTS-Realtime **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -6385,20 +5821,14 @@ qwen3-tts-flash-realtime-2025-09-18 qwen-tts-realtime -中国内地 - 10 100,000 qwen-tts-realtime-latest -中国内地 - qwen-tts-realtime-2025-07-15 -中国内地 - #### 新加坡 ##### **千问3-TTS-Instruct-Flash-Realtime** @@ -6489,14 +5919,10 @@ qwen3-tts-flash-realtime-2025-09-18 **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen-voice-enrollment -中国内地 - 180 #### 新加坡 @@ -6519,14 +5945,10 @@ qwen-voice-enrollment **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen-voice-design -中国内地 - 180 #### 新加坡 @@ -6549,36 +5971,22 @@ qwen-voice-design **模型名称** -**服务部署范围** - **提交作业接口RPS限制** cosyvoice-v3.5-plus -中国内地 - 3 cosyvoice-v3.5-flash -中国内地 - cosyvoice-v3-plus -中国内地 - cosyvoice-v3-flash -中国内地 - cosyvoice-v2 -中国内地 - cosyvoice-v1 -中国内地 - #### 新加坡 **模型名称** @@ -6605,14 +6013,10 @@ Qwen-Audio-TTS/CosyVoice声音复刻/设计共用一个模型，共用限流额 **模型名称** -**服务部署范围** - **提交作业接口RPS限制** voice-enrollment -中国内地 - 10 #### 新加坡 @@ -6635,14 +6039,10 @@ voice-enrollment **模型服务** -**服务部署范围** - **提交作业接口RPS限制** Sambert系列模型 -中国内地 - 20 ## **语音合成（文本转语音）-第三方模型** @@ -6653,8 +6053,6 @@ Sambert系列模型 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每分钟调用次数（RPM）** @@ -6665,32 +6063,24 @@ Sambert系列模型 MiniMax/speech-2.8-hd -中国内地 - 20 20,000 MiniMax/speech-02-hd -中国内地 - 20 20,000 MiniMax/speech-2.8-turbo -中国内地 - 20 20,000 MiniMax/speech-02-turbo -中国内地 - 20 20,000 @@ -6703,8 +6093,6 @@ MiniMax/speech-02-turbo **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -6717,16 +6105,12 @@ MiniMax/speech-02-turbo qwen3-livetranslate-flash -中国内地 - 100 100,000 qwen3-livetranslate-flash-2025-12-01 -中国内地 - #### 新加坡 **模型名称** @@ -6761,8 +6145,6 @@ qwen3-livetranslate-flash-2025-12-01 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -6775,24 +6157,16 @@ qwen3-livetranslate-flash-2025-12-01 qwen3.5-livetranslate-flash-realtime -中国内地 - 10 100,000 qwen3.5-livetranslate-flash-realtime-2026-05-19 -中国内地 - qwen3-livetranslate-flash-realtime -中国内地 - qwen3-livetranslate-flash-realtime-2025-09-22 -中国内地 - #### 新加坡 **模型名称** @@ -6837,42 +6211,28 @@ qwen3-livetranslate-flash-realtime-2025-09-22 **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-asr-flash-filetrans -中国内地 - 100 qwen3-asr-flash-filetrans-2025-11-17 -中国内地 - ##### **千问3-ASR-Flash** **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** qwen3-asr-flash -中国内地 - 100 qwen3-asr-flash-2026-02-10 -中国内地 - qwen3-asr-flash-2025-09-08 -中国内地 - #### 新加坡 ##### 千问3-ASR-Flash-Filetrans @@ -6939,24 +6299,16 @@ qwen3-asr-flash-2025-09-08-us **模型名称** -**服务部署范围** - **每秒钟调用次数（RPS）** qwen3-asr-flash-realtime -中国内地 - 20 qwen3-asr-flash-realtime-2026-02-10 -中国内地 - qwen3-asr-flash-realtime-2025-10-27 -中国内地 - #### 新加坡 **模型名称** @@ -6985,36 +6337,22 @@ qwen3-asr-flash-realtime-2025-10-27 **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** fun-asr -中国内地 - 600 fun-asr-2025-11-07 -中国内地 - fun-asr-2025-08-25 -中国内地 - fun-asr-mtl -中国内地 - fun-asr-mtl-2025-08-25 -中国内地 - fun-asr-flash-2026-06-15 -中国内地 - #### 新加坡 **模型名称** @@ -7065,36 +6403,22 @@ fun-asr-flash-2026-06-15 **模型名称** -**服务部署范围** - **提交作业接口RPS限制** fun-asr-realtime -中国内地 - 20 fun-asr-realtime-2026-02-28 -中国内地 - fun-asr-realtime-2025-11-07 -中国内地 - fun-asr-realtime-2025-09-15 -中国内地 - fun-asr-flash-8k-realtime -中国内地 - fun-asr-flash-8k-realtime-2026-01-28 -中国内地 - #### 新加坡 **模型名称** @@ -7119,44 +6443,28 @@ fun-asr-realtime-2025-11-07 **模型名称** -**服务部署范围** - **提交作业接口RPS限制** paraformer-realtime-v2 -中国内地 - 20 paraformer-realtime-v1 -中国内地 - paraformer-realtime-8k-v2 -中国内地 - paraformer-realtime-8k-v1 -中国内地 - **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** paraformer-v2 -中国内地 - 1,200 **模型名称** -**服务部署范围** - **每分钟调用次数（RPM）** **每分钟消耗Token数（TPM）** @@ -7165,40 +6473,30 @@ paraformer-v2 paraformer-v1 -中国内地 - 600 6,000,000 paraformer-mtl-v1 -中国内地 - 600 6,000,000 **模型名称** -**服务部署范围** - **提交作业接口RPS限制** **同时处理中任务数量（并发数）** paraformer-8k-v2 -中国内地 - 20 100 paraformer-8k-v1 -中国内地 - 10 500 @@ -7211,8 +6509,6 @@ paraformer-8k-v1 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每秒钟任务下发接口RPS限制** @@ -7221,57 +6517,43 @@ paraformer-8k-v1 happyhorse-1.1-t2v -中国内地 - -10 +5 5 happyhorse-1.1-i2v -中国内地 - -10 +5 5 happyhorse-1.1-r2v -中国内地 - -10 +5 5 happyhorse-1.0-t2v -中国内地 - -10 +5 5 happyhorse-1.0-i2v -中国内地 - -10 +5 5 happyhorse-1.0-r2v -中国内地 - -10 +5 5 happyhorse-1.0-video-edit -中国内地 - -10 +5 5 @@ -7291,7 +6573,7 @@ happyhorse-1.1-t2v 全球 -10 +5 5 @@ -7299,7 +6581,7 @@ happyhorse-1.1-i2v 全球 -10 +5 5 @@ -7307,7 +6589,7 @@ happyhorse-1.1-r2v 全球 -10 +5 5 @@ -7315,7 +6597,7 @@ happyhorse-1.0-t2v 全球 -10 +5 5 @@ -7323,7 +6605,7 @@ happyhorse-1.0-i2v 全球 -10 +5 5 @@ -7331,7 +6613,7 @@ happyhorse-1.0-r2v 全球 -10 +5 5 @@ -7339,7 +6621,7 @@ happyhorse-1.0-video-edit 全球 -10 +5 5 @@ -7359,7 +6641,7 @@ happyhorse-1.1-t2v 国际 -10 +5 5 @@ -7367,7 +6649,7 @@ happyhorse-1.1-i2v 国际 -10 +5 5 @@ -7375,7 +6657,7 @@ happyhorse-1.1-r2v 国际 -10 +5 5 @@ -7383,7 +6665,7 @@ happyhorse-1.0-t2v 国际 -10 +5 5 @@ -7391,7 +6673,7 @@ happyhorse-1.0-i2v 国际 -10 +5 5 @@ -7399,7 +6681,7 @@ happyhorse-1.0-r2v 国际 -10 +5 5 @@ -7407,7 +6689,7 @@ happyhorse-1.0-video-edit 国际 -10 +5 5 @@ -7427,7 +6709,7 @@ happyhorse-1.1-t2v 全球 -10 +5 5 @@ -7435,7 +6717,7 @@ happyhorse-1.1-i2v 全球 -10 +5 5 @@ -7443,7 +6725,7 @@ happyhorse-1.1-r2v 全球 -10 +5 5 @@ -7451,7 +6733,7 @@ happyhorse-1.0-t2v 全球 -10 +5 5 @@ -7459,7 +6741,7 @@ happyhorse-1.0-i2v 全球 -10 +5 5 @@ -7467,7 +6749,7 @@ happyhorse-1.0-r2v 全球 -10 +5 5 @@ -7475,7 +6757,51 @@ happyhorse-1.0-video-edit 全球 -10 +5 + +5 + +## **日本（东京）** + +**模型名称** + +**服务部署范围** + +**限流值（任何一个值超出即触发限流）** + +**任务提交接口RPS限制** + +**正在处理中的任务数（并发）** + +happyhorse-1.1-t2v + +全球 + +5 + +5 + +happyhorse-1.1-i2v + +全球 + +5 + +5 + +happyhorse-1.1-r2v + +全球 + +5 + +5 + +happyhorse-1.0-video-edit + +全球 + +5 5 @@ -7485,8 +6811,6 @@ happyhorse-1.0-video-edit **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每秒钟任务下发接口RPS限制** @@ -7495,232 +6819,174 @@ happyhorse-1.0-video-edit wan2.7-r2v-2026-06-12 -中国内地 - 5 5 wan2.7-t2v-2026-06-12 -中国内地 - 5 5 wan2.7-t2v-2026-04-25 -中国内地 - 5 5 wan2.7-t2v -中国内地 - 5 5 wan2.6-t2v -中国内地 - 5 5 wan2.5-t2v-preview -中国内地 - 5 5 wan2.2-t2v-plus -中国内地 - 2 2 wanx2.1-t2v-turbo -中国内地 - 2 2 wanx2.1-t2v-plus -中国内地 - 2 2 wan2.7-i2v-2026-04-25 -中国内地 - 5 5 wan2.7-i2v -中国内地 - 5 5 wan2.6-i2v-flash -中国内地 - 5 5 wan2.6-i2v -中国内地 - 5 5 wan2.5-i2v-preview -中国内地 - 5 5 wan2.2-i2v-flash -中国内地 - 2 2 wan2.2-i2v-plus -中国内地 - 2 2 wanx2.1-i2v-turbo -中国内地 - 2 2 wanx2.1-i2v-plus -中国内地 - 2 2 wan2.2-kf2v-flash -中国内地 - 2 2 wanx2.1-kf2v-plus -中国内地 - 2 2 wanx2.1-vace-plus -中国内地 - 2 2 wan2.7-videoedit -中国内地 - 5 5 wan2.7-r2v -中国内地 - 5 5 wan2.6-r2v-flash -中国内地 - 5 5 wan2.6-r2v -中国内地 - 5 5 wan2.2-s2v-detect -中国内地 - 5 同步接口无限制 wan2.2-s2v -中国内地 - 5 1 wan2.2-animate-move -中国内地 - 5 1 wan2.2-animate-mix -中国内地 - 5 1 @@ -8031,24 +7297,18 @@ wan2.6-r2v **模型名称** -**服务部署范围** - **任务下发接口RPS限制** **同时处理中任务数量** animate-anyone-detect-gen2 -中国内地 - 5 同步接口无限制 animate-anyone-template-gen2 -中国内地 - 5 1 @@ -8057,8 +7317,6 @@ animate-anyone-template-gen2 animate-anyone-gen2 -中国内地 - 5 1 @@ -8067,16 +7325,12 @@ animate-anyone-gen2 animate-anyone-detect -中国内地 - 5 1算力单元支持2并发 animate-anyone -中国内地 - 5 1算力单元支持1并发 @@ -8087,24 +7341,18 @@ animate-anyone **模型名称** -**服务部署范围** - **任务下发接口RPS限制** **同时处理中任务数量** emo-detect-v1 -中国内地 - 5 同步接口无限制 emo-v1 -中国内地 - 5 1 @@ -8117,24 +7365,18 @@ emo-v1 **模型名称** -**服务部署范围** - **任务下发接口RPS限制** **同时处理中任务数量** liveportrait-detect -中国内地 - 5 同步接口无限制 liveportrait -中国内地 - 5 1 @@ -8147,16 +7389,12 @@ liveportrait **模型名称** -**服务部署范围** - **任务下发接口RPS限制** **同时处理中任务数量** videoretalk -中国内地 - 1 1 @@ -8169,24 +7407,18 @@ videoretalk **模型名称** -**服务部署范围** - **任务下发接口RPS限制** **同时处理中任务数量** emoji-detect-v1 -中国内地 - 1 同步接口无限制 emoji-v1 -中国内地 - 1 1 @@ -8199,16 +7431,12 @@ emoji-v1 **模型名称** -**服务部署范围** - **任务下发接口RPS限制** **同时处理中任务数量** video-style-transform -中国内地 - 20 2 @@ -8223,8 +7451,6 @@ video-style-transform **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每分钟任务下发接口RPM限制** @@ -8233,8 +7459,6 @@ video-style-transform pixverse/pixverse-lipsync -中国内地 - 300 5 @@ -8243,20 +7467,14 @@ pixverse/pixverse-lipsync pixverse/pixverse-motioncontrol -中国内地 - 300 pixverse/pixverse-upscale -中国内地 - 300 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每秒钟任务下发接口RPS限制** @@ -8265,8 +7483,6 @@ pixverse/pixverse-upscale pixverse/pixverse-c1-t2v -中国内地 - 5 5 @@ -8275,20 +7491,12 @@ pixverse/pixverse-c1-t2v pixverse/pixverse-c1-it2v -中国内地 - pixverse/pixverse-c1-kf2v -中国内地 - pixverse/pixverse-c1-r2v -中国内地 - pixverse/pixverse-v6-t2v -中国内地 - 5 5 @@ -8297,20 +7505,12 @@ pixverse/pixverse-v6-t2v pixverse/pixverse-v6-it2v -中国内地 - pixverse/pixverse-v6-kf2v -中国内地 - pixverse/pixverse-v6-r2v -中国内地 - pixverse/pixverse-v5.6-t2v -中国内地 - 5 5 @@ -8319,24 +7519,16 @@ pixverse/pixverse-v5.6-t2v pixverse/pixverse-v5.6-it2v -中国内地 - pixverse/pixverse-v5.6-kf2v -中国内地 - pixverse/pixverse-v5.6-r2v -中国内地 - ### **可灵系列** ## **华北2（北京）** **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每秒钟任务下发接口RPS限制** @@ -8345,8 +7537,6 @@ pixverse/pixverse-v5.6-r2v kling/kling-v3-omni-video-generation -中国内地 - 5 10 @@ -8355,16 +7545,12 @@ kling/kling-v3-omni-video-generation kling/kling-v3-video-generation -中国内地 - ### **Vidu系列** ## **华北2（北京）** **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每秒钟任务下发接口RPS限制** @@ -8373,8 +7559,6 @@ kling/kling-v3-video-generation vidu/viduq3-ad\_reference2video -中国内地 - 5 5 @@ -8383,116 +7567,78 @@ vidu/viduq3-ad\_reference2video vidu/viduq3-drama\_reference2video -中国内地 - 5 vidu/viduq3-pro-fast\_img2video -中国内地 - 5 vidu/viduq3-turbo\_text2video -中国内地 - 5 vidu/viduq3-pro\_text2video -中国内地 - 5 vidu/viduq2\_text2video -中国内地 - 5 vidu/viduq3-turbo\_img2video -中国内地 - 5 vidu/viduq3-pro\_img2video -中国内地 - 5 vidu/viduq2-turbo\_img2video -中国内地 - 5 vidu/viduq2-pro\_img2video -中国内地 - 5 vidu/viduq2-pro-fast\_img2video -中国内地 - 5 vidu/viduq3-turbo\_start-end2video -中国内地 - 5 vidu/viduq3-pro\_start-end2video -中国内地 - 5 vidu/viduq2-turbo\_start-end2video -中国内地 - 5 vidu/viduq2-pro\_start-end2video -中国内地 - 5 vidu/viduq3-mix\_reference2video -中国内地 - 5 vidu/viduq3\_reference2video -中国内地 - 5 vidu/viduq3-turbo\_reference2video -中国内地 - 5 vidu/viduq2-pro\_reference2video -中国内地 - 5 vidu/viduq2\_reference2video -中国内地 - 5 ## **3D模型生成-第三方模型** @@ -8503,18 +7649,14 @@ vidu/viduq2\_reference2video **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** -**每秒钟任务下发接口RPS限制** +**每分钟任务下发接口RPM限制** **同时处理中任务数量（并发数）** Tripo/Tripo-H3.1 -中国内地 - 5 10 @@ -8523,8 +7665,6 @@ Tripo/Tripo-H3.1 Tripo/Tripo-P1.0 -中国内地 - 5 ## **向量模型** @@ -8535,8 +7675,6 @@ Tripo/Tripo-P1.0 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** **每分钟调用次数（RPM）** @@ -8547,8 +7685,6 @@ Tripo/Tripo-P1.0 qwen3.7-text-embedding -中国内地 - 1,800 1,000,000 @@ -8557,8 +7693,6 @@ text-embedding-v1 > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,800 1,200,000 @@ -8567,8 +7701,6 @@ text-embedding-v2 > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,800 1,200,000 @@ -8577,8 +7709,6 @@ text-embedding-v3 > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,800 1,200,000 @@ -8587,16 +7717,12 @@ text-embedding-v4 > 用[Batch API](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)调用服务时，不受限流限制。 -中国内地 - 1,800 1,200,000 text-embedding-async-v1 -中国内地 - 60 当前用户在系统通用文本向量异步作业排队中和运行中的作业数量不超过50个。 @@ -8605,8 +7731,6 @@ text-embedding-async-v1 text-embedding-async-v2 -中国内地 - 60 当前用户在系统通用文本向量异步作业排队中和运行中的作业数量不超过50个。 @@ -8651,8 +7775,6 @@ text-embedding-v3 **模型名称** -**服务部署范围** - **限流条件** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -8665,56 +7787,42 @@ text-embedding-v3 qwen3-vl-embedding -中国内地 - 2,400 1,200,000 qwen2.5-vl-embedding -中国内地 - 1,200 600,000 tongyi-embedding-vision-plus -中国内地 - 600 200,000 tongyi-embedding-vision-flash -中国内地 - 600 200,000 tongyi-embedding-vision-flash-2026-03-06 -中国内地 - 1,200 9,600,000 tongyi-embedding-vision-plus-2026-03-06 -中国内地 - 1,200 9,600,000 multimodal-embedding-v1 -中国内地 - 120 1,000,000 @@ -8727,8 +7835,6 @@ multimodal-embedding-v1 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -8741,24 +7847,18 @@ multimodal-embedding-v1 qwen3-rerank -中国内地 - 5,400 5,000,000,000 qwen3-vl-rerank -中国内地 - 600 9,000,000 gte-rerank-v2 -中国内地 - 5,040 4,980,000,000 @@ -8803,8 +7903,6 @@ gte-rerank-v2 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -8817,8 +7915,6 @@ gte-rerank-v2 farui-plus -中国内地 - 240 1,000,000 @@ -8829,8 +7925,6 @@ farui-plus **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -8843,8 +7937,6 @@ farui-plus tongyi-intent-detect-v3 -中国内地 - 1,200 1,000,000 @@ -8855,8 +7947,6 @@ tongyi-intent-detect-v3 **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -8869,24 +7959,18 @@ tongyi-intent-detect-v3 qwen-plus-character -中国内地 - 120 500,000 qwen-flash-character -中国内地 - 120 500,000 qwen-flash-character-2026-02-26 -中国内地 - 120 500,000 @@ -9009,8 +8093,6 @@ qwen-plus-character **模型名称** -**服务部署范围** - **限流条件（超出任一数值时触发限流）** > **以下为每分钟限流条件，服务可能按 RPS（RPM/60）与 TPS（TPM/60）限制** @@ -9023,16 +8105,12 @@ qwen-plus-character gui-plus -中国内地 - 80 540,000 gui-plus-2026-02-26 -中国内地 - 100 540,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/regions.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/regions.md index d9c3ff39..c6b3ce76 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/regions.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/regions.md @@ -1,9 +1,11 @@ -# 选择地域、服务部署范围和接入域名 +# 地域及接入域名 -调用百炼前先选择**地域**、**服务部署范围、接入域名：** +阿里云百炼提供多个地域的服务，各地域提供业务空间专属、Dashscope 等多种接入域名，并支持多种服务部署范围，以满足不同场景的接入需求。 + +## 地域概述 + +地域决定**接入点和数据存储位置**，就近选择可降低延迟。 -- 地域：决定**接入点和数据存储位置**，就近选择可降低延迟； - - 服务部署范围：决定**推理执行位置**，有数据合规需求选择特定地理边界的部署范围，无合规需求选择全球部署范围（推理资源池更大）； - 接入域名：影响**并发上限、超时等服务保障**，各地域具有独立的接入域名。 @@ -11,74 +13,48 @@ 一次完整的模型调用流程如下： -1. 应用经接入域名将请求发送到所选**地域**（如华北2-北京），请求数据存于该地域； +1. 应用经接入域名将请求发送到所选**地域**，请求数据存于该地域； 2. 接入地域将请求转发至**服务部署范围**内的推理节点完成计算（过程数据不持久化，传输全程加密）； 3. 推理结果回到接入地域存储，再响应给应用（用户静态数据始终存于所选地域）。 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7478182871/CAEQchiBgIDPq_PR9xkiIDJhZDdiNzAxMGFiODRhNmRiMDYxYjNjNGU2NTJkMDYw7466796_20260515102254.505.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/7931094871/CAEQchiBgIDh6dzL_BkiIDk2MmRhMjdmMGQyNTQ5N2ZiNzNlMGM4NjllMDBjMTUy7466796_20260515102254.505.svg) -## 选择地域和服务部署范围 - -按场景查表选地域和服务部署范围： - -**使用场景** +各地域支持的服务部署范围如下： **地域** **服务部署范围** -无数据驻留限制，追求更大推理资源池（跨多地域调度推理，自行确保跨境合法） - 美国（弗吉尼亚） -全球（任意可用节点，含中国境内及海外） +全球 -无数据驻留限制，追求更大推理资源池（跨多地域调度推理，自行确保跨境合法） +美国 德国（法兰克福） -全球（任意可用节点，含中国境内及海外） +全球 -无数据驻留限制，追求更大推理资源池（跨多地域调度推理，自行确保跨境合法） +欧盟 日本（东京） -全球（任意可用节点，含中国境内及海外） +全球 -要求数据不出中国内地 +日本 华北2（北京） -中国内地（限境内推理） - -要求数据不经过中国内地（会跨多地域调度推理，自行确保跨境合法） +中国内地 新加坡 -国际（除中国内地以外的全球节点） - -要求数据不出美国 - -美国（弗吉尼亚） - -美国（限境内推理） +国际 -要求数据不出欧盟 - -德国（法兰克福） - -欧盟（限境内推理） - -要求数据不出日本 - -日本（东京） - -日本（限境内推理） - -## 选择接入域名 +## 接入域名 百炼为模型推理 API 提供业务空间专属、Dashscope 和试用三种接入域名，适用于从试用体验到企业级生产的不同场景。推荐使用**业务空间专属域名**，各域名的核心差异如下： @@ -104,7 +80,7 @@ 推荐在生产环境中使用，具备更高并发承载能力与网络隔离性，保障大流量场景下的稳定、低延迟访问体验。 -存量业务兼容，建议[迁移至业务空间专属域名](#section-migrate-domain)。 +存量业务兼容，建议迁移至[业务空间专属域名](#section-migrate-domain)。 快速体验、功能验证，不建议用于生产环境。 @@ -247,13 +223,9 @@ HTTP、SSE 从 Dashscope 域名或试用域名迁移到业务空间专属域名只需两步，无需修改业务逻辑代码： -1. **获取业务空间专属域名**： +1. **获取业务空间专属域名**：在[业务空间管理](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management)页面，复制 **API Host** 列的内容。 - - 方式一：在[API Key 创建](https://bailian.console.aliyun.com/cn-beijing#/api-key)后的弹窗中，复制 **API Host** 。 - - - 方式二：在[业务空间管理](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management)页面，复制 **API Host** 列的内容。 - -2. **替换 Base URL 中的域名**：将原域名替换为业务空间专属域名。以华北2（北京）地域为例，`llm-xxx` 为业务空间 ID： +2. **替换请求地址中的域名**：将复制的 API Host（如 `llm-xxx.cn-beijing.maas.aliyuncs.com`）替换代码中原有的域名部分，以华北2（北京）地域为例，`llm-xxx` 为业务空间 ID： - OpenAI 兼容接口：从 `https://dashscope.aliyuncs.com/compatible-mode/v1` 替换为 `https://llm-xxx.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/what-is-model-studio.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/what-is-model-studio.md index 86dd07a9..9a83af9b 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/what-is-model-studio.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/get-started-with-models/what-is-model-studio.md @@ -223,7 +223,7 @@ A：百炼采用按量付费，本身**没有"自动扣费"开关**。以下措 - **清理计费资源：**删除不再使用的知识库；前往[模型部署](https://bailian.console.aliyun.com/?tab=model#/efm/model_deploy)页面，下线按算力时长计费的部署实例。 -- **开启"**[免费额度用完即停](https://help.aliyun.com/zh/model-studio/new-free-quota#d1cb80ac11i92)**"（仅限新用户且在免费额度有效期内）：**在模型详情页开启此开关，免费额度耗尽后服务自动停止，不会转为付费。仅适用于华北2（北京）地域（[中国内地服务部署范围](https://help.aliyun.com/zh/model-studio/regions/#080da663a75xh)），且须在免费额度有效期内。 +- **开启"**[免费额度用完即停](https://help.aliyun.com/zh/model-studio/new-free-quota#d1cb80ac11i92)**"（仅限新用户且在免费额度有效期内）：**在模型详情页开启此开关，免费额度耗尽后服务自动停止，不会转为付费。仅适用于华北2（北京）地域，且须在免费额度有效期内。 - **设置费用监控和预警：**查看 [账单详情](https://usercenter2.aliyun.com/finance/expense-report/expense-detail)和[模型监控](https://bailian.console.aliyun.com/?tab=model#/model-telemetry)，并设置[高额消费预警](https://usercenter2.aliyun.com/home/alarm-threshold)，及时发现异常消费。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-data-overview/model-log-backflow.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-data-overview/model-log-backflow.md new file mode 100644 index 00000000..ab920367 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-data-overview/model-log-backflow.md @@ -0,0 +1,330 @@ +# 日志回流 + +日志回流将 SLS（日志服务）推理日志转化为可用于模型微调或评测的结构化数据集。 + +## 功能概述 + +日志回流功能将 SLS（日志服务）中的推理日志数据回流到百炼平台，经过格式化处理后生成结构化数据集（JSONL 格式），可用于模型微调或模型评测。回流产出的是结构化数据，而非原始日志的直接副本。 + +### 支持范围 + +日志回流支持创建以下两类数据集： + +- **训练集**：训练场景为文本生成，训练方式支持 SFT（监督微调）、DPO（直接偏好优化）和 CPT（持续预训练）。 + +- **评测集**：支持文本生成场景。 + + +日志回流目前仅在**华北2（北京）**和**新加坡** Region 可用，其他 Region 不显示日志回流入口。单次回流上限为 10 万条，可多次回流到同一数据集的不同版本以积累更多数据。 + +日志回流支持**平台存储**（默认）和对象存储 **OSS 挂载**（需额外授权）两种存储方式。存储方式的差异和选择指引，请参见[创建日志回流数据集](#sec-create)。 + +回流生成的数据集可直接用于下游任务：训练集可用于[模型调优](https://help.aliyun.com/zh/model-studio/model-training-on-console#topic-2529531)，评测集用于模型评测。数据也支持后续进行[数据清洗](https://help.aliyun.com/zh/model-studio/data-processing#topic-2761062)。 + +## 开通并授权相关服务 + +使用日志回流前，请确认当前 Region 为华北2（北京）或新加坡（其他 Region 不显示日志回流入口），并在[模型监控](https://bailian.console.aliyun.com/#/model-telemetry)页面完成以下服务开通和权限授权。全部完成后授权配置抽屉自动关闭，进入日志回流表单。 + +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3463064871/p1086527.png) + +### 开启审计日志和推理日志 + +审计日志和推理日志各需完成三个步骤，共六个条件全部满足后才能使用日志回流。审计日志是推理日志的前置依赖，必须先完成审计日志的全部步骤。 + +**审计日志** + +1. 授权 SLS 服务关联角色：单击**立即授权**，授权 AliyunServiceRoleForSFMAccessSLS 角色。未授权时显示红色**未授权**标签。 + +2. 开通 SLS 日志服务：未开通时显示**未开通**状态和跳转链接，单击后前往 SLS 控制台完成开通。 + +3. 开启审计日志：单击**创建并开启审计日志**，系统创建 LogStore（日志库）实例并轮询等待就绪（最多 60 秒）。 + + +**推理日志** + +1. 授权推理日志的 SLS 服务关联角色。 + +2. 确认 SLS 日志服务已开通。 + +3. 开启推理日志：审计日志未开启时该按钮置灰，需先完成审计日志开启。 + + +**重要** + +开启必须按序：先审计日志，再推理日志。关闭必须反序：先关闭推理日志，再关闭审计日志。关闭日志后已有数据不可复原，操作前请确认不再需要日志数据。推理日志开启后 SLS 持续产生存储和读写费用，不再需要时应及时关闭。 + +### OSS 多角色授权（仅 OSS 挂载模式） + +选择 OSS 挂载存储方式时，需额外完成以下授权： + +1. 在弹出的授权弹窗中，勾选数据访问授权协议。 + +2. 单击**一键授权**，系统自动授权以下两个服务关联角色： + + - AliyunServiceRoleForAccessCusOss（OSS 写入角色） + + - AliyunServiceRoleForSFMDataHubOSSImport（DataHub 导入角色） + + +下表汇总了日志回流涉及的服务关联角色。 + +**角色名称** + +**用途** + +**授权时机** + +AliyunServiceRoleForSFMAccessSLS + +百炼访问 SLS 日志数据 + +审计日志和推理日志各授权一次 + +AliyunServiceRoleForAccessCusOss + +OSS 数据写入 + +选择 OSS 挂载存储时 + +AliyunServiceRoleForSFMDataHubOSSImport + +DataHub 数据导入 + +选择 OSS 挂载存储时 + +## 创建日志回流数据集 + +日志回流提供三个入口，均可进入配置表单创建数据集： + +### 模型监控列表页 + +在[**模型监控**列表页](https://bailian.console.aliyun.com/#/model-telemetry)顶部，单击**日志回流**。首次使用时先完成授权配置，授权通过后自动展示日志回流表单。 + +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3463064871/p1086529.png) + +### 模型监控详情页 + +在模型监控详情页的时间选择器区域，单击**日志回流**。从此入口进入时，表单自动预填当前页面的时间范围、API Key 和模型，且模型不可修改。 + +### 数据管理页 + +在**数据管理**页面新建数据集时，导入方式选择**日志回流**。该选项仅在训练集（文本生成 + SFT）或评测集（文本生成）时可见。 + +### 配置回流参数 + +进入日志回流表单后，按从上到下的顺序配置以下参数。部分参数有前置依赖：API Key 过滤需先选时间范围，模型选择需先选时间范围和 API Key。修改时间范围、数据类型、训练场景或训练方式会联动重置其他参数，建议严格按顺序填写。各参数取值说明见下表。 + +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3463064871/p1086595.png) + +**预估日志回流数据**：系统根据筛选条件显示预估的回流数据条数。超过 10 万条时红色警告，超出部分不会被回流。查询结果过多时，**确定**按钮禁用，需缩小筛选范围。 + +**说明** + +表单联动重置规则：修改时间范围会重置 API Key（回到**全部**）和模型选择（清空）；修改数据类型、训练场景或训练方式会重置存储位置和导入方式。 + +**重要** + +存储方式、数据类型和训练方式在创建后均不可更改，选择前请仔细确认。 + +**参数** + +**说明** + +**是否必填** + +**取值说明** + +回流位置 + +数据集的存储方式 + +是 + +平台存储（默认）或 OSS 挂载。评测集时 OSS 挂载禁用。创建后不可更改 + +数据集名称 + +数据集在列表中的显示名称 + +是 + +中文、英文、数字、下划线、斜杠、连字符，最大 50 字符。建议采用 功能场景\_模型名\_时间 格式命名。创建后不可修改 + +数据集描述 + +数据集用途补充说明 + +否 + +最大 200 字符 + +类型与格式 + +数据集用途类型 + +是 + +训练集或评测集。选择评测集时训练场景和训练方式隐藏。创建后不可更改 + +训练场景 + +训练场景类型（仅训练集显示） + +是 + +当前仅支持文本生成 + +训练方式 + +微调方法（仅训练集显示） + +是 + +SFT、DPO、CPT，选项由系统动态展示。创建后锁定 + +时间范围 + +回流日志的时间段 + +是 + +最近 30 天（含当天），精确到时分秒。修改会重置 API Key 和模型选择 + +API Key 过滤 + +按 API Key 筛选日志数据 + +是 + +全部（不过滤）、其他（排除已列出 Key）、或选择具体 Key（多选） + +模型选择 + +回流目标模型 + +是 + +最多 10 个，按能力类型过滤不匹配模型置灰 + +OSS 数据路径 + +数据存储的目标目录（仅 OSS 挂载模式显示） + +是 + +Bucket 需在同一 Region + +## 查看回流结果 + +提交日志回流任务后，在[**数据管理**列表页](https://bailian.console.aliyun.com/#/efm/data_ass)查看数据集和导入进度。 + +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3463064871/p1088579.png) + +### 列表页展示 + +日志回流创建的数据集在列表页中的导入方式显示为**日志回流**，存储位置根据创建时的选择显示为**平台存储**或**OSS 挂载**。 + +可在列表页查看任务的导入状态。任务失败时，可查看系统返回的具体失败原因。 + +平台存储模式下，导入完成后系统自动发布数据集版本，无需手动操作。 + +### 详情页信息 + +数据集详情页展示的信息根据存储方式有所不同： + +- **OSS 挂载**：展示发布状态、数据量、创建时间、FileID、数据类型、导入状态和 OSS 挂载地址。 + +- **平台存储（OSS 导入）**：展示发布状态、数据量、创建时间、FileID、数据类型、导入状态和 OSS 导入地址。 + +- 其他情况：展示发布状态、数据量、创建时间、FileID、数据类型和导入状态。 + + +## 追加日志回流数据 + +在[数据管理](https://bailian.console.aliyun.com/#/efm/data_ass)页面，向已有数据集追加新一批日志回流数据，有以下两种方式： + +### 导入数据页 + +进入已有数据集的**导入数据**页面，选择**日志回流**作为导入方式。表单参数与创建流程一致，详见[创建日志回流数据集](#sec-create)。此方式额外支持按工作空间过滤。 + +此方式适用于所有存储类型的数据集，包括 OSS 挂载数据集。 + +### 新增版本弹窗 + +在数据集详情页，单击**新增版本**，在弹窗中选择日志回流导入方式。仅平台存储数据集可用此方式。 + +**说明** + +OSS 挂载数据集不支持**新增版本**操作，按钮置灰。需通过**导入数据**页追加。 + +### 增量回流最佳实践 + +推荐分批次回流，针对不同时段或不同模型分别执行回流任务，逐步积累高质量训练集。每批次可精准选择表现良好的模型和业务高峰时段的数据，确保数据质量优于一次性大量回流。 + +## 常见问题 + +**API Key 过滤的「其他」包含哪些请求？** + +**问题**：API Key 过滤中**其他**包含哪些请求？ + +**其他**过滤掉已列出 Key 的请求，包含已删除的 Key 或其他工作空间产生的日志。**全部**则不按 Key 过滤。三种模式（全部/其他/逐个选择）的完整说明见[创建日志回流数据集](#sec-create)的 API Key 过滤参数。 + +**10 万条上限是数据集总量限制吗？** + +**问题**：单次回流上限 10 万条，是否意味着数据集总量也被限制在 10 万条？ + +不是。10 万条是单次回流的上限，并非数据集的总量上限。可以多次回流到同一数据集的不同版本，累计数据量不受此限制。例如分批次针对不同时段回流，每次 10 万条以内，最终数据集可以积累远超 10 万条的数据。 + +**平台存储和 OSS 挂载有什么区别？** + +**问题**：两种存储方式除了数据存储位置不同，还有哪些差异？ + +两种存储方式的核心差异对比如下。 + +**维度** + +**平台存储** + +**OSS 挂载** + +额外授权 + +无需额外授权 + +需授权两个服务关联角色并勾选数据访问协议 + +新增版本 + +支持 + +不支持（按钮置灰），需通过导入数据页追加 + +数据访问 + +通过控制台访问 + +可在 OSS Bucket 直接查看和管理 JSONL 文件 + +评测集 + +支持 + +不支持（选项禁用） + +自动发布 + +导入完成后自动发布版本 + +不自动发布 + +**审计日志和推理日志是同一个操作吗？** + +**问题**：开启日志需要分别操作审计日志和推理日志，两者有什么关系？ + +两者需分别开启，且必须先完成审计日志才能开启推理日志。操作步骤详见[开通并授权相关服务](#sec-authorize)。 + +**预估数据量为什么和实际回流结果不一致？** + +**问题**：提交前显示的预估数据量与回流完成后的实际条数有差异。 + +预估数据量是基于筛选条件的近似估算值，实际回流完成后的数据条数可能略有差异，这是正常现象。预估值用于帮助判断是否需要调整筛选条件（如超过 10 万条时提示缩小范围），不代表精确计数。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/model-deployment-introduction.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/model-deployment-introduction.md index fbf3abc6..d8fb82b2 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/model-deployment-introduction.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/model-deployment-introduction.md @@ -108,7 +108,7 @@ 1. 预付费按天计费。无法提前退费 -2. 如果单位时间内使用超出购买的吞吐量，将自动切换成百炼提供的[模型调用](https://help.aliyun.com/zh/model-studio/model-pricing)服务。 +2. 如果单位时间内使用超出购买的吞吐量，按创建时选择的溢出策略处理：自动溢出则切换为该模型的[模型调用](https://help.aliyun.com/zh/model-studio/model-pricing)按量付费，仅使用 PTU 容量则返回 429。 预付费购买后，若在首月内提前退订，日单价（≈ 月单价 / 30）将按 **1.2** 倍计费 @@ -137,9 +137,9 @@ - 后付费时，如果账户欠费，部署的资源将继续保留并计费 24 小时，在这 24 小时内服务仍可正常使用。超过 24 小时后系统停止计费，模型部署进入欠费状态，底层资源将被删除，但模型部署任务仍会保留。补足欠费后，系统将重新分配资源并恢复使用（恢复后继续产生费用）。如果您不希望继续产生费用，可删除模型部署任务，删除成功后将不再计费。 -当模型输入超过最长输入 Token 或 超出购买的 TPM 量时，相关调用将自动切换为当前模型的按量付费模式。此时，推理性能可能下降，将受业务空间中当前快照模型的公共流量的管控，[费用](https://help.aliyun.com/zh/model-studio/model-pricing)按模型调用（按量付费）标准计收。 +当模型输入超过最长输入 Token 时，相关调用将自动切换为当前模型的按量付费模式；超出购买的 TPM 量时，按创建时选择的溢出策略处理（「自动溢出」切换为按量付费，「仅使用 PTU 容量」返回 429）。此时，推理性能可能下降，将受业务空间中当前快照模型的公共流量的管控，[费用](https://help.aliyun.com/zh/model-studio/model-pricing)按模型调用（按量付费）标准计收。 -- 此时，调用 API 返回 Header 将包含：`x-dashscope-ptu-overflow:true`。 +- 此时（仅「自动溢出」策略下），调用 API 返回 Header 将包含：`x-dashscope-ptu-overflow:true`。 - TPM 统计请前往：[模型监控（北京）](https://bailian.console.aliyun.com/?tab=model#/model-telemetry)。 @@ -483,15 +483,33 @@ MU2 x 8 ¥240,288 +MU3 x 8 + +¥1,096 + +¥527,752 + 千问3.6-35B-A3B qwen3.6-35b-a3b -MU8 x 1 +MU1 x 8 -¥47 +¥432 -¥22,400 +¥208,944 + +MU2 x 8 + +¥504 + +¥240,288 + +MU3 x 8 + +¥1,096 + +¥527,752 MU9 x 1 @@ -509,6 +527,12 @@ MU1 x 2 ¥52,236 +MU3 x 8 + +¥1,096 + +¥527,752 + 千问3.6-Plus-2026-04-02 qwen3.6-plus-2026-04-02 @@ -529,12 +553,6 @@ PD分离模式：¥417,888 qwen3.5-397b-a17b -MU2 x 8 - -¥504 - -¥240,288 - MU3 x 8 MU3 x 16（PD分离模式） @@ -547,6 +565,12 @@ PD分离模式：¥2,192 PD分离模式：¥1,055,504 +MU6 x 16 + +¥400 + +¥193,424 + 千问3.5-122B-A10B qwen3.5-122b-a10b @@ -557,11 +581,11 @@ MU1 x 4 ¥104,472 -MU2 x 8 +MU6 x 16 -¥504 +¥400 -¥240,288 +¥193,424 千问3.5-35B-A3B @@ -579,10 +603,40 @@ MU2 x 8 ¥240,288 +MU3 x 8 + +¥1,096 + +¥527,752 + +MU9 x 1 + +¥51 + +¥24,600 + 千问3.5-27B qwen3.5-27b +MU2 x 8 + +¥504 + +¥240,288 + +MU3 x 8 + +¥1,096 + +¥527,752 + +MU8 x 1 + +¥47 + +¥22,400 + MU9 x 1 ¥51 @@ -599,6 +653,12 @@ MU1 x 2 ¥52,236 +MU2 x 2 + +¥126 + +¥60,072 + MU8 x 1 ¥47 @@ -625,12 +685,24 @@ MU1 x 2 qwen3.5-plus-2026-02-15 +MU1 x 8 + MU1 x 16（PD分离模式） +¥432 + PD分离模式：¥864 +¥208,944 + PD分离模式：¥417,888 +MU2 x 8 + +¥504 + +¥240,288 + MU3 x 8 MU3 x 16（PD分离模式） @@ -659,57 +731,25 @@ MU2 x 8 ¥240,288 -千问3-Next-80B-A3B-Instruct - -qwen3-next-80b-a3b-instruct - -MU1 x 2 - -¥108 - -¥52,236 - 千问3-32B qwen3-32b -MU1 x 4 - -¥216 - -¥104,472 - -MU6 x 4 - -¥100 - -¥48,356 - -千问3-30B-A3B - -qwen3-30b-a3b - -MU9 x 2 - -¥102 - -¥49,200 - -千问3-30B-A3B-Instruct-2507 +MU6 x 16 -qwen3-30b-a3b-instruct-2507 +¥400 -MU1 x 4 +¥193,424 -¥216 +千问3-30B-A3B-Thinking-2507 -¥104,472 +qwen3-30b-a3b-thinking-2507 -MU2 x 8 +MU1 x 2 -¥504 +¥108 -¥240,288 +¥52,236 千问3-8B @@ -759,12 +799,6 @@ MU1 x 2 ¥52,236 -MU5 x 1 - -¥21 - -¥10,139 - 千问3-Embedding-0.6B qwen3-embedding-0.6b @@ -837,11 +871,11 @@ MU5 x 1 qwen2.5-72b-instruct -MU1 x 4 +MU1 x 8 -¥216 +¥432 -¥104,472 +¥208,944 千问2.5-开源版-32B @@ -879,26 +913,6 @@ MU5 x 1 ¥10,139 -千问2.5-开源版-3B - -qwen2.5-3b-instruct - -MU5 x 1 - -¥21 - -¥10,139 - -千问-Flash-2025-07-28 - -qwen-flash-2025-07-28 - -MU1 x 4 - -¥216 - -¥104,472 - 千问-Plus-2025-07-28 qwen-plus-2025-07-28 @@ -955,6 +969,12 @@ GLM-5.1 glm-5.1 +MU2 x 8 + +¥504 + +¥240,288 + MU3 x 16（PD分离模式） PD分离模式：¥2,192 @@ -987,6 +1007,16 @@ PD分离模式：¥800 PD分离模式：¥386,848 +GLM-4.7-Flash + +glm-4.7-flash + +MU3 x 16（PD分离模式） + +PD分离模式：¥2,192 + +PD分离模式：¥1,055,504 + #### DeepSeek **模型名称** @@ -1007,11 +1037,11 @@ DeepSeek-v4-Flash deepseek-v4-flash -MU1 x 8 +MU3 x 8 -¥432 +¥1,096 -¥208,944 +¥527,752 DeepSeek-v3.2 @@ -1039,16 +1069,6 @@ PD分离模式：¥480,576 **最小计费：天** -MiniMax-M2.5 - -MiniMax-M2.5 - -MU1 x 16（PD分离模式） - -PD分离模式：¥864 - -PD分离模式：¥417,888 - Kimi-K2.5 kimi-k2.5 @@ -1091,15 +1111,21 @@ MU2 x 8 **最小计费：天** -千问3-VL-235B-A22B-Instruct +千问3-VL-32B-Instruct -qwen3-vl-235b-a22b-instruct +qwen3-vl-32b-instruct -MU1 x 4 +MU2 x 8 -¥216 +¥504 -¥104,472 +¥240,288 + +MU3 x 8 + +¥1,096 + +¥527,752 千问3-VL-8B-Instruct @@ -1111,6 +1137,12 @@ MU1 x 2 ¥52,236 +MU5 x 1 + +¥21 + +¥10,139 + 千问3-VL-4B-Instruct qwen3-vl-4b-instruct @@ -1131,6 +1163,16 @@ MU5 x 1 ¥10,139 +千问3-VL-Embedding-2B + +qwen3-vl-embedding-2b + +MU5 x 1 + +¥21 + +¥10,139 + 千问3-VL-Flash-2025-10-15 qwen3-vl-flash-2025-10-15 @@ -1161,16 +1203,6 @@ MU6 x 4 ¥48,356 -千问VL-OCR-2025-11-20 - -qwen-vl-ocr-2025-11-20 - -MU6 x 4 - -¥100 - -¥48,356 - #### 千问 Omni **模型名称** @@ -1251,6 +1283,14 @@ MU5 **元/千Token** +千问3.5-27B 邀测中 + +qwen3.5-27b + +¥0.0006 + +¥0.0048 + 千问3-32B qwen3-32b @@ -1281,6 +1321,16 @@ qwen3-8b 思考模式：¥0.005 +千问3-4B-Instruct-2507 + +qwen3-4b-instruct-2507 + +¥0.0003 + +非思考模式：¥0.0012 + +思考模式：¥0.003 + 千问2.5-开源版-72B qwen2.5-72b-instruct @@ -1313,6 +1363,14 @@ qwen2.5-7b-instruct ¥0.001 +千问2-开源版-7B + +qwen2-7b-instruct + +¥0.001 + +¥0.002 + #### 千问VL **基础模型** @@ -1359,6 +1417,14 @@ qwen2.5-vl-7b-instruct ¥0.005 +千问2.5-VL-3B-Instruct + +qwen2.5-vl-3b-instruct + +¥0.0012 + +¥0.0036 + 如果需要部署更多模型，请参考此[解决方案](https://www.aliyun.com/solution/tech-solution/deepseek-r1-for-platforms)并结合具体业务需求选择最适合的部署方案。 ## 部署方法 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/model-import.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/model-import.md index ec850a53..8e696bf6 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/model-import.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/model-import.md @@ -1,41 +1,35 @@ # 模型导入 -**[我的模型](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/model_center)**页面用于管理您创建和导入的模型。通过该页面，您可以将本地训练的 LoRA 模型从阿里云对象存储 OSS 导入到百炼平台。 +介绍百炼平台从阿里云 OSS 导入 LoRA 模型的全流程操作，涵盖首次 OSS 授权、模型文件准备、导入表单填写、导入模型管理与常见问题排查。 -## 使用前提 +## 模型导入概述 -导入前，请确保满足以下条件： +本篇介绍将本地训练的 LoRA 模型从阿里云对象存储 OSS 导入到百炼平台的全流程，覆盖首次 OSS 授权、模型文件准备、导入表单填写、模型查看管理与删除，以及导入失败、已失效等常见问题排查。导入成功后即可部署服务，部署、扩缩容与下线操作详见[部署运维](#)，通过 API 完成部署与调用详见[使用 API 进行模型部署](#)，部署前置概念与计费方案对比详见[模型部署简介](#)，本篇与三者构成概念、导入、部署、调用的完整链路。 -- **OSS Bucket 准备**：已创建 OSS Bucket，并为目标 Bucket 添加标签。 - - **说明** - - - 支持的 OSS Bucket 存储类型不包括归档、冷归档或深度冷归档。支持内容加密的 Bucket。支持私有的Bucket。 - - - 不支持访问OSS Bucket根目录下的文件，请您在OSS Bucket下选择已有的子目录或新建一个子目录供阿里云百炼访问。 - - - 支持导入任意大小的模型文件，导入后将使用阿里云百炼提供的免费存储空间。 - - -- **模型文件准备**：模型文件需符合[导入要求与限制](#h2-file-format-constraints)。模型文件夹需直接放在 OSS Bucket 中，系统会自动识别。 - +导入页基础模型字段由平台接口动态返回当前支持导入的基础模型清单，可能随版本更新，请以控制台可选列表为准。当前支持的基础模型如下： -## 支持导入的基础模型 +**当前支持的基础模型清单（点击展开）** -当前支持导入以下基础模型的 LoRA 微调版本： +以下清单由后端接口动态返回，可能随版本更新而调整，请以控制台实际可选项为准。 -模型系列 +**模型系列** -模型名称 +**模型名称** 千问3 千问3-32B +千问3 + 千问3-14B +千问3 + 千问3-8B +千问3 + 千问3-4B-Instruct-2507 千问3-VL @@ -46,249 +40,320 @@ 千问2.5-72B-Instruct +千问2.5 + 千问2.5-32B-Instruct +千问2.5 + 千问2.5-14B-Instruct +千问2.5 + 千问2.5-7B-Instruct 千问2.5-VL 千问2.5-VL-72B-Instruct +千问2.5-VL + 千问2.5-VL-7B-Instruct -## 操作步骤 +训练方式可选项取决于所选基础模型的声明，选择基础模型后训练方式字段自动渲染可选项并默认选中第一项。当前版本仅支持导入 LoRA 模型，全参微调模型不可导入。 -按照以下步骤将 LoRA 模型从 OSS 导入到百炼平台： +导入来源仅支持「从 OSS 导入」一项，表单中默认选中且无其他选项，暂不支持从其他渠道导入模型。 -1. 在**我的模型**页面，点击**导入模型**按钮。 - -2. 在导入模型页面中，填写以下信息： - - - **模型名称**：输入模型的显示名称，最多50个字符。 - - - **基础模型**：选择该 LoRA 模型对应的基础模型。 - - - **训练方式**：选择模型的训练方式。可选项取决于所选基础模型，选择基础模型后自动显示。 - - - **导入来源**：当前仅支持"从OSS导入"，系统已默认选中。 - - - **Bucket**：选择存储模型文件的 OSS Bucket。 - - - **模型目录**：在选定的 Bucket 中浏览并选择模型 Checkpoint 所在的目录。 - - - **模型加密**：为保障您的数据安全，平台会为导出的模型文件开启 OSS 服务端加密，使用 OSS 完全托管密钥进行加解密（SSE-OSS），加密算法为 AES256。 - -3. 确认信息无误后，点击**确定**提交导入请求。如需放弃导入，点击**取消**返回列表页。系统将自动验证模型文件格式和完整性，验证通过后开始导入。导入完成后，您可以在**[我的模型](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/model_center)**页面查看导入的模型，并进行部署、增量训练或删除等操作。 - - 导入后的模型状态包括：创建中（正在导入）、创建成功（导入成功，可部署）、创建失败（导入失败）、已失效（模型文件已变更）。列表页还展示模型名称/ID、基础模型、来源、支持部署方式和创建时间等信息。您可以通过页面顶部的搜索框按名称筛选模型。 - +导入成功后的模型状态为创建成功，可在[我的模型](https://bailian.console.aliyun.com/#/efm/model_center)列表中部署服务；从 OSS 导入的模型不支持增量训练，如需迭代请重新训练后再次导入。导入后将使用百炼提供的免费存储空间存放模型记录。 + +## 首次导入前完成 OSS 授权 -## 导入要求与限制 +首次从 OSS 导入模型前，须完成 OSS 服务关联角色授权，并为目标 Bucket 添加访问标签。授权通过导入页一键完成，授权后系统自动创建服务关联角色 AliyunServiceRoleForSFMDataHubOSSImport（服务名 datahub.sfm.aliyuncs.com，权限策略 AliyunServiceRolePolicyForSFMDataHubOSSImport），通常秒级生效。服务关联角色说明详见[OSS 服务关联角色](#)，主账号与子账号的概念与区别详见[主账号与子账号](#)。 -**重要** +在[导入模型页](https://bailian.console.aliyun.com/#/efm/model_center/import_model)的「导入来源」选择「从 OSS 导入」后，若未授权，OSS 字段区会提示「您还未授权OSS」并显示「前往授权」链接，确定按钮在未授权时不可用。主账号与子账号的授权路径不同，请按账号类型选择。 -重要：当前版本仅支持导入 LoRA（Low-Rank Adaptation）模型，不支持导入全参微调模型。 +【截图：ss-oss-auth-01.png — 导入页 OSS 字段区「您还未授权OSS」提示与「前往授权」入口，及授权弹窗】 -导入 LoRA 模型前，请确保满足以下要求： +### 使用主账号 -- **必需文件**：OSS Bucket 中需包含以下文件： +1. 在导入模型页「导入来源」选择「从 OSS 导入」后，页面提示「您还未授权OSS」，在提示栏右侧点击「前往授权」。 - - **adapter\_model.safetensors**：LoRA 适配器的权重文件，采用 SafeTensors 格式存储。 - - - **adapter\_config.json**：LoRA 适配器的配置文件，包含 rank、alpha 等关键参数信息。 - -- **rank 参数限制**：rank 值必须为 8、16、32 或 64 中的一个，且同一模型的所有 LoRA 层必须使用相同的 rank 值。 +2. 在弹出的对话框中点击「立即授权」，系统自动创建服务关联角色 AliyunServiceRoleForSFMDataHubOSSImport，通常秒级生效，服务高峰期可能稍有延迟。 + +3. 为目标 OSS Bucket 添加标签：访问 OSS 管理控制台的 Bucket 列表，找到目标 Bucket，悬停标签列图标并点击「前往编辑」→「创建标签」，添加标签名 bailian-datahub-access、标签值 read，保存。 -- **修改词汇表的模型**：如果训练过程中添加了新 token 或修改了原始词汇表（vocab），该模型无法导入。系统要求使用与基础模型完全一致的词汇表。 +4. 返回导入模型页，重新选择目标 Bucket 再尝试导入。百炼不支持访问 Bucket 根目录下的文件，须选择 Bucket 下已有的子目录或新建子目录。 -- **修改对话模板的模型**：如果训练过程中修改了 chat\_template 配置，该模型无法导入。系统仅支持使用与对应开源基础模型默认配置一致的 chat\_template。 + +### 使用子账号 + +1. 在导入模型页「导入来源」选择「从 OSS 导入」后，页面提示「您还未授权OSS」，点击「前往授权」。 + +2. 在弹出的对话框中点击「立即授权」，界面提示「授权失败：当前用户没有创建服务关联角色的权限」。对话框显示 Service Name 为 datahub.sfm.aliyuncs.com，服务关联角色名称为 AliyunServiceRoleForSFMDataHubOSSImport，所需用户权限为 ram:CreateServiceLinkedRole。须先由主账号授予子账号创建服务关联角色的权限，再由子账号完成授权。 - chat\_template 配置通常位于以下位置： +3. 由主账号授予子账号创建服务关联角色的权限： - - 模型的 config.json 文件中的 `chat_template` 字段。 + 1. 主账号登录 RAM 控制台，在左侧导航栏选择「权限管理 → 权限策略」，点击「创建权限策略」。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1185949671/p1049475.png) + 2. 选择「脚本编辑」，在 Effect、Action、Resource、Condition 中分别输入以下脚本内容，点击「确定」： - - tokenizer\_config.json 文件中的 `chat_template` 字段。 + ``` + { + "Action": [ + "ram:CreateServiceLinkedRole" + ], + "Resource": "*", + "Effect": "Allow", + "Condition": { + "StringEquals": { + "ram:ServiceName": "datahub.sfm.aliyuncs.com" + } + } + } + ``` + + 3. 输入权限策略名称（示例：服务关联角色）后点击「确定」。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1185949671/p1049476.png) + 4. 在左侧导航栏选择「身份管理 → 用户」，找到待授权的子账号，点击操作列**新增授权**。 -- **未冻结 VIT 的视觉语言模型**：对于 VL（Vision-Language）模型，必须冻结 Vision Transformer（VIT）部分。如果 LoRA adapter 中包含 visual 相关的权重参数（即未冻结 VIT），该模型无法导入。 + 5. 选择刚才创建的自定义权限策略，点击「确认新增授权」，子账号即拥有创建服务关联角色的权限。 + +4. 返回导入模型页点击「前往授权」，在弹出的对话框中点击「立即授权」，系统自动创建服务关联角色，通常秒级生效。 - 可以运行以下代码判断。 +5. 为目标 OSS Bucket 添加 bailian-datahub-access=read 标签（操作同主账号步骤 3），然后返回导入模型页重新选择目标 Bucket。百炼不支持访问 Bucket 根目录下的文件。 - ``` - from safetensors import safe_open - import argparse + +授权后须为目标 OSS Bucket 添加标签：标签名为 bailian-datahub-access，标签值为 read。该标签用于标记百炼可访问的 Bucket，未添加此标签的 Bucket 在下拉列表中不可选，须到[OSS 管理控制台](https://oss.console.aliyun.com/)添加标签后重新选择。 + +OSS Bucket 存储类型不支持归档、冷归档或深度冷归档，支持内容加密的 Bucket 与私有 Bucket。百炼不支持访问 Bucket 根目录下的文件，须选择 Bucket 下已有的子目录或新建子目录。 + +### 旧授权方式升级 + +若您此前使用的是旧授权方式，导入页 Bucket 字段下方会提示「建议转换为新的 Bucket 授权方式，提升安全性」并提供「直接转换」链接。点击后弹出确认框，确认即可升级为服务关联角色授权方式，升级不影响原有数据。 + +### 未开通 OSS 产品 + +若主账号尚未开通对象存储 OSS，导入页会提示「您还未开通OSS」并提供前往购买的链接。须由主账号前往 OSS 控制台开通 OSS 后返回导入页重新授权。 + +## 准备 LoRA 模型文件 + +导入前须将 LoRA 模型文件按以下要求存放在 OSS Bucket 的子目录中（不支持 Bucket 根目录），并在提交前通过系统的自动校验。模型文件须直接放在所选子目录下，系统会自动识别。 + +当前版本仅支持导入 LoRA 模型，不支持导入全参微调模型。 + +### 必需文件与目录结构 + +子目录中须包含以下文件：adapter\_model.safetensors（LoRA 适配器权重，SafeTensors 格式）、adapter\_config.json（含 rank、alpha 等参数的配置文件）、config.json（基础模型配置）。选中目录后系统会自动校验这些文件的格式与完整性。 + +### 训练参数约束 + +- **rank 取值**：rank 必须为 8、16、32 或 64 之一，且同一模型的所有 LoRA 层须使用相同的 rank 值，否则无法导入。 - def print_safetensor_structure(file_path): - print(f"Loading safetensor file: {file_path}") - print("="*80) - - with safe_open(file_path, framework="pt") as f: - keys = f.keys() - print(f"Found {len(keys)} tensors in the file:\n") - - for key in sorted(keys): - tensor = f.get_tensor(key) - shape = tuple(tensor.shape) - dtype = str(tensor.dtype) - device = tensor.device if hasattr(tensor, 'device') else 'cpu' - - lora_tag = " [LoRA]" if "lora_A" in key or "lora_B" in key else "" - - print(f"[{dtype:>14}] {shape} | {key} {lora_tag}") +- **词汇表不可修改**：训练中添加新 token 或修改原始词汇表的模型无法导入，须与基础模型词汇表完全一致。 - if __name__ == "__main__": - parser = argparse.ArgumentParser(description="Print structure of a .safetensors LoRA adapter.") - parser.add_argument("filepath", type=str, help="Path to the .safetensors file") - args = parser.parse_args() +- **对话模板不可修改**：训练中修改 chat\_template 的模型无法导入，须与基础模型默认配置一致。chat\_template 位于 config.json 或 tokenizer\_config.json 的 chat\_template 字段。 - print_safetensor_structure(args.filepath) - ``` +- **视觉模型须冻结 VIT**：视觉语言模型必须冻结 Vision Transformer 部分。若 LoRA 适配器中包含 visual 相关权重参数（即未冻结 VIT），该模型无法导入。 - 判断方法：检查 adapt\_model.safetensors 文件中是否包含 `visual` 相关的权重参数。如果文件中存在以 `visual` 开头的参数键（例如 `visual.encoder.layer.0...`），说明 VIT 部分未被冻结，该模型无法导入。 + +可在导入前运行以下脚本检查 adapter\_model.safetensors 是否含 visual 开头的参数键，以判断 VIT 是否冻结。 + +``` +from safetensors import safe_open +import argparse + +def print_safetensor_structure(file_path): + print(f"Loading safetensor file: {file_path}") + print("="*80) + + with safe_open(file_path, framework="pt") as f: + keys = f.keys() + print(f"Found {len(keys)} tensors in the file:\n") + + for key in sorted(keys): + tensor = f.get_tensor(key) + shape = tuple(tensor.shape) + dtype = str(tensor.dtype) + device = tensor.device if hasattr(tensor, 'device') else 'cpu' + + lora_tag = " [LoRA]" if "lora_A" in key or "lora_B" in key else "" + + print(f"[{dtype:>14}] {shape} | {key} {lora_tag}") + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description="Print structure of a .safetensors LoRA adapter.") + parser.add_argument("filepath", type=str, help="Path to the .safetensors file") + args = parser.parse_args() + + print_safetensor_structure(args.filepath) +``` + +判断方法：若脚本输出中存在以 visual 开头的参数键（如 visual.encoder.layer.0...），说明 VIT 部分未被冻结，该模型无法导入；若仅含 lora\_A、lora\_B 等 LoRA 相关参数键，则 VIT 已冻结，可正常导入。 + +【截图：ss-safetensors-out-01.png — safetensors 结构检查脚本输出示例，展示冻结 VIT 的 adapter 文件仅含 lora\_A、lora\_B 参数键】 + +### 提交前自动校验 + +在[导入页](https://bailian.console.aliyun.com/#/efm/model_center/import_model)选中模型目录后，系统会自动调用文件校验接口检查目录下模型文件的格式与完整性。校验失败会在目录字段下方显示红标提示并阻断提交，须按提示修正文件后再提交。常见失败原因如缺少必需文件，对应错误码 AvailableModelFileNotFound。 + +【截图：ss-dir-browser-01.png — 模型目录浏览器选中子目录后的文件自动校验结果】 + +## 导入 LoRA 模型 + +完成 OSS 授权并准备好模型文件后，在[百炼控制台·我的模型](https://bailian.console.aliyun.com/#/efm/model_center)页面点击右上角「导入模型」按钮进入创建页，按以下步骤完成导入。 + +1. 在「我的模型」页面，点击右上角的「导入模型」按钮进入导入模型创建页。 - ![冻结VIT的adapter文件示例](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1185949671/p1049184.png) +2. 按下方表格填写模型信息，其中基础模型须与 LoRA 训练时的基座一致，Bucket 须已添加 bailian-datahub-access=read 标签，模型目录选中后系统会自动校验文件。 - ![未冻结VIT的adapter文件示例（包含visual参数）](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1185949671/p1049183.png) +3. 确认信息无误后点击确定提交，系统自动验证文件格式和完整性，通过后开始导入；如需放弃点击取消返回列表页。导入后可在[管理导入的模型](#sec-manage)查看状态。 -## 常见问题 +导入模型表单各字段含义如下： -#### **为什么导入的模型与本地使用 vLLM、SGLang 推理的效果不一致？** +【截图：ss-import-form-01.png — 导入模型创建页 7 字段表单全貌】 -百炼平台的推理引擎参数设置可能与您本地使用的推理框架默认值不同。为确保效果一致，建议在调用 API 时调整以下参数： +**字段** -参数名称 +**说明** -推荐值（对应 vLLM 默认值） +**约束** -temperature +模型名称 -取值范围：\[0, 2)。设置为 1.0 等同于 vLLM 引擎默认值。 +输入模型的显示名称。 -top\_p +必填，最多 50 字符 -取值范围：(0, 1.0\]。设置为 1.0 等同于 vLLM 引擎默认值。 +基础模型 -top\_k +选择 LoRA 训练时的基座模型，须与训练基座一致。 -取值为 None 或大于 100 时，表示不启用 top\_k 策略，此时仅有 top\_p 策略生效。设置为 99 不支持全采样，该值接近 vLLM 默认值 0（全采样）。 +必填，下拉选择 -presence\_penalty +训练方式 -取值范围：\[-2.0, 2.0\]。设置为 0 等同于 vLLM 引擎默认值。 +可选项取决于所选基础模型，选择基础模型后自动渲染并默认选中第一项。 -repetition\_penalty（DashScope 协议） +必填，下拉选择 -提高 repetition\_penalty 可以降低模型生成的重复度，1.0 表示不做惩罚。取值范围：大于 0。设置为 1.0 等同于 vLLM 引擎默认值。 +导入来源 -说明：以上参数值基于 vLLM 引擎的默认配置。如果您的本地环境使用 SGLang 或其他推理框架，请参考对应框架的文档调整参数。 +当前仅支持「从 OSS 导入」，无其他选项。 -**首次从 OSS 向阿里云百炼导入文件，应该如何操作？** +只读，默认选中 -如果您是首次从 OSS 向阿里云百炼导入文件，请先按照界面提示完成授权，并为目标 OSS Bucket 添加`bailian-datahub-access`标签，然后再进行导入。 +Bucket -> 如果您尚不清楚主账号和子账号的概念和区别，请先阅读[权限管理](https://help.aliyun.com/zh/model-studio/permission-management-overview)。 +选择存放模型文件的 OSS Bucket，仅列出已添加 bailian-datahub-access=read 标签的 Bucket。 -## 使用主账号 +必填，下拉选择 -1. 单击**前往授权**。 - - 在**导入来源**中选择**从OSS导入**后，页面会显示"您还未授权OSS"的提示信息，在提示栏右侧找到 **前往授权** 链接。 - -2. 在弹出的对话框中，单击**确认授权**，系统将为您自动开通[OSS服务关联角色](https://help.aliyun.com/zh/model-studio/bailian-service-linked-role#32a41eac73z64)（必要条件）。 - - > 通常秒级生效，服务高峰期可能会稍有延迟。 - -3. 为目标 OSS Bucket 添加`bailian-datahub-access`标签。 - - > 该标签用于标记阿里云百炼可访问的 Bucket，未标记的 Bucket 阿里云百炼无法访问。 - - 1. 访问[OSS管理控制台](https://oss.console.aliyun.com/)，单击左侧导航栏中的****Bucket 列表****，即可查看您已创建的Bucket。 - - 2. 在待添加标签的Bucket**标签**列，悬停鼠标于![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1134056371/p903505.png)图标上，然后单击**前往编辑**。 - - 3. 单击**创建标签**。 - - 4. 单击**标签**，添加标签名为`bailian-datahub-access`，标签值为`read`的标签，然后单击**保存**。 - -4. 返回**导入模型**界面，重新选择目标 Bucket 再尝试导入。 - - > **请注意，阿里云百炼不支持访问保存在 Bucket 根目录下的文件。**请您选择 Bucket 下的现有文件夹或新建一个文件夹供阿里云百炼访问。 - +模型目录 -## 使用**子账号** +在选定 Bucket 中浏览并选择模型 Checkpoint 所在子目录，不支持选 Bucket 根目录。 -1. 单击**前往授权**。 - -2. 在弹出的对话框中，单击**确认授权**。界面会提示**授权失败**、**当前用户没有创建服务关联角色的权限**（因为当前子账号没有创建服务关联角色的权限。接下来需要先授予子账号创建服务关联角色的权限，再授予子账号通过阿里云百炼访问OSS的权限）。 - - 对话框中显示 Service Name 为 `datahub.sfm.aliyuncs.com`，服务关联角色名称为 `AliyunServiceRoleForSFMDataHubOSSImport`，执行该操作所需的用户权限为 `ram:CreateServiceLinkedRole`。 - -3. 授予子账号创建服务关联角色的权限。 - - 1. **需主账号登录**[RAM控制台](https://ram.console.aliyun.com/)，在左侧导航栏，选择**权限管理** > **权限策略**，然后单击界面上的**创建权限策略**。 - - 2. 在**脚本编辑**的`Effect`、`Action`、`Resource`、`Condition`中分别输入以下脚本中的对应内容后，单击**确定**。 - - ``` - { - "Action": [ - "ram:CreateServiceLinkedRole" - ], - "Resource": "*", - "Effect": "Allow", - "Condition": { - "StringEquals": { - "ram:ServiceName": "datahub.sfm.aliyuncs.com" - } - } - } - ``` - - 3. 输入权限策略名称后，单击**确定**。 - - 本示例中，权限策略名称为`服务关联角色`。 - - 4. 在左侧导航栏，选择**身份管理** > **用户**。在页面列表中找到待授权的子账号，然后单击子账号**操作**列的**添加权限**。 - - 5. 在权限策略中选择刚才创建的权限策略（自定义策略），单击**确认新增授权**。至此，子账号拥有了创建服务关联角色的权限。 - -4. 授权子账号通过阿里云百炼访问OSS。 - - 1. 返回**导入模型**界面，单击**前往授权**。 - - 在**导入来源**中选择**从OSS导入**后，界面提示**您还未授权OSS**。 - - 2. 在弹出的对话框中，单击**确认授权**，系统将为您自动开通[OSS服务关联角色](https://help.aliyun.com/zh/model-studio/bailian-service-linked-role#32a41eac73z64)（必要条件）。 - - > 通常秒级生效，服务高峰期可能会稍有延迟。 - -5. 为目标 OSS Bucket 添加`bailian-datahub-access`标签。 - - > 该标签用于标记阿里云百炼可访问的 Bucket，未标记的 Bucket 阿里云百炼无法访问。 +必填，树形选择 + +模型加密 + +平台自动为导出的模型文件开启 OSS 服务端加密（SSE-OSS），使用 OSS 完全托管密钥，加密算法为 AES256。 + +只读，平台强制 + +若尚未完成 OSS 授权，确定按钮不可用，须先完成[首次导入前完成 OSS 授权](#sec-authorize-oss)后再提交。导入成功后即可部署，部署操作详见[部署运维](#)。 + +导入后的模型状态包括创建中（正在导入）、创建成功（可部署）、创建失败（导入失败）和已失效（源文件已变更）。 + +## 管理导入的模型 + +导入提交后返回[我的模型](https://bailian.console.aliyun.com/#/efm/model_center)列表页，可查看与管理所有导入的模型。列表展示模型名称、模型 ID（附复制按钮）、基础模型、来源、支持部署方式、状态、创建时间与操作列，右上角提供导入模型入口与刷新按钮。 + +【截图：ss-my-model-list-01.png — 我的模型列表页，含名称、基础模型、来源、部署方式、状态、创建时间、操作列】 + +### 状态与流转 + +模型状态包括创建中、创建成功、创建失败和已失效。创建中表示正在导入；创建成功表示导入完成可部署；创建失败表示导入未成功；已失效表示创建成功后 OSS 源模型文件发生变更。列表对处于创建中状态的模型每 3 秒自动静默刷新，属正常行为，非接口异常。 + +创建失败状态旁附「详情」链接，悬停可查看失败错误码（如 AvailableModelFileNotFound）与对应的 oss://bucket/path 路径，用于定位失败文件。 + +已失效状态可悬停查看弹出框，展示「如下文件检测到更新」及发生变更的源文件名列表，提示须重新导入。 + +【截图：ss-status-popover-01.png — 已失效状态气泡弹出框文件变更列表与创建失败悬浮提示错误码】 + +### 操作列可用性 + +- **部署**：仅创建成功状态可点击，点击后跳转部署创建页，部署操作详见[部署运维](#)；其余状态或无可选部署方式时按钮不可用。 - 1. 访问[OSS管理控制台](https://oss.console.aliyun.com/)，单击左侧导航栏中的****Bucket 列表****，即可查看您已创建的Bucket。 - - 2. 在待添加标签的Bucket**标签**列，悬停鼠标于![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/1134056371/p903505.png)图标上，然后单击**前往编辑**。 - - 3. 单击**创建标签**。 - - 4. 单击**标签**，添加标签名为`bailian-datahub-access`，标签值为`read`的标签，然后单击**保存**。 - -6. 返回**导入模型**界面，重新选择目标 Bucket 再尝试导入。 +- **增量训练**：从 OSS 导入的模型不支持增量训练，按钮不可用。 - > **请注意，阿里云百炼不支持访问保存在 Bucket 根目录下的文件。**请您选择 Bucket 下的现有文件夹或新建一个文件夹供阿里云百炼访问。 +- **删除**：创建中状态不可删除；量化模型须前往模型压缩界面删除。 -**遇到“10041495”报错怎么办？** +在操作列点击删除并确认后，仅删除百炼侧的模型记录，不影响 OSS 源文件。其余状态删除时会调用删除接口清理记录。 -一般是由于主账号尚未开通对象存储服务 OSS，处理步骤： +删除操作不可恢复：仅移除百炼侧的模型记录，须重新导入才能恢复；不会删除 OSS 中的源文件。 -1. 需主账号前往[OSS管理控制台](https://oss.console.aliyun.com/)，按界面指引开通 OSS。 - -2. 返回阿里云百炼**导入模型**界面，再尝试授权。 +列表顶部搜索框可按模型名称筛选模型，支持清空重置。来源列按导入来源渲染：OSS 导入显示 oss://bucket/path，训练任务显示来源任务 ID（已删除则显示「训练任务已删除」），并标注全参、LoRA 或量化标签。通过 API 调用已部署模型详见[使用 API 进行模型部署](#)。 + +所有操作按单个模型进行，不支持批量删除或批量部署。 + +## 常见问题 + +汇总导入过程中常见的问题与处理方式。 + +**导入失败提示 AvailableModelFileNotFound 怎么办？** + +该错误表示模型目录文件校验未通过（格式或完整性问题），不是单纯重传文件可解决。请检查所选目录是否包含齐全合规的 adapter\_model.safetensors、adapter\_config.json、config.json，并确认 rank、词汇表、chat\_template 等约束均满足，修正后重新选择目录提交。 + +**模型状态显示「已失效」是怎么回事？** + +已失效表示该模型创建成功后，OSS 源模型文件发生了变更，属正常检测行为而非故障。将鼠标悬停在已失效状态上可查看发生变更的文件名列表，需重新导入模型方可恢复可用。 + +**遇到「10041495」报错怎么办？** + +一般是由于主账号尚未开通对象存储服务 OSS。须由主账号前往 OSS 管理控制台按界面指引开通 OSS，再返回百炼导入模型界面重新尝试授权。 + +**子账号授权 OSS 失败怎么办？** + +子账号授权失败是因为没有创建服务关联角色的权限，并非真失败。须先由主账号在 RAM 控制台创建自定义权限策略（操作为 ram:CreateServiceLinkedRole，针对服务 datahub.sfm.aliyuncs.com）并授予子账号，再由子账号重新点击授权。OSS 服务关联角色说明详见[OSS 服务关联角色](#)。 + +**Bucket 下拉列表中目标 Bucket 不可选怎么办？** + +这是授权要求而非故障。新授权方式下，未添加 bailian-datahub-access=read 标签的 Bucket 在下拉中不可选。须到 OSS 管理控制台为目标 Bucket 添加该标签后返回导入页重新选择。 + +**我的模型列表每隔几秒自动刷新是故障吗？** + +不是故障。列表检测到有处于创建中状态的模型时，会每 3 秒静默刷新以获取最新状态，无创建中状态时自动停止，属正常行为。 + +**为什么导入的模型与本地使用 vLLM、SGLang 推理的效果不一致？** + +百炼推理引擎的参数默认值可能与本地推理框架不同，并非模型导入有误。为对齐 vLLM 默认值，建议调用时参考下表设置参数；使用 SGLang 等其他框架请参考对应文档调整。该参数对照属部署与调用范畴，详见[使用 API 进行模型部署](#)。 + +**参数名称** + +**推荐值（对应 vLLM 默认值）** + +temperature + +取值范围 \[0, 2)，设置为 1.0 等同于 vLLM 引擎默认值。 + +top\_p + +取值范围 (0, 1.0\]，设置为 1.0 等同于 vLLM 引擎默认值。 + +top\_k + +取值为 None 或大于 100 时不启用 top\_k 策略，仅 top\_p 生效；设置为 99 不支持全采样，接近 vLLM 默认值 0（全采样）。 + +presence\_penalty + +取值范围 \[-2.0, 2.0\]，设置为 0 等同于 vLLM 引擎默认值。 + +repetition\_penalty（DashScope 协议） + +提高可降低生成重复度，1.0 表示不惩罚；取值范围大于 0，设置为 1.0 等同于 vLLM 引擎默认值。 + +**删除导入的模型会影响 OSS 中的源文件吗？** + +不会。删除仅移除百炼侧的模型记录，需重新导入方可恢复；OSS 源文件归您所有，百炼仅通过 bailian-datahub-access=read 标签读取访问，删除模型不会改动 OSS 中的任何文件。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/ptu-long-input-and-cache.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/ptu-long-input-and-cache.md index 37f2cbeb..230abbfc 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/ptu-long-input-and-cache.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-deployment-1/ptu-long-input-and-cache.md @@ -1,6 +1,6 @@ # 预置吞吐长输入与缓存 -本文介绍 PTU（预置吞吐）部署的长输入和前缀缓存能力，包括额度消耗规则、容量计算器使用方法和 API 响应字段说明。 +本文介绍 PTU（预置吞吐）部署的长输入和前缀缓存能力，包括额度消耗规则、**预置吞吐额度计算器**使用方法和 API 响应字段说明。 ## 功能概述 @@ -12,12 +12,12 @@ PTU 部署支持长输入请求（部分模型最高 200K token）和前缀缓 - 前缀缓存优惠：部分模型支持前缀缓存，命中缓存的输入 token 按折扣系数消耗额度（具体折扣率因模型而异），可降低多轮对话和重复前缀场景的额度消耗。 -- 自动转按量计费：超出 PTU 额度或输入超过模型上限（千问 128K / DeepSeek 64K）时，请求自动转为按量计费，无需修改调用代码。 +- 溢出策略：创建 PTU 时可选——自动溢出至按量计费（默认，业务不中断）或仅使用 PTU 容量（超出返回 429、不产生额外费用）。输入超过模型上限（千问 128K / DeepSeek 64K）仍自动转为按量计费。 **重要** -自动转按量计费后，费用按对应模型的按量付费单价计算。建议通过容量计算器合理规划 PTU 额度，避免意外费用。 +自动溢出策略下转为按量计费后，费用按对应模型的按量付费单价计算（仅使用 PTU 容量策略下超出返回 429、不产生额外费用）。建议通过**预置吞吐额度计算器**合理规划 PTU 额度，避免意外费用。 常见于长文档分析（合同、研报摘要）和多轮对话（客服、编程助手）等输入超 32K token 的场景。 @@ -44,6 +44,8 @@ glm-5.1 \[0, 32K)：输入 1.0 / 输出 1.0 \[32K, 200K\]：输入 1.33 / 输出 1.17 + + deepseek-v4-pro @@ -87,17 +89,17 @@ qwen3.7-plus-2026-05-26 输入合计 = 31.940 KTPM（比无缓存节省 43%） ``` -## 使用容量计算器估算额度 +## 使用**预置吞吐额度计算器**估算额度 **说明** 建议在创建或扩容前使用计算器评估长输入场景的额度需求，避免额度不足导致请求转为按量计费。购买上限以控制台实际展示为准。 -前提条件：已开通百炼服务并具备 PTU 部署权限。登录[百炼控制台](https://bailian.console.aliyun.com/#/efm/model_deploy/create)，在**模型部署** > **创建部署**页面（或在已有部署详情页单击**扩容**），选择可部署的PTU（预置吞吐）模型后，展开**容量计算器**。 +前提条件：已开通百炼服务并具备 PTU 部署权限。登录[百炼控制台](https://bailian.console.aliyun.com/#/efm/model_deploy/create)，在**模型部署** > **创建部署**页面（或在已有部署详情页单击**扩容**），选择可部署的PTU（预置吞吐）模型后，展开**预置吞吐额度计算器**。 ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/4645961871/p1082157.png) -容量计算器根据业务负载自动推荐 TPM 额度。填写以下参数后，计算器输出推荐的输入 TPM 和输出 TPM。 +**预置吞吐额度计算器**根据业务负载自动推荐额度。填写以下参数后，计算器输出建议购买的输入 KTPM 和输出 KTPM。 **参数** @@ -109,25 +111,25 @@ qwen3.7-plus-2026-05-26 业务高峰期每分钟的请求数。 -RPM 越大，建议购买的输入和输出 TPM 同比增大。 +RPM 越大，建议购买输入 KTPM 和输出 KTPM 同比增大。 平均输入长度（token） 每条请求的平均输入 token 数。 -输入越长，所处阶梯越高，系数越大，建议购买的输入 TPM 越高。不同模型的阶梯边界不同，以控制台实际展示为准。 +输入越长，所处阶梯越高，系数越大，建议购买输入 KTPM 越高。不同模型的阶梯边界不同，以控制台实际展示为准。 平均输出长度（token） 每条请求的平均输出 token 数。 -输出越长，系数可能越大，建议购买的输出 TPM 越高。 +输出越长，系数可能越大，建议购买输出 KTPM 越高。 -预估缓存命中率（%） +缓存命中率（%） 请求中重复前缀被缓存命中的比例。实际命中率取决于请求内容的重复程度，以运行结果为准。 -命中率越高，输入容量消耗越慢，建议购买的输入 TPM 越低。仅影响输入 TPM，不影响输出 TPM。 +命中率越高，输入容量消耗越慢，建议购买输入 KTPM 越低。仅影响输入 KTPM，不影响输出 KTPM。 ## API 响应字段说明 @@ -271,7 +273,7 @@ PTU 部署的运行监控通过百炼平台的模型监控功能实现，支持 - Token 用量与缓存命中：包含 `cached_tokens` 数据系列，可查看缓存命中量占总输入的比例。 -- 配额内/外调用次数：了解超出 PTU 额度后转为按量计费的请求占比。 +- 配额内/外调用次数：了解超出 PTU 额度后的请求占比（自动溢出策略下转为按量计费，仅使用 PTU 容量策略下返回 429）。 更多监控指标和操作方式，请参见[模型监控](https://help.aliyun.com/zh/model-studio/model-telemetry)。 @@ -280,7 +282,7 @@ PTU 部署的运行监控通过百炼平台的模型监控功能实现，支持 **Q: 超出 PTU 额度时会怎样？** -请求自动转为按量计费。API 响应中 `service_tier` 字段不返回或返回 `default`，同时响应头包含 `x-dashscope-ptu-overflow:true`。业务不会中断。 +取决于创建时选择的溢出策略：「自动溢出」策略下，请求自动转为按量计费，API 响应中 `service_tier` 字段不返回或返回 `default`，同时响应头包含 `x-dashscope-ptu-overflow:true`，业务不会中断；「仅使用 PTU 容量」策略下，超出请求返回 429 错误，不产生额外费用。 **Q: 单次输入超过模型上限时会怎样？** @@ -296,4 +298,4 @@ PTU 部署的运行监控通过百炼平台的模型监控功能实现，支持 **Q: 利用率为什么超过 100%？** -部分模型（如 glm-5.1）的长输入阶梯系数使实际额度消耗高于原始 token 数。利用率 = 折算后消耗 ÷ 购买额度。超过 100% 表示消耗速度超过购买额度，超出部分自动转为按量计费，不影响服务可用性。 +部分模型（如 glm-5.1）的长输入阶梯系数使实际额度消耗高于原始 token 数。利用率 = 折算后消耗 ÷ 购买额度。超过 100% 表示消耗速度超过购买额度，超出部分按溢出策略处理（自动溢出则转为按量计费、不影响服务可用性；仅使用 PTU 容量则返回 429）。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md index 4cb4529c..8c5b19fd 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md @@ -96,7 +96,7 @@ BLEU/ROUGE/余弦等算法 ## 创建评测维度模板 -登录[百炼控制台](https://bailian.console.aliyun.com/#/efm/model_evaluate/dimension_template)，在左侧导航栏选择**模型评测** > **评测维度**，单击**创建维度模板**进入创建页面。 +登录[百炼控制台](https://bailian.console.aliyun.com/#/efm/model_evaluate/dimension_template)，在左侧导航栏选择**模型评测** > **评测维度**，单击**创建评测维度**进入创建页面。 创建评测维度模板时，需先填写以下公共字段： @@ -189,7 +189,7 @@ BLEU/ROUGE/余弦等算法 4. 配置**Pass 标签**和**Fail 标签**：定义通过和未通过的分类标签，每个标签不超过 20 个字符。Pass 和 Fail 标签不可重复。 -5. 单击**完成**提交。 +5. 单击**保存**。 ### 大模型评估-数值型 @@ -206,7 +206,7 @@ BLEU/ROUGE/余弦等算法 5. 设置**通过阈值**：评分达到该值及以上判定为 Pass，步长 0.1。阈值随评分范围自动联动。 -6. 单击**完成**提交。 +6. 单击**保存**。 ### 规则评估-字符串匹配 @@ -219,7 +219,7 @@ BLEU/ROUGE/余弦等算法 3. 配置**模型输出**（右侧）：输入待匹配的文本，可使用变量。评测输入和模型输出至少一侧包含变量。 -4. 单击**完成**提交。 +4. 单击**保存**。 ### 规则评估-文本相似度 @@ -234,7 +234,7 @@ BLEU/ROUGE/余弦等算法 4. 设置**通过阈值**：范围 0~1，步长 0.01，相似度达到该值及以上判定为 Pass。 -5. 单击**完成**提交。 +5. 单击**保存**。 ### 人工评估-分类型 @@ -245,7 +245,7 @@ BLEU/ROUGE/余弦等算法 2. 配置**Fail 标签**：定义未通过的标签含义。Pass 和 Fail 标签不可重复。 -3. 单击**完成**提交。 +3. 单击**保存**。 ## 配置评分器Prompt diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/image-model.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/image-model.md index 5e614b57..eee5b087 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/image-model.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/image-model.md @@ -43,20 +43,22 @@ FLUX.2 - 写实人像和产品照片 -### 何时使用qwen-image-2.0-pro +### 何时使用Qwen Image - 需要使用负向提示词排除输出中的特定元素 - 需要每次调用生成最多6张图片变体（Wan标准模式最多支持4张） +- 需要复杂版面生成、小字精准渲染或多语言字体支持 → 选择 qwen-image-3.0-pro（邀测中） + ## 图片编辑 推荐使用`wan2.7-image-pro`，它支持多图参考（最多9张输入图片）、边界框交互式编辑以及角色一致性多图生成。详细使用方法请参见[图像编辑-千问](https://help.aliyun.com/zh/model-studio/qwen-image-edit-guide)和[图像编辑-万相2.7/2.6/2.5](https://help.aliyun.com/zh/model-studio/wan-image-edit)。 -### 何时使用qwen-image-2.0-pro +### 何时使用Qwen Image -如果编辑时需要使用负向提示词，请使用`qwen-image-2.0-pro`（生成和编辑使用同一个模型ID）。 +如果编辑时需要使用负向提示词，请使用`qwen-image-3.0-pro`（邀测中）或`qwen-image-2.0-pro`（生成和编辑使用同一个模型ID）。 ## 推荐模型 @@ -108,9 +110,9 @@ FLUX.2 2048x2048 -`qwen-image-2.0-pro` +`qwen-image-3.0-pro` -负向提示词、最多6张图片变体 +复杂版面生成、小字渲染、多语言字体 支持 @@ -272,6 +274,16 @@ qwen-image-2.0-pro的快速版本 **最大分辨率** +`qwen-image-3.0-pro` + +支持 + +支持 + +6 + +2048x2048 + `qwen-image-2.0-pro` 支持 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/s2s-model.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/s2s-model.md index 9a3c0b1a..dc841b66 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/s2s-model.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/s2s-model.md @@ -14,13 +14,13 @@ OpenAI GPT Realtime、Gemini 3.1 Live -`qwen3.5-omni-plus-realtime` +`qwen-audio-3.0-realtime-plus` 成本敏感对话 OpenAI gpt-4o-mini Realtime -`qwen3.5-omni-flash-realtime` +`qwen-audio-3.0-realtime-flash` 实时翻译 / 同传 @@ -74,7 +74,7 @@ Gemini 3.1 Live ## 实时还是文件模式？ -- **实时（WebSocket）**：适用于语音助手、呼叫中心、同声传译等实时语音交互场景。音频流式输入，语音流式输出。模型名称中包含`-realtime`。 +- **实时（WebSocket）**：适用于语音助手、呼叫中心、同声传译等实时语音交互场景。音频流式输入，语音流式输出。 - **文件模式（HTTP）**：可以用延迟换取更好的效果，适用于视频配音、播客翻译、离线内容处理等场景。文件模式下还支持 Function Calling、联网搜索、思考模式、视频上下文等附带能力（详见下方“S2S 单模型的附带能力”）。 @@ -91,13 +91,13 @@ Gemini 3.1 Live 语音助手 / 客服对话 -`qwen3.5-omni-plus-realtime` +`qwen-audio-3.0-realtime-plus` WebSocket 成本敏感的对话 -`qwen3.5-omni-flash-realtime` +`qwen-audio-3.0-realtime-flash` WebSocket @@ -127,27 +127,27 @@ WebSocket ## S2S 单模型的附带能力 -以下能力由 Qwen3.5-Omni / Qwen3-Omni 模型在 S2S 单模型路线下直接提供。Pipeline 路线中，对应能力需要由其中的 LLM 等组件分别支持。 +以下能力由 Qwen3.5-Omni / Qwen3-Omni 模型在 S2S 单模型路线下直接提供；其中 Function Calling 也可由 Qwen-Audio Realtime 提供。Qwen-Audio Realtime 不支持联网搜索和思考模式。Pipeline 路线中，对应能力需要由其中的 LLM 等组件分别支持。 ### Function Calling -让模型根据听到和看到的内容执行操作 -- 查询知识库、查询日程、触发工作流。使用Qwen3.5 Omni（WebSocket与HTTP模式） 或 Qwen3 Omni（HTTP模式）。 +让模型根据听到和看到的内容执行操作 -- 查询知识库、查询日程、触发工作流。使用 Qwen3.5 Omni（WebSocket 与 HTTP 模式）、Qwen3 Omni（HTTP 模式）或 Qwen-Audio Realtime（WebSocket 模式）。 **说明** -Qwen3.5-Omni/Qwen3-Omni实时（WebSocket）模式和Livetranslate模型不支持此功能。Qwen-Audio Realtime（WebSocket）支持Function Calling。 +Qwen3.5-Omni / Qwen3-Omni 实时（WebSocket）模式和 Livetranslate 模型不支持此功能。 ### 联网搜索 -让模型检索实时信息，回答关于时事、股价、天气等问题。使用Qwen3.5 Omni（HTTP和WebSocket），包括Plus和Flash系列。模型自主决定是否搜索。 +让模型检索实时信息，回答关于时事、股价、天气等问题。使用 Qwen3.5 Omni（HTTP 和 WebSocket），包括 Plus 和 Flash 系列。模型自主决定是否搜索。Qwen-Audio Realtime 不支持此功能。 **说明** -Qwen3-Omni-Flash和Livetranslate模型不支持此功能。 +Qwen3-Omni-Flash 和 Livetranslate 模型不支持此功能。 ### 思考模式 -当回答质量比延迟更重要时，使用Qwen3 Omni（HTTP模式）。模型在回复前会逐步推理，适用于视频分析、批量打标等场景。 +当回答质量比延迟更重要时，使用 Qwen3 Omni（HTTP 模式）。模型在回复前会逐步推理，适用于视频分析、批量打标等场景。Qwen-Audio Realtime 不支持此功能。 **说明** @@ -157,7 +157,7 @@ Qwen3-Omni-Flash和Livetranslate模型不支持此功能。 以下模型系列均支持语音翻译： -- **Qwen3.5-Livetranslate**：支持 60 种语言互译，其中 29 种支持音频+文本输出、31 种仅支持文本输出，覆盖中文、英语、法语、德语、俄语、日语、韩语、西班牙语、葡萄牙语、阿拉伯语等主流语种。。 +- **Qwen3.5-Livetranslate**：支持 60 种语言互译，其中 29 种支持音频+文本输出、31 种仅支持文本输出，覆盖中文、英语、法语、德语、俄语、日语、韩语、西班牙语、葡萄牙语、阿拉伯语等主流语种。 - **Qwen3-Livetranslate**：支持18种语言 + 5种中文方言，约3秒延迟，开箱即用。文件模式支持输入视频以获得上下文感知的翻译精度。其中7种语言仅输出文本（不输出语音）。 @@ -992,6 +992,50 @@ HTTP ## 所有模型 +### Qwen-Audio + +**模型** + +**API** + +**输入** + +**Function Calling** + +**联网搜索** + +**思考模式** + +**翻译** + +`qwen-audio-3.0-realtime-plus` + +WebSocket + +音频、文本 + +支持 + +\-- + +\-- + +\-- + +`qwen-audio-3.0-realtime-flash` + +WebSocket + +音频、文本 + +支持 + +\-- + +\-- + +\-- + ### Qwen3.5-Omni **模型** @@ -1256,50 +1300,6 @@ HTTP 18 -### Qwen-Audio - -**模型** - -**API** - -**输入** - -**Function Calling** - -**联网搜索** - -**思考模式** - -**翻译** - -`qwen-audio-3.0-realtime-plus` - -WebSocket - -音频、文本 - -支持 - -\-- - -\-- - -\-- - -`qwen-audio-3.0-realtime-flash` - -WebSocket - -音频、文本 - -支持 - -\-- - -\-- - -\-- - ### 旧版模型 以下模型不再更新，新项目建议使用Qwen3.5-Omni。 @@ -1356,6 +1356,8 @@ WebSocket 选定模型后，参考对应的调用文档： +- Qwen-Audio Realtime（WebSocket，实时语音对话）→ [实时语音对话（Qwen-Audio-Realtime）](https://help.aliyun.com/zh/model-studio/qwen-audio-realtime-user-guides) + - Qwen3.5-Omni / Qwen3-Omni（WebSocket，实时）→ [实时（Qwen-Omni-Realtime）](https://help.aliyun.com/zh/model-studio/realtime) - Qwen3.5-Omni / Qwen3-Omni（HTTP，文件）→ [非实时（Qwen-Omni）](https://help.aliyun.com/zh/model-studio/qwen-omni) @@ -1363,5 +1365,3 @@ WebSocket - Qwen3.5-Livetranslate（WebSocket，实时）→ [实时语音/音视频翻译-千问](https://help.aliyun.com/zh/model-studio/qwen3-5-livetranslate-flash-realtime) - Qwen3-Livetranslate（HTTP，文件）→ [音视频文件翻译-千问](https://help.aliyun.com/zh/model-studio/qwen3-livetranslate-flash) - -- Qwen-Audio Realtime（WebSocket，实时语音对话）→ [实时语音对话（Qwen-Audio-Realtime）](https://help.aliyun.com/zh/model-studio/qwen-audio-realtime-user-guides) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/text-generation-model.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/text-generation-model.md index 560fb0a8..35e48604 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/text-generation-model.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/text-generation-model.md @@ -4,7 +4,7 @@ ## 做 AI 编程或 Agent 开发（OpenClaw、Claude Code、Hermes 等）该选哪个模型？ -推荐 `qwen3.7-plus`——能力与成本均衡，完整工具调用支持，1M 上下文适合大型代码库。如需最强推理能力，可选择 `qwen3.7-max`。 +推荐 `qwen3.7-plus`——能力与成本均衡，完整工具调用支持，1M 上下文适合大型代码库。如需最强推理能力，可选择`qwen3.8-max-preview`（ Token Plan 可用），或 `qwen3.7-max`。 ## 从闭源模型迁移到百炼? @@ -18,7 +18,7 @@ GPT-5.5、Claude Opus 4.7、Gemini 3.1 Pro -`qwen3.7-max` +`qwen3.7-max`、`qwen3.8-max-preview`（仅 Token Plan 可用） 平衡 @@ -30,19 +30,19 @@ GPT-5.4、Claude Sonnet 4.6、Gemini 3 Pro GPT-5.4-mini、Claude Haiku 4.5、Gemini 3.1 Flash -`qwen3.6-flash`、`deepseek-v4-flash`、`MiniMax-M2.5` +`qwen3.7-flash`、`deepseek-v4-flash`、`MiniMax-M2.5` ## 应用场景 -聊天机器人、内容生成、摘要总结、文档处理等场景，推荐使用 `qwen3.7-plus`，能力与成本均衡，拥有100万上下文窗口和完整的内置工具。确认效果满足需求后，可以尝试 `qwen3.6-flash` 来降低成本，效果接近旗舰模型，且拥有相同的上下文长度和功能支持。如需最强推理能力，可选择 `qwen3.7-max`（百万 token 上下文），但成本较高。 +聊天机器人、内容生成、摘要总结、文档处理等场景，推荐使用 `qwen3.7-plus`，能力与成本均衡，拥有100万上下文窗口和完整的内置工具。确认效果满足需求后，可以尝试 `qwen3.7-flash` 来降低成本，效果接近旗舰模型，且拥有相同的上下文长度和功能支持。如需最强推理能力，可选择 `qwen3.7-max`（百万 token 上下文）；也可选择 `qwen3.8-max-preview`（ Token Plan 可用）。 ### 办公场景（非编程） 处理日常办公任务（文档撰写、邮件处理、会议纪要整理、数据分析等）时，推荐 `qwen3.7-plus`——能力与成本均衡，拥有 100 万上下文窗口，支持 Function Calling 和内置工具，适合文档摘要、内容生成等办公类任务。 -确认效果满足需求后，可尝试 `qwen3.6-flash` 降低成本，效果接近旗舰模型，且拥有相同的上下文长度和功能支持。 +确认效果满足需求后，可尝试 `qwen3.7-flash` 降低成本，效果接近旗舰模型，且拥有相同的上下文长度和功能支持。 -如需最强推理能力（如复杂数据分析、多步逻辑推演），可选择 `qwen3.7-max`，但成本较高。 +如需最强推理能力（如复杂数据分析、多步逻辑推演），可选择 `qwen3.7-max`，但成本较高；也可选择 `qwen3.8-max-preview`（仅 Token Plan 可用）。 处理超长文档（如同时审阅多份合同、大规模文献梳理）时，推荐 `qwen-long`——上下文窗口达 1000 万 Token，可完整处理大体量文档。 @@ -52,7 +52,7 @@ GPT-5.4-mini、Claude Haiku 4.5、Gemini 3.1 Flash 100万Token约相当于70万个汉字或10本小说。 -- 长文档或大型代码库：`qwen3.7-plus` / `qwen3.6-flash`（100万）。 +- 长文档或大型代码库：`qwen3.7-plus` / `qwen3.7-flash` （100万）。 - 常规任务：128k-256k已足够。 @@ -138,6 +138,22 @@ GPT-5.4-mini、Claude Haiku 4.5、Gemini 3.1 Flash 支持 +`qwen3.7-flash` + +查看快照版本 + +`qwen3.7-flash-2026-07-15` + +1M + +支持 + +支持 + +支持 + +支持 + `qwen3.6-flash` 查看快照版本 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/tts-model.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/tts-model.md index 65c53e25..e784e1ba 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/tts-model.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/tts-model.md @@ -54,7 +54,7 @@ ElevenLabs Multilingual v3 `qwen-audio-3.0-tts-plus`、`MiniMax/speech-2.8-hd` -`qwen-audio-3.0-tts-flash`（声音复刻）、`cosyvoice-v3.5-plus`（声音设计）、`MiniMax/speech-2.8-hd`（声音复刻） +`qwen-audio-3.0-tts-plus`（声音复刻）、`cosyvoice-v3.5-plus`（声音设计）、`MiniMax/speech-2.8-hd`（声音复刻） - **使用标准语音合成**：当内置音色库能满足需求，希望快速上手、无需额外配置时。 @@ -89,7 +89,7 @@ ElevenLabs Multilingual v3 推荐模型 -`qwen-audio-3.0-tts-flash`、`MiniMax/speech-2.8-hd` +`qwen-audio-3.0-tts-plus`、`qwen-audio-3.0-tts-flash`、`MiniMax/speech-2.8-hd` `cosyvoice-v3.5-plus`、`cosyvoice-v3.5-flash` @@ -145,7 +145,7 @@ Qwen-Audio-TTS WebSocket / HTTP -不支持 +支持 不支持 @@ -205,7 +205,7 @@ HTTP WebSocket / HTTP -不支持 +支持 不支持 @@ -225,7 +225,7 @@ WebSocket / HTTP - 系统音色（因音色而异）：中文（普通话）、英文 -- 声音复刻音色（方言通过指令控制功能进行设置）：中文（普通话、广东话、重庆话、东北话、甘肃话、贵州话、浙江话、河北话、河南话、湖北话、湖南话、江西话、宁波话、宁夏话、青岛话、陕西话、山西话、山东话、上海话、四川话、云南话）、英文、日语、韩语、德语、法语、意大利语、俄语、葡萄牙语、泰语、印尼语、马来语、越南语 +- 声音复刻音色（方言通过指令控制功能进行设置）：中文（普通话、广东话、重庆话、东北话、甘肃话、贵州话、浙江话、河北话、河南话、湖北话、湖南话、江西话、宁波话、宁夏话、青岛话、陕西话、山西话、山东话、上海话、四川话、云南话）、英语、日语、韩语、俄语、法语、德语、葡萄牙语、泰语、印尼语、越南语、西班牙语、意大利语、马来西亚语、菲律宾语、阿拉伯语 ### CosyVoice diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/video-generate-edit-model.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/video-generate-edit-model.md index 59bdcc23..6cd607e5 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/video-generate-edit-model.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/video-generate-edit-model.md @@ -156,8 +156,6 @@ Seedance 2.0、Runway Gen-4 ### HappyHorse 1.1 -以下模型适用于中国内地和国际部署范围。 - **模型ID** **类型** @@ -192,8 +190,6 @@ Seedance 2.0、Runway Gen-4 ### HappyHorse 1.0 -以下模型适用于中国内地和国际部署范围。 - **模型ID** **类型** @@ -236,8 +232,6 @@ Seedance 2.0、Runway Gen-4 ### Wan 2.7 -以下模型适用于中国内地和国际部署范围。 - **模型ID** **类型** @@ -304,8 +298,6 @@ Seedance 2.0、Runway Gen-4 ### Wan 2.6 -以下模型适用于中国内地和国际部署范围。 - **模型ID** **类型** @@ -372,8 +364,6 @@ Seedance 2.0、Runway Gen-4 ### Wan 2.5 -以下模型适用于中国内地和国际部署范围。 - **模型ID** **类型** @@ -400,8 +390,6 @@ Seedance 2.0、Runway Gen-4 ### Wan 2.2 -以下模型适用于中国内地和国际部署范围。 - **模型ID** **类型** @@ -460,8 +448,6 @@ wan-std / wan-pro 两种模式 ### Wan 2.1（推荐使用Wan 2.7） -以下模型适用于中国内地和国际部署范围。 - **模型ID** **类型** diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/vision-model.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/vision-model.md index da8325e3..9d006ab7 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/vision-model.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-experience/vision-model.md @@ -28,11 +28,11 @@ GPT-5.4、Claude Sonnet 4.6、Gemini 3 Pro GPT-5.4-mini、Gemini 3.1 Flash -`qwen3.6-flash` +`qwen3.7-flash` ## 图像与视频理解 -推荐从`qwen3.7-plus`开始，它是千问旗舰模型，支持1M上下文、最长2小时视频、Function Calling和内置工具等完整功能。当您的场景稳定后，可以尝试`qwen3.6-flash`来降低成本，它提供接近旗舰的效果，并支持相同的上下文长度和功能集。 +推荐从`qwen3.7-plus`开始，它是千问旗舰模型，支持1M上下文、最长2小时视频、Function Calling和内置工具等完整功能。当您的场景稳定后，可以尝试`qwen3.7-flash`来降低成本，它提供接近旗舰的效果，并支持相同的上下文长度和功能集。 ### 图像分辨率 @@ -40,7 +40,7 @@ GPT-5.4-mini、Gemini 3.1 Flash ### 视频支持 -- 最长2小时 / 2GB：`qwen3.7-plus`、`qwen3.6-plus`、`qwen3.6-flash`、`qwen3.5-plus`、`qwen3.5-flash` +- 最长2小时 / 2GB：`qwen3.7-plus`、`qwen3.6-plus`、`qwen3.7-flash`、`qwen3.6-flash`、`qwen3.5-plus`、`qwen3.5-flash` - 最长1小时 / 2GB：`qwen3-vl-plus`、`qwen3-vl-flash` @@ -53,7 +53,7 @@ GPT-5.4-mini、Gemini 3.1 Flash - Function Calling：Qwen3.7、Qwen3.6、Qwen3.5和Qwen3-VL系列模型均支持 -- 内置工具（联网搜索、代码执行，无需额外配置）：仅`qwen3.7-max-2026-06-08`、`qwen3.7-plus`、`qwen3.6-plus`、`qwen3.6-flash`、`qwen3.5-plus`、`qwen3.5-flash` +- 内置工具（联网搜索、代码执行，无需额外配置）：仅`qwen3.7-max-2026-06-08`、`qwen3.7-plus`、`qwen3.6-plus`、`qwen3.7-flash`、`qwen3.6-flash`、`qwen3.5-plus`、`qwen3.5-flash` ### 结构化输出 @@ -64,7 +64,7 @@ Qwen3.7、Qwen3.6、Qwen3.5和Qwen3-VL系列在非思考模式下支持此功能 ## OCR与文档提取 -`qwen3.5-ocr` 专为文档、表格、试卷和手写内容的文字提取而优化。您也可以使用`qwen3.7-plus`或`qwen3.6-flash`进行通用图片文字提取。 +`qwen3.5-ocr`专为文档、表格、试卷和手写内容的文字提取而优化。您也可以使用`qwen3.7-plus`或`qwen3.7-flash`进行通用图片文字提取。 ## 推荐模型 @@ -108,7 +108,7 @@ Qwen3.7、Qwen3.6、Qwen3.5和Qwen3-VL系列在非思考模式下支持此功能 支持 -`qwen3.6-flash` +`qwen3.7-flash` 1M @@ -148,8 +148,6 @@ Qwen3.7、Qwen3.6、Qwen3.5和Qwen3-VL系列在非思考模式下支持此功能 支持 -## 所有模型 - ### Qwen3.7 **模型 ID（Model ID）** @@ -232,6 +230,46 @@ Qwen3.7、Qwen3.6、Qwen3.5和Qwen3-VL系列在非思考模式下支持此功能 支持 +`qwen3.7-flash` + +文本、图像、视频 + +文本 + +1M + +64k + +256 + +64 + +支持 + +支持 + +支持 + +`qwen3.7-flash-2026-07-15` + +文本、图像、视频 + +文本 + +1M + +64k + +256 + +64 + +支持 + +支持 + +支持 + ### Qwen3.6 **模型 ID（Model ID）** @@ -354,6 +392,8 @@ Qwen3.7、Qwen3.6、Qwen3.5和Qwen3-VL系列在非思考模式下支持此功能 支持 +## 旧版模型 + ### Qwen3.5 **模型 ID（Model ID）** @@ -546,57 +586,34 @@ Qwen3.7、Qwen3.6、Qwen3.5和Qwen3-VL系列在非思考模式下支持此功能 #### Qwen3-VL -- `qwen3-vl-plus` - -- `qwen3-vl-plus-2026-01-25` +- `qwen3-vl-plus`及其快照版本 -- `qwen3-vl-flash` - -- `qwen3-vl-flash-2026-01-25` - - -#### Qwen2.5-VL - -- `qwen2.5-vl-72b-instruct` - -- `qwen2.5-vl-32b-instruct` - -- `qwen2.5-vl-7b-instruct` - -- `qwen2.5-vl-3b-instruct` +- `qwen3-vl-flash`及其快照版本 #### Qwen-Omni -- `qwen3-omni-flash` - -- `qwen3-omni-flash-2025-10-22` +- `qwen3-omni-flash`及其快照版本 - `qwen-omni-turbo`及其快照版本 #### Qwen-OCR -- `qwen-vl-ocr` +- `qwen-vl-ocr`及其快照版本 - `qwen-vl-ocr-latest` -- `qwen-vl-ocr-2025-07-14` - #### QVQ - `qvq-max` -- `qvq-max-2025-08-28` - - `qvq-plus` -- `qvq-plus-2025-08-27` - #### 旧版Qwen-VL -- `qwen-vl-max`及其快照版本 +- `qwen-vl-max` -- `qwen-vl-plus`及其快照版本 +- `qwen-vl-plus` diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-high-speed-inference/tpm-reservation.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-high-speed-inference/tpm-reservation.md index 8e525acd..ecfee151 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-high-speed-inference/tpm-reservation.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-high-speed-inference/tpm-reservation.md @@ -10,7 +10,7 @@ TPM 预留为指定模型锁定专属推理容量，确保业务高峰期不受 - 专属模型 code：创建 TPM 预留后，系统自动生成专属模型 code，您需要将 API 请求中的 `model` 参数替换为该 code。 -- 超额不中断：超出预留容量的请求自动降级为按量计费处理，无需修改代码。 +- 溢出策略：创建时可选超额处理方式——自动溢出至按量计费（默认，业务不中断）或仅使用预留容量（超出返回 429，不产生额外费用）。 ## **方案对比与选型** @@ -61,7 +61,7 @@ TPM 预留 **流量可预估、不能接受限流** -超出自动降级公共池按量，不中断 +可选：自动溢出按量（默认）/仅预留容量返回429 替换 model 参数即可 @@ -73,7 +73,7 @@ PTU（[模型部署](https://help.aliyun.com/zh/model-studio/model-deployment-in **高吞吐高性能** -超出转按量 +可选：自动溢出按量（默认）/仅PTU容量返回429 替换 model 参数即可 @@ -199,7 +199,7 @@ DeepSeek-v4-Pro `退款 = 降量部分预付费 - (降量部分预付费 × 已用时长/购买时长 × 1.5)` -- 超出保障额度后自动降级为标准按量计费，服务不中断。可在详情页**超额降级统计**中查看降级次数。 +- 溢出策略为「自动溢出」时：超出保障额度自动降级为标准按量计费，服务不中断，可在详情页**超额降级统计**查看降级次数；为「仅使用预留容量」时：超出返回 429，不产生额外费用。 - 服务到期后 2 小时内：实例仍为运行中，可继续调用，可续费；到期后 2~14 小时：实例已停止，不可调用，仍可续费；到期 14 小时后：实例已删除，不可恢复。 @@ -232,6 +232,7 @@ glm-5.1 \[0, 32K)：输入 1.0 / 输出 1.0 \[32K, 200K\]：输入 1.33 / 输出 1.17 + deepseek-v4-pro @@ -357,6 +358,14 @@ Qwen 系列 在输入框中输入天数，取值范围与购买时长一致。 + 溢出策略 + + 预留容量耗尽时，超出部分请求的处理方式。 + + 是 + + 自动溢出至按 token 付费（默认，超出转按量、业务不中断）/ 仅使用预留容量（超出返回 429、不产生额外费用） + 2. 确认参数后单击**立即购买**，在费用确认弹窗中核对费用，单击**确认支付**。 3. 在 TPM 预留详情页的**概览** Tab，找到**专属模型 code**，单击复制。 @@ -451,7 +460,7 @@ RPM 越大，建议购买的输入和输出 TPM 同比增大。 - 使用率趋势：可切换输入/输出方向，展示预留容量线和实际用量。 -- 超额降级统计：展示超出预留容量后被降级处理的次数。 +- 超额降级统计：展示超出预留容量后被降级处理的次数（仅「自动溢出」策略下产生降级）。 #### 监控 @@ -476,6 +485,10 @@ RPM 越大，建议购买的输入和输出 TPM 同比增大。 单击**扩缩容**，在弹窗中调整输入 TPM 和输出 TPM。 +**说明** + +输入 TPM 和输出 TPM 支持调整为 0：归 0 后不再产生容量费用，且专属模型 code 继续保留，避免因到期或退订导致 code 失效。但归 0 属于减配，已使用部分按 1.5 倍系数结算违约金（详见上方计费与使用说明）。 + #### 续费 单击**续订**，选择续费时长并完成支付。如已开启**到期自动续费**，系统在到期前一天 08:00 自动扣款续费。 @@ -524,7 +537,7 @@ RPM 越大，建议购买的输入和输出 TPM 同比增大。 **Q: 超出预留容量时会怎样？** -超出预留容量的请求自动降级为按量计费处理，服务不中断。可在详情页概览 Tab 的**超额降级统计**中查看降级次数和时间。频繁降级时建议扩容。 +取决于创建时选择的溢出策略：「自动溢出」策略下，超出预留容量的请求自动降级为按量计费，服务不中断，可在详情页概览 Tab 的**超额降级统计**查看降级次数和时间，频繁降级时建议扩容；「仅使用预留容量」策略下，超出请求返回 429 错误，不产生额外费用，频繁 429 时建议扩容。 **Q: 专属模型 code 怎么获取？** diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-monitoring/model-telemetry.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-monitoring/model-telemetry.md index 27e89c0c..a229ac38 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-monitoring/model-telemetry.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/model-monitoring/model-telemetry.md @@ -11,9 +11,9 @@ ## 支持的模型 -- **监控：普通监控**支持[选择模型](https://help.aliyun.com/zh/model-studio/models)中的所有模型，包括基于它们调优后的[自定义模型](https://help.aliyun.com/zh/model-studio/model-deployment-introduction#f17bf700c06k5)；**高级监控**支持北京、新加坡、弗吉尼亚地域下的所有模型**。** +- **监控：普通监控**支持[选择模型](https://help.aliyun.com/zh/model-studio/models)中的所有模型，包括基于它们调优后的[自定义模型](https://help.aliyun.com/zh/model-studio/model-deployment-introduction#f17bf700c06k5)；**高级监控**支持北京、上海、新加坡、弗吉尼亚地域下的所有模型**。** -- **告警功能：**支持北京、新加坡地域下的所有模型。 +- **告警功能：**支持北京、新加坡、弗吉尼亚地域下的所有模型。 ## **监控模型运行** @@ -22,9 +22,9 @@ > 列表记录按“模型 + 业务空间”维度生成。新模型在首次数据同步完成后自动加入列表（普通监控的延迟通常为小时级，请耐心等待；如需分钟级的数据洞察，请使用[高级监控](#2e5f2f0dffijg)）。 -列表顶部「监控数据看板」以卡片形式汇总**模型总量**、**总调用次数**、**总失败次数**、**平均调用时长**、**平均首包时长**。 +列表顶部「**监控数据**」以卡片形式汇总**模型总量**、**总调用次数**、**总失败次数**、**平均调用时长**、**平均首包时长**。 -「模型监控」表格列出各模型的**模型 Code**、**业务空间**、**调用总量**、**调用失败量**、**失败率**、**平均调用时长**、**平均首包时长**（除模型 Code、业务空间外均可排序），操作列提供**监控**、**日志**入口。 +「模型监控」表格列出各模型的**模型Code**、**业务空间**、**调用总量**、**调用失败量**、**失败率**、**平均调用时长**、**平均首Token延时**（除模型Code、业务空间外均可排序），操作列提供**监控**、**日志**入口。列表工具栏还提供**日志回流**入口，可将推理日志回流为训练数据集。 > 默认业务空间成员可查看所有业务空间的模型调用情况；子业务空间成员仅能查看当前空间的数据，无法切换查看其他业务空间数据。 @@ -80,7 +80,11 @@ ### **查看某次调用的 Token 消耗** -> 该功能目前仅适用于**华北2（北京）**地域的部分模型。 +> 该功能目前适用于**华北2（北京）**、新加坡地域的部分模型，弗吉尼亚地域同样支持。 + +**说明** + +**数据说明：**推理日志（高级监控）从调用发生到可查询存在分钟级延迟，请耐心等待；普通监控的用量汇总（如调用次数、Token 总量）延迟为小时级，高峰期可能达 1-2 小时。如遇无数据或查不到记录的情况，请先确认已等待足够的数据同步时间。仅记录**开启推理日志后**的调用数据，开通前的历史调用无法追溯。 1. 使用主账号（[或拥有足够权限的子账号](#f9d06146c0xe0)）登录，在目标业务空间的[模型监控（北京）](https://bailian.console.aliyun.com/?tab=model#/model-telemetry)页面，点击右上角的**模型监控配置**，按照指引依次开通审计日志和推理日志。 @@ -88,7 +92,7 @@ 2. 在模型监控列表中找到目标模型，点击其右侧**操作**列的**日志**。 -3. **日志**页签展示该模型的[实时推理](#f131611173sdx)调用记录，**用量**字段即为本次调用的Token消耗。 +3. **日志**页签以表格形式展示该模型的[实时推理](#f131611173sdx)调用记录，表格包含**Request ID/调用时间**、**调用时长**、**状态码**（支持筛选）、**错误码**、**用量**、**请求和响应**、**操作**等列。其中**用量**字段即为本次调用的Token消耗。 ### **创建异常消耗告警** @@ -100,10 +104,14 @@ **重要** -该功能目前仅适用于**华北2（北京）**地域的部分模型。 +该功能目前适用于**华北2（北京）**、新加坡地域的部分模型，弗吉尼亚地域同样支持。 模型监控支持查看模型的每一次对话，包括输入、输出及耗时，是故障排查和内容审计的关键工具。 +**说明** + +**数据说明：**日志从调用发生到可查询存在分钟级延迟，请耐心等待。如遇实时更新延迟或查不到记录的情况，请先确认已等待足够时间（普通监控用量汇总为小时级延迟，高级监控/推理日志为分钟级）。仅记录**开启推理日志后**的调用数据，开通前的历史调用无法追溯。 + ### **步骤一：开通日志** 使用主账号（[或拥有足够权限的子账号](#f9d06146c0xe0)）登录，在目标业务空间的[模型监控（北京）](https://bailian.console.aliyun.com/?tab=model#/model-telemetry)页面，点击右上角的**模型监控配置**，按照指引依次开通审计日志和推理日志。 @@ -116,7 +124,7 @@ 1. 在模型监控列表中找到目标模型，点击其右侧**操作**列的**日志**。 -2. **日志**页签展示该模型的[实时推理](#f131611173sdx)调用记录，**请求和响应**字段分别对应本次调用的输入与输出。 +2. **日志**页签以表格形式展示该模型的[实时推理](#f131611173sdx)调用记录，表格包含**Request ID/调用时间**、**调用时长**、**状态码**（支持筛选）、**错误码**、**用量**、**请求和响应**、**操作**等列。其中**请求和响应**字段分别对应本次调用的输入与输出。 **支持请求和响应的模型** @@ -152,14 +160,20 @@ - 三方模型：deepseek-v3.1、deepseek-v3.2、deepseek-v3.2-exp +并非所有模型都支持推理日志（请求/响应内容记录）。是否支持由模型本身决定，与模型是否为多模态无关。当所选模型不支持时，界面会显示**当前模型暂不支持日志**。 + +请注意：请求和响应内容**仅在开启推理日志后**才会被采集，开通前的历史调用不会补录。若某次调用的输出内容缺失或存在日志缺失，请先确认该模型是否支持推理日志，以及推理日志是否已在调用发生前完成开通。 + ## **建立主动告警** **重要** -该功能目前仅适用于新加坡和华北2（北京）地域。 +该功能目前适用于新加坡、北京和弗吉尼亚地域。 模型的静默失败（如超时、Token消耗突增），传统应用日志难以发现。模型监控支持对监控指标（如成本、失败率、响应延迟）设置告警。一旦指标出现异常，系统立即告警。 +模型告警页面包含**告警规则**和**告警历史**两个页签。**告警历史**页签可按告警时间、告警规则、告警等级、状态筛选查看历史告警记录，点击详情可查看告警详情。 + ### **步骤一：开启高级监控** 1. 使用主账号（[或拥有足够权限的子账号](#54ea9ba526ovz)）登录，在目标业务空间的模型监控（[北京](https://bailian.console.aliyun.com/?tab=model#/model-telemetry) 或 [新加坡](https://modelstudio.console.aliyun.com/?tab=dashboard#/model-telemetry)）页面，点击右上角的**模型监控配置**。 @@ -171,7 +185,7 @@ 1. 在模型告警（[北京](https://bailian.console.aliyun.com/?tab=model#/model-alert) 或[新加坡](https://modelstudio.console.aliyun.com/?tab=dashboard#/model-alert)）页面，点击右上角的**创建告警规则**。 -2. 在对话框中，选择要监控的模型和监控模板，确认无误后点击**创建**。当指定的监控指标（如调用统计或性能指标）出现异常时，系统将通知您的团队。 +2. 在对话框中，选择要监控的模型和监控模板，确认无误后点击**确定**。当指定的监控指标（如调用统计或性能指标）出现异常时，系统将通知您的团队。 - **通知方式：**支持短信、电子邮件、电话、钉钉群机器人、企业微信机器人及Webhook。 @@ -289,12 +303,20 @@ 模型非首包时长p99 + **TPS** + + model\_tps\_per\_request + + 单次请求输出 Token 速度（TPS），每秒生成 Token 数，衡量模型生成速度（仅高级监控支持） + **用量** model\_usage 模型用量总和 + **关于 TPS 指标：**`model_tps_per_request` 仅在**高级监控**中展示，高级监控为收费功能。TPS（每秒生成 Token 数）与非首包时长（每 Token 的平均生成耗时）呈倒数关系（TPS ≈ 1 ÷ 非首包时长均值）。排查响应慢的问题时，建议结合首 Token 延时（TTFT）、非首 Token 延时及输入 Token 量综合分析，单次调用总耗时还受输入长度、网络等因素影响，不能仅凭 TPS 判断。TPS 触发的是按请求维度的限流，区别于 TPM（每分钟 Token 数）的按账号维度限流。 + - **HTTP API：**`{HTTP API}`需替换为前面[步骤一](#title-tkb-ds1-4p5)获取的HTTP API地址。 - **Authorization：**需将阿里云账号的 `AccessKey:AccessKeySecret` 拼接后进行Base64编码，并以 `Basic 编码后字符串` 的形式提供。 @@ -530,6 +552,12 @@ 模型非首包时长p99 + **TPS** + + model\_tps\_per\_request + + 单次请求输出 Token 速度（TPS），每秒生成 Token 数，衡量模型生成速度（仅高级监控支持） + **用量** model\_usage @@ -715,6 +743,12 @@ } ``` +- **API Key 用量限额：**当前不支持为单个 API Key 设置月度或每日 Token 消耗上限并自动停服（即不支持硬性阻断，无法实现额度用尽后自动禁用 API 调用）。如需防止意外欠费，可采用以下替代方案： + + - **告警通知 + 手动禁用 Key：**开启**高级监控**（收费功能）后，可对 Token 消耗配置告警阈值，当消耗超出阈值时接收消费预警通知，再人工介入处理（如手动禁用对应 API Key）。普通监控仅支持基础用量查看，不支持配置告警。 + + - **免费额度用完即停：**适用于仅使用免费配额的场景，开启后免费额度耗尽时自动停止调用。 + ## **计费说明** diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-desc.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-desc.md new file mode 100644 index 00000000..8eaec477 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-desc.md @@ -0,0 +1,17 @@ +# AOQ SDK简介 + +阿里云AOQ SDK面向实时多模态场景，帮助开发者快速搭建基于阿里云实时多模态模型的解决方案。 + +欢迎使用AOQ SDK来实现您的业务需求，阿里云AOQ基于阿里云多年以来服务全球客户的深厚技术沉淀，以面向实时多模态场景向全球开发者开放的产品，致力于帮助全球开发者快速高效的搭建基于阿里云的实时多模态模型的解决方案。 + +## **API设计与回调机制** + +阿里云AOQ SDK的API设计遵循状态式API设计原则。客户只需要把想要的状态通过API传至SDK，SDK内部会进行状态判断，并在适当时机达成用户想要的目标状态，同时在执行时通过回调机制及时向客户反馈SDK当前执行状态，使客户使用我们的SDK时享受到无忧的体验。您只需设置期望的状态，SDK便会自动处理并通过回调通知您，无需重复调用，也不需要关心具体的调用时机和场景。 + +## **异常处理机制** + +遇到异常情况时，AOQ SDK会优先尝试内部解决，仅在内部无法恢复时，才会提示客户介入处理。目前这些无法恢复的内部错误主要分为物理限制（如网络问题、音频设备故障、视频设备故障、CPU或内存资源限制）和外部因素（token无效）两类。 + +## **下载指南** + +阿里云AOQ SDK是阿里云自研产品，请通过阿里云官网进行下载，详情请参见[SDK下载](https://help.aliyun.com/zh/model-studio/realtime-sdk-download)。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-audio-features.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-audio-features.md new file mode 100644 index 00000000..f675adaa --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-audio-features.md @@ -0,0 +1,1179 @@ +# 音频常用功能介绍 + +AOQ Client SDK 提供了完整的音频能力，覆盖音频采集、播放、编解码配置、扬声器管理、文件混音、外部音频流注入、音频帧数据回调等核心场景。本文档基于 Android（Java）、iOS（Objective-C）、Ohos（ArkTS）三个平台的公开 API，对音频常用功能进行统一介绍。 + +## **音频采集** + +音频采集用于打开设备麦克风，将实时音频数据送入 SDK 编码推流管线。SDK 支持两种采集模式： + +- **内部采集**（默认）：SDK 自动管理麦克风设备的打开、录音和关闭。 + +- **外部采集**：由应用自行管理麦克风，采集到的 PCM 数据通过外部音频流接口输入 SDK。 + + +### 配置参数 + +**参数** + +**类型** + +**默认值** + +**说明** + +isExternal + +bool + +false + +是否使用外部采集模式 + +isVoipMode + +bool + +false + +是否启用 VoIP 模式（硬件 AEC），移动端有效，采集播放参数先到为准 + +channel + +int + +1 + +采集通道数，支持 1（单声道）/ 2（立体声） + +### API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +开启采集 + +`startAudioCapture(config)` + +`startAudioCapture:config:` + +`startAudioCapture(config)` + +关闭采集 + +`stopAudioCapture()` + +`stopAudioCapture` + +`stopAudioCapture()` + +静音/取消静音 + +`muteAudioCapture(mute)` + +`muteAudioCapture:` + +`muteAudioCapture(mute)` + +### 使用示例 + +**Android** + +``` +AoqAudioCaptureConfig config = new AoqAudioCaptureConfig(); +config.isVoipMode = true; +config.channel = 1; +engine.startAudioCapture(config); +``` + +**iOS** + +``` +AoqAudioCaptureConfig *config = [[AoqAudioCaptureConfig alloc] init]; +config.isVoipMode = YES; +config.channel = 1; +[engine startAudioCapture:config]; +``` + +**Ohos** + +``` +const config: AoqAudioCaptureConfig = { isVoipMode: true, channel: 1 }; +engine.startAudioCapture(config); +``` + +## **音频播放** + +音频播放用于将接收到的远端音频数据渲染到本地扬声器或耳机。SDK 支持播放暂停/恢复（带淡入淡出）、打断当前轮音频通话等高级控制。 + +### 配置参数 + +**参数** + +**类型** + +**默认值** + +**说明** + +isVoipMode + +bool + +false + +是否启用 VoIP 模式（硬件AEC），移动端有效，采集播放参数先到为准 + +isDefaultSpeaker + +bool + +true + +是否默认使用扬声器（移动端有效，非VoIP时无效） + +isExternal + +bool + +false + +是否使用外部播放模式 + +channel + +int + +1 + +播放通道数，支持 1（单声道）/ 2（立体声） + +### API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +开始播放 + +`startAudioPlayer(config)` + +`startAudioPlayer:config:` + +`startAudioPlayer(config)` + +停止播放 + +`stopAudioPlayer()` + +`stopAudioPlayer` + +`stopAudioPlayer()` + +暂停播放 + +`pauseAudioPlayer(fadeMs)` + +`pauseAudioPlayer:` + +`pauseAudioPlayer(fadeMs)` + +恢复播放 + +`resumeAudioPlayer(fadeMs)` + +`resumeAudioPlayer:` + +`resumeAudioPlayer(fadeMs)` + +打断通话 + +`interruptAudioPlayer(trackType, fadeMs)` + +`interruptAudioPlayer:fadeMs:` + +`interruptAudioPlayer(trackType, fadeMs)` + +**说明** + +**fadeMs 参数**：暂停和恢复播放时的淡入/淡出时长（毫秒），设为 0 则立即切换。 + +## **扬声器管理** + +控制音频输出设备在扬声器和听筒之间切换。 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +切换扬声器 + +`enableSpeakerphone(enable)` + +`enableSpeakerphone:` + +`enableSpeakerphone(enable)` + +查询扬声器状态 + +`isSpeakerphoneEnabled()` + +`isSpeakerphoneEnabled` + +`isSpeakerphoneEnabled()` + +**说明** + +需要在 VoIP 模式下才允许切换，非 VoIP 时，`enableSpeakerphone` 调用有 OnError(AoqECAudioDeviceEarpieceRequiresVoipMode) 错误通知。 + +**说明** + +**iOS 特殊行为**：iPad 设备只有扬声器模式；当 AVAudioSession 不是 PlayAndRecord 类别时，也始终返回 YES。 + +## **音频编解码配置** + +设置音频上行（编码器）和下行（解码器）的编码格式、采样率、声道数和码率。表示推流/拉流的格式。 + +### 配置参数 + +**参数** + +**类型** + +**默认值** + +**说明** + +trackType + +AoqTrackType + +Audio + +音频轨道类型，当前只支持一条音频流 + +codecType + +AoqEncoderType + +AudioPCM + +编码类型：AudioPCM(1) 或 AudioOpus(2) + +sampleRate + +int + +48000 + +采样率，Opus 支持 8K/16K/48K，PCM 支持 8K/16K/32K/48K + +channel + +int + +1 + +声道数，支持 1（单声道）/ 2（立体声） + +bitrate + +int + +32000 + +码率（bps） + +### API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +设置编码参数 + +`setAudioEncoderConfig(config)` + +`setAudioEncoderConfig:` + +`setAudioEncoderConfig(config)` + +设置解码参数 + +`setAudioDecoderConfig(config)` + +`setAudioDecoderConfig:` + +`setAudioDecoderConfig(config)` + +### 支持的编码格式 + +**枚举值** + +**数值** + +**说明** + +AoqEncoderTypeAudioPCM + +1 + +PCM 裸音频 + +AoqEncoderTypeAudioOpus + +2 + +Opus 编码 + +## **音频文件混音** + +支持将本地音频文件混入当前音频流中一起推流和/或本地播放。每个音频文件通过业务自分配的 `fileId` 标识，可同时管理多个文件实例。 + +### 混音配置参数 + +**参数** + +**类型** + +**默认值** + +**说明** + +fileName + +String + +\- + +音频文件路径（含文件名） + +cycles + +int + +\-1 + +循环次数，-1 表示无限循环 + +startPosMs + +long + +0 + +起始播放位置（毫秒） + +publishVolume + +int + +100 + +推流音量 \[0-100\] + +playoutVolume + +int + +100 + +本地播放音量 \[0-100\] + +### API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +开始播放 + +`startAudioFile(fileId, config)` + +`startAudioFile:config:` + +`startAudioFile(fileId, config)` + +停止播放 + +`stopAudioFile(fileId)` + +`stopAudioFile:` + +`stopAudioFile(fileId)` + +暂停 + +`pauseAudioFile(fileId)` + +`pauseAudioFile:` + +`pauseAudioFile(fileId)` + +恢复 + +`resumeAudioFile(fileId)` + +`resumeAudioFile:` + +`resumeAudioFile(fileId)` + +获取文件时长 + +`getAudioFileDuration(fileId)` + +`getAudioFileDuration:` + +`getAudioFileDuration(fileId)` + +获取当前位置 + +`getAudioFileCurrentPosition(fileId)` + +`getAudioFileCurrentPosition:` + +`getAudioFileCurrentPosition(fileId)` + +设置播放位置 + +`setAudioFilePositionMillis(fileId, pos)` + +`setAudioFilePositionMillis:positionMillis:` + +`setAudioFilePositionMillis(fileId, pos)` + +设置音量 + +`setAudioFileVolume(fileId, type, vol)` + +`setAudioFileVolume:type:volume:` + +`setAudioFileVolume(fileId, type, vol)` + +获取音量 + +`getAudioFileVolume(fileId, type)` + +`getAudioFileVolume:type:` + +`getAudioFileVolume(fileId, type)` + +**说明** + +**音量方向（type）**：`AoqAudioStreamPublish(0)` 控制推流音量；`AoqAudioStreamPlayout(1)` 控制本地播放音量。 + +### 状态回调 + +**状态码** + +**数值** + +**说明** + +AoqAudioFileNone + +0 + +初始状态 + +AoqAudioFileStarted + +1 + +已开始播放 + +AoqAudioFileStopped + +2 + +已停止 + +AoqAudioFilePaused + +3 + +已暂停 + +AoqAudioFileResumed + +4 + +已恢复 + +AoqAudioFileEnded + +5 + +播放结束 + +AoqAudioFileBuffering + +6 + +缓冲中 + +AoqAudioFileBufferingEnd + +7 + +缓冲结束 + +AoqAudioFileFailed + +8 + +播放失败 + +## **外部音频流** + +外部音频流允许将应用生成的 PCM 音频数据注入到 SDK 的音频管线中，支持推流和/或本地播放。典型场景包括 TTS 语音合成输出、AI 模型音频输出、背景音效等。每个外部音频流通过业务自分配的 `streamId` 标识。 + +### 配置参数 + +**参数** + +**类型** + +**默认值** + +**说明** + +trackType + +AoqTrackType + +Audio + +音频轨道类型 + +codecType + +AoqEncoderType + +AudioPCM + +音频流格式 + +channels + +int + +1 + +声道数 + +sampleRate + +int + +48000 + +采样率，支持 8/12/16/24/32/44.1/48/64/88.2/96/176.4/192K + +playoutVolume + +int + +100 + +本地播放音量 \[0-100\] + +publishVolume + +int + +100 + +推流音量 \[0-100\] + +maxBufferDuration + +int + +600000 + +最大缓冲时长（毫秒），取值范围 \[100, ~\]，超过时 Push 失败 + +enable3A + +bool + +false + +输入 PCM 是否经过 3A 处理 + +### API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +新增外部音频流 + +`addAudioExternalStream(streamId, config)` + +`addAudioExternalStream:config:` + +`addAudioExternalStream(streamId, config)` + +输入音频数据 + +`pushAudioExternalStreamData(streamId, data)` + +`pushAudioExternalStreamData:data:` + +`pushAudioExternalStreamData(streamId, data)` + +设置音量 + +`setAudioExternalStreamVolume(streamId, type, vol)` + +`setAudioExternalStreamVolume:type:volume:` + +`setAudioExternalStreamVolume(streamId, type, vol)` + +获取音量 + +`getAudioExternalStreamVolume(streamId, type)` + +`getAudioExternalStreamVolume:type:` + +`getAudioExternalStreamVolume(streamId, type)` + +清空缓存 + +`clearAudioExternalStreamBuffer(streamId, fadeoutMs)` + +`clearAudioExternalStreamBuffer:fadeoutMs:` + +`clearAudioExternalStreamBuffer(streamId, fadeoutMs)` + +移除流 + +`removeAudioExternalStream(streamId)` + +`removeAudioExternalStream:` + +`removeAudioExternalStream(streamId)` + +### Push 数据最佳实践 + +- 需要循环调用 `pushAudioExternalStreamData`，保证数据 push 成功 + +- 返回错误码 110（缓冲区满）时短暂 Sleep 30ms 后重试，不要丢弃数据 + +- 引擎退出前先停止推送循环，再调用 `removeAudioExternalStream` + +- 实时采集每帧 10ms 长，有数据就调用 push；从文件解析每帧 40ms 长，间隔 30ms 调用 push 一次 + + +## **音频帧数据回调** + +音频帧回调允许开发者在音频管线的不同位置获取原始 PCM 数据，用于音频分析、自定义处理、录制等场景。 + +### 支持的数据源位置 + +**数据源** + +**枚举值** + +**说明** + +Captured + +0 + +采集后的原始音频数据（未经 3A 处理） + +ProcessCaptured + +1 + +经过 3A 处理后的音频数据，需要 Connect 成功后才回调数据 + +Publish + +2 + +即将推流的音频数据（需要 Connect 成功） + +Playback + +3 + +即将播放的音频数据（远端下行） + +### 回调配置参数 + +**参数** + +**类型** + +**默认值** + +**说明** + +sampleRate + +int + +48000 + +回调音频的采样率 + +channels + +int + +1 + +回调音频的声道数，支持 1/2 + +mode + +AoqAudioObserverMode + +ReadOnly + +只读(0)/读写(1) 模式 + +### 使用步骤 + +1. **注册观察者**：调用 `setAudioFrameObserver` 设置音频帧回调监听器 + +2. **启用数据源**：调用 `enableAudioFrameObserver` 选择需要监听的数据源位置，开启回调 + +3. **处理回调数据**：在回调函数中获取 PCM 数据 + + +### API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +注册观察者 + +`setAudioFrameObserver(listener)` + +`setAudioFrameObserver:` + +`setAudioFrameObserver(observer)` + +启用回调 + +`enableAudioFrameObserver(enabled, source, config)` + +`enableAudioFrameObserver:audioSource:config:` + +`enableAudioFrameObserver(enabled, source, config)` + +### 回调方法 + +**回调** + +**Android** + +**iOS** + +**Ohos** + +采集数据 + +`onCapturedAudioFrame(frame)` + +`onCapturedAudioFrame:` + +`onCapturedAudioFrame(frame)` + +3A 后数据 + +`onProcessCapturedAudioFrame(frame)` + +`onProcessCapturedAudioFrame:` + +`onProcessCapturedAudioFrame(frame)` + +推流数据 + +`onPublishAudioFrame(trackType, frame)` + +`onPublishAudioFrame:frame:` + +`onPublishAudioFrame(trackType, frame)` + +播放数据 + +`onPlaybackAudioFrame(frame)` + +`onPlaybackAudioFrame:` + +`onPlaybackAudioFrame(frame)` + +## **音频状态与路由** + +SDK 自动监测音频设备的状态变化和路由切换，并通过回调通知应用层。 + +### 设备状态码 + +**状态码** + +**值** + +**说明** + +AoqAudioDeviceNone + +0 + +初始状态 + +RecordStarting + +1 + +采集启动中 + +RecordStarted + +2 + +采集已启动 + +RecordStopping + +3 + +采集停止中 + +RecordStopped + +4 + +采集已停止 + +RecordFail + +5 + +采集失败 + +PlayStarting + +6 + +播放启动中 + +PlayStarted + +7 + +播放已启动 + +PlayStopping + +8 + +播放停止中 + +PlayStopped + +9 + +播放已停止 + +PlayFail + +10 + +播放失败 + +### 设备路由类型 + +**路由** + +**值** + +**说明** + +Default + +0 + +默认 + +Headset + +1 + +有线耳机 + +Earpiece + +2 + +听筒 + +HeadsetNoMic + +3 + +无麦克风耳机 + +SpeakerPhone + +4 + +扬声器 + +Usb + +5 + +USB 设备 + +Bluetooth + +6 + +蓝牙 SCO + +BluetoothA2dp + +7 + +蓝牙 A2DP + +### 回调对照 + +**回调** + +**Android** + +**iOS** + +**Ohos** + +设备状态变化 + +`onAudioDeviceStateChanged(state)` + +`onAudioDeviceStateChanged:` + +`onAudioDeviceStateChanged(state, reason)` + +路由变化 + +`onAudioDeviceRouteChanged(routeType)` + +`onAudioDeviceRouteChanged:` + +`onAudioDeviceRouteChanged(routeType)` + +设备中断 + +`onAudioDeviceInterrupted(interrupt)` + +`onAudioDeviceInterrupted:` + +`onAudioDeviceInterrupted(interrupt)` + +文件状态 + +`onAudioFileState(state)` + +`onAudioFileState:` + +`onAudioFileState(fileId, stateCode, errorCode)` + +## **音频错误码与警告码** + +### 音频错误码 + +**错误码** + +**值** + +**说明** + +AoqErrorCodeAudio + +100 + +通用音频错误 + +AudioExternalBufferFull + +110 + +外部缓冲区已满 + +AudioDevice + +120 + +设备通用错误 + +RecordingAuthFailed + +121 + +麦克风权限失败 + +RecordingOccupied + +122 + +麦克风被占用 + +RecordingBackgroundStart + +123 + +后台启动录音 + +RecordingStartFail + +124 + +录音启动失败 + +PlayoutOccupied + +125 + +播放设备被占用 + +PlayoutBackgroundStart + +126 + +后台启动播放 + +PlayoutStartFail + +127 + +播放启动失败 + +EarpieceRequiresVoipMode + +128 + +听筒需要启用 VoIP 模式 + +### 音频警告码 + +**警告码** + +**值** + +**说明** + +AoqWCAudio + +100 + +通用音频警告 + +AudioHowling + +101 + +啸叫检测 + +AudioDevice + +120 + +设备通用警告 + +MicEnumerateError + +121 + +麦克风枚举错误 + +MicStartTimeout + +122 + +麦克风启动超时 + +RecordingError + +123 + +录音错误 + +SpeakerEnumerateError + +124 + +扬声器枚举错误 + +SpeakerStartTimeout + +125 + +扬声器启动超时 + +PlayoutError + +126 + +播放错误 + +## **iOS 专有：AVAudioSession 控制** + +iOS 平台提供了 `setAudioSessionRestriction` 接口，可精细控制 SDK 对系统 AVAudioSession 的管理权限。 + +**控制项** + +**说明** + +SetCategory + +SDK 是否有权设置 Session 类别 + +ConfigureSession + +SDK 是否有权配置 Session 参数 + +DeactivateSession + +SDK 是否有权停用 Session + +ActivateSession + +SDK 是否有权激活 Session + +通过按位组合传入 restriction 值，可限制 SDK 对 AVAudioSession 的控制范围，避免与应用层其他音频组件冲突。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-connection-management.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-connection-management.md new file mode 100644 index 00000000..433f6552 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-connection-management.md @@ -0,0 +1,171 @@ +# 连接状态管理 + +本章节介绍 AOQ Client SDK 的连接状态机及其对应的 API 调用。 + +## **连接状态图** + +下图描述了 AOQ Client SDK 的连接状态迁移关系： + +AOQ Client SDK 连接状态迁移图 + +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3811364871/p1088081.svg) + +## **状态说明** + +**状态** + +**枚举值** + +**说明** + +链接中（Connecting） + +1 + +调用 `connect` 后进入，正在与 AI Service 建立连接 + +已连接（Connected） + +2 + +连接建立成功，可正常收发音视频和数据消息 + +失败（Failed） + +3 + +连接异常（鉴权失败/超时/服务端拒绝等），SDK 内部会自动迁移到已断开 + +已断开（Disconnected） + +0 + +初始状态/主动断开/异常断开后的终态 + +## **状态迁移规则** + +1. **App 调用** `connect` → 进入 `Connecting` 状态。 + +2. **链接成功** → 从 `Connecting` 迁移到 `Connected`。 + +3. **链接异常** → 从 `Connecting` 迁移到 `Failed`，随后 SDK 自动迁移到 `Disconnected`。 + +4. **链接异常** → 从 `Connected` 迁移到 `Failed`，随后 SDK 自动迁移到 `Disconnected`。 + +5. **App 调用** `disconnect` → 从 `Connected` 迁移到 `Disconnected`。 + + +**说明**：`Failed` 是瞬态，SDK 触发 `onConnectionStatusChange(Failed)` 后会自动迁移到 `Disconnected`，业务层无需手动调用 disconnect。 + +## **connect API** + +调用 `connect` 发起与 AI Service 的连接，传入由 AppServer allocate 接口返回的鉴权凭证。 + +### **方法签名** + +**Android：** + +``` +public abstract int connect(@NonNull AoqConnectConfig config); +``` + +**iOS：** + +``` +- (int)connect:(AoqConnectConfig * _Nonnull)config; +``` + +**Ohos：** + +``` +connect(config: AoqConnectConfig): number; +``` + +**返回值：**`0` 表示调用成功（异步建连）；`< 0` 表示失败。 + +### **行为说明** + +- 调用后触发 `onConnectionStatusChange(connecting)` 回调。 + +- 如果链接成功时触发 `onConnectionStatusChange(connected)` 回调。 + +- 如果链接失败时触发 `onConnectionStatusChange(failed)` 回调。 + + +## **disconnect API** + +调用 `disconnect` 主动断开与 AI Service 的连接。 + +### **方法签名** + +**Android：** + +``` +public abstract int disconnect(); +``` + +**iOS：** + +``` +- (int)disconnect; +``` + +**Ohos：** + +``` +disconnect(): number; +``` + +**返回值：**`0` 表示成功；`< 0` 表示失败。 + +### **行为说明** + +- 调用后触发 `onConnectionStatusChange(Disconnected)` 回调。 + +- 引擎不会自动释放，可重新调用 `connect` 进行重连。 + +- 未连接状态下调用 `disconnect` 是安全的，返回 0。 + + +## **onConnectionStatusChange 回调** + +连接状态变化时，SDK 通过此回调通知业务层。 + +**Android：** + +``` +public void onConnectionStatusChange( + @NonNull AoqClientEngine.AoqConnectionStatus status) {} +``` + +**iOS：** + +``` +- (void)onConnectionStatusChange:(AoqConnectionStatus)status; +``` + +**Ohos：** + +``` +onConnectionStatusChange?: (status: AoqConnectionStatus) => void; +``` + +### **示例代码** + +``` +func onConnectionStatusChange(_ status: AoqConnectionStatus) { + switch status { + case .connecting: + print("正在连接...") + case .connected: + print("连接成功") + // 连接成功后可发送 session.update + case .failed: + print("连接失败") + // SDK 会自动迁移到 disconnected，无需手动 disconnect + case .disconnected: + print("已断开") + // 可根据业务决定是否重连 + } +} +``` diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-capture.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-capture.md new file mode 100644 index 00000000..670fc6f0 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-capture.md @@ -0,0 +1,253 @@ +# 自定义音频采集 + +介绍如何使用 AOQ Client SDK 实现自定义音频采集功能，包括外部音频流的添加、PCM 数据推送和管理。 + +## **功能介绍** + +AOQ Client SDK 内部音频模块可满足应用中对基本音频功能的需求，但在特定场景中，SDK 内部的音频采集模块可能无法满足开发需求，需要实现自定义音频采集功能，例如： + +- 解决音频采集设备被占用问题。 + +- 需要从定制的采集系统、音频文件中获取音频数据后交给 SDK 传输。 + +- 需要将 AI TTS 生成的音频数据通过 SDK 推流传输。 + + +AOQ Client SDK 支持灵活的自定义采集功能，允许用户根据业务场景自行管理音频设备与音频源。外部音频流的数据会与内部采集的音频数据混音后一起推流发送。 + +## **示例代码** + +暂无 + +## **前提条件** + +- 已创建引擎实例（调用 `createEngine`）。 + +- 已成功连接服务器（`onConnectionStatusChange` 回调状态为 `AoqConnectionStatusConnected`）。 + + +## **功能实现** + +### **1\. 打开或关闭音频采集** + +需要先开启音频采集，外部音频流输入的数据会与内部采集数据混音后一起推流。如果不需要内部麦克风采集，可以设置 `isExternal=true` 关闭内部采集设备。 + +``` +// 方式一：开启内部采集，外部音频流数据会与麦克风数据混音推流 +AoqClientEngine.AoqAudioCaptureConfig config = new AoqClientEngine.AoqAudioCaptureConfig(); +config.isExternal = false; // 使用内部麦克风采集 +config.isVoipMode = false; +engine.startAudioCapture(config); + +// 方式二：关闭内部采集，仅推送外部音频流数据 +AoqClientEngine.AoqAudioCaptureConfig config = new AoqClientEngine.AoqAudioCaptureConfig(); +config.isExternal = true; // 不打开麦克风，由外部音频流提供数据 +engine.startAudioCapture(config); +``` + +### **2\. 连接成功后，添加外部音频流** + +在 `onConnectionStatusChange` 回调状态变为 `AoqConnectionStatusConnected` 后，调用 `addAudioExternalStream` 添加外部音频流。需要指定一个唯一的 `streamId` 用于后续推送数据和管理。 + +如果需要音频 3A 处理（回声消除、噪声抑制、自动增益），请配置 `AoqAudioExternalStreamConfig` 中的 `enable3A` 参数。 + +``` +// 在 onConnectionStatusChange 回调中确认连接成功后添加 +@Override +public void onConnectionStatusChange(AoqClientEngine.AoqConnectionStatus status) { + if (status == AoqClientEngine.AoqConnectionStatus.AoqConnectionStatusConnected) { + addExternalAudioStream(); + } +} + +private void addExternalAudioStream() { + AoqClientEngine.AoqAudioExternalStreamConfig config = new AoqClientEngine.AoqAudioExternalStreamConfig(); + config.sampleRate = 48000; // 采样率，需与实际音频数据一致 + config.channels = 1; // 声道数 + config.publishVolume = 100; // 推流音量 [0-100] + config.playoutVolume = 0; // 本地播放音量 [0-100]，0 表示不本地播放 + config.maxBufferDuration = 1000; // 最大缓冲时长（毫秒） + config.enable3A = true; // 是否对输入 PCM 进行 3A 处理 + + String streamId = "external_audio_1"; + int ret = engine.addAudioExternalStream(streamId, config); + if (ret == 0) { + mExternalStreamId = streamId; + } +} +``` + +**参数说明：** + +**参数** + +**类型** + +**默认值** + +**说明** + +trackType + +AoqTrackType + +AoqTrackTypeAudio + +音频轨道类型 + +codecType + +AoqEncoderType + +AoqEncoderTypeAudioPCM + +音频流格式 + +channels + +int + +1 + +声道数 + +sampleRate + +int + +48000 + +采样率（Hz） + +playoutVolume + +int + +100 + +播放音量 \[0-100\] + +publishVolume + +int + +100 + +推流音量 \[0-100\] + +maxBufferDuration + +int + +1000 + +最大缓冲时长（毫秒） + +enable3A + +boolean + +false + +是否对输入 PCM 进行 3A 处理 + +### **3\. 实现自采集模块或从文件获取 PCM 数据** + +自定义采集功能需要根据业务场景自行采集并处理音频数据，之后将数据传入 SDK 进行传输。常见的数据来源： + +- **麦克风采集**：通过 Android AudioRecord 采集 PCM 数据。 + +- **文件读取**：从本地 PCM/WAV 音频文件中解析获取 PCM 数据。 + +- **AI TTS**：从语音合成引擎获取 PCM 数据。 + +- **网络流**：从网络音频流中解码获取 PCM 数据。 + + +音频数据需要为 PCM 格式，并记录对应的采样率、声道数等参数，用于构造 `AoqAudioFrameData` 对象。 + +### **4\. 通过外部音频流 ID 推送音频数据到 SDK** + +调用 `pushAudioExternalStreamData` 接口，将采集到的 PCM 音频数据传入 SDK。 + +- 从硬件采集：建议采集 10ms 为一帧数据，采集到数据就 push 给 SDK。 + +- 从文件解析：建议 40ms 为一帧数据，每 push 一帧 Sleep 30ms 后 push 下一帧。 + +- 需要维护一个 `running` 标记，当引擎退出或 stream ID 被删除时退出推送循环。 + + +``` +// 成员变量：控制推送循环的运行标记 +private volatile boolean mPushRunning = false; + +// 推送单帧音频数据 + +private void pushAudioData(byte[] audioData, int bytesRead) { + + if (engine == null || mExternalStreamId == null || bytesRead <= 0) { + return; + } + + int channels = 1; + int bytesPerSample = 2; // 16bit PCM + int sampleRate = 48000; + + // 构造音频帧数据 + AoqClientEngine.AoqAudioFrameData frameData = new AoqClientEngine.AoqAudioFrameData(); + frameData.dataPtr = audioData; + frameData.dataSize = bytesRead; + frameData.numOfSamples = bytesRead / (channels * bytesPerSample); + frameData.bytesPerSample = bytesPerSample; + frameData.numOfChannels = channels; + frameData.samplesPerSec = sampleRate; + + // 推送数据，处理缓冲区满的情况 + int ret; + final int WAIT_MS = 30; + + do { + // 检查运行标记和 stream ID 是否仍有效 + if (!mPushRunning || mExternalStreamId == null) { + break; + } + ret = engine.pushAudioExternalStreamData(mExternalStreamId, frameData); + if (ret == 110) { // AoqErrorCodeAudioExternalBufferFull + try { + Thread.sleep(WAIT_MS); + } catch (InterruptedException e) { + break; + } + } else { + break; + } + } while (true); +} +``` + +**注意事项：** + +- 需要在连接成功且添加外部音频流之后再开始推送数据。 + +- 需要按照数据的实际长度设置 `AoqAudioFrameData` 的 `numOfSamples`。 + +- 调用 `pushAudioExternalStreamData` 时，可能出现内部缓冲区满（错误码 110）而导致失败，需要等待重试。 + +- 实时采集建议 10ms 一帧数据 push，有数据就调用 push，注意处理内部缓冲区满（错误码 110）。 + +- 从文件解析建议 40ms 一帧数据，间隔 30ms 调用 push，注意处理内部缓冲区满（错误码 110）。 + +- 引擎退出（`destroy`）或 stream ID 被移除前，必须先设置 `mPushRunning = false` 停止推送循环，避免在已释放的资源上操作。 + + +### **5\. 移除外部音频流** + +当不再需要发布自定义采集的音频时，先停止推送循环，再调用 `removeAudioExternalStream` 接口移除外部音频流。 + +``` +// 先停止推送 +stopPushAudio(); +// 再移除外部音频流 +engine.removeAudioExternalStream(mExternalStreamId); +mExternalStreamId = null; +``` diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-playback.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-playback.md new file mode 100644 index 00000000..1d94d8e7 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-playback.md @@ -0,0 +1,235 @@ +# 自定义音频播放 + +AOQ Client SDK 支持自定义音频播放功能，通过音频帧回调机制将解码后的 PCM 数据回调给应用层，由开发者自行实现音频渲染播放。 + +## **功能介绍** + +AOQ Client SDK 内部音频模块默认会将接收到的远端音频数据通过系统扬声器/听筒播放，但在特定场景中，SDK 内部的音频播放模块可能无法满足开发需求，需要实现自定义音频播放功能，例如： + +- 需要将接收到的音频数据输出到自定义的播放设备或音频处理管线。 + +- 需要对接收到的音频数据进行二次处理（如 AI 语音识别、音效处理等）。 + +- 解决音频播放设备被占用的问题。 + + +AOQ Client SDK 支持灵活的自定义播放功能，通过音频帧回调机制，将解码后的 PCM 数据回调给应用层，由开发者自行实现音频渲染播放。 + +## **示例代码** + +暂无 + +## **前提条件** + +- 已创建引擎实例（调用 `createEngine`）。 + +- 已成功连接服务器（`onConnectionStatusChange` 回调状态为 `AoqConnectionStatusConnected`）。 + + +## **功能实现** + +### **1\. 开启音频播放（外部模式）** + +调用 `startAudioPlayer` 时设置 `isExternal=true`，关闭 SDK 内部的音频渲染设备，由应用层自行处理音频播放。 + +``` +AoqClientEngine.AoqAudioPlaybackConfig config = new AoqClientEngine.AoqAudioPlaybackConfig(); +config.isExternal = true; // 关闭 SDK 内部播放，由应用层自行渲染 +config.channel = 1; // 声道数 +engine.startAudioPlayer(config); +``` + +**参数说明：** + +**参数** + +**类型** + +**默认值** + +**说明** + +isVoipMode + +boolean + +false + +是否启用 VoIP 模式（硬件AEC），移动端有效 + +isDefaultSpeaker + +boolean + +true + +是否默认扬声器，移动端有效 + +isExternal + +boolean + +false + +是否外部播放模式，true 时 SDK 不打开播放设备 + +channel + +int + +1 + +声道数 + +### **2\. 设置音频帧回调监听** + +调用 `setAudioFrameObserver` 设置音频帧数据回调监听器，实现 `onPlaybackAudioFrame` 回调方法接收播放 PCM 数据。 + +``` +engine.setAudioFrameObserver(new AoqClientListener.AoqAudioFrameListener() { + @Override + public void onPlaybackAudioFrame(@NonNull AoqClientEngine.AoqAudioFrameData frame) { + // 在此处理接收到的播放音频数据 + // frame.dataPtr: PCM 数据 + // frame.numOfSamples: 采样点数 + // frame.numOfChannels: 声道数 + // frame.samplesPerSec: 采样率 + // frame.bytesPerSample: 每采样点字节数 + playPcmData(frame); + } +}); +``` + +### **3\. 开启播放数据回调** + +调用 `enableAudioFrameObserver` 开启播放位置的音频帧回调，指定数据源为 `AoqAudioSourcePlayback`。 + +``` +AoqClientEngine.AoqAudioObserverConfig observerConfig = new AoqClientEngine.AoqAudioObserverConfig(); +observerConfig.sampleRate = 48000; // 回调音频采样率 +observerConfig.channels = 1; // 回调音频声道数 +observerConfig.mode = AoqClientEngine.AoqAudioObserverMode.AoqAudioObserverModeReadOnly; // 只读模式 + +engine.enableAudioFrameObserver( + true, // 开启回调 + AoqClientEngine.AoqAudioSource.AoqAudioSourcePlayback, // 播放数据源 + observerConfig +); +``` + +**参数说明：** + +**参数** + +**类型** + +**默认值** + +**说明** + +sampleRate + +int + +48000 + +回调音频采样率（Hz） + +channels + +int + +1 + +回调音频声道数 + +mode + +AoqAudioObserverMode + +AoqAudioObserverModeReadOnly + +读写模式 + +### **4\. 实现自定义音频渲染** + +在 `onPlaybackAudioFrame` 回调中接收到 PCM 数据后，由应用层自行实现音频渲染播放。常见的实现方式： + +- **Android AudioTrack**：通过 AudioTrack 将 PCM 数据写入系统音频设备播放。 + +- **AI 语音识别**：将 PCM 数据传入 ASR 引擎进行语音识别。 + +- **音效处理**：对 PCM 数据进行音效处理后再播放。 + +- **文件存储**：将接收到的音频数据保存到本地文件。 + + +``` +// 示例：使用 Android AudioTrack 播放 +private AudioTrack mAudioTrack; +private volatile boolean mPlayRunning = false; + +private void initAudioTrack(int sampleRate, int channels) { + int channelConfig = (channels == 2) + ? AudioFormat.CHANNEL_OUT_STEREO + : AudioFormat.CHANNEL_OUT_MONO; + int bufferSize = AudioTrack.getMinBufferSize( + sampleRate, channelConfig, AudioFormat.ENCODING_PCM_16BIT); + + mAudioTrack = new AudioTrack( + AudioManager.STREAM_VOICE_CALL, + sampleRate, + channelConfig, + AudioFormat.ENCODING_PCM_16BIT, + bufferSize, + AudioTrack.MODE_STREAM); + mAudioTrack.play(); + mPlayRunning = true; +} + +private void playPcmData(AoqClientEngine.AoqAudioFrameData frame) { + if (!mPlayRunning || mAudioTrack == null) { + return; + } + if (frame.dataPtr != null && frame.dataSize > 0) { + mAudioTrack.write(frame.dataPtr, 0, frame.dataSize); + } +} +``` + +**注意事项：** + +- `onPlaybackAudioFrame` 回调在 SDK 内部线程触发，回调中的 `frame.dataPtr` 仅在回调期间有效，异步使用需自行拷贝。 + +- AudioTrack.write 是阻塞操作，在回调中直接写入即可，SDK 内部会按节奏回调。 + +- 需要维护 `mPlayRunning` 标记，当引擎退出或停止播放时退出处理逻辑。 + + +### **5\. 停止自定义播放** + +当不再需要自定义播放时，先关闭音频帧回调，再停止播放设备，释放 AudioTrack 资源。 + +``` +// 1. 关闭播放位置的音频帧回调 +AoqClientEngine.AoqAudioObserverConfig observerConfig = new AoqClientEngine.AoqAudioObserverConfig(); +engine.enableAudioFrameObserver( + false, // 关闭回调 + AoqClientEngine.AoqAudioSource.AoqAudioSourcePlayback, + observerConfig +); + +// 2. 移除音频帧回调监听 +engine.setAudioFrameObserver(null); + +// 3. 停止 SDK 音频播放 +engine.stopAudioPlayer(); + +// 4. 释放 AudioTrack 资源 +mPlayRunning = false; +if (mAudioTrack != null) { + mAudioTrack.stop(); + mAudioTrack.release(); + mAudioTrack = null; +} +``` diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-video-input.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-video-input.md new file mode 100644 index 00000000..358f530f --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-video-input.md @@ -0,0 +1,647 @@ +# 自定义视频输入 + +介绍 AOQ Client SDK 自定义视频输入的两种模式：原始帧模式和编码帧模式，以及各模式的配置方法和示例代码。 + +## **功能介绍** + +AOQ Client SDK 内部视频模块可满足应用中对基本视频功能的需求，但在特定场景中，SDK 内部的视频采集模块可能无法满足开发需求，需要实现自定义视频采集功能，例如： + +- 解决摄像头设备被占用或不兼容问题。 + +- 需要从定制的采集系统、视频文件中获取视频数据后交给 SDK 传输。 + +- 需要将 AI 生成的画面、屏幕录制、虚拟摄像头等内容通过 SDK 推流传输。 + + +AOQ Client SDK 支持两种自定义视频采集模式： + +- **原始帧模式**：自行采集原始视频帧（BGRA、I420、NV12、NV21 等格式），通过 `pushExternalVideoCapturedFrame` 推送给 SDK 进行编码和传输。SDK 内部完成编码、传输等完整流程。 + +- **编码帧模式**：自行完成视频编码（目前支持 JPEG），通过 `pushExternalVideoEncodedFrame` 直推已编码数据给 SDK，跳过 SDK 内部编码器，直接打包发送。 + + +## **示例代码** + +暂无 + +## **前提条件** + +- 已创建引擎实例（调用 `createEngine`）。 + +- 已成功连接服务器（`onConnectionStatusChange` 回调状态为 `AoqConnectionStatusConnected`）。 + + +## **功能实现** + +根据业务场景选择以下两种模式之一。两种模式不可混用：同一时间只能使用其中一种推送接口。 + +## **模式一：原始帧模式** + +自行采集原始视频帧（BGRA、I420、NV12、NV21 等格式），推送给 SDK 进行编码和传输。SDK 内部完成编码、传输等完整流程。 + +### **1\. 配置视频编码参数** + +SDK 内部编码器会对推送的原始帧进行编码，可根据业务需要调整编码参数。 + +``` +AoqClientEngine.AoqVideoCodecConfig config = new AoqClientEngine.AoqVideoCodecConfig(); +config.width = 1280; +config.height = 720; +config.fps = 2; +config.bitrate = 500000; // 起始码率 500kbps +config.minBitrate = 128000; // 最小码率 128kbps +config.keyframeInterval = 2; +// isExternal 保持默认 false，SDK 内部编码 + +engine.setVideoEncoderConfig(config); +``` + +**参数说明：** + +**参数** + +**类型** + +**默认值** + +**说明** + +trackType + +AoqTrackType + +AoqTrackTypeVideo + +视频轨道类型 + +codecType + +AoqEncoderType + +AoqEncoderTypeVideoH264 + +编码器类型 + +width + +int + +720 + +编码宽度（像素） + +height + +int + +1280 + +编码高度（像素） + +fps + +int + +5 + +帧率 + +bitrate + +int + +500000 + +起始码率（bps） + +minBitrate + +int + +128000 + +最小码率（bps） + +keyframeInterval + +int + +2 + +关键帧间隔（秒） + +isExternal + +boolean + +false + +原始帧模式保持 false + +mirrorMode + +AoqMirrorMode + +AoqMirrorModeDisabled + +镜像模式 + +orientationMode + +AoqOrientationMode + +AoqOrientationModeAuto + +画面方向模式 + +### **2\. 以外部采集模式启动视频采集** + +调用 `startVideoCapture` 并设置 `isExternal=true`，告知 SDK 不打开摄像头，由外部源提供视频帧。这是原始帧模式的前置条件，未调用则 SDK 不会消费推送的帧数据。 + +``` +AoqClientEngine.AoqVideoCaptureConfig config = new AoqClientEngine.AoqVideoCaptureConfig(); +config.isExternal = true; // 不打开摄像头，由外部源推送视频帧 +// isExternal=true 时 width/height/fps 无效，实际分辨率和帧率由推送数据决定 +int ret = engine.startVideoCapture(config); +``` + +**参数说明：** + +**参数** + +**类型** + +**默认值** + +**说明** + +width + +int + +1280 + +采集宽度（`isExternal=true` 时无效） + +height + +int + +720 + +采集高度（`isExternal=true` 时无效） + +fps + +int + +15 + +采集帧率（`isExternal=true` 时无效） + +isExternal + +boolean + +false + +true：不打开摄像头，由外部源推送帧数据 + +cameraDirection + +AoqCameraDirection + +AoqCameraDirectionFront + +摄像头方向（`isExternal=true` 时无效） + +### **3\. 推送原始视频帧** + +调用 `pushExternalVideoCapturedFrame` 接口，将采集到的原始视频帧传入 SDK。SDK 内部完成编码和传输。 + +支持的视频帧格式：BGRA、I420、NV12、NV21、RGBA。Apple 平台额外支持 CVPixelBuffer 零拷贝格式。 + +#### **3.1 BGRA 格式** + +BGRA 为打包格式，每个像素 4 字节（Blue、Green、Red、Alpha），一帧数据量 = width x height x 4。 + +``` +// 构造 BGRA 视频帧 +AoqClientEngine.AoqVideoFrame frame = new AoqClientEngine.AoqVideoFrame(); +frame.format = AoqClientEngine.AoqVideoPixelFormat.AoqVideoPixelFormatBGRA; +frame.width = 1280; +frame.height = 720; + +frame.data = bgraBytes; // byte[]，长度 = width * height * 4 + +frame.timeStamp = System.currentTimeMillis(); + +int ret = engine.pushExternalVideoCapturedFrame( + AoqClientEngine.AoqTrackType.AoqTrackTypeVideo, frame); +``` + +#### **3.2 I420 格式** + +I420 为三平面格式（Y、U、V 分离），Y 平面大小 = width x height，U/V 平面各为 (width/2) x (height/2)。 + +``` +// 构造 I420 视频帧 +AoqClientEngine.AoqVideoFrame frame = new AoqClientEngine.AoqVideoFrame(); +frame.format = AoqClientEngine.AoqVideoPixelFormat.AoqVideoPixelFormatI420; +frame.width = 1280; +frame.height = 720; + +frame.dataY = yPlane; // byte[]，长度 = width * height + +frame.dataU = uPlane; // byte[]，长度 = (width/2) * (height/2) + +frame.dataV = vPlane; // byte[]，长度 = (width/2) * (height/2) + +frame.strideY = 1280; // Y 平面行字节数 +frame.strideU = 640; // U 平面行字节数 +frame.strideV = 640; // V 平面行字节数 +frame.timeStamp = System.currentTimeMillis(); + +int ret = engine.pushExternalVideoCapturedFrame( + AoqClientEngine.AoqTrackType.AoqTrackTypeVideo, frame); +``` + +#### **3.3 NV12 / NV21 格式** + +NV12 和 NV21 为半平面格式，Y 平面 + UV 交错平面。NV12 为 UV 交替排列，NV21 为 VU 交替排列。数据量 = width x height x 3 / 2，打包在 `data` 字段中。 + +``` +// 构造 NV12 视频帧（NV21 同理，修改 format 即可） +AoqClientEngine.AoqVideoFrame frame = new AoqClientEngine.AoqVideoFrame(); +frame.format = AoqClientEngine.AoqVideoPixelFormat.AoqVideoPixelFormatNV12; +frame.width = 1280; +frame.height = 720; + +frame.data = nv12Bytes; // byte[]，长度 = width * height * 3 / 2 + +frame.timeStamp = System.currentTimeMillis(); + +int ret = engine.pushExternalVideoCapturedFrame( + AoqClientEngine.AoqTrackType.AoqTrackTypeVideo, frame); +``` + +#### **3.4 CVPixelBuffer 格式（Apple 平台）** + +iOS / macOS 平台支持直接传递 `CVPixelBufferRef`，实现零拷贝传输，避免内存拷贝带来的性能开销。 + +``` +// iOS / macOS 平台 +let frame = AoqVideoFrame() +frame.format = .cvPixelBuffer +frame.width = 1280 +frame.height = 720 +frame.pixelBuffer = pixelBuffer // CVPixelBufferRef +frame.timeStamp = Int64(Date().timeIntervalSince1970 * 1000) + +// SDK 内部异步持有 pixelBuffer，需要额外 +1 引用计数 +// SDK 消费完毕后会自行释放 +let _ = Unmanaged.passRetained(pixelBuffer) + +engine.pushExternalVideoCapturedFrame(.video, frame: frame) +``` + +### **4\. 停止原始帧采集** + +当不再需要推送视频帧时，先停止推帧定时器，再调用 `stopVideoCapture` 关闭视频采集。 + +``` +// 1. 停止推帧定时器 +stopExternalFramePush(); +// 2. 停止视频采集 +engine.stopVideoCapture(); +``` + +## **模式二：编码帧模式** + +自行完成视频编码（目前支持 JPEG），直推已编码数据给 SDK，跳过 SDK 内部编码器，直接打包发送。此模式**不需要调用** `startVideoCapture` 等采集相关接口。 + +### **1\. 配置视频编码参数并启用外部编码** + +调用 `setVideoEncoderConfig` 并设置 `isExternal=true`，告知 SDK 跳过内部编码器，由外部提供已编码数据。 + +``` +AoqClientEngine.AoqVideoCodecConfig config = new AoqClientEngine.AoqVideoCodecConfig(); +config.width = 1280; +config.height = 720; +config.fps = 2; +config.isExternal = true; // 跳过内部编码，由外部推送已编码帧 + +engine.setVideoEncoderConfig(config); +``` + +设置完成后即可直接推送编码帧，无需调用 `startVideoCapture`。 + +### **2\. 推送编码视频帧** + +调用 `pushExternalVideoEncodedFrame` 接口，将已编码的视频数据直传给 SDK。目前仅支持 JPEG 编码格式。 + +``` +// 从 Bitmap 生成 JPEG 数据 +android.graphics.Bitmap bmp = android.graphics.Bitmap.createBitmap( + width, height, android.graphics.Bitmap.Config.ARGB_8888); +// ... 填充 Bitmap 内容 ... + +java.io.ByteArrayOutputStream baos = new java.io.ByteArrayOutputStream(); +bmp.compress(android.graphics.Bitmap.CompressFormat.JPEG, 85, baos); +bmp.recycle(); + +// 构造编码帧并推送 +AoqClientEngine.AoqVideoEncodedFrame frame = new AoqClientEngine.AoqVideoEncodedFrame(); +frame.codec = AoqClientEngine.AoqVideoCodecType.AoqVideoCodecTypeJPEG; +frame.data = baos.toByteArray(); +frame.width = width; +frame.height = height; +frame.timeStamp = System.currentTimeMillis(); + +int ret = engine.pushExternalVideoEncodedFrame( + AoqClientEngine.AoqTrackType.AoqTrackTypeVideo, frame); +``` + +**AoqVideoEncodedFrame 参数说明：** + +**参数** + +**类型** + +**默认值** + +**说明** + +codec + +AoqVideoCodecType + +AoqVideoCodecTypeJPEG + +编码格式，目前仅支持 JPEG + +data + +byte\[\] + +null + +编码后的数据 + +width + +int + +0 + +画面宽度（像素） + +height + +int + +0 + +画面高度（像素） + +timeStamp + +long + +0 + +时间戳（毫秒），为 0 时 SDK 使用本地时钟补充 + +### **3\. 停止编码帧推送** + +编码帧模式无需管理采集设备，停止推帧定时器即可。 + +``` +stopExternalFramePush(); +``` + +## **视频帧格式参考** + +### **AoqVideoFrame（原始帧模式使用）** + +**字段** + +**类型** + +**说明** + +format + +AoqVideoPixelFormat + +像素格式 + +width + +int + +画面宽度（像素） + +height + +int + +画面高度（像素） + +data + +byte\[\] + +打包格式数据（NV12/NV21/BGRA/RGBA） + +dataY + +byte\[\] + +I420 Y 平面数据 + +dataU + +byte\[\] + +I420 U 平面数据 + +dataV + +byte\[\] + +I420 V 平面数据 + +strideY + +int + +I420 Y 平面行字节数 + +strideU + +int + +I420 U 平面行字节数 + +strideV + +int + +I420 V 平面行字节数 + +textureId + +int + +Android 纹理 ID（TextureOES/Texture2D） + +transformMatrix + +float\[\] + +纹理变换矩阵（4x4 行优先） + +eglContext + +EGLContext + +Android 共享 EGL context（纹理模式使用） + +pixelBuffer + +CVPixelBufferRef + +Apple 零拷贝 CVPixelBuffer（仅 iOS/macOS） + +timeStamp + +long + +时间戳（毫秒），为 0 时 SDK 用本地时钟补充 + +### **AoqVideoPixelFormat 枚举值** + +**枚举值** + +**数值** + +**说明** + +AoqVideoPixelFormatUnknown + +0 + +未知格式 + +AoqVideoPixelFormatI420 + +1 + +I420 三平面格式 + +AoqVideoPixelFormatNV12 + +2 + +NV12 半平面格式（UV 交替） + +AoqVideoPixelFormatNV21 + +3 + +NV21 半平面格式（VU 交替） + +AoqVideoPixelFormatBGRA + +4 + +BGRA 打包格式 + +AoqVideoPixelFormatRGBA + +5 + +RGBA 打包格式 + +AoqVideoPixelFormatCVPixelBuffer + +6 + +Apple CVPixelBuffer（仅 iOS/macOS） + +AoqVideoPixelFormatTextureOES + +7 + +Android OES 外部纹理 + +AoqVideoPixelFormatTexture2D + +8 + +Android 2D 纹理 + +### **AoqVideoEncodedFrame（编码帧模式使用）** + +**字段** + +**类型** + +**说明** + +codec + +AoqVideoCodecType + +编码格式 + +data + +byte\[\] + +编码后的数据 + +width + +int + +画面宽度（像素） + +height + +int + +画面高度（像素） + +timeStamp + +long + +时间戳（毫秒），为 0 时 SDK 用本地时钟补充 + +### **AoqVideoCodecType 枚举值** + +**枚举值** + +**数值** + +**说明** + +AoqVideoCodecTypeJPEG + +0 + +JPEG 编码格式 + +## **注意事项** + +- 原始帧模式：必须先调用 `startVideoCapture(isExternal=true)` 再推送帧，否则 SDK 返回参数错误。 + +- 编码帧模式：只需调用 `setVideoEncoderConfig(isExternal=true)` 即可推送，**不需要**调用 `startVideoCapture`。 + +- 原始帧模式与编码帧模式不可混用：同一时间只能使用其中一种推送接口。 + +- 编码帧模式目前仅支持 JPEG 格式。 + +- 视频帧数据在推送后由 SDK 内部管理生命周期，调用方无需在推送后继续持有数据引用。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-media-stream-control.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-media-stream-control.md new file mode 100644 index 00000000..69b625b7 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-media-stream-control.md @@ -0,0 +1,115 @@ +# 媒体流发送管理 + +`enableSendMediaStream` 用于控制客户端是否向 AI 服务发送音频/视频媒体流，在 AOQ 协议接入场景中精确控制发送时机。 + +## **概述** + +`enableSendMediaStream` 用于控制客户端是否向 AI 服务发送音频/视频媒体流。在 AOQ 协议接入场景中，部分模型要求在收到 `session.updated` 确认后才能接收媒体数据，因此需要通过此接口精确控制发送时机。 + +## **API 定义** + +**iOS / Mac** + +``` +// iOS / Mac +func enableSendMediaStream(_ trackType: AoqTrackType, enable: Bool) +``` + +**Android** + +``` +// Android +void enableSendMediaStream(AoqTrackType trackType, boolean enable) +``` + +**OHOS (ArkTS)** + +``` +// OHOS (ArkTS) +enableSendMediaStream(trackType: AoqTrackType, enable: boolean): void +``` + +**参数说明：** + +**参数** + +**类型** + +**说明** + +trackType + +AoqTrackType + +媒体轨道类型：`.audio` 或 `.video` + +enable + +Bool + +`true` = 开启发送，`false` = 暂停发送 + +## **控制流程** + +典型的媒体流控制流程如下： + +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0575914871/p1088082.svg) + +## **使用示例（iOS Swift）** + +### **connect 前禁用发送** + +``` +// 连接前关闭音视频发送，避免模型未就绪时收到数据 +engine.enableSendMediaStream(.audio, enable: false) +engine.enableSendMediaStream(.video, enable: false) + +// 发起连接 +engine.connect(config) +``` + +### **收到 session.updated 后开启** + +``` +func onDataMsg(_ msg: AoqDataMsg) { + guard let obj = try? JSONSerialization.jsonObject(with: msg.data) as? [String: Any], + let type = obj["type"] as? String else { return } + + if type == "session.updated" { + // AI 侧已确认会话配置，开启媒体发送 + engine.enableSendMediaStream(.audio, enable: true) + engine.enableSendMediaStream(.video, enable: true) + } +} +``` + +## **注意事项** + +- **调用时机**：必须在 `createEngine` 之后调用，引擎未创建时调用无效。 + +- **默认行为**：如果不调用此接口，connect 成功后 SDK 会立即开始发送媒体流。 + +- **模型兼容性**：部分模型要求先收到 `session.updated` 再接收媒体数据，建议统一采用"先禁用、后开启"模式。 + +- **独立控制**：音频和视频可独立控制，例如仅发送音频不发送视频。 + + +## **常见场景** + +**场景** + +**操作** + +**说明** + +连接模型前 + +`enable(.audio, false)` + +等待 session.updated 再发送 + +收到 session.updated + +`enable(.audio, true)` + +AI 已就绪，开始发送 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-video-features.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-video-features.md new file mode 100644 index 00000000..f71bf047 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-video-features.md @@ -0,0 +1,1186 @@ +# 视频常用功能介绍 + +AOQ Client SDK 提供了完整的视频能力，覆盖视频采集、渲染显示、编码配置、帧数据回调、外部视频输入等核心场景。本文档基于 Android（Java）、iOS（Objective-C）、Ohos（ArkTS）三个平台的公开 API，对视频常用功能进行统一介绍。 + +## **1\. 视频采集** + +### 1.1 功能说明 + +视频采集用于打开设备摄像头，将实时视频帧数据送入 SDK 编码推流管线。SDK 支持两种采集模式： + +- **内部采集（默认）**：SDK 自动管理摄像头设备的打开、帧采集和关闭，支持前后置摄像头切换。 + +- **外部采集**：由应用自行管理摄像头或其他视频源，采集到的帧数据通过 `pushExternalVideoCapturedFrame` 接口输入 SDK。 + + +### 1.2 采集配置参数 + +**参数** + +**类型** + +**默认值** + +**说明** + +width + +int + +1280 + +采集宽度（像素），外部采集时无效 + +height + +int + +720 + +采集高度（像素），外部采集时无效 + +fps + +int + +15 + +采集帧率，外部采集时由送帧节奏决定 + +isExternal + +bool + +false + +是否使用外部采集模式 + +cameraDirection + +AoqCameraDirection + +Front(0) + +摄像头方向，外部采集时无效 + +### 1.3 摄像头方向枚举 + +**枚举值** + +**数值** + +**说明** + +AoqCameraDirectionFront + +0 + +前置摄像头 + +AoqCameraDirectionBack + +1 + +后置摄像头 + +### 1.4 API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +开启采集 + +startVideoCapture(config) + +startVideoCapture:config: + +startVideoCapture(config) + +关闭采集 + +stopVideoCapture() + +stopVideoCapture + +stopVideoCapture() + +切换摄像头 + +switchCamera(direction) + +switchCamera: + +switchCamera(direction) + +### 1.5 使用示例 + +**Android** + +``` +AoqVideoCaptureConfig config = new AoqVideoCaptureConfig(); +config.width = 1280; +config.height = 720; +config.fps = 15; +config.cameraDirection = AoqCameraDirection.AoqCameraDirectionFront; +engine.startVideoCapture(config); +``` + +**iOS** + +``` +AoqVideoCaptureConfig *config = [[AoqVideoCaptureConfig alloc] init]; +config.width = 1280; +config.height = 720; +config.fps = 15; +config.cameraDirection = AoqCameraDirectionFront; +[engine startVideoCapture:config]; +``` + +**Ohos** + +``` +let config: AoqVideoCaptureConfig = { + width: 1280, + height: 720, + fps: 15, + cameraDirection: AoqCameraDirection.AoqCameraDirectionFront +}; +engine.startVideoCapture(config); +``` + +## **2\. 视频渲染** + +### 2.1 功能说明 + +视频渲染用于将本地采集或远端接收的视频帧数据显示到屏幕上。SDK 支持设置本地预览窗口和远端渲染窗口，通过 `trackType` 区分视频流（Video）和屏幕共享流（Screen）。 + +### 2.2 渲染模式 + +**枚举值** + +**数值** + +**说明** + +AoqRenderModeAuto + +0 + +自动模式 + +AoqRenderModeStretch + +1 + +拉伸平铺，画面可能变形 + +AoqRenderModeFill + +2 + +填充黑边，画面完整显示 + +AoqRenderModeCrop + +3 + +裁剪模式，画面内容可能丢失 + +### 2.3 画布配置 + +**参数** + +**类型** + +**默认值** + +**说明** + +view + +平台视图 + +null + +渲染视图（Android: SurfaceView/TextureView, iOS: UIView, Ohos: XComponent） + +renderMode + +AoqRenderMode + +Auto(0) + +渲染显示模式 + +### 2.4 API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +设置本地预览 + +setLocalView(trackType, canvas) + +setLocalView:trackType:canvas: + +setLocalView(trackType, canvas) + +设置远端渲染 + +setRemoteView(trackType, canvas) + +setRemoteView:trackType:canvas: + +setRemoteView(trackType, canvas) + +**说明** + +**平台差异**：Android 使用 SurfaceView 或 TextureView 作为渲染容器；iOS 使用 UIView（内部通过 AoqRenderView 封装，支持 Metal 加速）；Ohos 使用 XComponent（通过 AoqXComponentController 管理 native 渲染视图）。 + +## **3\. 视频编码配置** + +### 3.1 功能说明 + +设置视频编码参数，包括编码格式、分辨率、帧率、码率、关键帧间隔、镜像和方向等。通过 `trackType` 区分视频轨道和屏幕共享轨道的编码配置。 + +### 3.2 编码配置参数 + +**参数** + +**类型** + +**默认值** + +**说明** + +trackType + +AoqTrackType + +Video(1) + +轨道类型：Video + +codecType + +AoqEncoderType + +VideoH264(3) + +编码格式 + +width + +int + +720 + +编码宽度 + +height + +int + +1280 + +编码高度 + +fps + +int + +5 + +编码帧率 + +bitrate + +int + +500000 + +目标码率（bps） + +minBitrate + +int + +128000 + +最小码率（bps） + +keyframeInterval + +int + +2 + +关键帧间隔（秒） + +mirrorMode + +AoqMirrorMode + +Disabled(0) + +镜像模式 + +orientationMode + +AoqOrientationMode + +Auto(0) + +方向模式 + +isExternal + +bool + +false + +外部编码模式（true 时由应用推送已编码帧） + +### 3.3 编码格式枚举 + +**枚举值** + +**数值** + +**说明** + +AoqEncoderTypeVideoH264 + +3 + +H.264 编码 + +AoqEncoderTypeVideoJpeg + +4 + +JPEG 编码（用于外部编码帧） + +### 3.4 镜像模式 + +**枚举值** + +**数值** + +**说明** + +AoqMirrorModeDisabled + +0 + +禁用镜像 + +AoqMirrorModeEnabled + +1 + +启用镜像 + +### 3.5 方向模式 + +**枚举值** + +**数值** + +**说明** + +AoqOrientationModeAuto + +0 + +自动方向 + +AoqOrientationModePortrait + +1 + +竖屏方向 + +AoqOrientationModeLandscape + +2 + +横屏方向 + +### 3.6 API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +设置编码参数 + +setVideoEncoderConfig(config) + +setVideoEncoderConfig: + +setVideoEncoderConfig(config) + +## **4\. 外部视频帧输入** + +### 4.1 功能说明 + +外部视频帧输入允许应用将自定义的视频帧数据推送到 SDK，用于外部采集或外部编码场景。支持两种推送方式： + +- **推送原始帧**：将未编码的像素数据（I420/NV12/NV21/BGRA/RGBA 等格式）推送给 SDK，由 SDK 进行编码。 + +- **推送已编码帧**：将已编码的数据（如 JPEG）直推给 SDK，SDK 不做二次编码，直接打包发送。 + + +通过 `trackType` 路由，`AoqTrackTypeVideo` 对应视频采集的外部帧，`AoqTrackTypeScreen` 对应屏幕共享的外部帧。 + +### 4.2 像素格式枚举 + +**枚举值** + +**数值** + +**说明** + +**平台支持** + +AoqVideoPixelFormatI420 + +1 + +I420 三平面 + +全平台 + +AoqVideoPixelFormatNV12 + +2 + +NV12 双平面 + +全平台 + +AoqVideoPixelFormatNV21 + +3 + +NV21 双平面 + +全平台 + +AoqVideoPixelFormatBGRA + +4 + +BGRA 打包 + +全平台 + +AoqVideoPixelFormatRGBA + +5 + +RGBA 打包 + +全平台 + +AoqVideoPixelFormatCVPixelBuffer + +6 + +Apple 零拷贝 + +仅 iOS + +AoqVideoPixelFormatTextureOES + +7 + +OES 纹理 + +仅 Android + +AoqVideoPixelFormatTexture2D + +8 + +2D 纹理 + +仅 Android + +### 4.3 原始视频帧数据结构 (AoqVideoFrame) + +**字段** + +**类型** + +**说明** + +format + +AoqVideoPixelFormat + +像素格式 + +width + +int + +宽度（像素） + +height + +int + +高度（像素） + +data + +byte\[\] / ArrayBuffer + +打包格式数据（NV12/NV21/BGRA/RGBA） + +dataY / dataU / dataV + +byte\[\] / ArrayBuffer + +I420 三平面数据 + +strideY / strideU / strideV + +int + +I420 三平面步长 + +textureId + +int + +纹理 ID（Android TextureOES/Texture2D 时有效） + +transformMatrix + +float\[16\] + +4x4 纹理变换矩阵（Android） + +eglContext + +EGLContext + +共享 EGL 上下文（Android） + +pixelBuffer + +CVPixelBufferRef + +Apple 零拷贝（iOS） + +timeStamp + +long + +时间戳（ms），0 时 SDK 用本地时钟补 + +### 4.4 已编码视频帧数据结构 (AoqVideoEncodedFrame) + +**字段** + +**类型** + +**默认值** + +**说明** + +codec + +AoqVideoCodecType + +JPEG(0) + +编码格式 + +data + +byte\[\] / ArrayBuffer + +\- + +编码后数据 + +width + +int + +\- + +宽度（像素） + +height + +int + +\- + +高度（像素） + +timeStamp + +long + +0 + +时间戳（ms） + +### 4.5 API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +推送原始帧 + +pushExternalVideoCapturedFrame(trackType, frame) + +pushExternalVideoCapturedFrame:frame: + +pushExternalVideoCapturedFrame(trackType, frame) + +推送已编码帧 + +pushExternalVideoEncodedFrame(trackType, frame) + +pushExternalVideoEncodedFrame:frame: + +pushExternalVideoEncodedFrame(trackType, frame) + +## **5\. 视频帧数据回调** + +### 5.1 功能说明 + +视频帧回调允许开发者在视频管线的不同位置获取原始帧数据，用于视频分析、自定义处理、录制等场景。支持只读和读写两种模式，读写模式下可修改帧数据并写回 SDK。 + +### 5.2 支持的数据源位置 + +**数据源** + +**枚举值** + +**说明** + +Captured + +0 + +采集后的视频数据（前处理前） + +PreEncode + +1 + +编码前的视频数据（前处理后） + +Remote + +2 + +远端解码后、渲染前的视频数据 + +### 5.3 回调配置参数 + +**参数** + +**类型** + +**默认值** + +**说明** + +format + +AoqVideoPixelFormat + +I420(1) + +期望回调的像素格式 + +alignment + +AoqVideoObserverAlignment + +Default(0) + +宽度对齐策略 + +mode + +AoqVideoObserverMode + +ReadOnly(0) + +只读(0)/读写(1) 模式 + +mirrorApplied + +bool + +false + +是否对回调数据应用镜像 + +### 5.4 宽度对齐枚举 + +**枚举值** + +**数值** + +**说明** + +AoqVideoObserverAlignmentDefault + +0 + +默认对齐 + +AoqVideoObserverAlignmentEven + +1 + +2 字节对齐 + +AoqVideoObserverAlignment4 + +2 + +4 字节对齐 + +AoqVideoObserverAlignment8 + +3 + +8 字节对齐 + +AoqVideoObserverAlignment16 + +4 + +16 字节对齐 + +### 5.5 使用步骤 + +1. **注册观察者**：调用 `setVideoFrameObserver` 设置视频帧回调监听器 + +2. **启用数据源**：调用 `enableVideoFrameObserver` 选择需要监听的数据源位置，开启回调 + +3. **处理回调数据**：在回调函数中获取帧数据（仅回调期间有效，异步使用需自行拷贝） + + +### 5.6 API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +注册观察者 + +setVideoFrameObserver(listener) + +setVideoFrameObserver: + +setVideoFrameObserver(observer) + +启用回调 + +enableVideoFrameObserver(enabled, source, config) + +enableVideoFrameObserver:videoSource:config: + +enableVideoFrameObserver(enabled, source, config) + +### 5.7 回调方法 + +**回调** + +**Android** + +**iOS** + +**Ohos** + +采集后数据 + +onCapturedVideoFrame(frame) + +onCapturedVideoFrame: + +onCapturedVideoFrame(frame) + +编码前数据 + +onPreEncodeVideoFrame(trackType, frame) + +onPreEncodeVideoFrame:frame: + +onPreEncodeVideoFrame(trackType, frame) + +远端数据 + +onRemoteVideoFrame(trackType, frame) + +onRemoteVideoFrame:frame: + +onRemoteVideoFrame(trackType, frame) + +**说明** + +回调方法返回 `true`/`YES` 表示数据已修改、需写回 SDK（仅 ReadWrite 模式且 I420 格式时生效）。 + +## **7\. 媒体流发送控制** + +### 7.1 功能说明 + +控制本地媒体流的发送开关，通过 `trackType` 路由到不同轨道（Audio/Video/Screen）。停用发送后，采集和编码继续运行，但数据不会发送到远端。 + +### 7.2 API 对照 + +**功能** + +**Android** + +**iOS** + +**Ohos** + +控制流发送 + +enableSendMediaStream(trackType, enable) + +enableSendMediaStream:enable: + +enableSendMediaStream(trackType, enable) + +### 7.3 轨道类型枚举 + +**枚举值** + +**数值** + +**说明** + +AoqTrackTypeAudio + +0 + +音频轨道 + +AoqTrackTypeVideo + +1 + +视频轨道 + +AoqTrackTypeData + +2 + +数据轨道 + +## **8\. 视频设备状态监控** + +### 8.1 功能说明 + +SDK 自动监测视频采集设备（摄像头）的状态变化，并通过 `onVideoDeviceStateChanged` 回调通知应用层。 + +### 8.2 设备状态码 + +**状态码** + +**值** + +**说明** + +AoqVideoDeviceNone + +0 + +初始状态 + +AoqVideoDeviceCaptureStarting + +1 + +采集启动中 + +AoqVideoDeviceCaptureStarted + +2 + +采集已启动 + +AoqVideoDeviceCaptureStopping + +3 + +采集停止中 + +AoqVideoDeviceCaptureStopped + +4 + +采集已停止 + +AoqVideoDeviceCaptureFail + +5 + +采集失败 + +### 8.3 回调对照 + +**回调** + +**Android** + +**iOS** + +**Ohos** + +设备状态变化 + +onVideoDeviceStateChanged(state) + +onVideoDeviceStateChanged: + +onVideoDeviceStateChanged(state) + +## **9\. 视频错误码与警告码** + +### 9.1 视频错误码 + +**错误码** + +**值** + +**说明** + +AoqErrorCodeVideo + +200 + +通用视频错误 + +VideoExternalBufferFull + +210 + +视频外部缓冲区已满 + +VideoDevice + +220 + +视频设备通用错误 + +CameraOpenFail + +221 + +摄像头打开失败 + +CameraAuthFailed + +222 + +摄像头权限被拒绝 + +CameraOccupied + +223 + +摄像头被占用 + +CameraRunningError + +224 + +摄像头运行错误 + +VideoCodec + +230 + +视频编解码通用错误 + +EncoderInitFail + +231 + +编码器初始化失败 + +VideoRender + +240 + +视频渲染通用错误 + +RenderCreateFail + +241 + +渲染器创建失败 + +RenderDrawError + +242 + +渲染绘制错误 + +Screen + +300 + +屏幕共享通用错误 + +**说明** + +Android 额外错误码：ScreenPermissionDenied(310) 屏幕共享权限被拒绝、ScreenForegroundServiceFailed(311) 前台服务启动失败。 + +### 9.2 视频警告码 + +**警告码** + +**值** + +**说明** + +AoqWCVideo + +200 + +通用视频警告 + +CameraEnumerateError + +201 + +摄像头枚举错误 + +EncoderSwitched + +202 + +编码器切换警告 + +RenderDowngrade + +203 + +渲染降级警告 + +## **附录：完整视频 API 方法列表** + +**分类** + +**方法名** + +**说明** + +采集控制 + +startVideoCapture + +打开视频采集设备 + +采集控制 + +stopVideoCapture + +关闭视频采集设备 + +采集控制 + +switchCamera + +切换前后置摄像头 + +渲染控制 + +setLocalView + +设置本地预览窗口 + +渲染控制 + +setRemoteView + +设置远端渲染窗口 + +编解码 + +setVideoEncoderConfig + +设置视频编码参数 + +外部输入 + +pushExternalVideoCapturedFrame + +推送原始视频帧 + +外部输入 + +pushExternalVideoEncodedFrame + +推送已编码视频帧 + +屏幕共享 + +startScreenCapture + +启动屏幕采集 + +屏幕共享 + +stopScreenCapture + +停止屏幕采集 + +流控制 + +enableSendMediaStream + +控制媒体流发送 + +帧回调 + +setVideoFrameObserver + +注册视频帧观察者 + +帧回调 + +enableVideoFrameObserver + +启用/禁用视频帧回调 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-aoq-omni-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-aoq-omni-realtime.md new file mode 100644 index 00000000..a291cc23 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-aoq-omni-realtime.md @@ -0,0 +1,520 @@ +# 通过AOQ使用qwen3.5-omni-plus-realtime实现实时通话 + +本文档说明如何在 Android、iOS、HarmonyOS 平台接入 AOQ Client SDK，实现 AOQ+qwen3.5-omni-plus-realtime 音视频通话功能。 + +## **SDK 获取** + +AOQ Client SDK 及音频 Opus 插件请参见[SDK下载](https://help.aliyun.com/zh/model-studio/realtime-sdk-download)。Opus 编码以独立插件形式提供，请根据您的场景按需引入。 + +## **SDK 导入** + +请根据不同平台将核心 SDK 产物导入工程依赖目录，并在工程配置中声明相关权限。 + +### **Android** + +将 `AoqClientSdk-release.aar` 放入工程 `app/libs/` 目录，将 `libPluginOpus.so` 按 ABI 放入 `app/libs/armeabi-v7a/` 和 `app/libs/arm64-v8a/`，并在 `app/build.gradle` 中： + +``` +android { + defaultConfig { + minSdk 21 + ndk { abiFilters 'armeabi-v7a', 'arm64-v8a' } + } + sourceSets { main { jniLibs.srcDirs = ['libs'] } } + packagingOptions { + // 避免与宿主工程的同名 so 冲突 + pickFirsts += ['lib/*/*.so'] + } +} + +dependencies { + implementation fileTree(dir: 'libs', include: ['*.aar']) +} +``` + +在 `AndroidManifest.xml` 声明权限： + +``` + + + + + +``` + +其中 `RECORD_AUDIO` 和 `CAMERA` 为运行时权限，应用需在运行时调用 Android `ActivityCompat.requestPermissions()` 方法，主动向 Android 系统申请用户授权。 + +### **iOS（framework）** + +1. 将 `AoqClientSdk.framework` 与 `PluginOpus.framework` 拖入 Xcode 工程，在 Target > General > Frameworks, Libraries, and Embedded Content 中选择 **Embed & Sign**。 + +2. 权限声明：在 Xcode 中选中您的 Target > Info > Custom iOS Target Properties，添加以下两项权限用途描述： + + **Key** + + **Value** + + `NSMicrophoneUsageDescription` + + 用于实时语音通话 + + `NSCameraUsageDescription` + + 用于实时视频通话 + +3. Swift 工程：`import AoqClientSdk`；Objective-C 工程：`#import `。 + + +### **HarmonyOS（har）** + +1. 将 `aoq-client-sdk.har` 放入工程 `libs/` 目录，将 `libPluginOpus.so` 按 ABI 放入 `entry/libs/armeabi-v7a/` 和 `entry/libs/arm64-v8a/`；并在 `entry/oh-package.json5` 中声明。 + +2. 在 `entry/src/main/module.json5` 添加权限： + + ``` + "requestPermissions": [ + { "name": "ohos.permission.INTERNET" }, + { "name": "ohos.permission.MICROPHONE", + "reason": "$string:perm_mic_reason", + "usedScene": { "abilities": ["EntryAbility"], "when": "inuse" } }, + { "name": "ohos.permission.CAMERA", + "reason": "$string:perm_camera_reason", + "usedScene": { "abilities": ["EntryAbility"], "when": "inuse" } } + ] + ``` + +3. 在 `EntryAbility` 中通过 `abilityAccessCtrl.createAtManager().requestPermissionsFromUser` 触发运行时授权。 + + +## **AppServer获取Token** + +请按照[Token鉴权](https://help.aliyun.com/zh/model-studio/realtime-token-authentication)的 AOQ 章节搭建获取 Token 的 AppServer。每次通话前，客户端需要向业务侧 AppServer 请求一次 Token。 + +## **实现 AI 音视频通话** + +![111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0075914871/p1088080.svg) + +### **创建引擎并设置回调** + +调用 `createEngine` 接口创建 `AoqClientEngine` 实例。 + +**iOS：** + +``` +let config = AoqCreateConfig() +config.workDir = workDir +engine = AoqClientEngine.createEngine(config, delegate: self) +``` + +实现 `AoqEngineDelegate` 协议监听 `onConnectionStatusChange`、`onDataMsg`、`onError` 等回调。 + +**Android：** + +``` +AoqCreateConfig config = new AoqCreateConfig(); +config.workDir = appCtx.getFilesDir().getAbsolutePath(); +engine = AoqClientEngine.createEngine(appCtx, config, this); +``` + +**HarmonyOS：** + +``` +const config: AoqCreateConfig = { workDir: context.filesDir, extras: '' }; +engine = AoqClientEngine.createEngine(config, this, context); +``` + +### **启动音视频采集与播放** + +调用 `startAudioCapture` 与 `startAudioPlayer` 启动本地音频采集与播放；调用 `startVideoCapture` 启动摄像头，并通过 `setLocalView` 将 SDK 渲染目标绑定到业务侧的预览控件。 + +**iOS：** + +``` +// 音频采集 +let capCfg = AoqAudioCaptureConfig() +capCfg.channel = 1; capCfg.isExternal = false +engine.startAudioCapture(capCfg) + +// 音频播放 +let playCfg = AoqAudioPlaybackConfig() +playCfg.channel = 1; playCfg.isExternal = false +engine.startAudioPlayer(playCfg) + +// 视频采集 +let vidCfg = AoqVideoCaptureConfig() +vidCfg.width = 720; vidCfg.height = 1280; vidCfg.fps = 15 +engine.startVideoCapture(vidCfg) + +// 为本地预览画面设置渲染视图 +let canvas = AoqVideoCanvas() +canvas.view = localPreview +canvas.renderMode = .crop +engine.setLocalView(.video, canvas: canvas) +``` + +**Android：** + +``` +// 音频采集 +AoqAudioCaptureConfig capCfg = new AoqAudioCaptureConfig(); +capCfg.channel = 1; capCfg.isExternal = false; +engine.startAudioCapture(capCfg); + +// 音频播放 +AoqAudioPlaybackConfig playCfg = new AoqAudioPlaybackConfig(); +playCfg.channel = 1; playCfg.isExternal = false; +engine.startAudioPlayer(playCfg); + +// 视频采集 +AoqVideoCaptureConfig vidCfg = new AoqVideoCaptureConfig(); +vidCfg.width = 720; vidCfg.height = 1280; vidCfg.fps = 15; +engine.startVideoCapture(vidCfg); + +// 为本地预览画面设置渲染视图 +AoqVideoCanvas canvas = new AoqVideoCanvas(); +canvas.view = localPreview; +canvas.renderMode = AoqRenderMode.AoqRenderModeCrop; +engine.setLocalView(AoqTrackType.AoqTrackTypeVideo, canvas); +``` + +**HarmonyOS：** + +``` +// 音频采集 +const capCfg: AoqAudioCaptureConfig = { channel: 1, isExternal: false }; +engine.startAudioCapture(capCfg); + +// 音频播放 +const playCfg: AoqAudioPlaybackConfig = { channel: 1, isExternal: false }; +engine.startAudioPlayer(playCfg); + +// 视频采集 +const vidCfg: AoqVideoCaptureConfig = { width: 720, height: 1280, fps: 15, isExternal: false }; +engine.startVideoCapture(vidCfg); + +// 为本地预览画面设置渲染视图 +const canvas: AoqVideoCanvas = { view: localCtrl, renderMode: AoqRenderMode.AoqRenderModeCrop }; +engine.setLocalView(AoqTrackType.AoqTrackTypeVideo, canvas); +``` + +### **获取连接凭证** + +由业务 AppServer 代理百炼请求，参见[Token鉴权](https://help.aliyun.com/zh/model-studio/realtime-token-authentication)。 + +### **设置编解码及建立连接** + +设置编解码参数后调用 `connect`。 + +注意：qwen3.5-omni-plus-realtime 要求客户端在收到服务端的 `session.updated` 之后才能开始发送媒体数据。为避免 `connect` 建联成功到 `session.updated` 到达之间的空档期误推媒体，在 `connect` 之前对上行音频与视频轨道分别调用 `enableSendMediaStream(trackType, false)`，将上行推流暂时关闭。WebSocket事件说明详见[客户端事件](https://help.aliyun.com/zh/model-studio/client-events)。 + +**iOS：** + +``` +// 音频编解码配置 +let encCfg = AoqAudioCodecConfig() +encCfg.codecType = .audioPCM; encCfg.sampleRate = 16000; encCfg.channel = 1 +engine.setAudioEncoderConfig(encCfg) +engine.setAudioDecoderConfig(encCfg) + +// connect 前关闭媒体发送，待 session.updated 后再开启 +engine.enableSendMediaStream(.audio, enable: false) +engine.enableSendMediaStream(.video, enable: false) + +// 建立连接 +let conn = AoqConnectConfig() +conn.token = token; conn.sid = sid; conn.certFingerprint = cert +conn.relayEndpoints = endpoints; conn.workspaceIdHash = workspaceIdHash + +let aTrack = AoqTrackParam(); aTrack.trackType = .audio +let vTrack = AoqTrackParam(); vTrack.trackType = .video +let dTrack = AoqTrackParam(); dTrack.trackType = .data +conn.publishTracks = [aTrack, vTrack, dTrack] +conn.subscribeTracks = [aTrack, dTrack] +engine.connect(conn) +``` + +**Android：** + +``` +// 音频编解码配置 +AoqAudioCodecConfig encCfg = new AoqAudioCodecConfig(); +encCfg.codecType = AoqEncoderType.AoqEncoderTypeAudioPCM; +encCfg.sampleRate = 16000; encCfg.channel = 1; +engine.setAudioEncoderConfig(encCfg); +engine.setAudioDecoderConfig(encCfg); + +// connect 前关闭媒体发送，待 session.updated 后再开启 +engine.enableSendMediaStream(AoqTrackType.AoqTrackTypeAudio, false); +engine.enableSendMediaStream(AoqTrackType.AoqTrackTypeVideo, false); + +// 建立连接 +AoqConnectConfig conn = new AoqConnectConfig(); +conn.token = token; conn.sid = sid; conn.certFingerprint = cert; +conn.relayEndpoints.addAll(endpoints); conn.workspaceIdHash = workspaceIdHash; + +AoqTrackParam aTrack = new AoqTrackParam(); aTrack.trackType = AoqTrackType.AoqTrackTypeAudio; +AoqTrackParam vTrack = new AoqTrackParam(); vTrack.trackType = AoqTrackType.AoqTrackTypeVideo; +AoqTrackParam dTrack = new AoqTrackParam(); dTrack.trackType = AoqTrackType.AoqTrackTypeData; +conn.publishTracks.add(aTrack); +conn.publishTracks.add(vTrack); +conn.publishTracks.add(dTrack); +conn.subscribeTracks.add(aTrack); +conn.subscribeTracks.add(dTrack); +engine.connect(conn); +``` + +**HarmonyOS：** + +``` +// 音频编解码配置 +const encCfg: AoqAudioCodecConfig = { + codecType: AoqEncoderType.AoqEncoderTypeAudioPCM, + sampleRate: 16000, channel: 1 +}; +engine.setAudioEncoderConfig(encCfg); +engine.setAudioDecoderConfig(encCfg); + +// connect 前关闭媒体发送，待 session.updated 后再开启 +engine.enableSendMediaStream(AoqTrackType.AoqTrackTypeAudio, false); +engine.enableSendMediaStream(AoqTrackType.AoqTrackTypeVideo, false); + +// 建立连接 +const conn: AoqConnectConfig = { + token, sid, certFingerprint: cert, + relayEndpoints: endpoints, + workspaceIdHash, + publishTracks: [ + { trackType: AoqTrackType.AoqTrackTypeAudio }, + { trackType: AoqTrackType.AoqTrackTypeVideo }, + { trackType: AoqTrackType.AoqTrackTypeData } + ], + subscribeTracks: [ + { trackType: AoqTrackType.AoqTrackTypeAudio }, + { trackType: AoqTrackType.AoqTrackTypeData } + ] +}; +engine.connect(conn); +``` + +**重要**：AOQ SDK 在建联后会默认发送媒体数据，此示例演示了连接模型时关闭媒体发送的能力。 + +### **配置 AI 会话** + +在 `onConnectionStatusChange(Connected)` 回调中通过 `sendDataMsg` 发送 `session.update` 消息（业务自定义 JSON，包含 modalities、voice、instructions、turn\_detection 等会话参数），完成会话握手，WebSocket事件说明详见[客户端事件](https://help.aliyun.com/zh/model-studio/client-events)。 + +**iOS：** + +``` +func onConnectionStatusChange(_ status: AoqConnectionStatus) { + if status == .connected { sendSessionUpdate() } +} + +private func sendSessionUpdate() { + let json = """ + { + // 该事件的id，由客户端生成 + "event_id": "event_ToPZqeobitzUJnt3QqtWg", + // 事件类型，固定为session.update + "type": "session.update", + // 会话配置 + "session": { + // 输出模态，支持设置为["text"]（仅输出文本）或["text","audio"]（输出文本与音频）。 + "modalities": [ + "text", + "audio" + ], + // 输出音频的音色 + "voice": "Ethan", + // 输入音频格式，当前仅支持设置为pcm。输入音频为16 kHz采样率的PCM音频流。 + "input_audio_format": "pcm", + // 输出音频格式，当前仅支持设置为pcm。输出音频为24 kHz采样率的PCM音频流。 + "output_audio_format": "pcm", + // 系统消息，用于设定模型的目标或角色。 + "instructions": "你是某五星级酒店的AI客服专员，请准确且友好地解答客户关于房型、设施、价格、预订政策的咨询。请始终以专业和乐于助人的态度回应，杜绝提供未经证实或超出酒店服务范围的信息。", + // 是否开启语音活动检测。若需启用，需传入一个配置对象，服务端将据此自动检测语音起止。 + // 设置为null表示由客户端决定何时发起模型响应。 + "turn_detection": { + // VAD类型，取值为server_vad或semantic_vad。使用qwen3.5-omni-realtime模型时推荐设为semantic_vad。 + "type": "semantic_vad", + // VAD检测阈值。建议在嘈杂的环境中增加，在安静的环境中降低。 + "threshold": 0.5, + // 检测语音停止的静音持续时间，超过此值后会触发模型响应 + "silence_duration_ms": 800 + } + } + } + """ + let msg = AoqDataMsg() + msg.data = json.data(using: .utf8)! + engine.send(msg) +} +``` + +**Android：** + +``` +@Override +public void onConnectionStatusChange(AoqConnectionStatus status) { + if (status == AoqConnectionStatus.AoqConnectionStatusConnected) { + sendSessionUpdate(); + } +} + +private void sendSessionUpdate() { + String sessionUpdateJson = /* 与 Swift 示例中相同的 session.update JSON */; + AoqDataMsg msg = new AoqDataMsg(); + msg.data = sessionUpdateJson.getBytes(StandardCharsets.UTF_8); + engine.sendDataMsg(msg); +} +``` + +**HarmonyOS：** + +``` +onConnectionStatusChange(status: AoqConnectionStatus): void { + if (status === AoqConnectionStatus.AoqConnectionStatusConnected) { + this.sendSessionUpdate(); + } +} + +private sendSessionUpdate(): void { + const sessionUpdateJson = /* 与 Swift 示例中相同的 session.update JSON */; + const msg: AoqDataMsg = { data: new TextEncoder().encode(sessionUpdateJson).buffer }; + this.engine.sendDataMsg(msg); +} +``` + +### **收到 session.updated 后开启媒体发送** + +在 `onDataMsg` 回调中解析下行消息，收到模型回复 `session.updated` 的时候，对上一步禁推的每个轨道类型调用 `enableSendMediaStream(trackType, true)` 放开推流。下面为代码示例，WebSocket事件说明详见[服务端事件](https://help.aliyun.com/zh/model-studio/server-events)。 + +**iOS：** + +``` +func onDataMsg(_ msg: AoqDataMsg) { + guard let obj = try? JSONSerialization.jsonObject(with: msg.data) as? [String: Any], + let type = obj["type"] as? String else { return } + if type == "session.updated" { + engine.enableSendMediaStream(.audio, enable: true) + engine.enableSendMediaStream(.video, enable: true) + } +} +``` + +**Android：** + +``` +@Override +public void onDataMsg(AoqDataMsg msg) { + if (msg == null || msg.data == null) return; + try { + JSONObject obj = new JSONObject(new String(msg.data, StandardCharsets.UTF_8)); + if ("session.updated".equals(obj.optString("type"))) { + engine.enableSendMediaStream(AoqTrackType.AoqTrackTypeAudio, true); + engine.enableSendMediaStream(AoqTrackType.AoqTrackTypeVideo, true); + } + } catch (JSONException ignored) {} +} +``` + +**HarmonyOS：** + +``` +onDataMsg(msg: AoqDataMsg): void { + if (!msg?.data) return; + try { + const text = new TextDecoder('utf-8').decode(new Uint8Array(msg.data)); + const obj = JSON.parse(text) as { type?: string }; + if (obj.type === 'session.updated') { + this.engine.enableSendMediaStream(AoqTrackType.AoqTrackTypeAudio, true); + this.engine.enableSendMediaStream(AoqTrackType.AoqTrackTypeVideo, true); + } + } catch (_) { /* 非 JSON，忽略 */ } +} +``` + +**重要** + +1. 模型必须在收到 `session.updated` 后才开启媒体流发送，否则 AI 侧可能还未准备好接收数据。 + +2. 建连时添加的音频轨道和视频轨道（即 AOQ 媒体通道）会自动将数据传输到服务端。 + + 1. 音频：通过音频轨道直接传输，无需发送 `input_audio_buffer.append` 事件。 + + 2. 视频：通过视频轨道发送画面帧，无需发送 `input_image_buffer.append` 事件。 + + +### **断开连接与销毁引擎** + +``` +engine.disconnect() +AoqClientEngine.destroy() +``` + +## **典型场景** + +### **打断（Barge-in）** + +- SDK 与百炼深度融合，支持百炼模型的打断消息会在新一轮对话开始时打断上一轮次。 + +- SDK 提供**本地播放器打断**接口 `interruptAudioPlayer`，当用户主动需要停止时可以调用打断 API 实现此功能。 + + +``` +// iOS +engine.interruptAudioPlayer(.audio, fadeMs: 100) +``` + +### **静音 / 取消静音** + +静音后 SDK 仍在采集音频，但只推送静音帧，`session` 不会中断。 + +``` +engine.muteAudioCapture(true); // 静音麦克风（采集仍在跑，但只送静音帧） +engine.muteAudioCapture(false); // 恢复 +``` + +### **切换前后摄像头** + +``` +// 传入期望切换到的方向枚举即可 +engine.switchCamera(AoqCameraDirection.AoqCameraDirectionFront); +engine.switchCamera(AoqCameraDirection.AoqCameraDirectionBack); +``` + +### **通话字幕与ASR结果显示** + +服务端通过下行数据消息推送 ASR 结果与 AI 文本回复。业务侧在 `onDataMsg` 回调中根据 `type` 字段分流即可。WebSocket事件说明详见[服务端事件](https://help.aliyun.com/zh/model-studio/server-events)。 + +## **注意事项** + +1. **单例语义**：`createEngine` 是单例，重复调用返回同一实例；`destroy` 后才能重新创建。多页面共用建议在 Application/Ability 级管理引擎生命周期。 + +2. **本地预览 View 类型**： + + - Android：`SurfaceView` 或 `TextureView`；其它类型不支持。 + + - iOS：任意 `UIView` 子类。 + + - HarmonyOS：请参考 SDK 文档。 + +3. **音频路由变化**：耳机插拔、蓝牙连接等会触发 `onAudioDeviceRouteChanged`，业务侧通常无需处理；如果 UI 上显示"扬声器/听筒"开关，需要根据该回调同步状态。 + +4. **后台续传**：如需通话切到后台后继续传音频，`Info.plist` 必须开启 `UIBackgroundModes = audio`，并在前台时正确激活 `AVAudioSession`（SDK 会处理大部分情况，业务侧用 `setAudioSessionRestriction:` 可精细控制是否让 SDK 接管）。 + + +## **Demo 示例下载** + +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/5762814871/p1087917.png) + +示例源码下载： + +**iOS：**[aoqdemo.zip](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/grpnos/aoqdemo.zip) + +## **相关文档** + +- AOQ Client SDK 详细 API：[AOQ SDK简介](https://help.aliyun.com/zh/model-studio/realtime-api-aoq-sdk-desc/) + +- qwen3.5-omni-plus-realtime 模型客户端事件：[客户端事件](https://help.aliyun.com/zh/model-studio/client-events) + +- qwen3.5-omni-plus-realtime 模型服务端事件：[服务端事件](https://help.aliyun.com/zh/model-studio/server-events) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-multimodal-dialog.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-multimodal-dialog.md new file mode 100644 index 00000000..e743e6b6 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-multimodal-dialog.md @@ -0,0 +1,512 @@ +# 通过WebRTC使用多模态交互套件实现实时通话 + +本文档说明如何在浏览器端通过 WebRTC + JavaScript 接入通义多模态交互套件（multimodal-dialog），实现与多模态 AI 应用的实时音视频交互。 + +**说明** + +多模态交互套件面向 **AI/AR 眼镜、学习机、智能机器人**等硬件场景，提供可视化应用配置、预置 Agent/插件、音色管理等完整业务能力。WebRTC 模式下音频通过 UDP 直接传输，内置回声消除和降噪，适合浏览器端低延迟交互场景。 + +## **前提条件及注意事项** + +1. 已在百炼控制台完成以下准备： + + - 已[配置 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)并将其[设置到环境变量](https://help.aliyun.com/zh/model-studio/configure-api-key-through-environment-variables)。 + + - 创建多模态交互应用，获取 **Workspace ID** 和 **App ID**（详情请参见[应用创建](https://help.aliyun.com/zh/model-studio/multimodal-app-creation)）。 + + - 在应用中完成模型、音色、提示词、Agent/插件等配置（详情请参见[应用配置](https://help.aliyun.com/zh/model-studio/multimodal-app-configuration)）。 + +2. 使用支持 WebRTC 的现代浏览器（Chrome、Edge、Firefox、Safari 等）。 + +3. 浏览器需要麦克风权限；如需视频交互，还需摄像头权限。 + +4. 浏览器无法直接向服务端发起 SDP 交换请求（受 CORS 限制），Demo 中需通过终端执行 curl 命令完成连接建立；正式产品中由业务后端代理时不存在此限制。 + + +## **实现实时通话** + +以下时序图展示了整个 WebRTC 实时通话的完整流程： + +WebRTC 多模态交互套件实时通话流程时序图 + +![1111](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6265914871/p1088078.svg) + +### **创建 RTCPeerConnection** + +调用浏览器原生 `RTCPeerConnection` 构造函数创建连接实例。服务端采用 ICE-lite 模式直连，无需配置 ICE 服务器。 + +``` +pc = new RTCPeerConnection(); +``` + +同时注册关键回调： + +``` +pc.onconnectionstatechange = () => { + if (pc.connectionState === 'connected') { + // 连接成功 + } else if (['failed', 'closed', 'disconnected'].includes(pc.connectionState)) { + // 连接断开，清理资源 + endSession(); + } +}; + +pc.ontrack = (e) => { + // 将远端音频流绑定到 audio 元素播放 + const remoteAudio = document.createElement('audio'); + remoteAudio.autoplay = true; + remoteAudio.srcObject = e.streams[0]; + document.body.appendChild(remoteAudio); +}; +``` + +### **获取本地媒体流** + +通过 `navigator.mediaDevices.getUserMedia` 获取麦克风权限（必须），以及摄像头权限（可选）。 + +**纯音频模式：** + +``` +const localStream = await navigator.mediaDevices.getUserMedia({ audio: true }); +``` + +**音视频模式（需要 AI 视觉理解时）：** + +``` +const localStream = await navigator.mediaDevices.getUserMedia({ + audio: true, + video: { + facingMode: { ideal: 'environment' }, // 后置摄像头 + frameRate: { ideal: 30, max: 30 }, + width: { ideal: 640 }, + height: { ideal: 480 } + } +}); +``` + +**说明** + +视频帧率根据场景调整。需要 AI 实时理解画面的场景（如物体识别、场景描述）建议 15-30 fps。 + +### **添加媒体轨道到 PeerConnection** + +将本地音频轨道添加到 PeerConnection。如果开启了视频，视频轨道一并添加。 + +``` +localStream.getTracks().forEach(track => pc.addTrack(track, localStream)); +``` + +**维持视频发送质量（可选）：** + +在弱网环境下，浏览器可能自动降低视频分辨率。可通过设置 sender 参数尽量维持： + +``` +const sender = pc.getSenders().find(s => s.track && s.track.kind === 'video'); +if (sender) { + const params = sender.getParameters(); + if (!params.encodings || params.encodings.length === 0) params.encodings = [{}]; + params.encodings[0].scaleResolutionDownBy = 1.0; + params.encodings[0].maxBitrate = 2500000; // 2.5 Mbps + params.encodings[0].maxFramerate = 30; + params.degradationPreference = 'maintain-resolution'; + await sender.setParameters(params); +} +``` + +### **创建 DataChannel** + +创建名为 `oai-events` 的 DataChannel，用于与服务端交换控制消息（run-task、事件通知等）。 + +``` +const dc = pc.createDataChannel('oai-events'); + +dc.onopen = () => { + console.log('DataChannel open'); + // DataChannel 就绪后发送 run-task + sendStartMessage(dc); +}; + +dc.onmessage = (e) => { + const evt = JSON.parse(e.data); + handleServerEvent(evt, dc); +}; +``` + +同时监听 `pc.ondatachannel` 以处理服务端主动创建的 DataChannel： + +``` +pc.ondatachannel = (event) => { + const ch = event.channel; + if (ch.label === 'txt' || ch.label === 'oai-events') { + ch.onopen = () => sendStartMessage(ch); + ch.onmessage = (e) => handleServerEvent(JSON.parse(e.data), ch); + } +}; +``` + +### **生成 Offer SDP** + +调用 `createOffer` 并 `setLocalDescription`，浏览器生成包含本地媒体能力描述的 Offer SDP。服务端采用 ICE-lite 模式，客户端无需等待 ICE candidates 收集完成，`setLocalDescription` 后即可发送。 + +``` +const offer = await pc.createOffer(); +await pc.setLocalDescription(offer); + +// offer.sdp 即为待发送的 Offer SDP 字符串 +``` + +### **交换 SDP（HTTP POST）** + +将 Offer SDP 通过 HTTP POST 发送到服务端 WebRTC 端点，服务端返回 Answer SDP。 + +**Endpoint 格式：**`{workspace_id}.{region}.maas.aliyuncs.com`，其中 `workspace_id` 为百炼工作空间 ID（如 `llm-xxxxxxxxxx`），`region` 为部署区域（如 `cn-beijing`）。创建工作空间后即可在百炼控制台获取。 + +**请求配置：** + +**配置项** + +**说明** + +请求地址 + +`POST https://{workspace_id}.{region}.maas.aliyuncs.com/api/v1/webrtc/inference?model=multimodal-dialog` + +Content-Type + +`application/sdp` + +请求头 + +`Authorization: Bearer {DASHSCOPE_API_KEY}` + +请求体 + +客户端生成的 Offer SDP 字符串 + +响应 + +成功：HTTP 200，返回服务端 Answer SDP 字符串 + +``` +const API_KEY = 'your-api-key'; // 百炼控制台获取 +const WORKSPACE_ID = '{workspace-id}'; // 百炼控制台获取 +const REGION = 'cn-beijing'; +const SIGNALING_URL = `https://${WORKSPACE_ID}.${REGION}.maas.aliyuncs.com/api/v1/webrtc/inference?model=multimodal-dialog`; + +const resp = await fetch(SIGNALING_URL, { + method: 'POST', + headers: { + 'Content-Type': 'application/sdp', + 'Authorization': `Bearer ${API_KEY}`, + }, + body: offer.sdp, +}); + +if (!resp.ok) throw new Error('SDP 交换失败: ' + resp.status); +const answerSdp = await resp.text(); +``` + +**说明** + +由于浏览器端受 CORS 限制无法直接请求服务端，Demo 中需用户手动在终端执行 curl 命令。正式产品中应通过业务后端代理此请求。 + +**curl 命令格式：** + +``` +curl -X POST 'https://{workspace_id}.{region}.maas.aliyuncs.com/api/v1/webrtc/inference?model=multimodal-dialog' \ + -H 'Content-Type: application/sdp' \ + -H 'Authorization: Bearer $DASHSCOPE_API_KEY' \ + --data-binary '' +``` + +### **设置 Answer SDP 建立连接** + +将从服务端获取的 Answer SDP 设置为远端描述，WebRTC 连接随即建立。 + +``` +await pc.setRemoteDescription({ type: 'answer', sdp: answerSdp }); +// 连接建立完成，pc.connectionState 将变为 'connected' +``` + +### **发送 run-task 启动会话** + +WebRTC 连接建立、DataChannel 就绪后，客户端发送 **run-task** 消息启动多模态会话。 + +``` +let currentTaskId = null; + +function sendStartMessage(channel) { + currentTaskId = generateTaskId(); + const msg = { + payload: { + input: { + workspace_id: '{workspace-id}', + app_id: '{app-id}', + directive: 'Start' + }, + task_group: 'aigc', + task: 'multimodal-generation', + function: 'generation', + model: 'multimodal-dialog', + parameters: { + client_info: { + user_id: '{user-id}', + device: { uuid: '{device-uuid}' }, + network: { ip: '{client-ip}' } + }, + upstream: { + mode: 'duplex', + sample_rate: '16000', + type: 'AudioAndVideo' // 'Audio' 或 'AudioAndVideo' + }, + dialog_attributes: { + vocabulary_id: '{vocabulary-id}' // 可选 + }, + downstream: { + voice: 'longanhuan', + sample_rate: 24000, + audio_format: 'pcm' + } + } + }, + header: { + streaming: 'duplex', + action: 'run-task', + task_id: currentTaskId + } + }; + channel.send(JSON.stringify(msg)); +} + +function generateTaskId() { + if (window.crypto && typeof window.crypto.randomUUID === 'function') { + return window.crypto.randomUUID().replace(/-/g, ''); + } + const r = () => Math.random().toString(16).slice(2); + return (Date.now().toString(16) + r() + r()).slice(0, 32); +} +``` + +**参数说明：** + +**参数路径** + +**类型** + +**说明** + +`payload.input.workspace_id` + +string + +百炼工作空间 ID，在控制台「应用管理」中获取 + +`payload.input.app_id` + +string + +多模态交互应用 ID，在控制台「应用管理」中获取 + +`payload.input.directive` + +string + +固定值 `Start` + +`payload.parameters.client_info.user_id` + +string + +业务系统中的用户标识，用于日志追踪 + +`payload.parameters.client_info.device.uuid` + +string + +设备唯一标识，用于设备维度的数据分析 + +`payload.parameters.upstream.mode` + +string + +交互模式：`duplex`（全双工）、`push2talk`（按住说话）、`tap2talk`（点击说话） + +`payload.parameters.upstream.type` + +string + +上行媒体类型：`Audio`（纯语音）或 `AudioAndVideo`（音视频） + +`payload.parameters.upstream.sample_rate` + +string + +上行音频采样率，通常 `16000` + +`payload.parameters.downstream.voice` + +string + +下行音色，可在百炼控制台音色列表中选取 + +`payload.parameters.downstream.sample_rate` + +number + +下行音频采样率，通常 `24000` + +`payload.parameters.dialog_attributes.vocabulary_id` + +string + +可选，热词表 ID，提升专有名词识别准确率 + +**说明** + +应用中的模型选择、提示词、Agent/插件、知识库等配置均在百炼控制台可视化完成，无需通过代码传入。run-task 只需指定 `workspace_id` 和 `app_id`，服务端会自动加载对应配置。 + +### **实时对话** + +run-task 发送成功后，进入实时对话状态： + +- **上行**：浏览器采集的音频/视频通过 RTP 协议自动发送到服务端 + +- **下行音频**：AI 语音回复通过 `ontrack` 回调接收并播放 + +- **下行事件**：通过 DataChannel 接收业务事件 + + +``` +function handleServerEvent(evt, channel) { + const type = evt.type || evt.header?.action; + + switch (type) { + case 'open_videochat': + // 服务端请求开启视频通道 + // 延迟数秒响应，确保视频处理通道就绪 + setTimeout(() => { + channel.send(JSON.stringify({ + payload: { + input: { text: '', type: 'prompt', directive: 'RequestToRespond' }, + parameters: { + biz_params: { + videos: [{ action: 'connect', type: 'voicechat_video_channel' }] + } + } + }, + header: { + streaming: 'duplex', + action: 'continue-task', + task_id: currentTaskId + } + })); + }, 3000); + break; + + default: + console.log('[服务端事件]', type, evt); + break; + } +} +``` + +**说明** + +**open\_videochat 机制**：当 `upstream.type` 设为 `AudioAndVideo` 时，客户端已通过 WebRTC 发送视频轨道。服务端在需要时（如 AI Agent 判断需要"看"画面）会推送 `open_videochat` 事件，客户端需回复 `continue-task` 确认视频通道建立。该机制允许按需开启视频处理，节省服务端资源。 + +**静音 / 取消静音：** + +``` +// 静音 +localStream.getAudioTracks().forEach(t => { t.enabled = false; }); +// 取消静音 +localStream.getAudioTracks().forEach(t => { t.enabled = true; }); +``` + +**开启/关闭视频：** + +``` +// 关闭视频 +localStream.getVideoTracks().forEach(t => { t.enabled = false; }); +// 开启视频 +localStream.getVideoTracks().forEach(t => { t.enabled = true; }); +``` + +### **结束会话与资源清理** + +通话结束后需要正确释放所有资源，避免内存泄漏和设备占用。 + +``` +function endSession() { + // 1. 关闭 DataChannel + if (dataChannel) { + dataChannel.close(); + dataChannel = null; + } + + // 2. 停止本地媒体流（释放麦克风/摄像头） + if (localStream) { + localStream.getTracks().forEach(t => t.stop()); + localStream = null; + } + + // 3. 关闭 PeerConnection + if (pc) { + pc.close(); + pc = null; + } + + // 4. 重置状态 + currentTaskId = null; +} +``` + +## **注意事项** + +1. **API Key 安全**：切勿将 API Key 硬编码在前端代码中。生产环境应通过后端服务代理 SDP 交换请求，API Key 仅存放在服务端。 + +2. **CORS 限制**：浏览器端无法直接调用百炼 API 进行 SDP 交换，正式产品中需要通过后端代理转发请求。 + +3. **HTTPS 要求**：`getUserMedia` 在非 localhost 环境下要求页面必须通过 HTTPS 提供服务。 + +4. **交互模式选择**：多模态套件支持三种交互模式——`duplex`（全双工，用户可随时打断）、`push2talk`（按住说话）、`tap2talk`（点击说话）。根据硬件形态选择合适的模式。 + +5. **视频帧率**：视频理解场景建议 15-30 fps；如果仅需偶尔拍照识别，可降低帧率节省带宽。 + +6. **浏览器兼容性**：推荐使用 Chrome 90+、Edge 90+、Firefox 85+、Safari 15+。 + +7. **单实例限制**：同一页面同时只应维护一个 `RTCPeerConnection` 实例，创建新会话前需先关闭旧连接。 + + +## **完整 Demo 示例下载** + +以下是一个完整的 HTML 页面 Demo，可直接在浏览器中运行体验多模态交互。由于浏览器 CORS 限制，SDP 交换通过 curl 命令手动完成。 + +[webrtc\_multimodel\_demo.html](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/ifirko/webrtc_multimodel_demo.html) + +**使用步骤：** + +1. 在浏览器中打开该文件。 + +2. 填写连接配置：Endpoint（格式为 `{workspace_id}.{region}.maas.aliyuncs.com`）、API Key、Workspace ID 和 App ID。 + +3. 如需视频交互，勾选“开启视频”。 + +4. 点击**开始会话**，允许浏览器访问麦克风（及摄像头）。 + +5. 如果浏览器能直接发起请求（无 CORS 限制），将自动完成连接；否则页面会显示 curl 命令，复制到终端执行后将返回的 Answer SDP 粘贴回页面即可。 + +6. 连接建立后，对着麦克风说话即可与多模态 AI 实时对话。 + +7. 通话结束后可点击“下载远端音频”保存 AI 回复的录音。 + + +## **相关文档** + +- [通义多模态交互开发套件产品概述](https://help.aliyun.com/zh/model-studio/multimodal-products-overview) + +- [多模态交互套件使用指南](https://help.aliyun.com/zh/model-studio/multimodal-guidelines/) + +- [多模态交互 SDK（Python/Java）GitHub 示例代码](https://github.com/aliyun/alibabacloud-bailian-speech-demo/tree/master/samples/conversation/multimodal_dialog) + +- [WebRTC API (MDN)](https://developer.mozilla.org/zh-CN/docs/Web/API/WebRTC_API) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-omni-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-omni-realtime.md new file mode 100644 index 00000000..b5d496b5 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-omni-realtime.md @@ -0,0 +1,388 @@ +# 通过WebRTC使用qwen3.5-omni-plus-realtime实现实时通话 + +本文档说明如何在浏览器端通过 WebRTC + JavaScript 接入百炼 Realtime API，实现与 qwen3.5-omni-plus-realtime 模型的实时音视频通话。 + +**说明** + +WebRTC 适合浏览器端、低延迟语音场景，音频通过 UDP 直接传输，内置回声消除和降噪。WebRTC 仅支持服务端 VAD 模式（`server_vad` 或 `semantic_vad`），不支持手动模式。 + +## **前提条件及注意事项** + +1. 已[配置 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)并将其[设置到环境变量](https://help.aliyun.com/zh/model-studio/configure-api-key-through-environment-variables)。 + +2. 使用支持 WebRTC 的现代浏览器（Chrome、Edge、Firefox、Safari 等）。 + +3. 浏览器需要麦克风权限；如需视频通话，还需摄像头权限。 + +4. 浏览器无法直接向服务端发起 SDP 交换请求（受 CORS 限制），Demo 中通过终端执行 curl 命令完成连接建立；正式使用时由业务 AppServer 代理完成，无此限制。 + + +## **实现 AI 音视频通话** + +以下时序图展示了整个 WebRTC 音视频通话的完整流程： + +WebRTC 音视频通话流程时序图 + +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/5665914871/p1088079.png) + +### **创建 RTCPeerConnection** + +调用浏览器原生 `RTCPeerConnection` 创建连接实例，无需配置 ICE 服务器（服务端会处理 NAT 穿透）。 + +``` +pc = new RTCPeerConnection({ iceServers: [ ] }); +``` + +注册关键回调： + +``` +// 连接状态监听 +pc.onconnectionstatechange = () => { + if (!pc) return; + if (pc.connectionState === 'connected') { + setStatus('已连接，请说话', 'connected'); + } else if (["failed", "closed", "disconnected"].includes(pc.connectionState)) { + endSession(true); + } +}; + +// 接收远端音频流并播放 + 启动录制 +pc.ontrack = async (e) => { + const stream = e.streams[0]; + ensureHiddenAudioEl(); + hiddenRemoteAudioEl.srcObject = stream; + try { await hiddenRemoteAudioEl.play(); } catch {} + startRecordingRemoteStream(stream); +}; +``` + +### **获取本地媒体流** + +通过一次 `getUserMedia` 调用获取所需的音频（必须）和视频（可选）。是否开启视频由用户勾选"开启视频"复选框决定。 + +``` +const wantVideo = !!sendVideoCheckbox.checked; + +const constraints = wantVideo + ? { + audio: true, + video: { + facingMode: { ideal: "user" }, + frameRate: { ideal: 30, max: 30 }, + width: { ideal: 640 }, + height: { ideal: 480 }, + } + } + : { audio: true }; + +localStream = await navigator.mediaDevices.getUserMedia(constraints); +``` + +**说明** + +音频和视频通过**同一次** `getUserMedia` 调用获取，而非分开请求。视频预览帧率为 30fps（本地流畅预览），发送帧率会通过 Canvas 降至 2fps。 + +### **添加媒体轨道到 PeerConnection** + +**添加音频轨道：** + +``` +localStream.getAudioTracks().forEach(t => { + pc.addTrack(t, localStream); + gatedAudioTracks.push(t); +}); +``` + +**添加视频轨道（可选，通过 Canvas 降帧至 2fps）：** + +Canvas 尺寸从摄像头实际分辨率动态获取，而非硬编码： + +``` +const sendFps = 2; +const settings = localStream.getVideoTracks()[0].getSettings(); +sendCanvas = document.createElement("canvas"); +sendCanvas.width = settings.width || 640; // 动态获取实际宽度 +sendCanvas.height = settings.height || 480; // 动态获取实际高度 +sendCanvasCtx = sendCanvas.getContext("2d", { alpha: false }); + +sendCanvasStream = sendCanvas.captureStream(sendFps); // 2fps +const lowFpsTrack = sendCanvasStream.getVideoTracks()[0]; +pc.addTrack(lowFpsTrack, sendCanvasStream); +gatedVideoTracks.push(lowFpsTrack); + +// requestAnimationFrame 循环：将摄像头画面绘制到 Canvas +const pump = () => { + if (!sendCanvasCtx || !sendCanvas) return; + try { sendCanvasCtx.drawImage(localVideo, 0, 0, sendCanvas.width, sendCanvas.height); } catch {} + sendRafId = requestAnimationFrame(pump); +}; +sendRafId = requestAnimationFrame(pump); +``` + +**媒体门控（关键）：** + +添加轨道后立即禁止发送，确保在收到 `session.created` 之前不推送媒体数据： + +``` +// 1. 禁用所有轨道的 enabled +gateMedia(false); // track.enabled = false + +// 2. 将 sender 的 track 替换为 null，彻底阻止发送 +audioSender = pc.getSenders().find(s => s.track?.kind === 'audio'); +videoSender = pc.getSenders().find(s => s.track?.kind === 'video'); +audioTrack = audioSender?.track; +videoTrack = videoSender?.track; +await audioSender?.replaceTrack(null); +await videoSender?.replaceTrack(videoTrack ? null : undefined); +``` + +**说明** + +等价于其他 SDK 中的 `enableSendMediaStream(false)`，必须在收到 `session.created` 后才恢复发送。 + +### **创建 DataChannel** + +创建名为 `oai-events` 的 DataChannel，用于与 AI 服务端交换会话控制事件。 + +``` +const dc = pc.createDataChannel('oai-events'); + +dc.onopen = () => console.log("DC open"); +dc.onmessage = (e) => { + handleDcMessage(e.data, dc); +}; + +// 同时监听服务端主动创建的 DataChannel +pc.ondatachannel = (event) => { + const ch = event.channel; + ch.onmessage = (e) => { + handleDcMessage(e.data, ch); + }; +}; +``` + +### **生成 Offer SDP** + +调用 `createOffer()` 并设置本地描述，等待 ICE 候选收集完成后获取完整的 Offer SDP。 + +``` +pc.onicegatheringstatechange = () => { + if (!pc) return; + if (pc.iceGatheringState === "complete" && pc.localDescription?.sdp) { + const sdp = pc.localDescription.sdp; + // ICE 收集完成，Offer SDP 可用 + // 自动生成 curl 命令供用户使用 + } +}; + +const offer = await pc.createOffer(); +await pc.setLocalDescription(offer); +``` + +**说明** + +必须等待 `iceGatheringState === "complete"` 后再使用 SDP，此时 SDP 中包含所有 ICE 候选信息。 + +### **交换 SDP（通过 curl 命令或业务 AppServer）** + +将 Offer SDP 发送到百炼服务端，获取 Answer SDP。Demo 中通过 curl 命令完成： + +``` +curl -X POST 'https://{endpoint}/api/v1/webrtc/realtime?model=qwen3.5-omni-plus-realtime' \ + -H 'Content-Type: application/sdp' \ + -H 'Authorization: Bearer $DASHSCOPE_API_KEY' \ + --data-binary '' +``` + +**说明** + +生产环境中，此步骤应由业务 AppServer 代理完成，避免前端暴露 API Key。`{endpoint}` 为 Realtime API 接入地址。 + +### **设置 Answer SDP 建立连接** + +将服务端返回的 Answer SDP 设置为远端描述，WebRTC 连接即开始建立。注意 SDP 格式需要规范化处理： + +``` +function normalizeSdpForSetRemote(sdp) { + sdp = String(sdp).trim().replace(/\r?\n/g, "\r\n"); + if (!sdp.endsWith("\r\n")) sdp += "\r\n"; + return sdp; +} + +const answerSdp = normalizeSdpForSetRemote(txt); +await pc.setRemoteDescription({ type: 'answer', sdp: answerSdp }); +``` + +**说明** + +SDP 规范要求行尾为 `\r\n`，`normalizeSdpForSetRemote` 负责处理不同来源的换行符兼容问题。 + +### **配置 AI 会话（session.update）** + +连接建立后，服务端通过 DataChannel 发送 `session.created` 事件。收到后需： + +1. 解除媒体门控，恢复音视频发送 + +2. 发送 `session.update` 配置会话参数 + + +**解除门控并恢复媒体：** + +``` +function handleDcMessage(data, channel) { + let obj; + try { obj = JSON.parse(data); } catch (err) { return; } + + if (obj?.type === "session.created") { + // 解除门控：恢复 track.enabled + gateMedia(true); + // 恢复 sender 的实际 track + if (audioSender) audioSender.replaceTrack(audioTrack); + if (videoSender && videoTrack) videoSender.replaceTrack(videoTrack); + // 发送会话配置 + sendUpdate(channel); + } +} +``` + +**session.update 消息体：** + +``` +const update = { + event_id: `event_${Date.now()}`, + type: "session.update", + session: { + input_audio_format: "pcm", + input_audio_transcription: { model: "qwen3-asr-flash-realtime" }, + instructions: "You are a helpful assistant.", + modalities: ["text", "audio"], + output_audio_format: "pcm", + smooth_output: false, + turn_detection: { + prefix_padding_ms: 500, + silence_duration_ms: 800, + threshold: 0.5, + type: "server_vad", + }, + }, +}; +if (channel && channel.readyState === "open") channel.send(JSON.stringify(update)); +``` + +**说明** + +`turn_detection.type` 可设为 `server_vad`（基于音量检测）或 `semantic_vad`（基于语义检测）。WebRTC 模式不支持手动 VAD。 + +### **实时对话** + +连接建立后，音视频通过 RTP 实时传输。远端 AI 语音通过 `ontrack` 回调接收并播放，同时使用 MediaRecorder 录制以便下载。 + +**接收远端音频并录制：** + +``` +pc.ontrack = async (e) => { + const stream = e.streams[0]; + ensureHiddenAudioEl(); + hiddenRemoteAudioEl.srcObject = stream; + try { await hiddenRemoteAudioEl.play(); } catch {} + startRecordingRemoteStream(stream); // 启动录制 +}; + +function startRecordingRemoteStream(remoteStream) { + const audioTracks = remoteStream.getAudioTracks(); + if (!audioTracks.length) return; + const audioStream = new MediaStream(audioTracks); + + recordedChunks = [ ]; + + mediaRecorder = new MediaRecorder(audioStream, { mimeType: 'audio/webm' }); + mediaRecorder.ondataavailable = (e) => { + if (e.data && e.data.size > 0) recordedChunks.push(e.data); + }; + mediaRecorder.onstop = () => { + audioBlob = new Blob(recordedChunks, { type: 'audio/webm' }); + // 录制结束后可下载 + }; + mediaRecorder.start(); +} +``` + +**DataChannel 事件统一展示：** + +所有通过 DataChannel 收发的事件（包括 `session.created`、`response.audio_transcript.done` 等）统一通过事件面板展示，支持展开查看完整 JSON： + +``` +function pushEventFromDataChannel(eventObj) { + const ts = eventObj.timestamp || nowTs(); + events.unshift({ event: eventObj, timestamp: ts }); + renderEvents(); +} +``` + +### **结束会话与资源清理** + +结束通话时需依次清理所有资源，顺序很重要： + +``` +function endSession(silent = false) { + // 1. 停止 Canvas 降帧循环 + if (sendRafId) cancelAnimationFrame(sendRafId); + sendRafId = 0; + if (sendCanvasStream) sendCanvasStream.getTracks().forEach(t => t.stop()); + sendCanvasStream = null; sendCanvasCtx = null; sendCanvas = null; + + // 2. 停止录制 + try { if (mediaRecorder && mediaRecorder.state !== "inactive") mediaRecorder.stop(); } catch {} + mediaRecorder = null; + + // 3. 停止本地媒体流 + if (localStream) { + localStream.getTracks().forEach(t => t.stop()); + localStream = null; + } + + // 4. 关闭 PeerConnection + if (pc) { try { pc.close(); } catch {} pc = null; } + + // 5. 清理远端音频元素 + if (hiddenRemoteAudioEl) { + try { hiddenRemoteAudioEl.pause(); } catch {} + hiddenRemoteAudioEl.srcObject = null; + hiddenRemoteAudioEl.remove(); + hiddenRemoteAudioEl = null; + } +} +``` + +**说明** + +结束后可通过"下载远端音频"按钮下载 AI 回复的录音（WebM 格式）。 + +## **注意事项** + +1. **媒体门控必须在 session.created 后解除**：在服务端发送 `session.created` 之前推送媒体数据会被丢弃，必须通过 `replaceTrack(null)` 彻底阻断发送。 + +2. **视频降帧通过 Canvas 实现**：本地预览 30fps，发送至服务端仅 2fps，通过 `captureStream(2)` 控制，节省带宽。 + +3. **SDP 格式规范化**：设置 Answer SDP 前必须确保行尾为 `\r\n`，否则 `setRemoteDescription` 可能失败。 + +4. **视频为可选功能**：用户未勾选视频时，仅请求音频权限，不会触发摄像头授权弹窗。 + +5. **远端音频自动录制**：通过 MediaRecorder 录制 AI 回复的音频流，会话结束后可下载 WebM 格式文件。 + +6. **WebRTC 仅支持服务端 VAD**：不支持 `manual` 模式，可选 `server_vad`（音量检测）或 `semantic_vad`（语义检测）。 + + +## **完整 demo 下载** + +完整示例代码请下载：[webrtc\_demo.html](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/ychtmj/webrtc_demo.html)。 + +## **相关文档** + +- [WebRTC API (MDN)](https://developer.mozilla.org/zh-CN/docs/Web/API/WebRTC_API) + +- [RTCPeerConnection (MDN)](https://developer.mozilla.org/zh-CN/docs/Web/API/RTCPeerConnection) + +- qwen3.5-omni-plus-realtime 模型客户端事件：[客户端事件](https://help.aliyun.com/zh/model-studio/client-events) + +- qwen3.5-omni-plus-realtime 模型服务端事件：[服务端事件](https://help.aliyun.com/zh/model-studio/server-events) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-overview.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-overview.md new file mode 100644 index 00000000..217e42ed --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-overview.md @@ -0,0 +1,180 @@ +# Realtime API简介 + +Realtime API 是一系列针对性能、延迟、抗弱网、对接成本、适配性提供多种对接方式的方法，供客户灵活选择。 + +## **概述** + +Realtime API 支持 **WebSocket**、**WebRTC** 和 **AOQ（AI over QUIC）**三种传输协议，开发者可以根据业务场景灵活选择。 + +**维度** + +**WebSocket** + +**WebRTC** + +**AOQ** + +适用场景 + +服务端集成、快速原型验证 + +浏览器端互动、传统音视频通话 + +AI 多模态实时交互、弱网场景、混合数据传输 + +浏览器兼容性 + +原生支持 + +原生支持 + +不支持 + +接入难度 + +极低 + +中等 + +低 + +弱网对抗 + +差 + +良好 + +极致 + +数据类型 + +文本/音频/图像 + +音视频 + 文本 + +音视频 + 文本 + +建连速度 + +慢 + +慢 + +快 + +回声消除/降噪 + +无，需客户端自行处理 + +内置 + +内置 + +AI 场景适配 + +基础，适合纯文本或低实时性场景 + +传统设计，AI 场景需额外适配 + +原生为 AI 多模态数据特征深度定制 + +端侧平台支持 + +全平台（任何支持 WebSocket 的环境） + +浏览器、移动端 + +Android / iOS / HarmonyOS + +开发者可根据实际需求选择协议方案： + +- **WebSocket 方案**：适合服务端集成、快速原型验证、对接入门槛要求极低的场景。通过 DashScope SDK 可快速实现实时语音对话。 + +- **WebRTC 方案**：适合需要浏览器原生支持、已有 WebRTC 基础设施的传统音视频通话场景，内置回声消除和降噪能力。 + +- **AOQ 方案**：适合对延迟、弱网对抗、多模态数据传输有极致要求的 AI 实时交互场景，同时内置回声消除和降噪能力，尤其是移动端原生应用。 + + +## **模型/应用支持力度** + +不同协议对模型和应用的支持情况如下： + +**模型/应用类型** + +**模型** + +**AOQ** + +**WebRTC** + +**WebSocket** + +实时全模态 + +qwen3.5-omni-plus-realtime + +支持 + +支持 + +支持 + +qwen3.5-omni-flash-realtime + +支持 + +支持 + +支持 + +qwen3.5-livetranslate-flash-realtime + +支持 + +支持 + +支持 + +多模态开发套件 + +multimodal-dialog + +不支持 + +支持 + +支持 + +实时语音识别 + +Fun-ASR系列模型 + +不支持 + +不支持 + +支持 + +实时语音合成 + +CosyVoice系列模型 + +不支持 + +不支持 + +支持 + +实时语音对话 + +qwen-audio-3.0-realtime-plus、qwen-audio-3.0-realtime-flash + +不支持 + +不支持 + +支持 + +**说明** + +模型的名称、上下文、价格、快照版本等信息请参见[阿里云百炼控制台](https://bailian.console.aliyun.com/cn-beijing#/home)；并发限流条件请参考[限流](https://help.aliyun.com/zh/model-studio/rate-limit)。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-connect-model.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-connect-model.md new file mode 100644 index 00000000..147179b9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-connect-model.md @@ -0,0 +1,291 @@ +# 实现接通模型/应用 + +介绍如何通过 AOQ、WebRTC、WebSocket 三种协议接入 Realtime API 模型或应用，包含各协议的连接流程、时序图和代码示例。 + +## **AOQ 接入** + +AOQ 基于 QUIC 协议深度定制，适合移动端原生应用，支持音频/视频/数据混合传输，内置极致抗弱网能力。以下以 iOS Demo 为例。 + +### **整体流程时序图** + +![AOQ中文1](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/5755914871/p1088073.jpg) + +### **创建引擎并设置回调** + +``` +let config = AoqCreateConfig() +config.workDir = workDir +config.enableDumpAudio = false +engine = AoqClientEngine.createEngine(config, delegate: self) +``` + +实现 `AoqEngineDelegate` 协议监听 `onConnectionStatusChange`、`onDataMsg`、`onError` 等回调。 + +### **启动音频采集与播放** + +``` +// 音频采集 +let capCfg = AoqAudioCaptureConfig() +capCfg.channel = 1; capCfg.isExternal = false +engine.startAudioCapture(capCfg) + +// 音频播放 +let playCfg = AoqAudioPlaybackConfig() +playCfg.channel = 1; playCfg.isExternal = false +engine.startAudioPlayer(playCfg) + +// 视频采集（可选） +let vidCfg = AoqVideoCaptureConfig() +vidCfg.width = 720; vidCfg.height = 1280; vidCfg.fps = 15 +engine.startVideoCapture(vidCfg) +``` + +### **获取连接凭证** + +由业务 AppServer 代理百炼请求，参考 [Token 鉴权](https://help.aliyun.com/zh/model-studio/realtime-token-authentication) 章节。 + +### **设置编解码及建立连接** + +设置编解码参数后调用 `connect`： + +``` +// 音频编解码配置 +let encCfg = AoqAudioCodecConfig() +encCfg.codecType = .audioPCM; encCfg.sampleRate = 16000; encCfg.channel = 1 +engine.setAudioEncoderConfig(encCfg) +engine.setAudioDecoderConfig(encCfg) + +// connect 前关闭媒体发送，待 session.updated 后再开启 +engine.enableSendMediaStream(.audio, enable: false) + +let config = AoqConnectConfig() +config.token = token +config.sid = sid +config.certFingerprint = certificate +config.relayEndpoints = relayEndpoints +config.workspaceIdHash = workspaceIdHash +config.publishTracks = [audioTrack, dataTrack] +config.subscribeTracks = [audioTrack, dataTrack] +engine.connect(config) +``` + +**重要** + +**重要**：AOQ SDK 在建联后会默认发送媒体数据，此示例演示了连接模型时关闭媒体发送的能力。 + +### **配置 AI 会话** + +连接成功后发送 `session.update` 的示例，详见[模型客户端事件参考](https://help.aliyun.com/zh/model-studio/client-events)： + +``` +func onConnectionStatusChange(_ status: AoqConnectionStatus) { + if status == .connected { sendSessionUpdate() } +} + +private func sendSessionUpdate() { + let json = """ + { + // 该事件的id，由客户端生成 + "event_id": "event_ToPZqeobitzUJnt3QqtWg", + // 事件类型，固定为session.update + "type": "session.update", + // 会话配置 + "session": { + // 输出模态，支持设置为["text"]（仅输出文本）或["text","audio"]（输出文本与音频）。 + "modalities": [ + "text", + "audio" + ], + // 输出音频的音色 + "voice": "Ethan", + // 输入音频格式，当前仅支持设置为pcm。输入音频为16 kHz采样率的PCM音频流。 + "input_audio_format": "pcm", + // 输出音频格式，当前仅支持设置为pcm。输出音频为24 kHz采样率的PCM音频流。 + "output_audio_format": "pcm", + // 系统消息，用于设定模型的目标或角色。 + "instructions": "你是某五星级酒店的AI客服专员，请准确且友好地解答客户关于房型、设施、价格、预订政策的咨询。请始终以专业和乐于助人的态度回应，杜绝提供未经证实或超出酒店服务范围的信息。", + // 是否开启语音活动检测。若需启用，需传入一个配置对象，服务端将据此自动检测语音起止。 + // 设置为null表示由客户端决定何时发起模型响应。 + "turn_detection": { + // VAD类型，取值为server_vad或semantic_vad。使用qwen3.5-omni-realtime模型时推荐设为semantic_vad。 + "type": "semantic_vad", + // VAD检测阈值。建议在嘈杂的环境中增加，在安静的环境中降低。 + "threshold": 0.5, + // 检测语音停止的静音持续时间，超过此值后会触发模型响应 + "silence_duration_ms": 800 + } + } + } + """ + let msg = AoqDataMsg() + msg.data = json.data(using: .utf8)! + engine.send(msg) +} +``` + +### **收到 session.updated 后开启媒体发送** + +收到模型回复 `session.updated` 的示例，详见[模型服务器事件参考](https://help.aliyun.com/zh/model-studio/server-events)： + +``` +func onDataMsg(_ msg: AoqDataMsg) { + guard let obj = try? JSONSerialization.jsonObject(with: msg.data) as? [String: Any], + let type = obj["type"] as? String else { return } + if type == "session.updated" { + engine.enableSendMediaStream(.audio, enable: true) + engine.enableSendMediaStream(.video, enable: true) + } +} +``` + +**重要** + +**重要**： + +1. 模型必须在收到 `session.updated` 后才开启媒体流发送，否则 AI 侧可能还未准备好接收数据。 + +2. 建连时添加的音频轨道和视频轨道（即 AOQ 媒体通道）会自动将数据传输到服务端。 + + 1. 音频：通过音频轨道直接传输，无需发送 `input_audio_buffer.append` 事件。 + + 2. 视频：通过视频轨道发送画面帧，无需发送 `input_image_buffer.append` 事件。 + + +### **断开连接与销毁引擎** + +``` +engine.disconnect() +AoqClientEngine.destroy() +``` + +## **WebRTC 接入** + +WebRTC 协议不提供 SDK，Web 端可以通过 JavaScript，其他端可以通过开源项目或者第三方支持标准 WebRTC 协议的 RTC 服务商进行接入。以下文档以 Web 端 JavaScript 为例进行介绍。 + +### **整体流程图** + +![AOQ中文2](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/5755914871/p1088074.jpg) + +### **建立连接** + +``` +# pip install aiortc aiohttp certifi +import asyncio, aiohttp, ssl, certifi +from aiortc import RTCPeerConnection, RTCConfiguration, RTCSessionDescription +from aiortc.mediastreams import AudioStreamTrack + +API_KEY = "your-api-key" +MODEL = "目标模型" +SIGNALING_URL = f"https://{{endpoint}}/api/v1/webrtc/realtime?model={MODEL}" + +async def connect(): + pc = RTCPeerConnection(RTCConfiguration(iceServers=[])) + + # 添加音频轨道，确保 Offer SDP 包含 m=audio（服务端必需） + pc.addTrack(AudioStreamTrack()) + + # 创建 DataChannel 以触发 SDP 协商（名称可自定义，服务端会通过名为 "txt" 的通道推送事件） + pc.createDataChannel("oai-events") + + # SDP 交换：创建 Offer 并发送到服务端 + offer = await pc.createOffer() + await pc.setLocalDescription(offer) + + async with aiohttp.ClientSession() as session: + async with session.post( + SIGNALING_URL, + ssl=ssl.create_default_context(cafile=certifi.where()), + data=offer.sdp.encode("utf-8"), + headers={ + "Content-Type": "application/sdp", + "Authorization": f"Bearer {API_KEY}", + }, + ) as resp: + if not resp.ok: + raise Exception(f"SDP 交换失败: {resp.status} {await resp.text()}") + answer_sdp = await resp.text() + + print("=== Offer SDP ===") + print(offer.sdp) + print("=== Answer SDP ===") + print(answer_sdp) + + # ICE 建连自动完成 + await pc.setRemoteDescription(RTCSessionDescription(sdp=answer_sdp, type="answer")) + print("WebRTC 连接已建立") + return pc +``` + +### **配置目标模型参数** + +监听模型返回的 DataChannel 消息保证交互时序： + +``` +pc.ondatachannel = (event) => { + const ch = event.channel; + ch.onmessage = (e) => { + let obj; + try { obj = JSON.parse(e.data); } + catch (err) { + return; + } + if (obj?.type === "session.created") { + sendUpdate(event.channel); + //开始推送音视频 + audioSender?.replaceTrack(audioTrack); + videoSender?.replaceTrack(videoTrack); + } + }; +}; +``` + +### **收发媒体数据** + +建连时添加的音频轨道和视频轨道（即 RTP 媒体通道）会自动将数据传输到服务端。 + +- 音频：通过音频轨道（RTP）直接传输，无需发送 `input_audio_buffer.append` 事件。 + +- 图片：通过视频轨道（RTP）发送画面帧，不支持 `input_image_buffer.append` 事件。 + + +**说明** + +WebRTC 仅支持服务端 VAD 模式（`server_vad` 或 `semantic_vad`），不支持手动模式。 + +### **Demo 源码** + +#### **前提条件** + +- 使用支持 WebRTC 的现代浏览器（Chrome、Edge、Firefox、Safari 等）。 + +- 浏览器需要麦克风权限。 + +- 浏览器无法直接向服务端发起建立连接的请求（受浏览器跨域安全策略限制），因此需要通过终端执行 curl 命令来完成连接建立。 + + +#### **运行示例** + +新建一个 HTML 文件，命名为 `webrtc_demo.html`，并将以下代码复制到文件中： + +[webrtc\_demo.html](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/crtwmi/webrtc_demo.html)。 + +在浏览器中打开此文件，按以下步骤操作： + +1. 点击开始会话，页面会自动生成 Offer SDP 和对应的 curl 命令。 + +2. 点击复制 curl 命令，在终端中执行。命令返回的内容即为 Answer SDP。 + +3. 将 Answer SDP 粘贴到页面的 Answer SDP 文本框中，点击设置 Answer 即可建立连接并开始语音对话。 + + +## **WebSocket 接入** + +可以通过 DashScope SDK 或者模型的 API 进行接入，详见： + +- [实时全模态](https://help.aliyun.com/zh/model-studio/realtime#bdaa43cdd7hsd) + +- [多模态开发套件](https://help.aliyun.com/zh/model-studio/multimodal-interaction-protocol/) + +- [实时语音识别](https://help.aliyun.com/zh/model-studio/fun-asr-realtime-websocket-api) + +- [实时语音合成](https://help.aliyun.com/zh/model-studio/cosyvoice-websocket-api) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-sdk-download.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-sdk-download.md new file mode 100644 index 00000000..69982edf --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-sdk-download.md @@ -0,0 +1,60 @@ +# SDK下载 + +本文提供AOQ SDK下载链接，介绍如何集成SDK，以及SDK相关信息。 + +## **AOQ SDK下载** + +**版本** + +**平台** + +**下载** + +**更新日期** + +**更新说明** + +v1.0.1 + +Android + +[AoqClientSdk-v1.0.1.aar](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/sqsjum/AoqClientSdk-v1.0.1.aar) + +[libPluginOpus.zip](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/oxzenn/libPluginOpus.zip) + +2026-07-09 + +1. 支持AOQ协议接入模型/应用 + +2. 支持音视频编解码参数 + +3. 支持音视频设备采集及播放 + +4. 支持自定义外部输入音频采集 + +5. 支持自定义外部输入视频采集 + +6. 支持自定义外部输入视频编码 + +7. 支持控制媒体流的发送 + + +iOS + +[AoqClientSdk-v1.0.1.zip](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/vlyumo/AoqClientSdk-v1.0.1.zip) + +[PluginOpus.framework.zip](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/rbvmfq/PluginOpus.framework.zip) + +Harmony + +[AoqClientSdk-v1.0.1.har](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/zwlikp/AoqClientSdk-v1.0.1.har) + +[libPluginOpus.zip](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260715/wvtyrh/libPluginOpus.zip) + +**重要** + +AOQ使用音频插件的方式加载Opus编解码器，如果需要Opus编解码时需要下载opus插件并加载到工程内。 + +## **WebSocket SDK下载** + +参见[安装SDK](https://help.aliyun.com/zh/model-studio/install-sdk)。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-token-authentication.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-token-authentication.md new file mode 100644 index 00000000..6830d3a4 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-token-authentication.md @@ -0,0 +1,358 @@ +# Token鉴权 + +介绍 Realtime API 的 Token 鉴权机制，包括 API Key 的获取方式以及 WebSocket、WebRTC、AOQ 三种协议的建连鉴权方法。 + +## **概述** + +Realtime API 使用 **API Key** 进行身份认证。无论您选择 WebSocket、WebRTC 还是 AOQ 协议接入，均通过 HTTP 请求头中的 `Authorization` 字段携带 Bearer Token 完成身份验证。 + +鉴权发生在**建连阶段**，连接建立后的音视频/数据传输无需重复鉴权。 + +三种协议的鉴权差异： + +**协议** + +**鉴权时机** + +**鉴权方式** + +**说明** + +WebSocket + +WebSocket 连接握手时 + +HTTP Header `Authorization: Bearer ` + +客户端或服务端直接携带 API Key 建连 + +WebRTC + +SDP 交换 HTTP 请求时 + +HTTP Header `Authorization: Bearer ` + +客户端或服务端携带 API Key 发起 SDP 交换 + +AOQ + +业务 AppServer 请求网关时 + +HTTP Header `Authorization: Bearer ` + +API Key 仅在服务端使用，客户端使用网关返回的 Token + +## **获取 API Key** + +### **步骤 1：开通百炼服务** + +1. 访问[阿里云百炼控制台](https://bailian.console.aliyun.com/cn-beijing#/home)并登录您的阿里云账号。 + +2. 如果是首次使用，按照页面提示完成服务开通。 + + +### **步骤 2：创建 API Key** + +1. 在控制台左侧导航栏中，选择 **API Key 管理**。 + +2. 点击 **创建 API Key**，选择关联的业务空间。 + +3. 创建完成后，请**立即复制并妥善保存** API Key。 + + +**重要** + +**安全提示**：API Key 是您访问服务的唯一凭证，请勿将其硬编码到客户端代码中或提交到代码仓库。建议通过环境变量或后端服务下发的方式管理。 + +## **建连鉴权详解** + +### **AOQ 协议鉴权** + +AOQ 采用**服务端代理鉴权**模式：API Key 仅在业务 AppServer 侧使用，客户端使用网关返回的临时 Token 建连，避免 API Key 暴露在客户端。 + +![Token鉴权](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3935914871/p1088069.jpg) + +#### **百炼网关请求 curl 示例** + +``` +curl -X POST \ + "https://{endpoint}/api/v1/webrtc/realtime?model=qwen3.5-omni-plus-realtime" \ + -H "Content-Type: application/json" \ + -H "Authorization: Bearer ${DASHSCOPE_API_KEY}" \ + -H "x-dashscope-rtc-transport: moq" \ + -d '{"clientIp": ${客户端真实IP}}' +``` + +#### **请求字段说明** + +**配置项** + +**值** + +**说明** + +endpoint + +根据业务情况选择接入域名 + +指定对应的接入域名，详情请参见[选择地域、服务部署范围和接入域名](https://help.aliyun.com/zh/model-studio/regions/) + +Content-Type + +`application/json` + +\- + +Authorization + +`Bearer ` + +必填 + +x-dashscope-rtc-transport + +`moq` + +**指定使用 AOQ 协议** + +clientIp + +选填。客户端真实公网 IP + +不填写时，使用请求百炼网关的 IP 作为客户端 IP；若填写，则以 clientIp 作为客户端 IP。Realtime API 会参考客户端 IP 提供最佳的 Relay 接入点信息 + +#### **响应示例** + +``` +{ + "sid": "1d06b55683db49bba67a407902f62d02:1782706970:69aecdc5...", + "aoqTokenForClient": "ecc1a46015d5496ca4ff7a48281eb739", + "clientRelayEndpoints": [{"endpoint": "121.199.XX.XX", "port": 8443}], + "clientRelayCertFingerprint": "sha256/99843495...", + "sidExpiresInSecs": 7200, + "extraInfo": {"workspaceIdHash": "2021b6f98cea4cff"} +} +``` + +#### **响应字段说明** + +**字段** + +**说明** + +sid + +会话唯一标识 + +aoqTokenForClient + +客户端连接令牌，传给 SDK 的 token 字段 + +clientRelayEndpoints + +Relay 接入点数组（endpoint + port） + +clientRelayCertFingerprint + +Relay TLS 证书指纹 + +sidExpiresInSecs + +会话过期时间（秒） + +extraInfo.workspaceIdHash + +工作区 ID 哈希 + +#### **AOQ Client SDK 连接示例** + +## **iOS (Swift)** + +``` +let resp = try JSONDecoder().decode(AllocateResponse.self, from: responseData) + +let config = AoqConnectConfig() +config.token = resp.aoqTokenForClient +config.sid = resp.sid +config.certFingerprint = resp.clientRelayCertFingerprint +config.relayEndpoints = resp.clientRelayEndpoints.map { item in + let ep = AoqRelayEndpoint() + ep.endpoint = item.endpoint + ep.port = item.port + return ep +} +config.workspaceIdHash = resp.extraInfo?.workspaceIdHash ?? "" + +let audioTrack = AoqTrackParam() +audioTrack.trackType = .audio +let dataTrack = AoqTrackParam() +dataTrack.trackType = .data +config.publishTracks = [audioTrack, dataTrack] +config.subscribeTracks = [audioTrack, dataTrack] + +engine.connect(config) +``` + +## **Android (Java)** + +``` +JSONObject obj = new JSONObject(responseText); +AoqClientEngine.AoqConnectConfig cfg = new AoqClientEngine.AoqConnectConfig(); +cfg.token = obj.optString("aoqTokenForClient", ""); +cfg.sid = obj.optString("sid", ""); +cfg.certFingerprint = obj.optString("clientRelayCertFingerprint", ""); + +JSONArray arr = obj.optJSONArray("clientRelayEndpoints"); +if (arr != null) { + for (int i = 0; i < arr.length(); i++) { + JSONObject o = arr.optJSONObject(i); + AoqClientEngine.AoqRelayEndpoint ep = new AoqClientEngine.AoqRelayEndpoint(); + ep.endpoint = o.optString("endpoint", ""); + ep.port = o.optInt("port", 0); + cfg.relayEndpoints.add(ep); + } +} + +JSONObject ext = obj.optJSONObject("extraInfo"); +cfg.workspaceIdHash = ext != null ? ext.optString("workspaceIdHash", "") : ""; + +AoqClientEngine.AoqTrackParam audio = new AoqClientEngine.AoqTrackParam(); +audio.trackType = AoqClientEngine.AoqTrackType.AoqTrackTypeAudio; +AoqClientEngine.AoqTrackParam data = new AoqClientEngine.AoqTrackParam(); +data.trackType = AoqClientEngine.AoqTrackType.AoqTrackTypeData; +cfg.publishTracks.add(audio); +cfg.publishTracks.add(data); +cfg.subscribeTracks.add(audio); +cfg.subscribeTracks.add(data); + +engine.connect(cfg); +``` + +## **OHOS (ArkTS)** + +``` +const obj = JSON.parse(responseText) as Record; +const cfg: AoqConnectConfig = { + token: String(obj['aoqTokenForClient'] ?? ''), + sid: String(obj['sid'] ?? ''), + certFingerprint: String(obj['clientRelayCertFingerprint'] ?? ''), + relayEndpoints: (obj['clientRelayEndpoints'] as Array).map(item => ({ + endpoint: String(item['endpoint'] ?? ''), + port: Number(item['port'] ?? 0) + })), + workspaceIdHash: String((obj['extraInfo'] as any)?.['workspaceIdHash'] ?? ''), + publishTracks: [ + { trackType: AoqTrackType.AoqTrackTypeAudio }, + { trackType: AoqTrackType.AoqTrackTypeData } + ], + subscribeTracks: [ + { trackType: AoqTrackType.AoqTrackTypeAudio }, + { trackType: AoqTrackType.AoqTrackTypeData } + ] +}; +engine.connect(cfg); +``` + +**说明** + +`clientIp` 为请求体中的非必填字段。不填写时，使用请求百炼网关的 IP 作为客户端 IP；若填写，则以 clientIp 作为客户端 IP。建议由业务 AppServer 在服务端获取客户端真实 IP 后填入，以获得最佳的 Relay 接入点。 + +## **WebRTC 协议鉴权** + +WebRTC 通过 HTTP POST 请求完成 SDP 交换，鉴权在此阶段完成。客户端将 Offer SDP 发送给服务端，服务端返回 Answer SDP。 + +**配置项** + +**值** + +**说明** + +请求方法 + +POST + +\- + +请求地址 + +`https://{endpoint}/api/v1/webrtc/realtime?model={model_name}` + +替换 endpoint 和 model\_name + +Content-Type + +`application/sdp` + +请求体为 SDP 字符串 + +Authorization + +`Bearer ` + +必填 + +响应 + +HTTP 200，返回 Answer SDP + +失败返回 4xx + +**说明** + +WebRTC 功能目前为白名单开放，请联系商务经理获取 Endpoint。 + +``` +const pc = new RTCPeerConnection(); +const stream = await navigator.mediaDevices.getUserMedia({ audio: true }); +stream.getAudioTracks().forEach(t => pc.addTrack(t, stream)); +pc.createDataChannel('oai-events'); + +const offer = await pc.createOffer(); +await pc.setLocalDescription(offer); + +// 等待 ICE 收集完成后发送 +const resp = await fetch(API_URL, { + method: 'POST', + headers: { + 'Content-Type': 'application/sdp', + 'Authorization': `Bearer ${API_KEY}`, + }, + body: pc.localDescription.sdp, +}); +const answerSdp = await resp.text(); +await pc.setRemoteDescription({ type: 'answer', sdp: answerSdp }); +``` + +## **WebSocket 协议鉴权** + +WebSocket 鉴权最为简单，客户端在建立 WebSocket 连接时直接通过 HTTP Header 携带 API Key。 + +**配置项** + +**值** + +**说明** + +连接地址 + +`wss://dashscope.aliyuncs.com/api-ws/v1/realtime?model={model_name}` + +华北2（北京） + +Authorization + +`Bearer ` + +必填 + +``` +import websocket, os +API_KEY = os.getenv("DASHSCOPE_API_KEY") +URL = "wss://dashscope.aliyuncs.com/api-ws/v1/realtime?model=qwen3.5-omni-plus-realtime" +ws = websocket.WebSocketApp(URL, header=["Authorization: Bearer " + API_KEY]) +ws.run_forever() +``` + +**说明** + +您也可以使用 [DashScope SDK](https://help.aliyun.com/zh/model-studio/install-sdk) 方式接入。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/release-notes/model-release-notes.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/release-notes/model-release-notes.md index a108688f..5867b61c 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/release-notes/model-release-notes.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/release-notes/model-release-notes.md @@ -33,6 +33,30 @@ **功能说明** +7月21日 + +平台功能 + +记忆库商业化通知 + +记忆库商业化通知[了解详情](https://www.aliyun.com/notice/118464) + +7月16日 + +平台功能 + +企业知识库（旧）下线通知 + +企业知识库（旧）下线通知[了解详情](https://www.aliyun.com/notice/118448) + +7月16日 + +平台功能 + +Managed Agent商业化通知 + +Managed Agent商业化通知[了解详情](https://www.aliyun.com/notice/118456) + 7月14日 平台功能 @@ -177,7 +201,7 @@ Coding Plan Coding Plan 联网搜索 MCP 升级 -Coding Plan 联网搜索 MCP 升级 Streamable HTTP 协议，前 2000 次免费，[了解详情](https://help.aliyun.com/zh/model-studio/web-search-for-coding-plan) +Coding Plan 联网搜索 MCP 升级 Streamable HTTP 协议，前 2000 次免费，[了解详情](https://help.aliyun.com/zh/model-studio/web-search-mcp) 6月10日 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/release-notes/newly-released-models.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/release-notes/newly-released-models.md index 1fffce5d..eaba4c3a 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/release-notes/newly-released-models.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/release-notes/newly-released-models.md @@ -8,18 +8,38 @@ **时间** -**服务部署范围** - **模型规格** **功能说明** +推理模型 + +2026-07-25 + +qwen3.7-flash、qwen3.7-flash-2026-07-15 + +Qwen3.7-Flash模型，相较3.6-Flash全面提升多模态理解与Agent执行能力。重点强化多模态基础能力、万物识别能力更强，真实世界感知与空间智能进一步提升，Search Agent、CI Agent等多模态Agent场景能力显著升级、端到端任务执行更稳定，多模态Coding能力优化、vibe coding 体验更加流畅。 + +图像生成 + +2026-07-21 + +qwen-image-3.0-pro + +Qwen-Image-3.0-Pro 系列模型邀测上线，支持长文本输入与图中图密集排版，能够一次性精准生成报纸、分镜、菜单及试卷等复杂版面；具备10像素小字精准渲染能力，生动还原微表情、毛孔与发丝等摄影级细节，并支持 12 国语言、多种字体及主流网页、游戏界面的高保真仿真。[千问-图像生成与编辑3.0](https://help.aliyun.com/zh/model-studio/qwen-image-generation-and-editing-api-reference) + +文生文与视觉理解 + +2026-07-17 + +kimi/kimi-k3 + +Kimi K3 是 Kimi 迄今能力最强的旗舰模型，拥有 2.8 万亿参数，原生支持视觉理解，并拥有 100 万 token 上下文窗口，面向长程编程、知识工作和推理等前沿智能场景而设计。[Kimi-月之暗面](https://help.aliyun.com/zh/model-studio/kimi-api-by-moonshot-ai) + 视频对口型 2026-07-15 -中国内地 - pixverse/pixverse-lipsync 爱诗视频对口型模型，输入视频和音频，生成口型与音频同步的视频。 @@ -28,8 +48,6 @@ pixverse/pixverse-lipsync 2026-07-15 -中国内地 - pixverse/pixverse-motioncontrol 爱诗视频动作模仿模型，输入视频和参考动作视频，生成模仿参考动作的视频。 @@ -38,8 +56,6 @@ pixverse/pixverse-motioncontrol 2026-07-15 -中国内地 - pixverse/pixverse-upscale 爱诗视频超清模型，将低分辨率视频提升至更高分辨率。 @@ -48,8 +64,6 @@ pixverse/pixverse-upscale 2026-07-14 -中国内地 - qwen-audio-3.0-realtime-plus、qwen-audio-3.0-realtime-flash Qwen-Audio端到端实时语音大模型兼顾语音推理能力与双工对话节奏，在保持流畅、自然的实时交互体验的同时，通过并行推理、全向流式等工程优化，有效控制端到端响应时延。[实时语音对话（Qwen-Audio-Realtime）](https://help.aliyun.com/zh/model-studio/qwen-audio-realtime-user-guides) @@ -58,18 +72,14 @@ Qwen-Audio端到端实时语音大模型兼顾语音推理能力与双工对话 2026-07-14 -中国内地 - qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash -Qwen-Audio-TTS语音合成模型上线，新增更多小语种和中文方言支持，增强了指令遵循与细粒度标签控制能力，音质和表现力全面提升。其中 Plus 版本面向高品质专业场景，Flash 版本面向低延迟实时交互场景，首包延时控制在 200ms 以内。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) +Qwen-Audio-TTS语音合成模型上线，新增更多小语种和中文方言支持，增强了指令遵循与细粒度标签控制能力，音质和表现力全面提升。其中 Plus 版本面向高品质专业场景，Flash 版本面向低延迟实时交互场景。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) 文生图/参考生图 2026-07-13 -中国内地 - vidu/vidu-image\_reference2image、vidu/viduq3-fast\_reference2image、vidu/viduq2-pro\_reference2image、vidu/viduq2-fast\_reference2image 由生数科技提供Vidu系列图片生成API服务，多图参考，精准还原，高速高质。 [Vidu-图像生成](https://help.aliyun.com/zh/model-studio/vidu-image-generation-api-reference) @@ -78,8 +88,6 @@ vidu/vidu-image\_reference2image、vidu/viduq3-fast\_reference2image、vidu/vidu 2026-07-13 -中国内地 - vidu/viduq3-ad\_reference2video、vidu/viduq3-drama\_reference2video Vidu-参考生视频系列模型（广告、短剧方向），支持传入参考图片和视频，生成对应场景的视频内容。[Vidu-参考生视频](https://help.aliyun.com/zh/model-studio/vidu-reference-to-video-api-reference) @@ -88,8 +96,6 @@ Vidu-参考生视频系列模型（广告、短剧方向），支持传入参考 2026-07-13 -中国内地 - vidu/viduq3-pro-fast\_img2video Vidu-图生视频模型，根据输入图像和文本提示词快速生成视频。[Vidu-图生视频-基于首帧](https://help.aliyun.com/zh/model-studio/vidu-image-to-video-api-reference) @@ -98,8 +104,6 @@ Vidu-图生视频模型，根据输入图像和文本提示词快速生成视频 2026-07-01 -中国内地 - wan2.7-r2v-2026-06-12 万相2.7参考生视频模型快照版本，支持主体参考和音色定制，并可输入单张多宫格故事板直接生成剧本化视频。[万相2.7-参考生视频](https://help.aliyun.com/zh/model-studio/wan-video-to-video-api-reference) @@ -108,8 +112,6 @@ wan2.7-r2v-2026-06-12 2026-06-25 -中国内地 - qwen-image-2.0-pro-2026-06-22 Qwen-Image-2.0系列模型最新快照，融合图片生成与编辑能力。相较于4月22日快照，文字渲染能力进一步增强，支持最长1k token的指令输入；真实质感与写实场景细节刻画更加细腻；语义遵循能力更强。[千问-文生图](https://help.aliyun.com/zh/model-studio/qwen-image-api)、[千问-图像编辑](https://help.aliyun.com/zh/model-studio/qwen-image-edit-api) @@ -118,8 +120,6 @@ Qwen-Image-2.0系列模型最新快照，融合图片生成与编辑能力。相 2026-06-22 -中国内地 - happyhorse-1.1-t2v HappyHorse 1.1系列文生视频模型，支持有声视频生成，可生成3~15秒、720P/1080P视频。[HappyHorse-文生视频](https://help.aliyun.com/zh/model-studio/happyhorse-text-to-video-api-reference) @@ -128,8 +128,6 @@ HappyHorse 1.1系列文生视频模型，支持有声视频生成，可生成3~1 2026-06-22 -中国内地 - happyhorse-1.1-i2v HappyHorse 1.1系列图生视频模型，支持有声视频生成，可生成3~15秒、720P/1080P视频。[HappyHorse-图生视频-基于首帧](https://help.aliyun.com/zh/model-studio/happyhorse-image-to-video-api-reference) @@ -138,8 +136,6 @@ HappyHorse 1.1系列图生视频模型，支持有声视频生成，可生成3~1 2026-06-22 -中国内地 - happyhorse-1.1-r2v HappyHorse 1.1系列参考生视频模型，支持多参考图输入生成有声视频，可生成3~15秒、720P/1080P视频。[HappyHorse-参考生视频](https://help.aliyun.com/zh/model-studio/happyhorse-reference-to-video-api-reference) @@ -148,8 +144,6 @@ HappyHorse 1.1系列参考生视频模型，支持多参考图输入生成有声 2026-06-18 -中国内地 - kimi/kimi-k2.7-code-highspeed 月之暗面直供的高速编程模型，与 kimi/kimi-k2.7-code 功能完全一致，速度提升5~6倍。[Kimi-月之暗面](https://help.aliyun.com/zh/model-studio/kimi-api-by-moonshot-ai) @@ -158,8 +152,6 @@ kimi/kimi-k2.7-code-highspeed 2026-06-16 -中国内地 - qwen3.5-ocr 千问文字提取模型，基于Qwen3.5架构，速度更快，效果更强。上下文长度扩展至128K，支持多轮对话。信息抽取能力大幅提升，覆盖多种国内外卡证。[文字提取](https://help.aliyun.com/zh/model-studio/qwen-vl-ocr) @@ -168,8 +160,6 @@ qwen3.5-ocr 2026-06-15 -中国内地 - kimi/kimi-k2.7-code 月之暗面直供的 Kimi K2.7 Code 模型，以编码为中心的智能体模型，专为长程软件工程任务优化，仅支持思考模式。擅长跨多文件重构、功能实现、长会话调试等复杂工作流。[Kimi-月之暗面](https://help.aliyun.com/zh/model-studio/kimi-api-by-moonshot-ai) @@ -178,8 +168,6 @@ kimi/kimi-k2.7-code 2026-06-10 -中国内地 - qwen3.7-max-2026-06-08 Qwen3.7系列中规模最大、综合能力最强的Max模型，相较于5月20日快照增加了视觉模态理解能力，能够感知真实世界场景，具备多模态交互混合智能体能力。 @@ -188,8 +176,6 @@ Qwen3.7系列中规模最大、综合能力最强的Max模型，相较于5月20 2026-06-08 -中国内地 - pixverse/pixverse-v6-r2v 爱诗V6-参考生视频模型，参考多张图像生成视频。相较于v5.6全面升级，通用场景推荐使用。支持将一张多宫格分镜拼接图一键转化为视频。[爱诗-参考生视频](https://help.aliyun.com/zh/model-studio/pixverse-reference-to-video-api-reference) @@ -198,8 +184,6 @@ pixverse/pixverse-v6-r2v 2026-06-01 -中国内地 - qwen3.7-plus、qwen3.7-plus-2026-05-26 千问3.7Plus系列，在强大文本能力的基础上全面升级了视觉-语言能力，同时保持了在编码、工具使用和生产力工作流方面的完整智能体能力。其核心特色为多模态交互混合智能体能力，能够感知真实世界场景、读取屏幕并操作 GUI、基于视觉参考生成代码、端到端导航移动应用。 @@ -208,8 +192,6 @@ qwen3.7-plus、qwen3.7-plus-2026-05-26 2026-05-29 -中国内地 - vanchin/deepseek-v4-pro 由快手万擎直供的 DeepSeek 模型推理服务。[DeepSeek-快手万擎](https://help.aliyun.com/zh/model-studio/deepseek-api-by-vanchin) @@ -218,8 +200,6 @@ vanchin/deepseek-v4-pro 2026-05-29 -中国内地 - stepfun/step-3.7-flash 阶跃星辰直供的 Step 3.7 Flash 模型，Flash 档位新一代旗舰。在搜索、Agent、编码、多模态四大方向全面升级，深度检索与多模态图搜能力大幅增强，Agent 与工具调用能力显著提升，编码能力在 Flash 档位多项基准中表现优异，多模态理解能力对标头部旗舰。 @@ -228,8 +208,6 @@ stepfun/step-3.7-flash 2026-05-25 -中国内地 - qwen3.7-max-preview、qwen3.7-max-2026-05-17 Qwen Max 系列模型。仅支持纯文本输入，仅支持思考模式。 @@ -238,8 +216,6 @@ Qwen Max 系列模型。仅支持纯文本输入，仅支持思考模式。 2026-05-21 -中国内地 - qwen3.7-max、qwen3.7-max-2026-05-20 Qwen Max 系列新一代旗舰模型。仅支持纯文本输入，默认开启思考模式，支持显式缓存，在编程、办公与生产力、长周期自主执行方面均能出色胜任各项任务。 @@ -248,8 +224,6 @@ Qwen Max 系列新一代旗舰模型。仅支持纯文本输入，默认开启 2026-05-19 -中国内地 - qwen3.5-livetranslate-flash-realtime、qwen3.5-livetranslate-flash-realtime-2026-05-19 一款多语言音视频实时翻译模型，可识别 60 种语言，并实时翻译为 29 种语言的音频。[实时语音/音视频翻译-千问](https://help.aliyun.com/zh/model-studio/qwen3-5-livetranslate-flash-realtime) @@ -258,8 +232,6 @@ qwen3.5-livetranslate-flash-realtime、qwen3.5-livetranslate-flash-realtime-2026 2026-05-19 -中国内地 - xiaomi/mimo-v2.5-pro 小米直供的 MiMo-V2.5-Pro 模型，在通用智能体能力、复杂软件工程以及长程任务等方面提升显著。[MiMo-小米](https://help.aliyun.com/zh/model-studio/mimo) @@ -268,8 +240,6 @@ xiaomi/mimo-v2.5-pro 2026-05-19 -中国内地 - ZHIPU/GLM-5.1、ZHIPU/GLM-5 智谱直供的 glm-5、glm-5.1模型，适用于智能交互、企业应用及开发辅助等场景。[GLM-智谱](https://help.aliyun.com/zh/model-studio/glm-zhipu) @@ -278,8 +248,6 @@ ZHIPU/GLM-5.1、ZHIPU/GLM-5 2026-05-06 -中国内地 - fun-music-v1 百聆音乐生成大模型（Fun音乐大模型）支持输入开放性歌曲的创作要求或歌词，生成整首男/女声演唱的中文或英文歌曲。歌曲通俗易懂，情绪由浅入深，是人类灵感与大模型能力的完美结合。[音乐生成](https://help.aliyun.com/zh/model-studio/fun-music/) @@ -288,8 +256,6 @@ fun-music-v1 2026-04-29 -中国内地 - deepseek-v4-pro deepseek-v4-pro `cached_token` 单价调整为 **1 元/百万 token**，标准 `input_token` 单价不变。详见[上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)。 @@ -298,8 +264,6 @@ deepseek-v4-pro `cached_token` 单价调整为 **1 元/百万 token**，标准 ` 2026-04-29 -中国内地 - vidu/viduq3-mix\_reference2video、vidu/viduq3\_reference2video、vidu/viduq3-turbo\_reference2video Vidu-参考生视频系列模型，支持传入参考图片和文本提示词，将图片中的主体角色融合到提示词描述的场景中，生成流畅的视频内容。[Vidu-参考生视频](https://help.aliyun.com/zh/model-studio/vidu-reference-to-video-api-reference) @@ -308,8 +272,6 @@ Vidu-参考生视频系列模型，支持传入参考图片和文本提示词， 2026-04-29 -中国内地 - vidu/viduq2-pro-fast\_img2video Vidu-首帧生视频系列模型，根据输入图像和文本提示词，生成一段流畅的视频。[Vidu-图生视频-基于首帧](https://help.aliyun.com/zh/model-studio/vidu-image-to-video-api-reference) @@ -318,8 +280,6 @@ Vidu-首帧生视频系列模型，根据输入图像和文本提示词，生成 2026-04-29 -中国内地 - Tripo/Tripo-H3.1 Tripo 高精度3D生成模型，支持最高200万面，可实现文生3D、单图生3D和多图生3D任务。[Tripo-3D模型生成](https://help.aliyun.com/zh/model-studio/tripo-3d-generation-api-reference) @@ -328,8 +288,6 @@ Tripo 高精度3D生成模型，支持最高200万面，可实现文生3D、单 2026-04-29 -中国内地 - Tripo/Tripo-P1.0 Tripo 专业3D生成模型，最高2万面，生成速度较快，可实现文生3D、单图生3D和多图生3D任务。[Tripo-3D模型生成](https://help.aliyun.com/zh/model-studio/tripo-3d-generation-api-reference) @@ -338,8 +296,6 @@ Tripo 专业3D生成模型，最高2万面，生成速度较快，可实现文 2026-04-27 -中国内地 - happyhorse-1.0-t2v HappyHorse系列文生视频模型，支持有声视频生成，可生成3~15秒、720P/1080P视频。[HappyHorse-文生视频](https://help.aliyun.com/zh/model-studio/happyhorse-text-to-video-api-reference) @@ -348,8 +304,6 @@ HappyHorse系列文生视频模型，支持有声视频生成，可生成3~15秒 2026-04-27 -中国内地 - happyhorse-1.0-i2v HappyHorse系列图生视频模型，支持有声视频生成，可生成3~15秒、720P/1080P视频。[HappyHorse-图生视频-基于首帧](https://help.aliyun.com/zh/model-studio/happyhorse-image-to-video-api-reference) @@ -358,8 +312,6 @@ HappyHorse系列图生视频模型，支持有声视频生成，可生成3~15秒 2026-04-27 -中国内地 - happyhorse-1.0-r2v HappyHorse系列参考生视频模型，支持多参考图输入生成有声视频，可生成3~15秒、720P/1080P视频。[HappyHorse-参考生视频](https://help.aliyun.com/zh/model-studio/happyhorse-reference-to-video-api-reference) @@ -368,8 +320,6 @@ HappyHorse系列参考生视频模型，支持多参考图输入生成有声视 2026-04-27 -中国内地 - happyhorse-1.0-video-edit HappyHorse系列视频编辑模型，支持对视频进行编辑处理。[HappyHorse-视频编辑](https://help.aliyun.com/zh/model-studio/happyhorse-video-edit-api-reference) @@ -378,8 +328,6 @@ HappyHorse系列视频编辑模型，支持对视频进行编辑处理。[HappyH 2026-07-01 -中国内地 - wan2.7-t2v-2026-06-12 万相2.7-文生视频模型快照版本，模型能力与wan2.7-t2v一致。[万相2.7-文生视频](https://help.aliyun.com/zh/model-studio/text-to-video-api-reference) @@ -388,8 +336,6 @@ wan2.7-t2v-2026-06-12 2026-04-26 -中国内地 - wan2.7-t2v-2026-04-25 万相2.7-文生视频模型快照版本，模型能力与wan2.7-t2v一致。[万相2.7-文生视频](https://help.aliyun.com/zh/model-studio/text-to-video-api-reference) @@ -398,8 +344,6 @@ wan2.7-t2v-2026-04-25 2026-04-26 -中国内地 - wan2.7-i2v-2026-04-25 万相2.7-图生视频模型快照版本，模型能力与wan2.7-i2v一致。[万相2.7-图生视频](https://help.aliyun.com/zh/model-studio/image-to-video-general-api-reference) @@ -408,8 +352,6 @@ wan2.7-i2v-2026-04-25 2026-04-26 -中国内地 - kimi/kimi-k2.6 kimi/kimi-k2.6 是 Kimi 最新最智能的模型，月之暗面直供，在通用 Agent、代码、视觉理解等综合能力得到全面提升，同时支持文本、图片与视频输入、思考与非思考模式、对话与 Agent 任务。[Kimi-月之暗面](https://help.aliyun.com/zh/model-studio/kimi-api-by-moonshot-ai) @@ -418,8 +360,6 @@ kimi/kimi-k2.6 是 Kimi 最新最智能的模型，月之暗面直供，在通 2026-04-24 -中国内地 - deepseek-v4-pro、deepseek-v4-flash DeepSeek-V4系列模型，阿里直供。deepseek-v4-pro为旗舰模型，deepseek-v4-flash为轻量级高速模型。[DeepSeek-阿里云](https://help.aliyun.com/zh/model-studio/deepseek-api) @@ -428,8 +368,6 @@ DeepSeek-V4系列模型，阿里直供。deepseek-v4-pro为旗舰模型，deepse 2026-04-23 -中国内地 - qwen-image-2.0-pro-2026-04-22 Qwen-Image-2.0系列模型，实现了图片生成和图片编辑的融合。相较于3月3日快照，本模型在画面质感，尤其是纹理细节、光影、材质上有明显跃升；支持多语言的图内文字生成；艺术风格表现更加均衡。[千问-文生图](https://help.aliyun.com/zh/model-studio/qwen-image-api)、[千问-图像编辑](https://help.aliyun.com/zh/model-studio/qwen-image-edit-api) @@ -438,8 +376,6 @@ Qwen-Image-2.0系列模型，实现了图片生成和图片编辑的融合。相 2026-04-23 -中国内地 - qwen3.5-plus-2026-04-20 Qwen3.5原生视觉语言系列Plus模型新快照，相较2月15日快照Agentic coding能力大幅提升，推理速度显著提升；知识、推理与长上下文能力保持较高水准，适合编码智能体、生产工作流和高吞吐场景。 @@ -448,8 +384,6 @@ Qwen3.5原生视觉语言系列Plus模型新快照，相较2月15日快照Agenti 2026-04-23 -中国内地 - qwen3.6-27b Qwen3.6系列27B原生视觉语言Dense模型，相较3.5-27B重点提升Agentic coding能力，STEM与推理能力进一步增强；视觉模态方面空间智能、物体定位与检测能力显著增强，视频理解、文档OCR及视觉Agent能力稳步提升。 @@ -458,8 +392,6 @@ Qwen3.6系列27B原生视觉语言Dense模型，相较3.5-27B重点提升Agentic 2026-04-21 -中国内地 - kimi-k2.6 Kimi最新最智能的模型，具备更强更稳的长程代码编写能力，指令遵循和自我纠错能力显著提升。同时支持文本、图片与视频输入、思考与非思考模式、对话与 Agent 任务。[Kimi-阿里云](https://help.aliyun.com/zh/model-studio/kimi-api) @@ -468,8 +400,6 @@ Kimi最新最智能的模型，具备更强更稳的长程代码编写能力， 2026-04-20 -中国内地 - qwen3.6-max-preview Qwen3.6系列规模最大的闭源模型，Coding能力进一步提升、Agent执行更加高效。仅支持纯文本输入，支持思考模式（默认开启），支持显式缓存和Function Calling。[概述](https://help.aliyun.com/zh/model-studio/text-generation) @@ -480,8 +410,6 @@ Qwen3.6系列规模最大的闭源模型，Coding能力进一步提升、Agent 2026-04-16 -中国内地 - qwen3.6-flash、qwen3.6-flash-2026-04-16、qwen3.6-35b-a3b Qwen3.6 原生视觉语言 Flash 系列模型，在整体性能上较 Qwen3.5-Flash 显著提升。重点增强了智能体编程能力（在多项代码智能体基准上大幅超越前代）、数学推理和代码推理能力；在视觉能力方面，空间智能显著增强，其中物体定位和目标检测表现尤为突出。[概述](https://help.aliyun.com/zh/model-studio/text-generation) @@ -490,8 +418,6 @@ Qwen3.6 原生视觉语言 Flash 系列模型，在整体性能上较 Qwen3.5-Fl 2026-04-16 -中国内地 - fun-asr、fun-asr-2025-11-07 Fun-ASR 实时语音识别能力升级： @@ -511,8 +437,6 @@ Fun-ASR 实时语音识别能力升级： 2026-04-14 -中国内地 - glm-5.1 智谱GLM-5.1模型，专为长程任务设计，支持 200K 上下文，最大输出可达 128K Token。通过强大的逻辑推理、长文本理解及代码生成能力，在多项基准测试中表现优异，适用于智能交互、企业应用及开发辅助等场景。[GLM-阿里云](https://help.aliyun.com/zh/model-studio/glm) @@ -521,8 +445,6 @@ glm-5.1 2026-04-14 -中国内地 - vanchin/deepseek-v3.2-think、vanchin/deepseek-v3.1-terminus、vanchin/deepseek-r1、vanchin/deepseek-v3 由快手万擎直供的 DeepSeek 模型推理服务。[DeepSeek-快手万擎](https://help.aliyun.com/zh/model-studio/deepseek-api-by-vanchin) @@ -531,8 +453,6 @@ vanchin/deepseek-v3.2-think、vanchin/deepseek-v3.1-terminus、vanchin/deepseek- 2026-04-14 -中国内地 - vanchin/deepseek-ocr 由快手万擎直供的 DeepSeek OCR 模型推理服务。[DeepSeek-快手万擎](https://help.aliyun.com/zh/model-studio/deepseek-api-by-vanchin) @@ -541,8 +461,6 @@ vanchin/deepseek-ocr 2026-04-13 -中国内地 - pixverse/pixverse-c1-t2v 爱诗C1-文生视频模型，基于文本提示词生成视频。支持提示词智能分镜，打斗与特效强化升级，支持真人或动漫风格。[爱诗-文生视频](https://help.aliyun.com/zh/model-studio/pixverse-text-to-video-api-reference) @@ -551,8 +469,6 @@ pixverse/pixverse-c1-t2v 2026-04-13 -中国内地 - pixverse/pixverse-c1-it2v 爱诗C1-首帧生视频模型，根据输入图像和文本提示词生成视频。支持提示词智能分镜，打斗与特效强化升级，支持真人或动漫风格。[爱诗-图生视频-基于首帧](https://help.aliyun.com/zh/model-studio/pixverse-image-to-video-api-reference) @@ -561,8 +477,6 @@ pixverse/pixverse-c1-it2v 2026-04-13 -中国内地 - pixverse/pixverse-c1-kf2v 爱诗C1-首尾帧生视频模型，基于首帧图像、尾帧图像和文本提示词生成平滑过渡的视频。支持提示词智能分镜，打斗与特效强化升级，支持真人或动漫风格。[爱诗-图生视频-基于首尾帧](https://help.aliyun.com/zh/model-studio/pixverse-keyframe-to-video-api-reference) @@ -571,8 +485,6 @@ pixverse/pixverse-c1-kf2v 2026-04-13 -中国内地 - pixverse/pixverse-c1-r2v 爱诗C1-参考生视频模型，参考多张图像生成视频。支持将一张多宫格分镜拼接图一键转化为视频。[爱诗-参考生视频](https://help.aliyun.com/zh/model-studio/pixverse-reference-to-video-api-reference) @@ -581,8 +493,6 @@ pixverse/pixverse-c1-r2v 2026-04-03 -中国内地 - wan2.7-videoedit 万相2.7-视频编辑模型，支持指令编辑与视频迁移任务，可修改视频局部或整体画面，同时支持多图参考替换及动作、特效、运镜的复刻。[万相2.7-视频编辑](https://help.aliyun.com/zh/model-studio/wan-video-editing-api-reference) @@ -591,8 +501,6 @@ wan2.7-videoedit 2026-04-03 -中国内地 - wan2.7-i2v 万相2.7-图生视频模型，支持多模态输入（文本/图像/音频/视频），可完成首帧生视频、首尾帧生视频、视频续写三大任务。[万相2.7-图生视频](https://help.aliyun.com/zh/model-studio/image-to-video-general-api-reference) @@ -601,8 +509,6 @@ wan2.7-i2v 2026-04-03 -中国内地 - wan2.7-t2v 万相2.7-文生视频模型，新增分辨率档位与宽高比自定义设置，支持灵活适配不同创作场景与平台发布需求。[万相2.7-文生视频](https://help.aliyun.com/zh/model-studio/text-to-video-api-reference) @@ -611,8 +517,6 @@ wan2.7-t2v 2026-04-03 -中国内地 - wan2.7-r2v 万相2.7参考生视频模型，支持主体参考和音色定制，并可输入单张多宫格故事板直接生成剧本化视频。[万相2.7-参考生视频](https://help.aliyun.com/zh/model-studio/wan-video-to-video-api-reference) @@ -621,8 +525,6 @@ wan2.7-r2v 2026-04-02 -中国内地 - qwen3.6-plus、qwen3.6-plus-2026-04-02 千问3.6-Plus，代码开发能力重点升级（Agentic Coding、前端编程等），Vibe Coding体验显著提升；泛化场景推理能力进一步增强；多模态方面，万物识别、OCR、物体定位等能力显著提升；同时修复了Qwen3.5-Plus上线后的已知问题。使用方法与qwen3.5-plus一致。[概述](https://help.aliyun.com/zh/model-studio/text-generation) @@ -631,8 +533,6 @@ qwen3.6-plus、qwen3.6-plus-2026-04-02 2026-04-01 -中国内地 - wan2.7-image-pro、wan2.7-image 万相2.7-图像生成与编辑模型，支持文生图、文生组图、图生组图、图像编辑、多图参考生成、交互式编辑，在文字渲染、主体一致性、复杂指令遵循表现更优。Pro系列支持4K输出；加速版兼顾效果与响应速度。[万相-图像生成与编辑2.7](https://help.aliyun.com/zh/model-studio/wan-image-generation-and-editing-api-reference) @@ -641,8 +541,6 @@ wan2.7-image-pro、wan2.7-image 2026-03-30 -中国内地 - qwen3.5-omni-plus、qwen3.5-omni-plus-2026-03-15、qwen3.5-omni-flash、qwen3.5-omni-flash-2026-03-15 最新一代全模态大模型，支持长视频分析、会议纪要、字幕输出、安全审核、音视频交互；支持音视频内容的深度理解与生成描述，支持 113 种语言识别和 36 种语言的音频生成，可处理 3 小时音频及1 小时视频输入，支持联网搜索及指令来控制输出音频的音量、语速、情绪。[非实时（Qwen-Omni）](https://help.aliyun.com/zh/model-studio/qwen-omni) @@ -651,8 +549,6 @@ qwen3.5-omni-plus、qwen3.5-omni-plus-2026-03-15、qwen3.5-omni-flash、qwen3.5- 2026-03-30 -中国内地 - qwen3.5-omni-plus-realtime、qwen3.5-omni-plus-realtime-2026-03-15、qwen3.5-omni-flash-realtime、qwen3.5-omni-flash-realtime-2026-03-15 千问最新推出的实时多模态模型，相比于上一代的 Qwen3-Omni-Flash-Realtime：模型智力大幅提升，与 Qwen3.5-Plus 智能水平相当。原生支持联网搜索（WebSearch），支持语音打断和控制；支持 113 种语种和方言的语音识别，以及 36 种语种和方言的语音生成。[实时（Qwen-Omni-Realtime）](https://help.aliyun.com/zh/model-studio/realtime) @@ -661,8 +557,6 @@ qwen3.5-omni-plus-realtime、qwen3.5-omni-plus-realtime-2026-03-15、qwen3.5-omn 2026-03-29 -中国内地 - pixverse/pixverse-v6-t2v 爱诗v6-文生视频模型，基于文本提示词，生成一段流畅的视频。[爱诗-文生视频](https://help.aliyun.com/zh/model-studio/pixverse-text-to-video-api-reference) @@ -671,8 +565,6 @@ pixverse/pixverse-v6-t2v 2026-03-29 -中国内地 - pixverse/pixverse-v6-it2v 爱诗v6-首帧生视频模型，根据输入图像和文本提示词，生成一段流畅的视频。[爱诗-图生视频-基于首帧](https://help.aliyun.com/zh/model-studio/pixverse-image-to-video-api-reference) @@ -681,8 +573,6 @@ pixverse/pixverse-v6-it2v 2026-03-29 -中国内地 - pixverse/pixverse-v6-kf2v 爱诗v6-首尾帧生视频模型，基于首帧图像、尾帧图像和文本提示词，生成一段平滑过渡的视频。[爱诗-图生视频-基于首尾帧](https://help.aliyun.com/zh/model-studio/pixverse-keyframe-to-video-api-reference) @@ -691,8 +581,6 @@ pixverse/pixverse-v6-kf2v 2026-03-27 -中国内地 - vidu/viduq3-pro\_text2video、vidu/viduq3-turbo\_text2video、vidu/viduq2\_text2video Vidu-文生视频系列模型。基于文本提示词，生成一段流畅的视频。[Vidu-文生视频](https://help.aliyun.com/zh/model-studio/vidu-text-to-video-api-reference) @@ -701,8 +589,6 @@ Vidu-文生视频系列模型。基于文本提示词，生成一段流畅的视 2026-03-27 -中国内地 - vidu/viduq3-pro\_img2video、vidu/viduq3-turbo\_img2video、vidu/viduq2-pro\_img2video、vidu/viduq2-turbo\_img2video Vidu-首帧生视频系列模型，根据输入图像和文本提示词，生成一段流畅的视频。[Vidu-图生视频-基于首帧](https://help.aliyun.com/zh/model-studio/vidu-image-to-video-api-reference) @@ -711,8 +597,6 @@ Vidu-首帧生视频系列模型，根据输入图像和文本提示词，生成 2026-03-27 -中国内地 - vidu/viduq3-pro\_start-end2video、vidu/viduq3-turbo\_start-end2video、vidu/viduq2-pro\_start-end2video、vidu/viduq2-turbo\_start-end2video Vidu-首尾帧生视频系列模型，基于首帧图像、尾帧图像和文本提示词，生成一段平滑过渡的视频。[Vidu-图生视频-基于首尾帧](https://help.aliyun.com/zh/model-studio/vidu-keyframe-to-video-api-reference) @@ -721,8 +605,6 @@ Vidu-首尾帧生视频系列模型，基于首帧图像、尾帧图像和文本 2026-03-27 -中国内地 - vidu/viduq2-pro\_reference2video、vidu/viduq2\_reference2video Vidu-参考生视频系列模型，支持传入参考图片和文本提示词，将图片中的主体角色融合到提示词描述的场景中，生成流畅的视频内容。[Vidu-参考生视频](https://help.aliyun.com/zh/model-studio/vidu-reference-to-video-api-reference) @@ -731,8 +613,6 @@ Vidu-参考生视频系列模型，支持传入参考图片和文本提示词， 2026-03-24 -中国内地 - kling/kling-v3-image-generation、kling/kling-v3-omni-image-generation 可灵V3图像生成系列模型，支持文生图、参考图生图两种任务。[可灵-图像生成](https://help.aliyun.com/zh/model-studio/kling-image-generation-api-reference) @@ -741,8 +621,6 @@ kling/kling-v3-image-generation、kling/kling-v3-omni-image-generation 2026-03-24 -中国内地 - kling/kling-v3-video-generation、kling/kling-v3-omni-video-generation 可灵V3视频生成系列模型，支持文生视频、图生视频-基于首帧、图生视频-基于首尾帧、参考生视频以及视频编辑。[可灵-视频生成](https://help.aliyun.com/zh/model-studio/kling-video-generation-api-reference/) @@ -751,8 +629,6 @@ kling/kling-v3-video-generation、kling/kling-v3-omni-video-generation 2026-03-20 -中国内地 - pixverse/pixverse-v5.6-r2v 参考多张图像生视频。[爱诗-参考生视频](https://help.aliyun.com/zh/model-studio/pixverse-reference-to-video-api-reference) @@ -761,8 +637,6 @@ pixverse/pixverse-v5.6-r2v 2026-03-20 -中国内地 - pixverse/pixverse-v5.6-kf2v 基于首帧和尾帧生视频。[爱诗-图生视频-基于首尾帧](https://help.aliyun.com/zh/model-studio/pixverse-keyframe-to-video-api-reference) @@ -771,8 +645,6 @@ pixverse/pixverse-v5.6-kf2v 2026-03-20 -中国内地 - pixverse/pixverse-v5.6-it2v 基于首帧生视频。[爱诗-图生视频-基于首帧](https://help.aliyun.com/zh/model-studio/pixverse-image-to-video-api-reference) @@ -781,8 +653,6 @@ pixverse/pixverse-v5.6-it2v 2026-03-20 -中国内地 - pixverse/pixverse-v5.6-t2v 文生视频系列模型。[爱诗-文生视频](https://help.aliyun.com/zh/model-studio/pixverse-text-to-video-api-reference) @@ -791,8 +661,6 @@ pixverse/pixverse-v5.6-t2v 2026-03-19 -中国内地 - tongyi-embedding-vision-plus-2026-03-06、tongyi-embedding-vision-flash-2026-03-06 相较于 tongyi-embedding-vision 主线模型，tongyi-embedding-vision-plus-2026-03-06 / flash-2026-03-06 视觉向量化能力全面升级，基于 Qwen3 底座，支持多分辨率（0-3 档可调）、多维度（Plus 64-1152 维 / Flash 64-768 维）、30+ 语言，以及文本/图片/视频融合向量化，IPC/自驾/电商/视频营销等场景图像搜索效果大幅提升。 @@ -801,8 +669,6 @@ tongyi-embedding-vision-plus-2026-03-06、tongyi-embedding-vision-flash-2026-03- 2026-03-18 -中国内地 - gui-plus-2026-02-26 相较于`gui-plus`，`gui-plus-2026-02-26`模型能力全面升级，支持思考与非思考模式，模型在处理跨平台、多 APP 任务的效果上得到大幅提升。[界面交互](https://help.aliyun.com/zh/model-studio/gui-automation) @@ -811,8 +677,6 @@ gui-plus-2026-02-26 2026-03-05 -中国内地 - fun-asr-realtime-2026-02-28 Fun-ASR实时语音识别新增快照模型，较 fun-asr-realtime-2025-11-07 效果更优。[实时语音识别-Fun-ASR/Paraformer](https://help.aliyun.com/zh/model-studio/real-time-speech-recognition) @@ -821,8 +685,6 @@ Fun-ASR实时语音识别新增快照模型，较 fun-asr-realtime-2025-11-07  2026-03-03 -中国内地 - qwen-image-2.0、qwen-image-2.0-2026-03-03、qwen-image-2.0-pro、qwen-image-2.0-pro-2026-03-03 千问-Image2.0系列，同时支持图像生成和编辑。Pro系列文字渲染、真实质感、语义遵循能力更强；加速版兼顾效果与响应速度。[千问-文生图](https://help.aliyun.com/zh/model-studio/qwen-image-api)、[千问-图像编辑](https://help.aliyun.com/zh/model-studio/qwen-image-edit-api) @@ -831,8 +693,6 @@ qwen-image-2.0、qwen-image-2.0-2026-03-03、qwen-image-2.0-pro、qwen-image-2.0 2026-03-03 -中国内地 - qwen3-asr-flash-2026-02-10 千问录音文件识别新增快照模型，较 qwen3-asr-flash-2025-09-08 效果更优。[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide) @@ -841,8 +701,6 @@ qwen3-asr-flash-2026-02-10 2026-03-02 -中国内地 - cosyvoice-v3.5-plus、cosyvoice-v3.5-flash CosyVoice3.5 模型上线，专注声音复刻与设计，支持指令控制语音合成效果。[实时语音合成-CosyVoice /Sambert](https://help.aliyun.com/zh/model-studio/text-to-speech) @@ -851,8 +709,6 @@ CosyVoice3.5 模型上线，专注声音复刻与设计，支持指令控制语 2026-02-24 -中国内地 - MiniMax-M2.5 稀宇科技（MiniMax）推出的新模型，响应速度快，擅长编程、办公等任务。[使用方法](https://help.aliyun.com/zh/model-studio/minimax-api) @@ -861,8 +717,6 @@ MiniMax-M2.5 2026-02-24 -中国内地 - qwen3.5-flash、qwen3.5-flash-2026-02-23、qwen3.5-122b-a10b、qwen3.5-27b、qwen3.5-35b-a3b 阿里巴巴推出的最新模型千问3.5-Flash和开源模型，支持文本、图像和视频输入，响应速度快，综合表现接近qwen3.5-plus，支持内置[工具调用](https://help.aliyun.com/zh/model-studio/tool-calls/)。[概述](https://help.aliyun.com/zh/model-studio/text-generation) @@ -871,8 +725,6 @@ qwen3.5-flash、qwen3.5-flash-2026-02-23、qwen3.5-122b-a10b、qwen3.5-27b、qwe 2026-02-20 -中国内地 - qwen3-coder-next Qwen3系列新一代开源代码生成模型，支持多轮工具交互，提升了对仓库级别代码的理解能力和对AI编程工具的适配性。[代码能力（Qwen-Coder）](https://help.aliyun.com/zh/model-studio/qwen-coder) @@ -881,8 +733,6 @@ Qwen3系列新一代开源代码生成模型，支持多轮工具交互，提升 2026-02-18 -中国内地 - glm-5 智谱推出的最新模型，专为编程与智能体场景打造，擅长复杂的系统工程与长程Agent任务。[GLM-阿里云](https://help.aliyun.com/zh/model-studio/glm) @@ -891,8 +741,6 @@ glm-5 2026-02-16 -中国内地 - qwen3.5-plus、qwen3.5-plus-2026-02-15、qwen3.5-397b-a17b 阿里巴巴推出的最新模型千问3.5-Plus和开源模型，支持文本、图像和视频输入，在语言理解、逻辑推理、代码生成、智能体任务、图像理解、视频理解、图形用户界面（GUI）等多种任务中表现卓越，支持内置[工具调用](https://help.aliyun.com/zh/model-studio/tool-calls/)。[概述](https://help.aliyun.com/zh/model-studio/text-generation) @@ -901,8 +749,6 @@ qwen3.5-plus、qwen3.5-plus-2026-02-15、qwen3.5-397b-a17b 2026-02-13 -中国内地 - qwen3-asr-flash-realtime-2026-02-10 千问实时语音识别新增最新快照模型，较 qwen3-asr-flash-realtime-2025-10-27 效果更优。[实时语音识别](https://help.aliyun.com/zh/model-studio/real-time-speech-recognition-user-guide) @@ -911,8 +757,6 @@ qwen3-asr-flash-realtime-2026-02-10 2026-02-12 -中国内地 - fun-asr-flash-8k-realtime、fun-asr-flash-8k-realtime-2026-01-28 新增基于Fun-ASR大模型架构的小尺寸ASR模型，专为8kHz场景优化，适合对成本敏感的客户。[实时语音识别-Fun-ASR/Paraformer](https://help.aliyun.com/zh/model-studio/real-time-speech-recognition) @@ -921,8 +765,6 @@ fun-asr-flash-8k-realtime、fun-asr-flash-8k-realtime-2026-01-28 2026-02-10 -中国内地 - qwen3-tts-instruct-flash、qwen3-tts-instruct-flash-2026-01-26 千问语音合成上线Instruct（指令控制）模型，支持通过自然语言指令精准控制合成效果。[非实时语音合成](https://help.aliyun.com/zh/model-studio/non-realtime-tts-user-guide) @@ -931,8 +773,6 @@ qwen3-tts-instruct-flash、qwen3-tts-instruct-flash-2026-01-26 2026-02-10 -中国内地 - qwen3-tts-vd-2026-01-26 千问语音合成上线声音设计模型，可通过文本描述创建定制化音色。[非实时语音合成](https://help.aliyun.com/zh/model-studio/non-realtime-tts-user-guide) @@ -941,8 +781,6 @@ qwen3-tts-vd-2026-01-26 2026-02-10 -中国内地 - qwen3-tts-vc-2026-01-22 千问语音合成上线声音复刻模型，可基于真实音频样本快速复刻音色。[非实时语音合成](https://help.aliyun.com/zh/model-studio/non-realtime-tts-user-guide) @@ -951,8 +789,6 @@ qwen3-tts-vc-2026-01-22 2026-02-04 -中国内地 - qwen3-tts-instruct-flash-realtime、qwen3-tts-instruct-flash-realtime-2026-01-22 千问实时语音合成新增Instruct（指令控制）模型，支持通过自然语言指令精准控制合成效果。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) @@ -961,8 +797,6 @@ qwen3-tts-instruct-flash-realtime、qwen3-tts-instruct-flash-realtime-2026-01-22 2026-02-02 -中国内地 - wan2.6-r2v-flash 基于参考视频和图像的角色形象，生成多镜头视频，支持自动配音。[万相2.7-参考生视频](https://help.aliyun.com/zh/model-studio/wan-video-to-video-api-reference) @@ -971,8 +805,6 @@ wan2.6-r2v-flash 2026-01-30 -中国内地 - kimi-k2.5 由月之暗面（Moonshot AI）公司推出的视觉理解模型，在代码生成、视觉理解等通用智能任务中表现突出。同时支持图像、视频与文本输入、对话与 Agent 任务。[Kimi-阿里云](https://help.aliyun.com/zh/model-studio/kimi-api) @@ -981,8 +813,6 @@ kimi-k2.5 2026-01-29 -中国内地 - tongyi-xiaomi-analysis-flash、tongyi-xiaomi-analysis-pro 通义晓蜜对话分析专注于对话信息抽取、场景分类、满意度判定等分析需求，擅长处理复杂业务逻辑的质检规则，支持自定义分析标准，具备强大的多轮对话理解和语义推理能力。[对话分析（Tongyi-Xiaomi-Analysis）](https://help.aliyun.com/zh/model-studio/dialogue-analysis) @@ -991,8 +821,6 @@ tongyi-xiaomi-analysis-flash、tongyi-xiaomi-analysis-pro 2026-01-28 -中国内地 - qwen3-asr-flash-filetrans、qwen3-asr-flash-filetrans-2025-11-17 千问3-ASR-Flash-Filetrans系列模型现已支持词级别时间戳，通过设置新参数 `enable_words`，获取毫秒级的词/字对齐信息，并体验更符合语义的精细化断句。[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide) @@ -1001,8 +829,6 @@ qwen3-asr-flash-filetrans、qwen3-asr-flash-filetrans-2025-11-17 2026-01-27 -中国内地 - qwen3-max-2026-01-23 相较于 2025 年 9 月 23 日的快照版本，有效融合了思考模式与非思考模式，显著提升了模型的整体性能。在思考模式下，模型集成了 Web 搜索、网页信息提取和代码解释器三项工具，通过在思考过程中引入外部工具，在复杂问题上实现更高的准确率。[OpenAI兼容-Responses](https://help.aliyun.com/zh/model-studio/compatibility-with-openai-responses-api) @@ -1011,8 +837,6 @@ qwen3-max-2026-01-23 2026-01-27 -中国内地 - MiniMax-M2.1 MiniMax-M2.1是MiniMax推出的旗舰级开源大模型，聚焦真实世界复杂任务，其核心优势在于多语言编程能力以及作为Agent解决复杂任务的能力。[使用方法](https://help.aliyun.com/zh/model-studio/minimax-api) @@ -1021,8 +845,6 @@ MiniMax-M2.1是MiniMax推出的旗舰级开源大模型，聚焦真实世界复 2026-01-23 -中国内地 - qwen3-vl-flash-2026-01-22 千问VL的全新快照版模型，有效融合了思考模式与非思考模式，相较于 2025 年 10 月 15 日的快照版本，显著提升了模型的整体性能，在通用视觉识别、安防、巡店、巡检、拍照解题等业务场景中实现了更高准确率的推理。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1031,8 +853,6 @@ qwen3-vl-flash-2026-01-22 2026-01-17 -中国内地 - wan2.6-i2v-flash 支持生成有声与无声视频，两类视频按各自计费规则独立计费；同时具备多镜头叙事能力与音频处理能力。[万相-图生视频-基于首帧（2.1-2.6）](https://help.aliyun.com/zh/model-studio/legacy-image-to-video-api-reference/) @@ -1041,8 +861,6 @@ wan2.6-i2v-flash 2026-01-17 -中国内地 - qwen-image-edit-max、qwen-image-edit-max-2026-01-16 千问图像编辑模型Max系列，具备更稳定、丰富的编辑能力，增强了工业设计与几何推理能力，并提升了角色一致性与编辑的精准度。[图像编辑-千问](https://help.aliyun.com/zh/model-studio/qwen-image-edit-guide) @@ -1051,8 +869,6 @@ qwen-image-edit-max、qwen-image-edit-max-2026-01-16 2026-01-16 -中国内地 - qwen3-tts-vc-realtime-2026-01-15 千问实时语音合成新增最新快照模型，[声音复刻（Qwen）](https://help.aliyun.com/zh/model-studio/qwen-tts-voice-cloning)效果进一步优化，较 qwen3-tts-vc-realtime-2025-11-27 更自然、更贴近原声。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) @@ -1061,8 +877,6 @@ qwen3-tts-vc-realtime-2026-01-15 2026-01-12 -中国内地 - qwen-image-plus-2026-01-09 千问图像生成的全新快照版模型，为qwen-image-max的蒸馏加速版，支持快速生成高质量图像。[千问-文生图](https://help.aliyun.com/zh/model-studio/qwen-image-api) @@ -1071,8 +885,6 @@ qwen-image-plus-2026-01-09 2026-01-12 -中国内地 - deepseek-v3.2 deepseek-v3.2 模型支持隐式缓存与显式缓存，可提升响应速度，并在不影响回复效果的前提下降低使用成本。[上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache) @@ -1081,8 +893,6 @@ deepseek-v3.2 模型支持隐式缓存与显式缓存，可提升响应速度， 2026-01-06 -中国内地 - qwen3-asr-flash、qwen3-asr-flash-2025-09-08 千问3-ASR-Flash支持OpenAI兼容模式。[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide) @@ -1091,8 +901,6 @@ qwen3-asr-flash、qwen3-asr-flash-2025-09-08 2026-01-05 -中国内地 - cosyvoice-v3-flash 语音合成CosyVoice新增24个音色（详情请参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)）： @@ -1116,8 +924,6 @@ cosyvoice-v3-flash 2025-12-31 -中国内地 - qwen-image-max、qwen-image-max-2025-12-30 千问图像生成模型Max系列，相较于Plus系列提升了图像的真实感与自然度，有效降低了AI合成痕迹，在人物质感、纹理细节和文字渲染等方面表现突出。[千问-文生图](https://help.aliyun.com/zh/model-studio/qwen-image-api) @@ -1126,8 +932,6 @@ qwen-image-max、qwen-image-max-2025-12-30 2025-12-29 -中国内地 - glm-4.7 智谱最新旗舰模型，编程能力更强，多步骤推理与执行更稳定。模型总参数达355B，支持复杂任务的分步规划、代码生成与工具协同；问答流畅，写作沉浸感强，创意角色扮演表现突出。[GLM-阿里云](https://help.aliyun.com/zh/model-studio/glm) @@ -1136,8 +940,6 @@ glm-4.7 2025-12-23 -中国内地 - qwen-image-edit-plus-2025-12-15 千问图像编辑发布的最新快照模型，相较于上一版本提升了角色一致性、工业设计能力和几何推理能力，并优化了编辑后的图片与原图在空间布局、纹理和风格上的匹配度，编辑效果更精准。[图像编辑-千问](https://help.aliyun.com/zh/model-studio/qwen-image-edit-guide) @@ -1146,8 +948,6 @@ qwen-image-edit-plus-2025-12-15 2025-12-19 -中国内地 - z-image-turbo 轻量级文生图模型，可快速生成高质量图像，支持中英双语渲染、复杂语义理解和多风格题材，并可灵活适配多种分辨率与宽高比。[文生图Z-Image](https://help.aliyun.com/zh/model-studio/z-image-api-reference) @@ -1156,8 +956,6 @@ z-image-turbo 2025-12-19 -中国内地 - qwen3-vl-plus-2025-12-19 千问VL的全新快照版模型，指令遵循能力更强，具有更低的延迟。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1166,8 +964,6 @@ qwen3-vl-plus-2025-12-19 2025-12-19 -中国内地 - qwen3-asr-flash-filetrans、qwen3-asr-flash-filetrans-2025-11-17、qwen3-asr-flash、qwen3-asr-flash-2025-09-08 新增捷克语、丹麦语等共 9 种语言的语音识别支持。[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide) @@ -1176,8 +972,6 @@ qwen3-asr-flash-filetrans、qwen3-asr-flash-filetrans-2025-11-17、qwen3-asr-fla 2025-12-17 -中国内地 - qwen3-asr-flash-realtime、qwen3-asr-flash-realtime-2025-10-27 新增捷克语、丹麦语等共 9 种语言的语音识别支持。[实时语音识别](https://help.aliyun.com/zh/model-studio/real-time-speech-recognition-user-guide) @@ -1186,8 +980,6 @@ qwen3-asr-flash-realtime、qwen3-asr-flash-realtime-2025-10-27 2025-12-17 -中国内地 - qwen3-asr-flash、qwen3-asr-flash-2025-09-08 支持任意采样率和声道的音频。[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide) @@ -1196,8 +988,6 @@ qwen3-asr-flash、qwen3-asr-flash-2025-09-08 2025-12-17 -中国内地 - fun-asr-mtl、fun-asr-mtl-2025-08-25 支持对中、英、日、韩等共 31 种语言的语音识别，尤其适合东南亚出海场景。[录音文件识别-Fun-ASR/Paraformer](https://help.aliyun.com/zh/model-studio/recording-file-recognition) @@ -1206,8 +996,6 @@ fun-asr-mtl、fun-asr-mtl-2025-08-25 2025-12-16 -中国内地 - wan2.6-t2i 新增同步接口。支持在总像素面积与宽高比约束内，自由选尺寸。[万相-文生图V2](https://help.aliyun.com/zh/model-studio/text-to-image-v2-api-reference) @@ -1216,8 +1004,6 @@ wan2.6-t2i 2025-12-16 -中国内地 - wan2.6-image 支持图像编辑和图文混合输出。[万相-图像生成与编辑2.6](https://help.aliyun.com/zh/model-studio/wan-image-generation-api-reference) @@ -1226,8 +1012,6 @@ wan2.6-image 2025-12-16 -中国内地 - wan2.6-i2v 新增多镜头叙事能力，支持音频能力，支持自动配音，或传入自定义音频文件。[万相-图生视频-基于首帧（2.1-2.6）](https://help.aliyun.com/zh/model-studio/legacy-image-to-video-api-reference/) @@ -1236,8 +1020,6 @@ wan2.6-i2v 2025-12-16 -中国内地 - wan2.6-r2v 基于参考视频的角色形象和音色，生成多镜头视频，支持自动配音。[万相2.7-参考生视频](https://help.aliyun.com/zh/model-studio/wan-video-to-video-api-reference) @@ -1246,8 +1028,6 @@ wan2.6-r2v 2025-12-16 -中国内地 - wan2.6-t2v 新增多镜头叙事能力，支持音频能力，支持自动配音，或传入自定义音频文件。[文生视频](https://help.aliyun.com/zh/model-studio/text-to-video-api-reference) @@ -1256,8 +1036,6 @@ wan2.6-t2v 2025-12-16 -中国内地 - qwen-voice-design 千问发布声音设计模型，通过文本描述生成定制化音色。结合qwen3-tts-vd-realtime-2025-12-16模型使用生成语音，覆盖 10 种语言。[声音设计（Qwen）](https://help.aliyun.com/zh/model-studio/qwen-tts-voice-design) @@ -1266,8 +1044,6 @@ qwen-voice-design 2025-12-16 -中国内地 - qwen3-tts-vd-realtime-2025-12-16（快照版） 千问实时语音合成发布全新快照版模型，可使用[声音设计（Qwen）](https://help.aliyun.com/zh/model-studio/qwen-tts-voice-design)生成的音色进行低延迟、高稳定性的实时合成；支持多语言输出；能根据文本自动调节语气，并优化复杂文本的合成表现。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) @@ -1276,8 +1052,6 @@ qwen3-tts-vd-realtime-2025-12-16（快照版） 2025-12-12 -中国内地 - fun-asr、fun-asr-2025-11-07 录音文件识别-Fun-ASR功能更新： @@ -1289,8 +1063,6 @@ fun-asr、fun-asr-2025-11-07 2025-12-11 -中国内地 - cosyvoice-v3-flash、cosyvoice-v3-plus CosyVoice模型音色更新： @@ -1304,8 +1076,6 @@ CosyVoice模型音色更新： 2025-12-04 -中国内地 - qwen3-omni-flash-2025-12-01 千问Omni发布的最新快照模型，支持的音色增加至49种，模型的指令跟随能力大幅升级，能高效理解文本、图像、音频、视频。[非实时（Qwen-Omni）](https://help.aliyun.com/zh/model-studio/qwen-omni) @@ -1314,8 +1084,6 @@ qwen3-omni-flash-2025-12-01 2025-12-04 -中国内地 - qwen3-omni-flash-realtime**\-**2025-12-01 千问Omni 实时版发布的最新快照模型，提供了低延迟的多模态交互能力，支持的音色增加至49种，模型的指令跟随能力和交互体验大幅升级。[实时（Qwen-Omni-Realtime）](https://help.aliyun.com/zh/model-studio/realtime) @@ -1324,8 +1092,6 @@ qwen3-omni-flash-realtime**\-**2025-12-01 2025-12-04 -中国内地 - qwen3-livetranslate-flash、qwen3-livetranslate-flash-2025-12-01 千问3-LiveTranslate-Flash 是音视频翻译模型，支持 18 种语言（包括中文、英文、俄文、法文等）互译，可结合视觉上下文提升翻译准确性，并输出文本与语音。[音视频文件翻译-千问](https://help.aliyun.com/zh/model-studio/qwen3-livetranslate-flash) @@ -1334,8 +1100,6 @@ qwen3-livetranslate-flash、qwen3-livetranslate-flash-2025-12-01 2025-12-04 -中国内地 - deepseek-v3.2 DeepSeek-V3.2是引入DeepSeek Sparse Attention（一种稀疏注意力机制）的正式版模型，也是DeepSeek推出的首个将思考融入工具使用的模型，同时支持思考模式与非思考模式的工具调用。 @@ -1346,8 +1110,6 @@ DeepSeek-V3.2是引入DeepSeek Sparse Attention（一种稀疏注意力机制） 2025-11-28 -中国内地 - qwen-mt-lite 千问基础级文本翻译大模型，支持31个语种互译，相较于qwen-mt-flash响应更快，成本更低，适用于等对延迟敏感的场景。[翻译能力（Qwen-MT）](https://help.aliyun.com/zh/model-studio/machine-translation) @@ -1356,8 +1118,6 @@ qwen-mt-lite 2025-11-27 -中国内地 - qwen-voice-enrollment 千问发布声音复刻模型，仅需 5 秒以上音频即可快速生成高相似度声音。结合qwen3-tts-vc-realtime-2025-11-27模型使用，可高保真复刻并实时输出某人的声音，覆盖 11 种语言。[声音复刻（Qwen）](https://help.aliyun.com/zh/model-studio/qwen-tts-voice-cloning) @@ -1366,8 +1126,6 @@ qwen-voice-enrollment 2025-11-27 -中国内地 - qwen3-tts-vc-realtime-2025-11-27（快照版） 千问实时语音合成发布全新快照版模型，可使用[声音复刻（Qwen）](https://help.aliyun.com/zh/model-studio/qwen-tts-voice-cloning)生成的音色进行低延迟、高稳定性的实时合成；支持多语言输出；能根据文本自动调节语气，并优化复杂文本的合成表现。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) @@ -1376,8 +1134,6 @@ qwen3-tts-vc-realtime-2025-11-27（快照版） 2025-11-27 -中国内地 - qwen3-tts-flash-realtime-2025-11-27（快照版） 千问实时语音合成发布全新快照版模型，低延迟且稳定性高；音色更丰富，同一音色支持多语言输出；能根据文本自动调节语气，并提升复杂文本的合成表现。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) @@ -1386,8 +1142,6 @@ qwen3-tts-flash-realtime-2025-11-27（快照版） 2025-11-27 -中国内地 - qwen3-tts-flash-2025-11-27（快照版） 千问语音合成发布全新快照版模型，音色更丰富；同一音色支持多语言输出；可自适应文本调节语气，并优化复杂文本的合成能力。[非实时语音合成](https://help.aliyun.com/zh/model-studio/non-realtime-tts-user-guide) @@ -1396,8 +1150,6 @@ qwen3-tts-flash-2025-11-27（快照版） 2025-11-21 -中国内地 - qwen-vl-ocr-2025-11-20（快照版） 千问文字提取模型，该快照版基于Qwen3-VL架构，大幅提升文档解析、文字定位能力。[文字提取](https://help.aliyun.com/zh/model-studio/qwen-vl-ocr) @@ -1406,8 +1158,6 @@ qwen-vl-ocr-2025-11-20（快照版） 2025-11-20 -中国内地 - qwen3-asr-flash-filetrans、qwen3-asr-flash-filetrans-2025-11-17（快照版） 千问录音文件识别发布了新模型，专为音频文件的异步转写设计，支持最长12小时录音。[非实时语音识别](https://help.aliyun.com/zh/model-studio/non-realtime-speech-recognition-user-guide) @@ -1416,8 +1166,6 @@ qwen3-asr-flash-filetrans、qwen3-asr-flash-filetrans-2025-11-17（快照版） 2025-11-20 -中国内地 - gui-plus GUI-Plus 可基于屏幕截图和自然语言指令来解析用户意图，并转换为标准化的图像用户界面（GUI）操作（如点击、输入、滚动等），供外部系统决策或执行。相较于千问VL系列模型，提升了GUI操作的准确性。[界面交互](https://help.aliyun.com/zh/model-studio/gui-automation) @@ -1426,8 +1174,6 @@ GUI-Plus 可基于屏幕截图和自然语言指令来解析用户意图，并 2025-11-19 -中国内地 - fun-asr-realtime-2025-11-07（快照版） Fun-ASR实时语音识别发布了全新快照版模型，优化远场语音活动检测（VAD）以提升识别准确率与稳定性，并在原有中英文识别基础上新增支持中文多地方言及日语。[实时语音识别-Fun-ASR/Paraformer](https://help.aliyun.com/zh/model-studio/real-time-speech-recognition) @@ -1436,8 +1182,6 @@ Fun-ASR实时语音识别发布了全新快照版模型，优化远场语音活 2025-11-19 -中国内地 - fun-asr-2025-11-07（快照版） Fun-ASR录音文件识别发布了全新快照版模型，优化远场语音活动检测（VAD）以提升识别准确率与稳定性，并在原有中英文识别基础上新增支持中文多地方言及日语。[录音文件识别-Fun-ASR/Paraformer](https://help.aliyun.com/zh/model-studio/recording-file-recognition) @@ -1446,8 +1190,6 @@ Fun-ASR录音文件识别发布了全新快照版模型，优化远场语音活 2025-11-19 -中国内地 - cosyvoice-v3-flash 较之前版本提升了发音准确性、音色相似度，并且增加了更多小语种支持（德、西、法、意、俄）。[实时语音合成-CosyVoice /Sambert](https://help.aliyun.com/zh/model-studio/text-to-speech) @@ -1456,8 +1198,6 @@ cosyvoice-v3-flash 2025-11-11 -中国内地 - kimi-k2-thinking 由月之暗面（Moonshot AI）公司推出的思考模型，具有通用智能体能力和推理能力，擅长深度推理，并能通过多步工具调用解决各类难题。[Kimi-阿里云](https://help.aliyun.com/zh/model-studio/kimi-api) @@ -1466,8 +1206,6 @@ kimi-k2-thinking 2025-11-10 -中国内地 - qwen-mt-flash 相较于qwen-mt-turbo支持流式增量输出，整体性能表现有所提升。[翻译能力（Qwen-MT）](https://help.aliyun.com/zh/model-studio/machine-translation) @@ -1476,8 +1214,6 @@ qwen-mt-flash 2025-11-03 -中国内地 - qwen3-max-preview qwen3-max-preview 模型的思考模式：在整体推理能力上显著提升，尤其在智能体编程、常识推理，以及数学、科学和通用任务方面表现更优。[深度思考](https://help.aliyun.com/zh/model-studio/deep-thinking) @@ -1486,8 +1222,6 @@ qwen3-max-preview 模型的思考模式：在整体推理能力上显著提升 2025-10-31 -中国内地 - qwen-image-edit-plus、qwen-image-edit-plus-2025-10-30 在qwen-image-edit的基础上优化了推理性能与系统稳定性，大幅缩短图像生成与编辑的响应时间，且支持单次请求返回多张图片。[图像编辑-千问](https://help.aliyun.com/zh/model-studio/qwen-image-edit-guide) @@ -1496,8 +1230,6 @@ qwen-image-edit-plus、qwen-image-edit-plus-2025-10-30 2025-10-27 -中国内地 - qwen3-asr-flash-realtime、qwen3-asr-flash-realtime-2025-10-27 千问实时语音识别大模型具备自动语种识别功能，可识别 11 种语音类型，并能在复杂音频环境下较为准确地转录。[实时语音识别](https://help.aliyun.com/zh/model-studio/real-time-speech-recognition-user-guide) @@ -1506,8 +1238,6 @@ qwen3-asr-flash-realtime、qwen3-asr-flash-realtime-2025-10-27 2025-10-21 -中国内地 - qwen3-vl-32b-thinking、qwen3-vl-32b-instruct Qwen3-VL系列 32B 的Dense模型，文档识别与理解、空间感知与万物识别能力、视觉2D检测与空间推理能力均表现出色，适合通用场景下的复杂感知任务。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1516,8 +1246,6 @@ Qwen3-VL系列 32B 的Dense模型，文档识别与理解、空间感知与万 2025-10-21 -中国内地 - glm-4.6 GLM新一代旗舰模型，核心能力较glm-4.5全面提升。总参数量为3550 亿，激活参数320亿，上下文窗口扩展至200K。[GLM-阿里云](https://help.aliyun.com/zh/model-studio/glm) @@ -1526,8 +1254,6 @@ GLM新一代旗舰模型，核心能力较glm-4.5全面提升。总参数量为3 2025-10-16 -中国内地 - qwen3-vl-flash、qwen3-vl-flash-2025-10-15 Qwen3系列小尺寸视觉理解模型，实现思考模式和非思考模式的有效融合，相较于开源版Qwen3-VL-30B-A3B，效果更优，响应速度更快。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1536,8 +1262,6 @@ Qwen3系列小尺寸视觉理解模型，实现思考模式和非思考模式的 2025-10-14 -中国内地 - qwen3-vl-8b-thinking、qwen3-vl-8b-instruct Qwen3-VL系列 8B 的Dense模型，占用显存更低，能够完成多模态理解与推理；支持长视频长文档等超长上下文、视觉2D/3D定位；全面空间感知与万物识别能力。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1546,8 +1270,6 @@ Qwen3-VL系列 8B 的Dense模型，占用显存更低，能够完成多模态理 2025-10-03 -中国内地 - qwen3-vl-30b-a3b-thinking、qwen3-vl-30b-a3b-instruct 基于Qwen3-VL新一代开源模型，响应速度快，具备更强多模态理解与推理、视觉智能体、长视频长文档等超长上下文支持能力；全面升级空间感知与万物识别能力，胜任复杂现实任务。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1556,8 +1278,6 @@ qwen3-vl-30b-a3b-thinking、qwen3-vl-30b-a3b-instruct 2025-09-30 -中国内地 - deepseek-v3.2-exp 混合推理架构模型，同时支持思考模式与非思考模式，引入稀疏注意力机制，旨在提升处理长文本时的训练与推理效率，价格低于 deepseek-v3.1。详情参见[DeepSeek-阿里云](https://help.aliyun.com/zh/model-studio/deepseek-api)。 @@ -1566,8 +1286,6 @@ deepseek-v3.2-exp 2025-09-23 -中国内地 - wan2.5-i2v-preview 新增音频能力，支持自动配音，或传入自定义音频文件，实现音画同步。[图生视频-基于首帧](https://help.aliyun.com/zh/model-studio/legacy-image-to-video-api-reference/) @@ -1576,8 +1294,6 @@ wan2.5-i2v-preview 2025-09-23 -中国内地 - wan2.5-t2v-preview 新增音频能力，支持自动配音，或传入自定义音频文件，实现音画同步。[文生视频](https://help.aliyun.com/zh/model-studio/text-to-video-api-reference) @@ -1586,8 +1302,6 @@ wan2.5-t2v-preview 2025-09-23 -中国内地 - wan2.5-i2i-preview 支持输入文本、单图或多图实现基于主体一致性的图像编辑、多图融合和组图生成。[通用图像编辑2.5](https://help.aliyun.com/zh/model-studio/wan2-5-image-edit-api-reference) @@ -1596,8 +1310,6 @@ wan2.5-i2i-preview 2025-09-23 -中国内地 - wan2.5-t2i-preview 取消单边限制，在总像素面积与宽高比约束内，自由选尺寸。[文生图V2](https://help.aliyun.com/zh/model-studio/text-to-image-v2-api-reference) @@ -1606,8 +1318,6 @@ wan2.5-t2i-preview 2025-09-23 -中国内地 - qwen3-vl-plus、qwen3-vl-plus-2025-09-23、qwen3-vl-235b-a22b-thinking、qwen3-vl-235b-a22b-instruct Qwen3系列视觉理解模型，实现思考模式和非思考模式的有效融合，视觉智能体能力达到世界顶尖水平。此版本在视觉编码、空间感知、多模态思考等方向全面升级；视觉感知与识别能力大幅提升。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1616,8 +1326,6 @@ Qwen3系列视觉理解模型，实现思考模式和非思考模式的有效融 2025-09-23 -中国内地 - qwen3-max、qwen3-max-2025-09-23 相较qwen3-max-preview版本，在智能体编程与工具调用方向进行了专项升级。本次发布的正式版模型达到领域SOTA水平，适配场景更加复杂的智能体需求。模型列表 @@ -1626,8 +1334,6 @@ qwen3-max、qwen3-max-2025-09-23 2025-09-23 -中国内地 - qwen3-livetranslate-flash-realtime-2025-09-22 qwen3-livetranslate-flash-realtime 是一款多语言音视频实时翻译模型，可识别 18 种语言，并实时翻译为 10 种语言的音频。[实时语音/音视频翻译-千问](https://help.aliyun.com/zh/model-studio/qwen3-5-livetranslate-flash-realtime) @@ -1636,8 +1342,6 @@ qwen3-livetranslate-flash-realtime 是一款多语言音视频实时翻译模 2025-09-23 -中国内地 - qwen-image-plus 在复杂文本渲染方面表现突出，特别是中英文文本渲染，可实现复杂的图文混合布局，相较于qwen-image更具价格优势。[文生图（Qwen-Image)](https://help.aliyun.com/zh/model-studio/qwen-image-api) @@ -1646,8 +1350,6 @@ qwen-image-plus 2025-09-23 -中国内地 - qwen3-coder-plus-2025-09-23 基于 Qwen3 的代码生成模型，相较上一版本（7月22日快照）在下游任务效果和工具调用方面鲁棒性有所提升，代码安全性增强。[代码能力（Qwen-Coder）](https://help.aliyun.com/zh/model-studio/qwen-coder) @@ -1656,8 +1358,6 @@ qwen3-coder-plus-2025-09-23 2025-09-23 -中国内地 - tongyi-embedding-vision-plus、tongyi-embedding-vision-flash 以Qwen系列大语言模型为基座构建，围绕视觉向量化能力进行增强训练，支持文字、图像、视频三种模态。[文本与多模态向量化](https://help.aliyun.com/zh/model-studio/embedding) @@ -1666,8 +1366,6 @@ tongyi-embedding-vision-plus、tongyi-embedding-vision-flash 2025-09-23 -中国内地 - fun-asr-realtime、fun-asr-realtime-2025-09-15 集成了创新的 RAG 技术，支持大规模热词自定义、ITN 规范化、标点预测等多维功能，显著提升了整体识别准确率和语境贴合度，同时支持中英文自由切换，具备更强的噪声鲁棒性，适应多样复杂环境。[实时语音识别-Fun-ASR/Paraformer](https://help.aliyun.com/zh/model-studio/real-time-speech-recognition) @@ -1676,8 +1374,6 @@ fun-asr-realtime、fun-asr-realtime-2025-09-15 2025-09-22 -中国内地 - qwen3-tts-flash-realtime、qwen3-tts-flash-realtime-2025-09-18 千问最新的实时语音合成大模型，不仅拥有17种高表现力的拟人音色，且能低延迟高稳定地合成音频；同时支持多种语言、方言。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) @@ -1686,8 +1382,6 @@ qwen3-tts-flash-realtime、qwen3-tts-flash-realtime-2025-09-18 2025-09-22 -中国内地 - qwen3-tts-flash、qwen3-tts-flash-2025-09-18 千问最新的离线语音合成大模型，不仅拥有17种高表现力的拟人音色，且能低延迟高稳定地合成音频；同时支持多种语言、方言。[非实时语音合成](https://help.aliyun.com/zh/model-studio/non-realtime-tts-user-guide) @@ -1696,8 +1390,6 @@ qwen3-tts-flash、qwen3-tts-flash-2025-09-18 2025-09-22 -中国内地 - qwen3-omni-flash-realtime、qwen3-omni-flash-realtime-2025-09-15 Qwen3系列多模态大模型的实时版，提供了低延迟的多模态交互能力，支持音频的流式输入，支持输出文本和音频，同时支持多种音色和语言。[实时（Qwen-Omni-Realtime）](https://help.aliyun.com/zh/model-studio/realtime) @@ -1706,8 +1398,6 @@ Qwen3系列多模态大模型的实时版，提供了低延迟的多模态交互 2025-09-22 -中国内地 - qwen3-omni-flash、qwen3-omni-flash-2025-09-15 Qwen3系列多模态模型，能高效理解文本、图像、音频和视频，支持119种语言的文本交互，具备卓越的指令跟随与系统提示定制能力，可轻松设定角色和对话风格，可用于构建语音助手、多媒体分析和内容创作应用。[非实时（Qwen-Omni）](https://help.aliyun.com/zh/model-studio/qwen-omni) @@ -1716,8 +1406,6 @@ Qwen3系列多模态模型，能高效理解文本、图像、音频和视频， 2025-09-22 -中国内地 - qwen3-omni-30b-a3b-captioner 以千问3-Omni为基座的开源模型，无需任何提示，自动为复杂语音、环境声、音乐、影视声效等生成精准、全面的描述，能识别说话人情绪、音乐元素（如风格、乐器）、敏感信息等，适用于音频内容分析、安全审核、意图识别、音频剪辑等多个领域。[音频理解-Qwen3-Omni-Captioner](https://help.aliyun.com/zh/model-studio/qwen3-omni-captioner) @@ -1726,8 +1414,6 @@ qwen3-omni-30b-a3b-captioner 2025-09-19 -中国内地 - wan2.2-animate-move 支持将模板视频中角色的动作和表情，迁移至单张静态人物图片上，生成人物动作视频。[万相-图生动作](https://help.aliyun.com/zh/model-studio/wan-animate-move-api) @@ -1736,8 +1422,6 @@ wan2.2-animate-move 2025-09-19 -中国内地 - wan2.2-animate-mix 能够依据人物图片和参考视频，将视频中的主角替换为图片中的角色，同时保留原视频的场景、光照和色调，实现无缝人物替换。[万相-视频换人](https://help.aliyun.com/zh/model-studio/wan-animate-mix-api) @@ -1746,8 +1430,6 @@ wan2.2-animate-mix 2025-09-12 -中国内地 - qwen-vl-ocr-2025-08-18 文字定位能力全面升级，通用文字识别、信息抽取能力均有提升。[文字提取](https://help.aliyun.com/zh/model-studio/qwen-vl-ocr#e91318ce09pxi)。 @@ -1756,8 +1438,6 @@ qwen-vl-ocr-2025-08-18 2025-09-12 -中国内地 - wan2.2-kf2v-flash 指令理解与运镜控制更准，稳定性全面提升。[图生视频-基于首尾帧](https://help.aliyun.com/zh/model-studio/legacy-image-to-video-by-first-and-last-frame-api-reference) @@ -1766,8 +1446,6 @@ wan2.2-kf2v-flash 2025-09-11 -中国内地 - qwen-plus-2025-09-11 属于 Qwen3 系列模型，相较于qwen-plus-2025-07-28，在思考模式下提升了指令遵循能力、总结回复更加精简，详见[深度思考](https://help.aliyun.com/zh/model-studio/deep-thinking)。在非思考模式下中文理解与逻辑推理能力得到增强，详见[概述](https://help.aliyun.com/zh/model-studio/text-generation)。 @@ -1776,8 +1454,6 @@ qwen-plus-2025-09-11 2025-09-11 -中国内地 - qwen3-next-80b-a3b-thinking、qwen3-next-80b-a3b-instruct 基于Qwen3的新一代开源模型，thinking模型相较于qwen3-235b-a22b-thinking-2507提升了指令遵循能力，总结回复更加精简，详见[深度思考](https://help.aliyun.com/zh/model-studio/deep-thinking)。instruct模型相较于qwen3-235b-a22b-instruct-2507增强了中文理解、逻辑推理及文本生成能力，详见[概述](https://help.aliyun.com/zh/model-studio/text-generation)。 @@ -1786,8 +1462,6 @@ qwen3-next-80b-a3b-thinking、qwen3-next-80b-a3b-instruct 2025-09-08 -中国内地 - cosyvoice-v3-plus、cosyvoice-v3 语音合成CosyVoice新增cosyvoice-v3-plus和cosyvoice-v3模型，较之前版本在自然度、音质、韵律、情感表现力上有更好的表现。使用方法请参见[实时语音合成-CosyVoice /Sambert](https://help.aliyun.com/zh/model-studio/text-to-speech)。 @@ -1796,8 +1470,6 @@ cosyvoice-v3-plus、cosyvoice-v3 2025-09-08 -中国内地 - qwen3-asr-flash（稳定版，等同qwen3-asr-flash-2025-09-08） qwen3-asr-flash-2025-09-08（快照版） @@ -1808,8 +1480,6 @@ Qwen3-ASR模型基于Qwen3-Omni模型基座训练而成，支持多语言识别 2025-09-05 -中国内地 - qwen3-max-preview 基于Qwen3的Qwen-Max模型（预览版），相较Qwen 2.5系列整体通用能力有大幅度提升，中英文通用文本理解能力、复杂指令遵循能力、主观开放任务能力、多语言能力、工具调用能力均显著增强；模型知识幻觉更少。千问 Max @@ -1818,8 +1488,6 @@ qwen3-max-preview 2025-08-26 -中国内地 - wan2.2-s2v-detect、wan2.2-s2v 万相数字人模型支持基于单张图片和音频，生成动作自然的说话、唱歌或表演视频，不限制形象画幅，支持肖像、全身或半身的人物图像。[万相-数字人](https://help.aliyun.com/zh/model-studio/wan-s2v-overview/) @@ -1828,8 +1496,6 @@ wan2.2-s2v-detect、wan2.2-s2v 2025-08-22 -中国内地 - qwen-mt-image 千问图像翻译模型支持将11种语言图片的文字翻译成中文或英文，能精准保留原始排版与内容信息，并提供术语定义、敏感词过滤、图像主体检测等功能。[千问-图像翻译](https://help.aliyun.com/zh/model-studio/qwen-mt-image-api) @@ -1838,8 +1504,6 @@ qwen-mt-image 2025-08-22 -中国内地 - deepseek-v3.1 混合推理架构模型，同时支持思考模式与非思考模式。详情参见[DeepSeek-阿里云](https://help.aliyun.com/zh/model-studio/deepseek-api)。 @@ -1848,8 +1512,6 @@ deepseek-v3.1 2025-08-22 -中国内地 - qwen-deep-research 千问深入研究模型，它可以拆解复杂问题，结合互联网搜索进行推理分析并生成研究报告。详情请参见[深入研究（Qwen-Deep-Research）](https://help.aliyun.com/zh/model-studio/qwen-deep-research) @@ -1858,8 +1520,6 @@ qwen-deep-research 2025-08-22 -中国内地 - fun-asr（稳定版，等同fun-asr-2025-08-25） fun-asr-2025-08-25（快照版） @@ -1870,8 +1530,6 @@ Fun-ASR 是通义实验室推出的端到端语音识别大模型。它基于先 2025-08-19 -中国内地 - qwen-image-edit 千问图像编辑模型支持精准的中英双语文字编辑、调色、细节增强、风格迁移、增删物体、改变位置和动作等操作，可实现复杂的图文编辑。[图像编辑-千问](https://help.aliyun.com/zh/model-studio/qwen-image-edit-guide) @@ -1880,8 +1538,6 @@ qwen-image-edit 2025-08-15 -中国内地 - qwen-vl-plus-2025-08-15 视觉理解模型。在物体识别与定位、多语言处理的能力上有显著提升。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1890,8 +1546,6 @@ qwen-vl-plus-2025-08-15 2025-08-13 -中国内地 - qwen-vl-max-2025-08-13 视觉理解模型。视觉理解指标全面提升，数学、推理、物体识别、多语言处理能力显著增强。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1900,8 +1554,6 @@ qwen-vl-max-2025-08-13 2025-08-13 -中国内地 - qwen-image 千问文生图模型在复杂文本渲染方面表现突出，特别是中英文文本渲染，可实现复杂的图文混合布局**。**[文生图（Qwen-Image)](https://help.aliyun.com/zh/model-studio/qwen-image-api) @@ -1910,8 +1562,6 @@ qwen-image 2025-08-11 -中国内地 - wan2.2-i2v-flash 万相2.2极速版模型。相较 2.1 模型，新版本在画面细节表现和运动稳定性方面均有显著提升，生成速度提升达 50%。[首帧生视频](https://help.aliyun.com/zh/model-studio/legacy-image-to-video-api-reference/) @@ -1920,8 +1570,6 @@ wan2.2-i2v-flash 2025-08-06 -中国内地 - qwen-vl-plus-2025-07-10 视觉理解模型。相较于上一版模型，进一步提升监控视频内容的理解能力。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -1930,8 +1578,6 @@ qwen-vl-plus-2025-07-10 2025-08-05 -中国内地 - qwen3-coder-flash、qwen3-coder-flash-2025-07-28 千问Coder系列速度最快、成本最低的模型。[代码能力（Qwen-Coder）](https://help.aliyun.com/zh/model-studio/qwen-coder) @@ -1940,8 +1586,6 @@ qwen3-coder-flash、qwen3-coder-flash-2025-07-28 2025-08-05 -中国内地 - qwen-flash、qwen-flash-2025-07-28 千问系列速度最快、成本极低的模型，适合简单任务。模型列表 @@ -1950,8 +1594,6 @@ qwen-flash、qwen-flash-2025-07-28 2025-07-30 -中国内地 - qwen-plus-2025-07-28 属于 Qwen3 系列模型，相较于上一版模型，将上下文长度提高到了1,000,000。思考模式请参见[深度思考](https://help.aliyun.com/zh/model-studio/deep-thinking)，非思考模式请参见[概述](https://help.aliyun.com/zh/model-studio/text-generation)。 @@ -1960,8 +1602,6 @@ qwen-plus-2025-07-28 2025-07-30 -中国内地 - qwen3-30b-a3b-thinking-2507 qwen3-30b-a3b-instruct-2507 @@ -1972,8 +1612,6 @@ qwen3-30b-a3b-instruct-2507 2025-07-28 -中国内地 - wan2.2-i2v-plus 相较 2.1 模型，新版本在画面细节表现和运动稳定性方面均有显著提升，生成速度提升达 50%。[首帧生视频](https://help.aliyun.com/zh/model-studio/legacy-image-to-video-api-reference/) @@ -1982,8 +1620,6 @@ wan2.2-i2v-plus 2025-07-28 -中国内地 - wan2.2-t2v-plus 相较 2.1 模型，新版本在画面细节表现和运动稳定性方面均有显著提升，生成速度提升达 50%。[文生视频](https://help.aliyun.com/zh/model-studio/text-to-video-api-reference) @@ -1992,8 +1628,6 @@ wan2.2-t2v-plus 2025-07-28 -中国内地 - wan2.2-t2i-flash、wan2.2-t2i-plus 相较 2.1 模型，新版本在创意性、稳定性、写实质感上全面升级，生成速度提升达 50%。[文生图V2版](https://help.aliyun.com/zh/model-studio/text-to-image-v2-api-reference) @@ -2002,8 +1636,6 @@ wan2.2-t2i-flash、wan2.2-t2i-plus 2025-07-26 -中国内地 - qwen-doc-turbo 千问数据挖掘模型，可以提取文档中的结构化信息并用于数据标注和内容审核等领域。详情请参见[数据挖掘（Qwen-Doc）](https://help.aliyun.com/zh/model-studio/data-mining-qwen-doc)。 @@ -2012,8 +1644,6 @@ qwen-doc-turbo 2025-07-24 -中国内地 - qwen3-235b-a22b-thinking-2507、qwen3-235b-a22b-instruct-2507 是qwen3-235b-a22b的升级版。thinking模型逻辑能力、通用能力、知识增强及创作能力均有大幅提升，适用于高难度强推理场景，参见[深度思考](https://help.aliyun.com/zh/model-studio/deep-thinking)。instruct模型创作能力与模型安全性均有提升，参见[概述](https://help.aliyun.com/zh/model-studio/text-generation)。 @@ -2022,8 +1652,6 @@ qwen3-235b-a22b-thinking-2507、qwen3-235b-a22b-instruct-2507 2025-07-23 -中国内地 - qwen3-coder、qwen3-coder-plus-2025-07-22 基于 Qwen3 的代码生成模型，具有强大的Coding Agent能力，擅长工具调用和环境交互，代码能力卓越的同时兼具通用能力。[代码能力（Qwen-Coder）](https://help.aliyun.com/zh/model-studio/qwen-coder) @@ -2032,8 +1660,6 @@ qwen3-coder、qwen3-coder-plus-2025-07-22 2025-06-26 -中国内地 - qwen-tts-2025-05-22 qwen-tts 模型的2025年5月22日快照版本。新增北京话、吴语和四川话三种音色。[非实时语音合成](https://help.aliyun.com/zh/model-studio/non-realtime-tts-user-guide) @@ -2042,8 +1668,6 @@ qwen-tts 模型的2025年5月22日快照版本。新增北京话、吴语和四 2025-06-24 -中国内地 - qwen-plus、qwen-turbo qwen-plus 已更新，支持[思考模式](https://help.aliyun.com/zh/model-studio/deep-thinking)，与 qwen-plus-2025-04-28 能力相同；qwen-turbo 已更新，支持[思考模式](https://help.aliyun.com/zh/model-studio/deep-thinking)，与 qwen-turbo-2025-04-28 能力相同。 @@ -2052,8 +1676,6 @@ qwen-plus 已更新，支持[思考模式](https://help.aliyun.com/zh/model-stud 2025-06-23 -中国内地 - cosyvoice-v2 cosyvoice-v2 全新升级，新增多种音色，不仅丰富了语言支持范围（如粤语、韩语、日语），还引入了更多风格化音色（如英式英语、美式英语）。详情请参见[CosyVoice音色列表](https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list)。 @@ -2062,8 +1684,6 @@ cosyvoice-v2 全新升级，新增多种音色，不仅丰富了语言支持范 2025-06-04 -中国内地 - text-embedding-v4 text-embedding-v4为text-embedding-v3的升级版模型，属于[Qwen3-Embedding](https://qwenlm.github.io/zh/blog/qwen3-embedding/)系列。相较于上一版模型，新版模型涵盖了更多自然语言以及多种编程语言，并新增2048及1536向量维度的选择。[文本与多模态向量化](https://help.aliyun.com/zh/model-studio/embedding) @@ -2072,8 +1692,6 @@ text-embedding-v4为text-embedding-v3的升级版模型，属于[Qwen3-Embedding 2025-06-04 -中国内地 - deepseek-r1-0528 deepseek-r1-0528为deepseek-r1的升级版模型，相较于上一版模型，新版模型在复杂推理任务中的表现有了显著提升。在数学、编程与通用逻辑等多个基准测评中取得了优异成绩。[DeepSeek-阿里云](https://help.aliyun.com/zh/model-studio/deepseek-api) @@ -2082,8 +1700,6 @@ deepseek-r1-0528为deepseek-r1的升级版模型，相较于上一版模型， 2025-06-03 -中国内地 - qvq-max-2025-05-15 视觉推理模型。较上一个版本，增强了在数学、编程、视觉分析、创作以及通用任务方面的能力。[视觉推理](https://help.aliyun.com/zh/model-studio/visual-reasoning) @@ -2092,8 +1708,6 @@ qvq-max-2025-05-15 2025-06-03 -中国内地 - qvq-plus、qvq-plus-latest、qvq-plus-2025-05-15 视觉推理模型。支持视觉输入及思维链输出，继qvq-max模型后推出的plus版本，相较于qvq-max模型，qvq-plus系列模型推理速度更快，效果和成本更均衡。[视觉推理](https://help.aliyun.com/zh/model-studio/visual-reasoning) @@ -2102,8 +1716,6 @@ qvq-plus、qvq-plus-latest、qvq-plus-2025-05-15 2025-06-03 -中国内地 - cosyvoice-v2 SSML（Speech Synthesis Markup Language，语音合成标记语言）不仅能指定语音合成读什么内容，还能精细控制其朗读方式，包括断句分词、发音、语速、停顿、语调、音量等语音特征，甚至支持添加背景音乐，从而实现更加自然、富有表现力的语音输出。详细使用方法请参见[SSML标记语言介绍](https://help.aliyun.com/zh/model-studio/introduction-to-cosyvoice-ssml-markup-language)。 @@ -2112,8 +1724,6 @@ SSML（Speech Synthesis Markup Language，语音合成标记语言）不仅能 2025-05-19 -中国内地 - wanx2.1-vace-plus 通用视频编辑模型。模型具备多模态输入能力，融合图片、视频与文本提示词，可执行图生视频（参考图像主体或背景生成视频）、视频重绘（从输入视频中提取运动特征生成视频）等多种任务。[通用视频编辑](https://help.aliyun.com/zh/model-studio/legacy-wanx-vace-api-reference) @@ -2122,8 +1732,6 @@ wanx2.1-vace-plus 2025-05-15 -中国内地 - aitryon-plus AI试衣模型。相较于基础版aitryon模型，aitryon-plus模型在图片清晰度、服饰纹理细节和logo还原效果等方面均有提升，但生成耗时较长，适用于对时效性要求不高的场景。[AI试衣-Plus版](https://help.aliyun.com/zh/model-studio/aitryon-plus-api) @@ -2132,8 +1740,6 @@ AI试衣模型。相较于基础版aitryon模型，aitryon-plus模型在图片 2025-05-15 -中国内地 - qwen-vl-plus-2025-05-07 视觉理解模型。模型在数学、推理、监控视频内容的理解方面的能力有显著提升。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -2142,8 +1748,6 @@ qwen-vl-plus-2025-05-07 2025-04-30 -中国内地 - qwen-vl-ocr-2025-04-13 基于Qwen-VL训练的OCR专有大模型，提供强大的图文识别能力；该模型新增六种内置的OCR任务，分别是：通用识别、文档解析、表格解析、信息抽取、多语言识别、公式识别。[文字提取](https://help.aliyun.com/zh/model-studio/qwen-vl-ocr) @@ -2152,8 +1756,6 @@ qwen-vl-ocr-2025-04-13 2025-04-28 -中国内地 - **Qwen3商业版模型** qwen-plus-2025-04-28、qwen-turbo-2025-04-28 @@ -2181,8 +1783,6 @@ Qwen3 模型支持思考模式和非思考模式，您可以通过 `enable_think 2025-04-24 -中国内地 - cosyvoice-v2 用户仅需提供10~20秒的音频，即可迅速生成高度相似且听感自然的定制声音。[声音复刻/设计](https://help.aliyun.com/zh/model-studio/voice-replica-1/) @@ -2191,8 +1791,6 @@ cosyvoice-v2 2025-04-18 -中国内地 - qwen-vl-max-2025-04-08 视觉理解模型。数学和推理能力有所提升，回复风格面向人类偏好进行调整，模型回复详实程度和格式清晰度明显改善。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -2201,8 +1799,6 @@ qwen-vl-max-2025-04-08 2025-04-18 -中国内地 - wanx2.1-kf2v-plus 基于输入的首帧和尾帧图片，模型能够根据提示词生成一段丝滑流畅的动态视频。[首尾帧生视频](https://help.aliyun.com/zh/model-studio/legacy-image-to-video-by-first-and-last-frame-api-reference) @@ -2211,8 +1807,6 @@ wanx2.1-kf2v-plus 2025-04-04 -中国内地 - qwen-vl-max-2025-04-02 视觉理解模型。在解决复杂数学问题方面，准确性显著提高，回复风格面向人类偏好进行大幅调整，尤其是数学、逻辑推理、知识问答等客观类问题，模型回复详实程度和格式清晰度明显改善。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -2221,8 +1815,6 @@ qwen-vl-max-2025-04-02 2025-03-28 -中国内地 - qvq-max、qvq-max-latest、qvq-max-2025-03-25 视觉推理模型。支持视觉输入及思维链输出，在数学、编程、视觉分析、创作以及通用任务上都表现出更强的能力。[视觉推理](https://help.aliyun.com/zh/model-studio/visual-reasoning) @@ -2231,8 +1823,6 @@ qvq-max、qvq-max-latest、qvq-max-2025-03-25 2025-03-26 -中国内地 - qwen-omni-turbo-2025-03-26 千问全新多模态理解生成大模型，支持文本、图像、语音与视频输入，并输出文本与音频，提供了4种自然对话音色。使用方法请参见[非实时（Qwen-Omni）](https://help.aliyun.com/zh/model-studio/qwen-omni)。 @@ -2241,8 +1831,6 @@ qwen-omni-turbo-2025-03-26 2025-03-26 -中国内地 - qwen2.5-omni-7b 千问全新多模态理解生成大模型，支持文本、图像、语音与视频输入，并输出文本与音频，提供了2种自然对话音色。使用方法请参见[非实时（Qwen-Omni）](https://help.aliyun.com/zh/model-studio/qwen-omni)。 @@ -2251,8 +1839,6 @@ qwen2.5-omni-7b 2025-03-25 -中国内地 - wanx2.1-imageedit 通用图像编辑模型。通过一句话指令实现多样化的图像编辑功能，如扩图、去文字水印、图像修复、图像风格迁移等。[万相-通用图像编辑2.1](https://help.aliyun.com/zh/model-studio/wanx-image-edit-api-reference) @@ -2261,8 +1847,6 @@ wanx2.1-imageedit 2025-03-24 -中国内地 - qwen2.5-vl-32b-instruct  视觉理解模型。在数学问题的解答方面达到了接近Qwen2.5VL-72B的水平，回复风格面向人类偏好进行大幅调整，尤其是数学、逻辑推理、知识问答等客观类问题，模型回复详实程度和格式清晰度明显改善。[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision) @@ -2271,8 +1855,6 @@ qwen2.5-vl-32b-instruct  2025-03-06 -中国内地 - qwq-plus、qwq-plus-latest、qwq-plus-2025-03-05 基于 Qwen2.5 模型训练的 QwQ 推理模型，通过强化学习大幅度提升了模型推理能力。模型数学代码等核心指标（AIME 24/25、LiveCodeBench）以及部分通用指标（IFEval、LiveBench等）达到DeepSeek-R1 满血版水平。[深度思考](https://help.aliyun.com/zh/model-studio/deep-thinking) @@ -2281,8 +1863,6 @@ qwq-plus、qwq-plus-latest、qwq-plus-2025-03-05 2025-03-06 -中国内地 - qwq-32b 基于 Qwen2.5-32B 模型训练的 QwQ 推理模型，通过强化学习大幅度提升了模型推理能力。模型数学代码等核心指标（AIME 24/25、LiveCodeBench）以及部分通用指标（IFEval、LiveBench等）达到DeepSeek-R1 满血版水平，各指标均显著超过同样基于 Qwen2.5-32B 的 DeepSeek-R1-Distill-Qwen-32B。[深度思考](https://help.aliyun.com/zh/model-studio/deep-thinking) @@ -2291,8 +1871,6 @@ qwq-32b 2025-03-03 -中国内地 - gummy-realtime-v1 gummy-chat-v1 @@ -2303,8 +1881,6 @@ Gummy大模型支持实时语音识别与翻译，能够精准识别中、英、 2025-02-25 -中国内地 - wanx2.1-i2v-turbo 相较于wanx2.1-i2v-plus模型，turbo模型生成速度更快，耗时仅为plus模型的三分之一，性价比更高。使用方法请参见[首帧生视频](https://help.aliyun.com/zh/model-studio/legacy-image-to-video-api-reference/)。 @@ -2313,8 +1889,6 @@ wanx2.1-i2v-turbo 2025-02-14 -中国内地 - qwen-omni-turbo qwen-omni-turbo-latest @@ -2327,8 +1901,6 @@ Qwen-Omni 系列模型支持输入多种模态的数据，包括视频、音频 2025-02-07 -中国内地 - qwen-vl-max-2025-01-25 视觉理解模型。属于[Qwen2.5-VL](https://qwenlm.github.io/blog/qwen2.5-vl/)系列模型，相较于上一版模型，扩展上下文至128k，显著增强图像和视频的理解能力。 @@ -2337,8 +1909,6 @@ qwen-vl-max-2025-01-25 2025-02-07 -中国内地 - qwen-vl-plus-2025-01-25 视觉理解模型。属于[Qwen2.5-VL](https://qwenlm.github.io/blog/qwen2.5-vl/)系列模型，相较于上一版模型，扩展上下文至128k，显著增强图像和视频的理解能力。 @@ -2347,8 +1917,6 @@ qwen-vl-plus-2025-01-25 2025-01-27 -中国内地 - deepseek-v3 deepseek-r1 @@ -2366,8 +1934,6 @@ DeepSeek系列模型是由深度求索（DeepSeek）公司推出的大语言模 2025-01-27 -中国内地 - qwen2.5-vl-3b-instruct qwen2.5-vl-7b-instruct @@ -2389,8 +1955,6 @@ qwen2.5-vl-72b-instruct 2025-01-27 -中国内地 - qwen-max-2025-01-25 qwen2.5-14b-instruct-1m @@ -2406,8 +1970,6 @@ qwen2.5-7b-instruct-1m 2025-01-22 -中国内地 - emoji-detect-v1 emoji-v1 @@ -2419,8 +1981,6 @@ emoji-v1 2025-01-17 -中国内地 - qwen-plus-2025-01-12 - 相对于qwen-plus-2024-12-20模型，中英文整体能力有提升，中英常识、阅读理解能力提升较为显著，在不同语言、方言、风格之间自然切换的能力有显著改善，中文指令遵循能力显著提升。使用方法请参见[qwen-plus-2025-01-12](https://bailian.console.aliyun.com/#/model-market/detail/qwen-plus-0112)。 @@ -2430,8 +1990,6 @@ qwen-plus-2025-01-12 2025-01-17 -中国内地 - wanx2.1-i2v-plus - 输入图片作为视频首帧，再根据提示词生成视频。使用方法请参见[首帧生视频](https://help.aliyun.com/zh/model-studio/legacy-image-to-video-api-reference/)。 @@ -2441,8 +1999,6 @@ wanx2.1-i2v-plus 2025-01-17 -中国内地 - wanx2.0-t2i-turbo - 擅长质感人像与创意设计，速度中等，性价比高。使用方法请参见[文生图V2系列模型](https://help.aliyun.com/zh/model-studio/text-to-image-v2-api-reference)。 @@ -2452,8 +2008,6 @@ wanx2.0-t2i-turbo 2025-01-13 -中国内地 - qwen-vl-plus-2025-01-02 - 相较于qwen-vl-plus-0809模型，大幅提升指令跟随、图像理解和数学能力。使用方法请参见[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)。 @@ -2463,8 +2017,6 @@ qwen-vl-plus-2025-01-02 2025-01-08 -中国内地 - wanx2.1-t2v-turbo wanx2.1-t2v-plus @@ -2478,8 +2030,6 @@ wanx2.1-t2v-plus 2025-01-08 -中国内地 - wanx2.1-t2i-turbo wanx2.1-t2i-plus @@ -2491,8 +2041,6 @@ wanx2.1-t2i-plus 2025-01-07 -中国内地 - qwen-vl-max-2024-12-30 - 该模型丰富了知识库，图像识别和理解能力进一步提升，能够解析复杂的视觉内容。使用方法请参见[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)。 @@ -2502,8 +2050,6 @@ qwen-vl-max-2024-12-30 2025-01-02 -中国内地 - qwen-audio-asr qwen-audio-asr-latest @@ -2517,8 +2063,6 @@ qwen-audio-asr-2024-12-04 2024-12-25 -中国内地 - qvq-72b-preview - 专注于提升视觉推理能力，尤其在数学推理领域。使用方法请参见[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)。 @@ -2528,8 +2072,6 @@ qvq-72b-preview 2024-12-25 -中国内地 - qwen-mt-plus qwen-mt-turbo @@ -2541,8 +2083,6 @@ qwen-mt-turbo 2024-12-18 -中国内地 - qwen2-vl-72b-instruct - 在多个视觉理解基准测试中取得了最先进的成绩，显著增强多模态任务的处理能力。使用方法请参见[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)。 @@ -2552,8 +2092,6 @@ qwen2-vl-72b-instruct 2024-12-12 -中国内地 - tongyi-intent-detect-v3 - 千问意图理解模型，能够在百毫秒级时间内快速、准确地解析用户意图，并选择合适工具来解决用户问题。详情请参见[意图理解](https://help.aliyun.com/zh/model-studio/intent-detect-capability)。 @@ -2563,8 +2101,6 @@ tongyi-intent-detect-v3 2024-12-10 -中国内地 - videoretalk - 支持根据人物视频和音频生成对口型视频，详情请参见[快速开始](https://help.aliyun.com/zh/model-studio/videos/videoretalk-quick-start)。 @@ -2574,8 +2110,6 @@ videoretalk 2024-12-10 -中国内地 - animate-anyone-gen2 animate-anyone-detect-gen2 @@ -2589,8 +2123,6 @@ animate-anyone-template-gen2 2024-12-10 -中国内地 - cosyvoice-v1 - 用户仅需提供10~20秒的音频，即可迅速生成高度相似且听感自然的定制声音。详情请参见[声音复刻/设计](https://help.aliyun.com/zh/model-studio/voice-replica-1/)。 @@ -2600,8 +2132,6 @@ llama系列 2024-12-09 -中国内地 - llama3.3-70b-instruct - 新增第三方大模型llama3.3系列中70B参数的模型。详情请参见[Llama（输入文本和图像）](https://help.aliyun.com/zh/model-studio/api-details-of-llama3-2-text-image)。 @@ -2611,8 +2141,6 @@ llama3.3-70b-instruct 2024-12-09 -中国内地 - qwen-audio-turbo-latest qwen-audio-turbo-2024-12-04 @@ -2624,8 +2152,6 @@ qwen-audio-turbo-2024-12-04 2024-11-28 -中国内地 - qwq-32b-preview - 本模型专注于增强 AI 推理能力。详情请参见QWQ。 @@ -2635,10 +2161,6 @@ qwq-32b-preview 中国内地 -qwen-plus-2024-11-25 - -qwen-plus-2024-11-27 - - 相较于qwen-plus-0919模型中英文回复详实程度显著提升，更加符合用户偏好；模型角色扮演能力显著增强；模型中文的文本创作能力显著提升；中英文指令遵循能力提升；修复了RAG场景下引用角标的生成问题。详情请参见[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)。 @@ -2646,8 +2168,6 @@ qwen-plus-2024-11-27 中国内地 -qwen-turbo-2024-11-01 - - 上下文长度扩展至一百万Token。详情请参见[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)。 @@ -2655,8 +2175,6 @@ qwen-turbo-2024-11-01 2024-11-15 -中国内地 - qwen-vl-max-2024-11-19 - 增强了图像理解效果，改善生成重复文本的情况。详情请参见[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)。 @@ -2666,9 +2184,6 @@ qwen-vl-max-2024-11-19 qwen-vl-max-2024-10-30 -- 加强了多语言理解能力。详情请参见[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)。 - - 中国内地 - qwen-vl-ocr @@ -2678,15 +2193,10 @@ qwen-vl-max-2024-10-30 - qwen-vl-ocr-latest -- qwen-vl-ocr是专用于OCR的模型；在表格、试题等类型图像的文字提取能力大幅提升。详情请参见[文字提取](https://help.aliyun.com/zh/model-studio/qwen-vl-ocr)。 - - 文生文 2024-11-12 -中国内地 - - 闭源版本qwen-coder-plus等模型 - 开源版本qwen2.5-coder-32b-instruct等模型 @@ -2699,8 +2209,6 @@ qwen-vl-max-2024-10-30 2024-11-7 -中国内地 - - liveportrait-detect - liveportrait @@ -2713,8 +2221,6 @@ qwen-vl-max-2024-10-30 2024-11-7 -中国内地 - - emo-detect-v1 - emo-v1 @@ -2727,8 +2233,6 @@ llama系列 2024-11-5 -中国内地 - - llama3.2-90b-vision-instruct - llama3.2-11b-vision @@ -2741,8 +2245,6 @@ llama系列 2024-10-29 -中国内地 - qwen2-vl-2b-instruct - 扩展上下文至32k，大幅提升图像理解能力。详情请参见[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)。 @@ -2752,8 +2254,6 @@ qwen2-vl-2b-instruct 2024-10-23 -中国内地 - qwen2-vl-7b-instruct - 新增qwen2-vl-7b-instruct模型，详情请参见[图像与视频理解](https://help.aliyun.com/zh/model-studio/vision)。 @@ -2763,8 +2263,6 @@ qwen2-vl-7b-instruct 2024-09-18 -中国内地 - qwen-long - 新增qwen-long模型，详情请参见[长上下文（Qwen-Long）](https://help.aliyun.com/zh/model-studio/long-context-qwen-long) @@ -2774,8 +2272,6 @@ qwen-long 2024-08-27 -中国内地 - qwen-vl-max-0809 - 本模型为qwen-vl-max的2024年8月9日快照版本，将在9月9日更新至qwen-vl-max主版本，快照版本维护到下个快照版本发布时间（待定）后一个月。点击[千问VL API详情](https://help.aliyun.com/zh/document_detail/2712587.html)查看完整信息。 @@ -2785,8 +2281,6 @@ qwen-vl-max-0809 2024-08-16 -中国内地 - Qwen2-Math系列模型 - 具有强大的数学解题能力，点击[通过API调用Qwen2-Math模型](https://help.aliyun.com/zh/document_detail/2844170.html)进行体验。 @@ -2796,8 +2290,6 @@ Qwen2-Math系列模型 2024-08-07 -中国内地 - FLUX文生图模型 - FLUX文生图模型是由 Black Forest Labs 开源的高质量文本到图像生成模型，它在多个维度上展现了卓越性能，尤其在文本引导的图像生成、多主体场景构建、以及精细的手部细节生成等方面，实现了显著的提升，为文生图领域设定了新的技术标杆。点击[文生图FLUX](https://help.aliyun.com/zh/model-studio/videos/flux)进行了解。 @@ -2807,8 +2299,6 @@ llama系列 2024-07-23 -中国内地 - - Llama3.1-8b-instruct - Llama3.1-70b-instruct @@ -2823,8 +2313,6 @@ llama系列 2024-07-10 -中国内地 - text-embedding-v3 - **text-embedding-v3**模型是**text-embedding-v2**模型的升级版本，主打**高性能、低成本、支持50+多语言、超长文本**。 @@ -2849,8 +2337,6 @@ text-embedding-v3 2024-07-05 -中国内地 - - qwen-turbo-0206 - qwen-turbo-0624 @@ -2863,8 +2349,6 @@ text-embedding-v3 2024-07-05 -中国内地 - - qwen-plus-0206 - qwen-plus-0624 @@ -2877,8 +2361,6 @@ text-embedding-v3 2024-07-02 -中国内地 - Minimax大语言模型 - 新增Minimax大语言模型，点击[MiniMax大语言模型](https://help.aliyun.com/zh/model-studio/videos/minimax-llm/)查看详情。 @@ -2888,8 +2370,6 @@ Minimax大语言模型 2024-06-07 -中国内地 - qwen2系列开源大语言模型 - 与qwen1.5相比，qwen2在语言理解、语言生成、多语言能力、编码、推理等基准测试中超越了大多数开源模型。点击[模型介绍](https://help.aliyun.com/zh/document_detail/2804542.html)查看详情。 @@ -2899,8 +2379,6 @@ qwen2系列开源大语言模型 2024-06-03 -中国内地 - 零一万物大语言模型 零一万物大语言模型是千亿参数大语言模型，是LM SYS榜单TOP10上唯一国产大模型。具备超强的问答、推理及文本生能力，完整内容请点击[零一万物大语言模型](https://help.aliyun.com/zh/model-studio/videos/yi-large-llm/)查看。 @@ -2909,8 +2387,6 @@ qwen2系列开源大语言模型 2024-04-28 -中国内地 - qwen1.5-110b-chat - qwen1.5-110b-chat模型时千问1.5对外开源的110B规模参数量的经过人类指令对齐的chat模型。点击[通义千问开源模型](https://help.aliyun.com/zh/document_detail/2713159.html)查看详情。 @@ -2928,6 +2404,26 @@ qwen1.5-110b-chat **功能说明** +推理模型 + +2026-07-25 + +国际 + +qwen3.7-flash、qwen3.7-flash-2026-07-15 + +Qwen3.7-Flash模型，相较3.6-Flash全面提升多模态理解与Agent执行能力。重点强化多模态基础能力、万物识别能力更强，真实世界感知与空间智能进一步提升，Search Agent、CI Agent等多模态Agent场景能力显著升级、端到端任务执行更稳定，多模态Coding能力优化、vibe coding 体验更加流畅。 + +图像生成 + +2026-07-21 + +国际 + +qwen-image-3.0-pro + +Qwen-Image-3.0-Pro 系列模型邀测上线，支持长文本输入与图中图密集排版，能够一次性精准生成报纸、分镜、菜单及试卷等复杂版面；具备10像素小字精准渲染能力，生动还原微表情、毛孔与发丝等摄影级细节，并支持 12 国语言、多种字体及主流网页、游戏界面的高保真仿真。[千问-图像生成与编辑3.0](https://help.aliyun.com/zh/model-studio/qwen-image-generation-and-editing-api-reference) + 语音合成 2026-07-14 @@ -2936,7 +2432,7 @@ qwen1.5-110b-chat qwen-audio-3.0-tts-plus、qwen-audio-3.0-tts-flash -Qwen-Audio-TTS语音合成模型上线，新增更多小语种和中文方言支持，增强了指令遵循与细粒度标签控制能力，音质和表现力全面提升。其中 Plus 版本面向高品质专业场景，Flash 版本面向低延迟实时交互场景，首包延时控制在 200ms 以内。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) +Qwen-Audio-TTS语音合成模型上线，新增更多小语种和中文方言支持，增强了指令遵循与细粒度标签控制能力，音质和表现力全面提升。其中 Plus 版本面向高品质专业场景，Flash 版本面向低延迟实时交互场景。[实时语音合成](https://help.aliyun.com/zh/model-studio/realtime-tts-user-guide) 参考生视频 @@ -2956,7 +2452,7 @@ wan2.7-r2v-2026-06-12 kimi-k2.7-code -Kimi K2.7 Code 模型新加坡地域上线。以编码为中心的智能体模型，专为长程软件工程任务优化，仅支持思考模式。[](#) +Kimi K2.7 Code 模型新加坡地域上线。以编码为中心的智能体模型，专为长程软件工程任务优化，仅支持思考模式。[Kimi-阿里云](https://help.aliyun.com/zh/model-studio/kimi-api) 图像生成 @@ -3852,6 +3348,16 @@ qwen3-asr-flash-realtime、qwen3-asr-flash-realtime-2025-10-27 **功能说明** +推理模型 + +2026-07-17 + +美国 + +qwen3.6-flash-us + +Qwen3.6 原生视觉语言 Flash 系列模型，在整体性能上较 Qwen3.5-Flash 显著提升。重点增强了智能体编程能力（在多项代码智能体基准上大幅超越前代）、数学推理和代码推理能力；在视觉能力方面，空间智能显著增强，其中物体定位和目标检测表现尤为突出。 + 文生文 2026-07-14 @@ -4942,6 +4448,46 @@ qwen-flash、qwen-flash-2025-07-28 **功能说明** +视频生成 + +2026-07-24 + +全球 + +happyhorse-1.1-t2v + +支持有声视频生成，可生成3~15秒、720P/1080P视频。[模型详情](https://help.aliyun.com/zh/model-studio/happyhorse-text-to-video-api-reference) + +视频生成 + +2026-07-24 + +全球 + +happyhorse-1.1-i2v + +支持有声视频生成，可生成3~15秒、720P/1080P视频。[模型详情](https://help.aliyun.com/zh/model-studio/happyhorse-image-to-video-api-reference) + +视频生成 + +2026-07-24 + +全球 + +happyhorse-1.1-r2v + +支持有声视频生成，可生成3~15秒、720P/1080P视频。[模型详情](https://help.aliyun.com/zh/model-studio/happyhorse-reference-to-video-api-reference) + +视频生成 + +2026-07-24 + +全球 + +happyhorse-1.0-video-edit + +支持视频编辑，可对视频进行风格化编辑和增强。[模型详情](https://help.aliyun.com/zh/model-studio/happyhorse-video-edit-api-reference) + 文生文 2026-07-14 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/security-and-compliance/content-security.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/security-and-compliance/content-security.md index 9244e72f..e7b5761d 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/security-and-compliance/content-security.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/security-and-compliance/content-security.md @@ -327,6 +327,41 @@ curl -X POST https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation 结果查询页面包含筛选栏（支持按条件、文本内容、时间范围搜索）和结果表格。表格列包括**文本内容**、**服务Service**、**风险等级**、**返回标签（释义）**、**反馈结果**、**请求时间**和**操作**。示例中两条记录分别对应`bailian_query_check`（请求检查）和`bailian_response_check`（响应检查）服务，均被标记为**高风险**，返回标签为`contraband_act（疑似违禁行为）:100`，每行可单击**详情**或**反馈**进行操作。 +## 知识库检索场景安全拦截 + +使用知识库检索问答功能时，上传到知识库的文档内容经检索后会作为上下文注入模型输入。如果知识库文档中包含违规或敏感内容，这些内容同样会触发 AI 安全护栏，导致请求被拦截并返回 `DataInspectionFailed` 错误。 + +如果您在知识库检索场景中遇到安全策略拦截，建议按以下步骤排查： + +1. 检查上传到知识库的文档内容，确认是否包含政治敏感、涉黄涉暴、违法不良、个人隐私数据或可能诱导违规的表述。 + +2. 简化或改写应用的提示词，减少可能触发安全拦截的表述。 + +3. 如问题仍未解决，收集完整报错信息（HTTP 状态码、错误码、错误消息），提交工单申请加白，具体流程见下方[内容加白申请流程](#section-whitelist-apply)。 + + +## 内容加白申请流程 + +如果您的业务场景需要豁免特定内容的安全拦截，可通过提交工单申请内容加白。安全部门审核通过后，指定内容将被添加至豁免名单。 + +申请加白前，请准备以下信息： + +- HTTP 状态码（如 `400`） + +- 错误码（如 `DataInspectionFailed` 或 `data_inspection_failed`） + +- 完整错误消息（如 `Input data may contain inappropriate content.`） + + +**申请步骤**： + +1. 收集上述完整报错信息及问题复现场景描述（包括知识库 ID、调用的模型名称、触发拦截的操作）。 + +2. 提交工单，在工单中提供报错信息、使用场景及需要加白的具体内容。 + +3. 等待安全部门完成审核，审核通过后加白处理生效。 + + ## 计费说明 在百炼控制台开通 AI 安全护栏产品的 SLR 授权，并通过百炼配置启用该产品策略后，系统将根据实际调用量计费。计费方式为按 Token 数量后付费，每日费用按当日实际使用量结算；未调用服务时不产生费用。计费规则详见[计费概述](https://help.aliyun.com/zh/document_detail/2872706.html#0529545b91g7c)。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/security-and-compliance/permission-management-overview.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/security-and-compliance/permission-management-overview.md index e0f5e9f2..9c33367e 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/security-and-compliance/permission-management-overview.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/security-and-compliance/permission-management-overview.md @@ -125,7 +125,7 @@ API Key 管理 ## **业务空间权限管理** -百炼按地理区域划分资源和业务空间，**单个业务空间不能跨地域存在。即使各个地域的默认业务空间，也是不同的空间**。点击前往全局管理菜单（[北京](https://bailian.console.aliyun.com/?tab=globalset#/efm/business_management) | [新加坡](https://modelstudio.console.aliyun.com/?tab=globalset#/efm/business_management) | [弗吉尼亚](https://modelstudio.console.aliyun.com/us-east-1?tab=globalset#/efm/business_management) | [法兰克福](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management)）。 +百炼按地理区域划分资源和业务空间，**单个业务空间不能跨地域存在。即使各个地域的默认业务空间，也是不同的空间**。单击前往全局管理菜单（[北京](https://bailian.console.aliyun.com/?tab=globalset#/efm/business_management) | [新加坡](https://modelstudio.console.aliyun.com/?tab=globalset#/efm/business_management) | [弗吉尼亚](https://modelstudio.console.aliyun.com/us-east-1?tab=globalset#/efm/business_management) | [法兰克福](https://modelstudio.console.aliyun.com/eu-central-1?tab=globalset#/efm/business_management)）。 同时百炼的业务空间是进行**精细化权限管理**的**最小**管理单元，它可管理： @@ -205,7 +205,7 @@ API Key 管理 > 默认业务空间无法设置此限制，所有模型均可调用，且无法限流。 - 在**模型列表**中，通过**模型调用**列的开关控制模型授权状态，并在**当前空间限流**列分别设置请求数限流值与Token限流值及对应的时间单位。 + 在**模型列表**中，通过**模型调用**列的开关控制模型授权状态，并在**当前空间限流**列分别设置请求数限流值与Token限流值及对应的时间单位（支持 1 秒、2 秒、3 秒、5 秒、10 秒、30 秒、60 秒）。未设置限流时，该空间将共享账号级别的总限流配额。 - **限制模型训练**：管理某个模型可否在该业务空间进行调优（通过控制台和API）和调优后部署**。** @@ -219,14 +219,14 @@ API Key 管理 在**模型列表**页签中，**模型授权**区域的**模型部署**列显示各模型的部署授权状态，通过 toggle 开关可将状态切换为**已授权**或**未授权**。 -- **用户（账号）控制台权限管理**：管理某个 RAM 用户是否能使用该业务空间**控制台**的功能以及能使用该业务空间控制台的哪些功能。但无法限制归属该用户的 API Key 的调用。 +- **用户（账号）控制台权限管理**：管理某个 RAM 用户是否能使用该业务空间**控制台**的功能以及能使用该业务空间控制台的哪些功能。控制台页面权限不影响该用户通过 API Key 调用模型（API Key 的权限由所属业务空间的模型授权决定）。 > 阿里云主账号无须设置，可以访问所有业务空间的所有页面。 在**编辑权限**弹窗中，选择**模型**页签，在权限列表中找到并勾选**模型体验-操作**权限项。左侧导航菜单中的**模型体验**与权限列表中的**模型体验-操作**相对应。 -### **API-Key 权限** +### **API Key 权限** 单个 API Key 只能归属一个地域内的一个业务空间和一个用户，且不能转移给其他业务空间或其他用户。API Key 的可调用的功能和模型限流与**归属业务空间**的权限保持一致**，**不受**用户（账号）控制台权限管理**的影响，也无需为不同模型（如文生文、文生图、语音合成）创建不同的API Key。 @@ -268,9 +268,9 @@ API Key 的状态随归属用户（账号）操作的变化： **华北2（北京）**地域的 API Key 支持设置。 -**管理 API-Key**：可以通过百炼控制台**左侧导航栏**中的**权限管理**页签内，为 RAM 用户添加 API-Key 权限。赋予对应 RAM 用户**创建、删除、查看该空间下所有 API-Key** 的权限。 +**管理 API Key**：可以通过百炼控制台**左侧导航栏**中的**权限管理**页签内，为 RAM 用户添加 API Key 权限。赋予对应 RAM 用户**创建、删除、查看该空间下所有 API Key** 的权限。 -在**编辑权限**弹窗中，切换到**其他**页签，勾选**API-Key**即可。 +在**编辑权限**弹窗中，切换到**其他**页签，勾选**API Key**即可。 ### **OpenAPI 接口权限** @@ -280,7 +280,7 @@ RAM 用户默认无权调用百炼**应用**的数据、知识库、Prompt工程 - [AliyunBailianDataFullAccess](https://help.aliyun.com/zh/ram/developer-reference/aliyunbailiandatafullaccess)：可调用百炼应用 [API目录](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-dir/)下的所有API。 -- [AliyunBailianDataReadOnlyAccess](https://help.aliyun.com/zh/ram/developer-reference/aliyunbailiandatareadonlyaccess)：可调用百炼应用 [API目录](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-dir/)下的**只读类**API，例如[DescribeFile - 查询文件状态](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-describefile)、[GetIndexJobStatus - 查询知识库创建任务状态](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getindexjobstatus)等。 +- [AliyunBailianDataReadOnlyAccess](https://help.aliyun.com/zh/ram/developer-reference/aliyunbailiandatareadonlyaccess)：可调用百炼应用 [API目录](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-dir/)下的**只读类**API，例如[查询文件状态](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-describefile)、[查询知识库创建任务状态](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-getindexjobstatus)等。 ### **应用于生产环境** @@ -305,7 +305,7 @@ RAM 用户默认无权调用百炼**应用**的数据、知识库、Prompt工程 - 将主账号总配额按比例分配给各业务空间，并预留一部分作为缓冲，以应对突发流量。 - **示例：**账号总配额为 1000 QPM，分配方案如下： + **示例：**账号总配额为 1000 QPM，分配方案。 - `project-prod-workspace`: 600 QPM (60%) @@ -673,9 +673,9 @@ AliyunBSSOrderAccess 2. **批量推理-操作** 权限，用于支持 [批量推理](https://help.aliyun.com/zh/model-studio/batch-inference)功能。 - 3. **模型观测-操作** 权限，用于查看模型调用、评测的 Token 消耗量。 + 3. **模型监控-操作** 权限，用于查看模型调用、评测的 Token 消耗量。 -3. **若需要通过百炼的 API 调用，**需要为 RAM 用户在对应业务空间创建或分配 API Key，更多细节请参考本文的：[API-Key 权限](#f2704153a055r)。（需要超级管理员或业务空间管理员操作） +3. **若需要通过百炼的 API 调用，**需要为 RAM 用户在对应业务空间创建或分配 API Key，更多细节请参考本文的：[API Key 权限](#f2704153a055r)。（需要超级管理员或业务空间管理员操作） ### **设置控制台模型调优权限** @@ -696,14 +696,14 @@ AliyunBSSOrderAccess 6. **数据管理-操作** 权限，用于管理调优数据集。 - 7. **模型观测-操作** 权限，用于查看模型调用、评测的 Token 消耗量。 + 7. **模型监控-操作** 权限，用于查看模型调用、评测的 Token 消耗量。 ### **设置 API 模型调优权限** 1. 若不使用**默认业务空间**，需保证业务空间为特定模型开通了[模型调优（训练）](#c180b853793v3)权限。（需要超级管理员操作） -2. 为 RAM 用户在对应业务空间创建或分配 API Key，更多细节请参考本文的：[API-Key 权限](#f2704153a055r)。（需要超级管理员或业务空间管理员操作） +2. 为 RAM 用户在对应业务空间创建或分配 API Key，更多细节请参考本文的：[API Key 权限](#f2704153a055r)。（需要超级管理员或业务空间管理员操作） ## **常见问题** @@ -714,8 +714,27 @@ AliyunBSSOrderAccess ### **2\. 如何使用子业务空间调用模型？** -无需特殊设置，使用子业务空间的 API-Key 即可。 +无需特殊设置，使用子业务空间的 API Key 即可。 ### **3\. 如何使用特定业务空间的应用？** 使用 API 管理、调用特定业务空间的应用需要同时设置 [APP ID 和 Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id)。 + +### **4\. 已配置 AliyunBailianFullAccess 但仍无法管理业务空间怎么办？** + +RAM 权限（如 AliyunBailianFullAccess）与百炼平台内部的业务空间级别权限是两套独立的权限体系。RAM 权限仅控制对百炼控制台的访问权限，而业务空间内的管理权限（如编辑空间、管理成员等）需要由主账号或超级管理员在百炼控制台中单独授权。 + +如果子账号已拥有 AliyunBailianFullAccess 权限但仍提示无权限操作业务空间，请按以下步骤处理： + +1. 使用主账号登录百炼控制台。 + +2. 单击右上角设置图标，在左侧导航栏中选择**账号管理**。 + +3. 找到需要授权的子账号，单击**权限管理**。 + +4. 为该子账号添加对应业务空间的管理权限。 + + +### **5\. 如何删除业务空间？** + +访问[阿里云百炼控制台](https://bailian.console.aliyun.com/cn-beijing?spm=5176.29619931.J__Z58Z6CX7MY__Ll8p1ZOR.1.7dd7521cmX1pAh&tab=model#/model-market)，在页面右上角选择目标地域，进入[业务空间管理](https://bailian.console.aliyun.com/?tab=globalset#/efm/business_management)页面，在操作列选择删除，**删除后不可恢复**。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/after-sales-service-scope.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/after-sales-service-scope.md new file mode 100644 index 00000000..54844525 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/after-sales-service-scope.md @@ -0,0 +1,57 @@ +# 阿里云百炼平台售后服务范围说明 + +## **阿里云百炼平台售后服务范围说明** + +欢迎您使用阿里云百炼。本《**阿里云百炼平台售后服务范围说明》是对您使用阿里云百炼相关产品和服务时适用的售后服务范围的说明。** + +1\. 在您购买的服务期限内，我们将为您提供如下售后基础服务，即通过官网、电话及阿里云APP提供7×24的电话咨询（95187、400电话）、智能在线和标准工单支持。支持范围包括： + +（1）关于阿里云百炼模型服务与产品功能、架构的咨询； + +（2）使用、配置阿里云百炼模型服务的最佳实践； + +（3）阿里云百炼模型服务的使用咨询、技术问题及故障诊断； + +（4）阿里云百炼API及阿里云百炼官方SDK问题的故障诊断； + +（5）与阿里云百炼管理控制台相关的问题； + +（6）与阿里云相关的账号问题咨询支持； + +（7）与阿里云相关的财务、合同及计费问题的咨询支持。 + +2\. 阿里云百炼将以阿里云官网页面公布的[客户服务权益](https://www.aliyun.com/service/customer-service-benefits?spm=5176.support-home.J_3451238410.1.12d1156fPBBxO0)标准向您提供相应的售后服务支持。 + +3\. 阿里云百炼同时提供付费版的售后增值服务（包括支持计划等），该等服务需在阿里云官网订购后生效使用。您还可通过阿里云获得其他付费的售后服务，具体详见阿里云的网站相关页面的收费售后服务内容。如您的项目需要更深度的技术支持（如业务代码编写指导、定制化集成方案等），建议联系阿里云商务经理沟通定制化服务方案。 + +4\. 为了方便您的生产或使用，如您选择将阿里云百炼服务与外部（非阿里云百炼平台上的）第三方工具或产品进行对接，阿里云将尽商业上合理的努力为您提供第三方工具在接入阿里云百炼模型推理服务过程中的方向性建议，但针对非阿里云百炼服务相关的问题，我们无法提供专业意见。 + +（1）我们提供建议的范围包括： + +（i）确认阿里云百炼API接口及服务端的可用状态； + +（ii）阿里云百炼官方API调用示例及SDK使用说明参考； + +（iii）协助核查阿里云百炼服务端调用明细和计费记录； + +（iv）基本的连通性测试建议（如通过curl等标准工具测试阿里云百炼服务地址的可达性）。 + +**（2）我们提供建议的范围不包括：** + +（i）第三方工具（如Cursor、Windsurf、Cline、OpenClaw等）的安装、部署、配置、升级及日常使用指导； + +（ii）第三方工具的产品功能、交互逻辑及内部实现问题的排查； + +（iii）其他云厂商、企业或社区提供的产品或服务的配置与运维； + +（iv）用户业务代码的编写、调试与实现； + +（v）用户本地环境（含内网、代理、VPN、防火墙、操作系统等）导致的连通性或兼容性问题的排查； + +（vi）第三方工具内部显示的Token数量、费用预估值或调用统计与阿里云计费数据之间的差异解释； + +（vii）所有第三方工具的安装、补丁更新、测试、故障诊断、优化等日常运维服务； + +（viii）基于阿里云百炼模型服务原生能力之上的第三方自建业务相关支持。 + +5\. 但请您注意，除我们另有书面说明外，第三方工具不构成我们的代理、受托或联合服务主体，**我们不对外部第三方工具的任何陈述、承诺或行为承担责任。**您知悉并确认，阿里云仅负责阿里云百炼平台自身的运营维护，即百炼服务端的技术架构、API接口、计量计费系统、控制台功能等；**第三方工具的运行维护（如AI编程工具的安装配置、开源代理框架的部署调优等）由您及相应工具提供方负责。**当您的问题出现在百炼模型服务的使用过程中，但其原因、责任范围或依赖关系已超出阿里云百炼平台本身可直接提供支持和保障的范围时——通常涉及您侧系统、外部第三方服务、网络环境、账号权限、业务流程或非标集成，阿里云将协助进行初步排查。若经排查确认问题来源于非阿里云侧，阿里云将给予方向性建议并引导您联系相应的服务主体。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/faq-about-alibaba-cloud-model-studio.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/faq-about-alibaba-cloud-model-studio.md index fff36d4d..53d67f33 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/faq-about-alibaba-cloud-model-studio.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/faq-about-alibaba-cloud-model-studio.md @@ -208,6 +208,10 @@ 等待时间取决于您的具体限流值（RPS/RPM）。例如，如果您的限流是120 RPM（每分钟查询数），即每秒2次请求。如果您在0.2秒内连续提交了2次请求，第3次请求就会被限流，您需要等待大约0.8秒后才能再次成功提交。 +17. **qwen-plus-latest 这个模型具体对应哪个系列？是 Qwen3.7 还是 Qwen3.5？** + + qwen-plus-latest 是 qwen-plus 的最新版本，属于 Qwen3 系列，而非 Qwen3.5 或 Qwen3.7 系列。另外 Qwen3.5、Qwen3.7 等是独立的模型系列，与 Qwen3 系列并列，并非 Qwen3 的子版本。 + ## **模型幻觉问题** diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-h3-1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-h3-1.md new file mode 100644 index 00000000..90ca2526 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-h3-1.md @@ -0,0 +1,207 @@ +# Tripo/Tripo-H3.1 + +Tripo H3.1 是 Tripo 推出的高精度 3D 生成模型，专为需要极致视觉质量与细节表现的创作者设计。模型通过核心算法升级与模块优化，参数规模达 200 亿级，支持十亿体素级三维分辨率与最高 200 万面多边形生成。在保持高精度几何与真实纹理的同时，Tripo H3.1 对输入参考图的还原度与对齐度进一步提升，在角色形体、面部细节与几何文字等复杂结构上实现更稳定、细致的表达，适用于高质量视觉制作与 3D 打印等高精度资产生产场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**3D-Generation** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +文生3D（标准版+无贴图） + +0.7 + +每次 + +单图生3D（标准版+无贴图） + +1.4 + +每次 + +多图生3D（标准版+无贴图） + +1.4 + +每次 + +文生3D（标准版+带标清贴图） + +1.4 + +每次 + +单图生3D（标准版+带标清贴图） + +2.1 + +每次 + +多图生3D（标准版+带标清贴图） + +2.1 + +每次 + +文生3D（标准版+带高清贴图） + +2.1 + +每次 + +单图生3D（标准版+带高清贴图） + +2.8 + +每次 + +多图生3D（标准版+带高清贴图） + +2.8 + +每次 + +文生3D（超清版+无贴图） + +2.1 + +每次 + +单图生3D（超清版+无贴图） + +2.8 + +每次 + +多图生3D（超清版+无贴图） + +2.8 + +每次 + +文生3D（超清版+带标清贴图） + +2.8 + +每次 + +单图生3D（超清版+带标清贴图） + +3.5 + +每次 + +多图生3D（超清版+带标清贴图） + +3.5 + +每次 + +文生3D（超清版+带高清贴图） + +3.5 + +每次 + +单图生3D（超清版+带高清贴图） + +4.2 + +每次 + +多图生3D（超清版+带高清贴图） + +4.2 + +每次 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +5 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-p1-0.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-p1-0.md new file mode 100644 index 00000000..25a22d33 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-3d-generation/tripo-p1-0.md @@ -0,0 +1,153 @@ +# Tripo/Tripo-P1.0 + +Tripo P1.0 是面向实时应用与生产管线的 3D 生成模型，专为需要干净拓扑和引擎可用网格的开发者与创作者设计。模型可在约 2 秒内生成具备专业级拓扑结构的 3D 资产，适用于游戏、Web3D 与各类实时交互场景。针对 UGC 内容生产中对“速度”和“开箱即用”的需求，Tripo P1.0 在保证质量的同时大幅提升生成效率，使资产能够快速接入实时引擎与开发流程。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**3D-Generation** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +文生3D（无贴图） + +2.1 + +每次 + +单图生3D（无贴图） + +2.8 + +每次 + +多图生3D（无贴图） + +2.8 + +每次 + +文生3D（带标清贴图） + +2.8 + +每次 + +单图生3D（带标清贴图） + +3.5 + +每次 + +多图生3D（带标清贴图） + +3.5 + +每次 + +文生3D（带高清贴图） + +3.5 + +每次 + +单图生3D（带高清贴图） + +4.2 + +每次 + +多图生3D（带高清贴图） + +4.2 + +每次 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +5 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/gte-rerank-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/gte-rerank-v2.md new file mode 100644 index 00000000..22abd779 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/gte-rerank-v2.md @@ -0,0 +1,109 @@ +# gte-rerank-v2 + +gte-rerank-v2是通义实验室研发的多语言文本统一排序模型，面向全球多个主流语种，提供高水平的文本排序服务。通常用于语义检索、RAG等场景，可以简单、有效地提升文本检索的效果。给定查询 (Query) 和一系列候选文本 (documents)，模型会根据与查询的语义相关性从高到低对候选文本进行排序。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30000 + +最大输出长度 + +0 + +上下文长度 + +30000 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +文本输入 + +0.8 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +5040 + +TPM（每分钟tokens） + +4,980,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/multimodal-embedding-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/multimodal-embedding-v1.md new file mode 100644 index 00000000..fa0fcc79 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/multimodal-embedding-v1.md @@ -0,0 +1,115 @@ +# multimodal-embedding-v1 + +通义实验室基于预训练多模态大模型构建的多模态向量模型。该模型根据用户的输入生成高维连续向量，这些输入可以是文本、图片或视频。多模态向量在可应用于图片搜索、文搜图、视频搜索、图片分类和视频内容审核等下游任务中。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +512 + +最大输出长度 + +0 + +上下文长度 + +0 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片输入 + +0.9 + +每百万tokens + +文本输入 + +0.7 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen2-5-vl-embedding.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen2-5-vl-embedding.md new file mode 100644 index 00000000..368c9f7f --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen2-5-vl-embedding.md @@ -0,0 +1,115 @@ +# qwen2.5-vl-embedding + +基于Qwen2.5-VL底座训练的统一多模态向量模型，支持文本、图片、视频单模态/混合模态输入，输出统一表征向量，适用于跨模态检索、图搜、视频检索、图像聚类、复杂多模态信息检索、打标等场景 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片输入 + +1.8 + +每百万tokens + +文本输入 + +0.7 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +TPM（每分钟tokens） + +600,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-7-text-embedding.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-7-text-embedding.md new file mode 100644 index 00000000..38595a34 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-7-text-embedding.md @@ -0,0 +1,109 @@ +# qwen3.7-text-embedding + +是通义实验室基于Qwen3.7训练的多语言文本统一向量模型，相较text-embedding-v4版本在文本检索、聚类、分类性能大幅提升；在MTEB多语言、中英、Code检索等评测任务上效果提升20%；支持256~2560维用户自定义向量维度。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +131072 + +最大输出长度 + +— + +上下文长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +文本输入 + +0.5 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +24000 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-rerank.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-rerank.md new file mode 100644 index 00000000..d0f6e41d --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-rerank.md @@ -0,0 +1,141 @@ +# qwen3-rerank + +基于Qwen LLM底座训练的文本排序模型，对输入的Query和候选Docs进行相关性排序，支持100+语种和长文本输入，适用于文本检索、RAG等场景，效果对齐开源Qwen3-Rerank系列模型 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30000 + +最大输出长度 + +0 + +上下文长度 + +30000 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +文本输入 + +0.5 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +文本输入 + +0.749 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +5400 + +TPM（每分钟tokens） + +5,000,000,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +5400 + +TPM（每分钟tokens） + +5,000,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-embedding.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-embedding.md new file mode 100644 index 00000000..dd65f4d8 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-embedding.md @@ -0,0 +1,115 @@ +# qwen3-vl-embedding + +基于Qwen3-VL底座训练的统一多模态向量模型，支持文本、图片、视频单模态/混合模态输入，输出统一表征向量，适用于跨模态检索、图搜、视频检索、图像聚类、复杂多模态信息检索、打标等场景 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**—** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +32000 + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片输入 + +1.8 + +每百万tokens + +文本输入 + +0.7 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +2400 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-rerank.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-rerank.md new file mode 100644 index 00000000..06444037 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/qwen3-vl-rerank.md @@ -0,0 +1,115 @@ +# qwen3-vl-rerank + +Qwen3-VL-Rerank重排模型，它能够深入理解文本、图片、视频的丰富多模态信息。在初步检索获得结果后，Qwen3-VL-Rerank 能够运用其先进的跨模态关联能力，对候选项目进行智能化的二次排序，将最相关的结果置于显要位置。通用用于提升跨模态搜索的准确率、优化图搜和视频检索的精准度、辅助图像聚类的分组质量、以及实现复杂多模态信息的高效检索和精确打标。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +120000 + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片输入 + +1.8 + +每百万tokens + +文本输入 + +0.7 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +TPM（每分钟tokens） + +9,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v1.md new file mode 100644 index 00000000..91fac5d0 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v1.md @@ -0,0 +1,93 @@ +# text-embedding-async-v1 + +通用文本向量的批处理接口，通过这个接口客户可以以文本方式一次性的提交大批量的向量计算请求，在系统完成所有的计算之后，大模型服务平台会将结果信息存储在结果文件中供客户下载解析。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +文本输入 + +0.7 + +每百万tokens diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v2.md new file mode 100644 index 00000000..7a9d3d98 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-async-v2.md @@ -0,0 +1,93 @@ +# text-embedding-async-v2 + +通用文本向量的批处理接口，通过这个接口客户可以以文本方式一次性的提交大批量的向量计算请求，在系统完成所有的计算之后，大模型服务平台会将结果信息存储在结果文件中供客户下载解析。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +文本输入 + +0.7 + +每百万tokens diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v1.md new file mode 100644 index 00000000..8e422ae0 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v1.md @@ -0,0 +1,115 @@ +# text-embedding-v1 + +通用文本向量，是通义实验室基于LLM底座的多语言文本统一向量模型，面向全球多个主流语种，提供高水准的向量服务，帮助开发者将文本数据快速转换为高质量的向量数据。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +向量输入（Batch File） + +0.35 + +每百万tokens + +文本输入 + +0.7 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1800 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v2.md new file mode 100644 index 00000000..fbe32319 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v2.md @@ -0,0 +1,115 @@ +# text-embedding-v2 + +通用文本向量，是通义实验室基于LLM底座的多语言文本统一向量模型，面向全球多个主流语种，提供高水准的向量服务，帮助开发者将文本数据快速转换为高质量的向量数据。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +向量输入（Batch File） + +0.35 + +每百万tokens + +文本输入 + +0.7 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1800 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v3.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v3.md new file mode 100644 index 00000000..e6292896 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v3.md @@ -0,0 +1,147 @@ +# text-embedding-v3 + +通用文本向量，是通义实验室基于LLM底座的多语言文本统一向量模型，面向全球多个主流语种，提供高水准的向量服务，帮助开发者将文本数据快速转换为高质量的向量数据。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +向量输入（Batch File） + +0.25 + +每百万tokens + +文本输入 + +0.5 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +文本输入 + +0.514 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1800 + +TPM（每分钟tokens） + +1,200,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +6000 + +TPM（每分钟tokens） + +24,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v4.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v4.md new file mode 100644 index 00000000..a405f587 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/text-embedding-v4.md @@ -0,0 +1,147 @@ +# text-embedding-v4 + +通用文本向量V4版本，是通义实验室基于Qwen3训练的多语言文本统一向量模型，相较V3版本在文本检索、聚类、分类性能大幅提升；在MTEB多语言、中英、Code检索等评测任务上效果提升15%~40%；支持64~2048维用户自定义向量维度。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +向量输入（Batch File） + +0.25 + +每百万tokens + +文本输入 + +0.5 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +文本输入 + +0.514 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1800 + +TPM（每分钟tokens） + +1,200,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +1800 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-flash.md new file mode 100644 index 00000000..a4f2ae78 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-flash.md @@ -0,0 +1,271 @@ +# tongyi-embedding-vision-flash + +Embedding-Vision是基于LLM底座的视觉多模态表征模型，具有以视觉为中心、领域性能优异（电商、 安防、相册/图库、自驾等）、高性价比的特点。兼容文本、图像、视频3种模态，可应用于以图搜图、以文搜图、以文搜视频，以视频搜视频等下游任务场景。本模型（tongyi-embedding-vision-flash）是轻量化版本，在视觉向量化上具备极高性价比。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片输入 + +0.15 + +每百万tokens + +文本输入 + +0.15 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片输入 + +0.2 + +每百万tokens + +文本输入 + +0.6 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +TPM（每分钟tokens） + +200,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +600 + +TPM（每分钟tokens） + +200,000 + +## 快照版本 + +### tongyi-embedding-vision-flash-2026-03-06 + +Tongyi-Embedding-Vision是基于LLM底座的视觉多模态表征模型，支持文本、图像、视频3种模态，具有以视觉为中心、全场景性能优异、高性价比的特点，适用于以图搜图、以文搜图、以文搜视频、以视频搜视频、以文搜文、以文搜图文等下游多样化任务场景。本模型（tongyi-embedding-vision-flash）是轻量化版本，具备极高性价比。 2026-03-06版本在保留极致性价比优势的同时，基于Qwen3底座实现了效果与功能全面升级，包括全场景性能提升、多分辨率模式/多向量维度/多语言能力/融合向量等能力的支持。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片输入 + +0.15 + +每百万tokens + +文本输入 + +0.15 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +TPM（每分钟tokens） + +9,600,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-plus.md new file mode 100644 index 00000000..620504a1 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-embedding-reranking/tongyi-embedding-vision-plus.md @@ -0,0 +1,271 @@ +# tongyi-embedding-vision-plus + +Embedding-Vision是基于LLM底座的视觉多模态表征模型，具有以视觉为中心、领域性能优异（电商、 安防、相册/图库、自驾等）、高性价比的特点。兼容文本、图像、视频3种模态，可应用于以图搜图、以文搜图、以文搜视频，以视频搜视频等下游任务场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片输入 + +0.5 + +每百万tokens + +文本输入 + +0.5 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片输入 + +0.6 + +每百万tokens + +文本输入 + +0.6 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +TPM（每分钟tokens） + +200,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +600 + +TPM（每分钟tokens） + +200,000 + +## 快照版本 + +### tongyi-embedding-vision-plus-2026-03-06 + +Tongyi-Embedding-Vision是基于LLM底座的视觉多模态表征模型，支持文本、图像、视频3种模态，具有以视觉为中心、全场景性能优异、高性价比的特点，适用于以图搜图、以文搜图、以文搜视频、以视频搜视频、以文搜文、以文搜图文等下游多样化任务场景。 2026-03-06版本在保留极致性价比优势的同时，基于Qwen3底座实现了效果与功能全面升级，包括全场景性能提升、多分辨率模式/多向量维度/多语言能力/融合向量等能力的支持。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片输入 + +0.5 + +每百万tokens + +文本输入 + +0.5 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +TPM（每分钟tokens） + +9,600,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-parsing-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-parsing-v1.md new file mode 100644 index 00000000..187a3545 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-parsing-v1.md @@ -0,0 +1,93 @@ +# aitryon-parsing-v1 + +图片分割模型是AI试衣OutfitAnyone的辅助模型，可对模特图、服饰图进行分割，用于试衣图片的前后处理。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片检测 + +0.004 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-plus.md new file mode 100644 index 00000000..3e62e864 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-plus.md @@ -0,0 +1,93 @@ +# aitryon-plus + +aitryon-plus是一款效果出众的虚拟试衣图片生成模型，可基于服饰平拍图片以及人物正面全身照，输出服饰的人物试衣效果图片。 相较于aitryon模型，aitryon-plus模型在图片清晰度、服饰纹理细节和logo还原效果等方面均有提升，但生成耗时较长，适用于对时效性要求不高的场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-refiner.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-refiner.md new file mode 100644 index 00000000..0a12c174 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon-refiner.md @@ -0,0 +1,179 @@ +# aitryon-refiner + +图片精修是对AI试衣生成的效果图进行二次生成，输出还原度更高的精修试衣效果图。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +**图片生成数量<=25** + +计费项 + +价格（元） + +单位 + +图片生成 + +0.3 + +每张 + +**25<图片生成数量<=125** + +计费项 + +价格（元） + +单位 + +图片生成 + +0.275 + +每张 + +**125<图片生成数量<=250** + +计费项 + +价格（元） + +单位 + +图片生成 + +0.25 + +每张 + +**250<图片生成数量<=1250** + +计费项 + +价格（元） + +单位 + +图片生成 + +0.225 + +每张 + +**1250<图片生成数量<=2500** + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +**2500<图片生成数量<=25000** + +计费项 + +价格（元） + +单位 + +图片生成 + +0.175 + +每张 + +**25000<图片生成数量** + +计费项 + +价格（元） + +单位 + +图片生成 + +0.15 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon.md new file mode 100644 index 00000000..d2f7596c --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/aitryon.md @@ -0,0 +1,93 @@ +# aitryon + +aitryon是一款性能出众的虚拟试衣图片生成模型，可基于服饰平拍图片以及人物正面全身照，输出服饰的人物试衣效果图片。aitryon模型可在较短时间内生成试衣图片，适用于对时效性要求较高的场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/facechain-facedetect.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/facechain-facedetect.md new file mode 100644 index 00000000..05289818 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/facechain-facedetect.md @@ -0,0 +1,79 @@ +# facechain-facedetect + +对用户上传的人物图像进行检测，判断其中所包含的人脸是否符合facechain微调所需的标准，检测维度包括人脸数量、大小、角度、光照、清晰度等多维度，支持图像组输入，并返回每张图像对应的检测结果。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**—** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-erase-completion.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-erase-completion.md new file mode 100644 index 00000000..9292ad58 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-erase-completion.md @@ -0,0 +1,79 @@ +# image-erase-completion + +图像擦除补全通过指定图像mask中要删除的人体、宠物、物品、文字、水印等图像区域，在保留背景的同时移除图像中的一个或多个人物、物体、文字等元素，此功能不支持输入prompt的消除。擦除补全技术结合了计算机视觉、AIGC inpainting等先进技术，可以在多种场景下应用，从而满足用户对隐私保护、内容创作和图像编辑等方面需求。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-out-painting.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-out-painting.md new file mode 100644 index 00000000..27667566 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/image-out-painting.md @@ -0,0 +1,79 @@ +# image-out-painting + +图像画面大模型，对输入图像进行画面自由扩展，支持旋转画面，支持按照扩展系数和扩展像素数两种方式进行扩图。用户可以通过指定宽度、高度画面扩展比例或者左、右、上、下的扩展的像素值来控制画面扩展，可用于创意娱乐、辅助作图、画面设计、影视后期制作等场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-image-generation.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-image-generation.md new file mode 100644 index 00000000..e65e80f1 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-image-generation.md @@ -0,0 +1,111 @@ +# kling/kling-v3-image-generation + +支持最多10张参考图，可锁定主体、元素和色调，保证风格一致。融合风格转绘、人像/角色参考、多图融合及局部重绘，操作灵活。人像细节真实，整体画面细腻丰富，色彩氛围兼具影视感。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Text** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成（1K） + +0.2 + +每张 + +图片生成（2K） + +0.2 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-omni-image-generation.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-omni-image-generation.md new file mode 100644 index 00000000..eaa4867e --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/kling-v3-omni-image-generation.md @@ -0,0 +1,117 @@ +# kling/kling-v3-omni-image-generation + +解锁影视级叙事画面，新增系列组图及2K/4K直出。深度解析提示词视听元素，精确响应创作指令。支持自由多参考图及全面效果升级，适合分镜、剧情概念图及场景设定。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Text** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成（1K） + +0.2 + +每张 + +图片生成（2K） + +0.2 + +每张 + +图片生成（4K） + +0.4 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-facechain-generation.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-facechain-generation.md new file mode 100644 index 00000000..9965c8a3 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-facechain-generation.md @@ -0,0 +1,93 @@ +# facechain-generation + +基于人物形象训练已经得到的形象，可以继续通过人物生成写真模型完成该形象的写真生成，支持多种预设风格，包括证件照、商务写真等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.18 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-image-instance-segmentation.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-image-instance-segmentation.md new file mode 100644 index 00000000..b6357bac --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-image-instance-segmentation.md @@ -0,0 +1,79 @@ +# image-instance-segmentation + +人物实例分割运用了检测和分割技术，不仅能够在图像中识别出不同的对象，而且还能准确地画出每一个对象边界的像素级掩码（mask）。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-qwen-image-edit.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-qwen-image-edit.md new file mode 100644 index 00000000..23603d8d --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/model-qwen-image-edit.md @@ -0,0 +1,133 @@ +# qwen-image-edit + +千问系列首个图像编辑模型，成功将Qwen-Image的文本渲染能力拓展到编辑任务上。支持精准的中英双语文字编辑、视觉外观与语义双重编辑、具备强大的跨基准性能表现。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.3 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.330266 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0-pro.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0-pro.md new file mode 100644 index 00000000..02dbc8a2 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0-pro.md @@ -0,0 +1,537 @@ +# qwen-image-2.0-pro + +Qwen-Image-2.0系列满血版模型，实现了图片生成和图片编辑的融合；具备更专业的文字渲染1k token指令支持能力、更细腻的真实质感，细腻刻画写实场景、更强的语义遵循能力。满血版具备2.0系列最强的文字渲染能力和真实质感。该模型版本功能等同于快照模型 qwen-image-2.0-pro-2026-04-22。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.550443 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +2 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +2 + +## 快照版本 + +### qwen-image-2.0-pro-2026-06-22 + +Qwen-Image-2.0系列满血版模型，实现了图片生成和图片编辑的融合；具备更专业的文字渲染1k token指令支持能力、更细腻的真实质感，细腻刻画写实场景、更强的语义遵循能力。满血版具备2.0系列最强的文字渲染能力和真实质感。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.550443 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +2 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +2 + +### qwen-image-2.0-pro-2026-04-22 + +Qwen-Image-2.0系列满血版模型，实现了图片生成和图片编辑的融合。相较于3月3日快照，本模型在画面质感，尤其是纹理细节、光影、材质上有明显跃升；支持多语言的图内文字生成；艺术风格表现更加均衡。该版本为2026年4月22日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.550443 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +2 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +2 + +### qwen-image-2.0-pro-2026-03-03 + +Qwen-Image-2.0系列满血版模型，实现了图片生成和图片编辑的融合；具备更专业的文字渲染1k token指令支持能力、更细腻的真实质感，细腻刻画写实场景、更强的语义遵循能力。满血版具备2.0系列最强的文字渲染能力和真实质感。该版本为2026年3月3日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.550443 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +2 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +2 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0.md new file mode 100644 index 00000000..e9323f34 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-2-0.md @@ -0,0 +1,269 @@ +# qwen-image-2.0 + +Qwen-Image-2.0系列加速版模型，实现了图片生成和图片编辑的融合；具备更专业的文字渲染1k token指令支持能力、更细腻的真实质感，细腻刻画写实场景、更强的语义遵循能力。加速版有效实现了模型效果和性能的最佳平衡。该模型版本功能等同于快照模型 qwen-image-2.0-2026-03-03。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.256873 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 快照版本 + +### qwen-image-2.0-2026-03-03 + +Qwen-Image-2.0系列加速版模型，实现了图片生成和图片编辑的融合；具备更专业的文字渲染1k token指令支持能力、更细腻的真实质感，细腻刻画写实场景、更强的语义遵循能力。加速版有效实现了模型效果和性能的最佳平衡。该版本为2026年3月3日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.256873 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-3-0-pro.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-3-0-pro.md new file mode 100644 index 00000000..9ba4fec6 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-3-0-pro.md @@ -0,0 +1,103 @@ +# qwen-image-3.0-pro + +内容丰实：支持最大 4.5k token 输入，支持图中图密集信息排版，让报纸、分镜、菜单、试卷等复杂版面一次生成。 细节真实：支持 10px 小字精准渲染，微表情、毛孔、发丝等细节生动还原，逼近真实摄影的质感。 知识厚实：支持 12 国语言、20+ 字体原生渲染，主流网页、游戏、直播等界面仿真，外部知识全纳入。 Qwen-Image-3.0-Pro 不只是在追求"好看"，更在追求“好用”——让图像生成真正成为可落地的生产力工具。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +1 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-max.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-max.md new file mode 100644 index 00000000..1105208a --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-max.md @@ -0,0 +1,269 @@ +# qwen-image-edit-max + +千问图像编辑模型Max系列，提供更稳定、更丰富的编辑能力：提升工业设计与几何推理能力；提升角色一致性；减轻偏移问题；集成Lora能力，可以进行更多功能的图像编辑。该模型版本功能等同于快照模型 qwen-image-edit-max-2026-01-16。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.550443 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +2 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +2 + +## 快照版本 + +### qwen-image-edit-max-2026-01-16 + +千问图像编辑模型Max系列，提供更稳定、更丰富的编辑能力：提升工业设计与几何推理能力；提升角色一致性；减轻偏移问题；集成Lora能力，可以进行更多功能的图像编辑。此版本为2026年1月16日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.550443 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +2 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +2 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-plus.md new file mode 100644 index 00000000..d24783b6 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-edit-plus.md @@ -0,0 +1,403 @@ +# qwen-image-edit-plus + +千问系列图像编辑Plus模型，在首版Edit模型基础上进一步优化了推理性能与系统稳定性，大幅缩短图像生成与编辑的响应时间；支持单次请求返回多张图片，显著提升用户体验。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 快照版本 + +### qwen-image-edit-plus-2025-12-15 + +千问系列图像编辑Plus模型，相较10月30日快照提升角色一致性、工业设计能力、几何推理能力；同时集成例如打光等Lora能力、减轻偏移问题。此版本为2025年12月15日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 + +### qwen-image-edit-plus-2025-10-30 + +千问系列图像编辑Plus模型，在首版Edit模型基础上进一步优化了推理性能与系统稳定性，大幅缩短图像生成与编辑的响应时间；支持单次请求返回多张图片，显著提升用户体验。此版本为2025年10月30日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-max.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-max.md new file mode 100644 index 00000000..ec275344 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-max.md @@ -0,0 +1,253 @@ +# qwen-image-max + +千问图像生成模型Max系列，在各类生成任务中表现出色，相较Plus系列大幅度降低生成图片的AI感，提升图像真实性；具备更真实的人物质感、更细腻的自然纹理、更美观的文字渲染。该模型版本功能等同于快照模型 qwen-image-max-2025-12-30。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.550443 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +2 + +TPM（每分钟tokens） + +1,000,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +2 + +TPM（每分钟tokens） + +1,000,000 + +## 快照版本 + +### qwen-image-max-2025-12-30 + +千问图像生成模型Max系列，在各类生成任务中表现出色，相较Plus系列大幅度降低生成图片的AI感，提升图像真实性；具备更真实的人物质感、更细腻的自然纹理、更美观的文字渲染。此版本为2025年12月30日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.550443 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-plus.md new file mode 100644 index 00000000..01173503 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image-plus.md @@ -0,0 +1,273 @@ +# qwen-image-plus + +千问系列图像生成模型，参数规模200亿。具备卓越的文本渲染能力，在复杂文本渲染、各类生成与编辑任务重表现出色，在多个公开基准测试中获得SOTA，模型性能大幅提升。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 快照版本 + +### qwen-image-plus-2026-01-09 + +千问系列图像生成模型，具备卓越的文本渲染能力，在复杂文本渲染、各类生成与编辑任务重表现出色。此版本为2026年1月9日快照，为Qwen-Image-Max的蒸馏加速版，可以更快速地生成高质量图片。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image.md new file mode 100644 index 00000000..3d2249d9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-image.md @@ -0,0 +1,137 @@ +# qwen-image + +千问系列首个图像生成模型，参数规模200亿。具备卓越的文本渲染能力，在复杂文本渲染、各类生成与编辑任务重表现出色，在多个公开基准测试中获得SOTA。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.25 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.256873 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +TPM（每分钟tokens） + +1,000,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-mt-image.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-mt-image.md new file mode 100644 index 00000000..90fdd451 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/qwen-mt-image.md @@ -0,0 +1,105 @@ +# qwen-mt-image + +专注做图片翻译的模型服务，能将中、英、日等11个语言的图片翻译到指定的语言，精准还原图片排版和内容信息，支持术语定义、敏感词过滤、商品主体检测等自定义功能，提供灵活、准确、高效的图像本地化服务。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.003 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/shoemodel-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/shoemodel-v1.md new file mode 100644 index 00000000..19782b35 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/shoemodel-v1.md @@ -0,0 +1,79 @@ +# shoemodel-v1 + +鞋靴模特支持输入多视角鞋靴系列图片，同时对输入模特模板图的鞋子区域进行鞋靴AI试穿，实现模特鞋靴布局重绘生成，最终生成图片的效果, 布局自然、细节丰富、画面细腻、试穿结果逼真。可用于模特商品图设计、新鞋AI试穿、模特穿戴布局重绘等场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/vidu-image-reference2image.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/vidu-image-reference2image.md new file mode 100644 index 00000000..ac70168e --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/vidu-image-reference2image.md @@ -0,0 +1,117 @@ +# vidu/vidu-image_reference2image + +输入0-14张参考图片或文本描述，支持参考生图、文生图、图片编辑，对中英文字的精准渲染、UI/图表等设计细节的像素级还原，适合制作海报、信息图等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成（1K） + +0.625 + +每张 + +图片生成（2K） + +1 + +每张 + +图片生成（4K） + +1.46875 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-fast-reference2image.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-fast-reference2image.md new file mode 100644 index 00000000..d78c99b7 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-fast-reference2image.md @@ -0,0 +1,105 @@ +# vidu/viduq2-fast_reference2image + +输入0-14张参考图片或文本描述，支持参考生图、文生图、图片编辑，语义理解能力大幅提升，支持更多风格。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成（1K） + +0.28125 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-pro-reference2image.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-pro-reference2image.md new file mode 100644 index 00000000..6114ecc1 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq2-pro-reference2image.md @@ -0,0 +1,117 @@ +# vidu/viduq2-pro_reference2image + +输入0-14张参考图片或文本描述，支持参考生图、文生图、图片编辑，擅长处理复杂逻辑，具备超强上下文一致性和工业级稳定性。适合专业设计、漫剧制作等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成（1K） + +0.9375 + +每张 + +图片生成（2K） + +0.9375 + +每张 + +图片生成（4K） + +1.71875 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq3-fast-reference2image.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq3-fast-reference2image.md new file mode 100644 index 00000000..ba54bf3e --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/viduq3-fast-reference2image.md @@ -0,0 +1,117 @@ +# vidu/viduq3-fast_reference2image + +输入0-14张参考图片或文本描述，支持参考生图、文生图、图片编辑，主打高速高质与低成本，成本比Pro降低约50%。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成（1K） + +0.46875 + +每张 + +图片生成（2K） + +0.78125 + +每张 + +图片生成（4K） + +1.09375 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/virtualmodel-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/virtualmodel-v2.md new file mode 100644 index 00000000..3137808f --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/virtualmodel-v2.md @@ -0,0 +1,79 @@ +# virtualmodel-v2 + +虚拟模特可以对上传的真人或者人台实拍商品展示图进行智能生成，将其中的模特和背景替换为心仪的内容，在保持人物姿态不变的情况下，使用虚拟模特对商品进行更加精美、多样的展示。支持各种与模特产生互动的商品，如手持小商品、服装、鞋靴、配饰等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-plus.md new file mode 100644 index 00000000..db20527f --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-plus.md @@ -0,0 +1,109 @@ +# wan2.1-t2i-plus + +Wan2.1 Text-to-Image Plus version, Generate more image details. Upgraded in image beauty, realism, and artistry. Stronger semantic understanding ability, rich style generalization ability, supports up to 2 million pixel generation, supports smart prompt rewriting. + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.366962 + +每张 + +## 限流 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-turbo.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-turbo.md new file mode 100644 index 00000000..0ce22ceb --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-1-t2i-turbo.md @@ -0,0 +1,109 @@ +# wan2.1-t2i-turbo + +Wan2.1 Text-to-Image Turbo version, faster generation speed. Upgraded in image beauty, realism, and artistry. Stronger semantic understanding ability, rich style generalization ability, supports up to 2 million pixel generation, supports smart prompt rewriting. + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.183481 + +每张 + +## 限流 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-flash.md new file mode 100644 index 00000000..47a662c8 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-flash.md @@ -0,0 +1,133 @@ +# wan2.2-t2i-flash + +全新升级的万相2.2文生图，更快的生成速度。在生成图像创意性、稳定性、写实质感方面全面升级，指令遵循更强，原生支持多种风格。支持最大200万像素生成，支持智能提示词改写等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.14 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.183481 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-plus.md new file mode 100644 index 00000000..1e6851e5 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-2-t2i-plus.md @@ -0,0 +1,133 @@ +# wan2.2-t2i-plus + +全新升级的万相2.2文生图，更丰富的画面细节。在生成图像创意性、稳定性、写实质感方面全面升级，指令遵循更强，原生支持多种风格。支持最大200万像素生成，支持智能提示词改写等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.366962 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-i2i.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-i2i.md new file mode 100644 index 00000000..a427eb0b --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-i2i.md @@ -0,0 +1,269 @@ +# wan2.5-i2i-preview + +万相2.5-图像编辑-Preview，全新升级模型架构。支持指令控制实现丰富的图像编辑能力，指令遵循能力进一步提升，支持高一致性保持的多图参考生成，文字生成表现优异。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 快照版本 + +### wan2.5-i2i-preview + +万相2.5-图像编辑-Preview，全新升级模型架构。支持指令控制实现丰富的图像编辑能力，指令遵循能力进一步提升，支持高一致性保持的多图参考生成，文字生成表现优异。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-t2i.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-t2i.md new file mode 100644 index 00000000..11aa6e7d --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-5-t2i.md @@ -0,0 +1,269 @@ +# wan2.5-t2i-preview + +万相2.5-文生图-Preview，全新升级模型架构。画面美学、设计感、真实质感显著提升，精准指令遵循，擅长中英文和小语种文字生成，支持复杂结构化长文本和图表、架构图等内容生成。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 快照版本 + +### wan2.5-t2i-preview + +万相2.5-文生图-Preview，全新升级模型架构。画面美学、设计感、真实质感显著提升，精准指令遵循，擅长中英文和小语种文字生成，支持复杂结构化长文本和图表、架构图等内容生成。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-image.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-image.md new file mode 100644 index 00000000..607786ff --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-image.md @@ -0,0 +1,189 @@ +# wan2.6-image + +万相2.6-图像生成，全能图像生成模型，支持图文一体化推理生成，具备多图创意融合、商用级一致性、美学要素迁移与镜头光影精确控制，全面提升图像生成的一致性、可控性和表现力。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image**　**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +## 德国（法兰克福） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 美国（弗吉尼亚） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 德国（法兰克福） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 美国（弗吉尼亚） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-t2i.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-t2i.md new file mode 100644 index 00000000..78736bec --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-6-t2i.md @@ -0,0 +1,189 @@ +# wan2.6-t2i + +万相2.6-文生图，画面质感、美学表现、指令遵循升级，在艺术风格精准控制、真实感人像、长文本生图及广泛历史文化IP覆盖上均表现出卓越能力，可生成高质量且富有表现力的视觉内容。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.220177 + +每张 + +## 德国（法兰克福） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 美国（弗吉尼亚） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 德国（法兰克福） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +## 美国（弗吉尼亚） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +60 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image-pro.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image-pro.md new file mode 100644 index 00000000..bdbe604e --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image-pro.md @@ -0,0 +1,133 @@ +# wan2.7-image-pro + +万相2.7-图像生成与编辑旗舰版模型，支持文生图、文生组图、图生组图、图像编辑、多图参考生成、交互式编辑，在文字渲染、主体一致性、复杂指令遵循上都有更强表现。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.5 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.562065 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image.md new file mode 100644 index 00000000..c105e3ed --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wan2-7-image.md @@ -0,0 +1,161 @@ +# wan2.7-image + +万相2.7-图像生成与编辑，支持文生图、文生组图、图生组图、图像编辑、多图参考生成、交互式编辑，在文字渲染、主体一致性、复杂指令遵循上都有更强表现 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.224826 + +每张 + +## 日本（东京） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +300 + +## 日本（东京） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +300 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-background-generation-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-background-generation-v2.md new file mode 100644 index 00000000..f7c4c8a0 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-background-generation-v2.md @@ -0,0 +1,93 @@ +# wanx-background-generation-v2 + +图像背景生成可以基于输入的前景图像素材拓展生成背景信息，实现自然的光影融合效果，与细腻的写实画面生成。支持文本描述、图像引导等多种方式，同时支持对生成的图像智能添加文字内容。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.08 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-poster-generation-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-poster-generation-v1.md new file mode 100644 index 00000000..418ae55c --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-poster-generation-v1.md @@ -0,0 +1,79 @@ +# wanx-poster-generation-v1 + +创意海报生成，您的创意海报魔法工厂！它能够根据你的要求自动生成海报的背景和文字排版，支持多种海报风格，从宣传到祝福，让每一张海报都成为你的个性宣言。无需设计基础，轻松制作出彩作品，让创意触手可及。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-sketch-to-image-lite.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-sketch-to-image-lite.md new file mode 100644 index 00000000..b5621c86 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-sketch-to-image-lite.md @@ -0,0 +1,93 @@ +# wanx-sketch-to-image-lite + +万相-涂鸦作画通过手绘任意内容加文字描述，即可生成精美的涂鸦绘画作品，作品中的内容在参考手绘线条的同时，兼顾创意性和趣味性。涂鸦作画支持扁平插画、油画、二次元、3D卡通和水彩5种风格，可用于创意娱乐、辅助设计、儿童教学等场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.06 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-style-repaint-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-style-repaint-v1.md new file mode 100644 index 00000000..e4b64c62 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-style-repaint-v1.md @@ -0,0 +1,93 @@ +# wanx-style-repaint-v1 + +人像风格重绘可以将输入的人物图像进行多种风格化的重绘生成，使新生成的图像在兼顾原始人物相貌的同时，带来不同风格的绘画效果。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.12 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-v1.md new file mode 100644 index 00000000..505665a9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-v1.md @@ -0,0 +1,93 @@ +# wanx-v1 + +万相-文本生成图像大模型，支持中英文双语输入，重点风格包括但不限于水彩、油画、中国画、素描、扁平插画、二次元、3D卡通 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.16 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-virtualmodel.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-virtualmodel.md new file mode 100644 index 00000000..76587b56 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-virtualmodel.md @@ -0,0 +1,79 @@ +# wanx-virtualmodel + +虚拟模特可以对上传的真人或者人台实拍商品展示图进行智能生成，将其中的模特和背景替换为心仪的内容，在保持人物姿态不变的情况下，使用虚拟模特对商品进行更加精美、多样的展示。支持各种与模特产生互动的商品，如手持小商品、服装、鞋靴、配饰等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-x-painting.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-x-painting.md new file mode 100644 index 00000000..0a5ebe7e --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx-x-painting.md @@ -0,0 +1,79 @@ +# wanx-x-painting + +万相-图像局部重绘是基于自研的Composer组合生成框架的AI绘画创作大模型后置处理链路，能够根据用户输入的原始图片和意涂抹图中局部区域和prompt提示词文字内容，生成符合语义描述的多样化风格的局部重绘图像。通过知识重组与可变维度扩散模型，加速收敛并提升最终生成图片的效果, 布局自然、细节丰富、画面细腻、结果逼真。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-0-t2i-turbo.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-0-t2i-turbo.md new file mode 100644 index 00000000..e70449c1 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-0-t2i-turbo.md @@ -0,0 +1,105 @@ +# wanx2.0-t2i-turbo + +Wan2.0-T2I-Turbo，更擅长质感人像和创意设计画作生成，在图像美观度、真实感、艺术性上全面升级，支持最大200万像素生成，支持智能提示词改写等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.04 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-imageedit.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-imageedit.md new file mode 100644 index 00000000..6a1133ee --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-imageedit.md @@ -0,0 +1,93 @@ +# wanx2.1-imageedit + +万相-通义图像编辑，支持预设编辑任务与指令式编辑，包含多种局部/全图编辑能力，如图像风格化、线稿生图、局部重绘、参考图生成、图像外扩、图像超分等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.14 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-plus.md new file mode 100644 index 00000000..0405cd39 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-plus.md @@ -0,0 +1,105 @@ +# wanx2.1-t2i-plus + +万相2.1-文生图-Plus，更丰富的画面细节，在图像美观度、真实感、艺术性上全面升级，更强的语义理解能力、丰富的风格泛化性、支持最大200万像素生成，支持智能提示词改写等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.2 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-turbo.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-turbo.md new file mode 100644 index 00000000..a73233ff --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wanx2-1-t2i-turbo.md @@ -0,0 +1,105 @@ +# wanx2.1-t2i-turbo + +万相2.1-文生图-Turbo，更快的生成速度，在图像美观度、真实感、艺术性上全面升级，更强的语义理解能力、丰富的风格泛化性、支持最大200万像素生成，支持智能提示词改写等。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.14 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-semantic.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-semantic.md new file mode 100644 index 00000000..d499104c --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-semantic.md @@ -0,0 +1,93 @@ +# wordart-semantic + +WordArt锦书-文字变形可以对输入的文字边缘轮廓进行创意变形，根据提示词内容进行边缘变化，实现一种字体的更多种创意用法，返回带有文字内容的黑底白色mask图。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.24 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-texture.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-texture.md new file mode 100644 index 00000000..67bfaace --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/wordart-texture.md @@ -0,0 +1,93 @@ +# wordart-texture + +WordArt锦书-文字纹理生成可以对输入的文字内容或文字图片进行创意设计，根据提示词内容对文字添加材质和纹理，实现立体凸显或场景融合的效果，生成效果精美、风格多样的艺术字，结合背景可以直接作为文字海报使用。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Image** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成 + +0.08 + +每张 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/z-image-turbo.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/z-image-turbo.md new file mode 100644 index 00000000..2c4d6679 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-image-generation/z-image-turbo.md @@ -0,0 +1,145 @@ +# z-image-turbo + +Z-Image-Turbo是在Artificial Analysis评测中荣登文生图开源模型世界第一的高效图像生成模型,仅用60亿参数和8步推理就能生成媲美大规模商业模型的照片级真实感图像,并在中英双语文本渲染、复杂语义理解和多样化主题生成上表现卓越。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text**　**Image** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +图片生成（标准） + +0.1 + +每张 + +图片生成（思考） + +0.2 + +每张 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +图片生成（标准） + +0.220177 + +每张 + +图片生成（思考） + +0.110089 + +每张 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +120 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo-realtime.md new file mode 100644 index 00000000..d172c6a7 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo-realtime.md @@ -0,0 +1,609 @@ +# qwen-omni-turbo-realtime + +千问全新多模态理解生成大模型实时版，适合实时音频交互场景。支持音频伴随文本、图像、视频混合输入理解，具备语音和文本同时流式生成能力，提供了4种自然对话音色。该模型版本功能等同于快照模型 qwen-omni-turbo-realtime-2025-05-08。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30720 + +最大输出长度 + +2048 + +上下文长度 + +32768 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.6 + +每百万tokens + +输入：音频 + +25 + +每百万tokens + +输入：图片/视频 + +6 + +每百万tokens + +输出：文本（输入仅包含文本时） + +6.4 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +18 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +50 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.982 + +每百万tokens + +输入：音频 + +32.586 + +每百万tokens + +输入：图片/视频 + +6.165 + +每百万tokens + +输出：文本（输入仅包含文本时） + +7.853 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +18.495 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +65.246 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +10,000 + +## 动态更新 + +### qwen-omni-turbo-realtime-latest + +千问全新多模态理解生成大模型实时版，此版本为动态更新版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30720 + +最大输出长度 + +2048 + +上下文长度 + +32768 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.6 + +每百万tokens + +输入：音频 + +25 + +每百万tokens + +输入：图片/视频 + +6 + +每百万tokens + +输出：文本（输入仅包含文本时） + +6.4 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +18 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +50 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.982 + +每百万tokens + +输入：音频 + +32.586 + +每百万tokens + +输入：图片/视频 + +6.165 + +每百万tokens + +输出：文本（输入仅包含文本时） + +7.853 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +18.495 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +65.246 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +10,000 + +## 快照版本 + +### qwen-omni-turbo-realtime-2025-05-08 + +千问全新多模态理解生成大模型实时版，适合实时音频交互场景。支持音频伴随文本、图像、视频混合输入理解，具备语音和文本同时流式生成能力，提供了4种自然对话音色。此版本为2025年5月8日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30720 + +最大输出长度 + +2048 + +上下文长度 + +32768 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.6 + +每百万tokens + +输入：音频 + +25 + +每百万tokens + +输入：图片/视频 + +6 + +每百万tokens + +输出：文本（输入仅包含文本时） + +6.4 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +18 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +50 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.982 + +每百万tokens + +输入：音频 + +32.586 + +每百万tokens + +输入：图片/视频 + +6.165 + +每百万tokens + +输出：文本（输入仅包含文本时） + +7.853 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +18.495 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +65.246 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +10,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo.md new file mode 100644 index 00000000..e5af0ff6 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen-omni-turbo.md @@ -0,0 +1,821 @@ +# qwen-omni-turbo + +千问全新多模态理解生成大模型，支持文本, 图像，语音，视频输入理解和混合输入理解，具备文本和语音同时流式生成能力，多模态内容理解速度显著提升，提供了4种自然对话音色。该模型版本功能等同于快照模型 qwen-omni-turbo-2025-03-26。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30720 + +最大输出长度 + +2048 + +上下文长度 + +32768 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +0.4 + +每百万tokens + +输入：音频 + +25 + +每百万tokens + +输入：图片/视频 + +1.5 + +每百万tokens + +输出：文本（输入仅包含文本时） + +1.6 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +4.5 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +50 + +每百万tokens + +输入：图片/视频（缓存命中） + +0.3 + +每百万tokens + +输入：文本（缓存命中） + +0.08 + +每百万tokens + +输入：音频（缓存命中） + +5 + +每百万tokens + +输入：文本（Batch File） + +0.2 + +每百万tokens + +输入：音频（Batch File） + +12.5 + +每百万tokens + +输入：图片/视频（Batch File） + +0.75 + +每百万tokens + +输出：文本+音频（Batch File） + +25 + +每百万tokens + +输出：文本（Batch File，输入包含图片/音频/视频时） + +2.25 + +每百万tokens + +输出：文本（Batch File，输入仅包含文本时） + +0.8 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +0.514 + +每百万tokens + +输入：音频 + +32.586 + +每百万tokens + +输入：图片/视频 + +1.541 + +每百万tokens + +输出：文本（输入仅包含文本时） + +1.982 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +4.624 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +65.246 + +每百万tokens + +输入：图片/视频（缓存命中） + +0.294 + +每百万tokens + +输入：文本（缓存命中） + +0.11 + +每百万tokens + +输入：音频（缓存命中） + +6.532 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 动态更新 + +### qwen-omni-turbo-latest + +千问全新多模态理解生成大模型，支持文本, 图像，语音，视频输入理解和混合输入理解，具备文本和语音同时流式生成能力，多模态内容理解速度显著提升，提供了4种自然对话音色，此版本为动态更新版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30720 + +最大输出长度 + +2048 + +上下文长度 + +32768 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +0.4 + +每百万tokens + +输入：音频 + +25 + +每百万tokens + +输入：图片/视频 + +1.5 + +每百万tokens + +输出：文本（输入仅包含文本时） + +1.6 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +4.5 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +50 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +0.514 + +每百万tokens + +输入：音频 + +32.586 + +每百万tokens + +输入：图片/视频 + +1.541 + +每百万tokens + +输出：文本（输入仅包含文本时） + +1.982 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +4.624 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +65.246 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen-omni-turbo-2025-03-26 + +千问全新多模态理解生成大模型，支持文本, 图像，语音，视频输入理解和混合输入理解，具备文本和语音同时流式生成能力，多模态内容理解速度显著提升，提供了4种自然对话音色，此版本为2025年3月26日的快照版本，相比1月19日的快照版本在视觉能力上有大幅提升。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30720 + +最大输出长度 + +2048 + +上下文长度 + +32768 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +0.4 + +每百万tokens + +输入：音频 + +25 + +每百万tokens + +输入：图片/视频 + +1.5 + +每百万tokens + +输出：文本（输入仅包含文本时） + +1.6 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +4.5 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +50 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +0.514 + +每百万tokens + +输入：音频 + +32.586 + +每百万tokens + +输入：图片/视频 + +1.541 + +每百万tokens + +输出：文本（输入仅包含文本时） + +1.982 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +4.624 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +65.246 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +### qwen-omni-turbo-2025-01-19 + +千问全模态理解生成大模型，支持文本, 图像，语音，视频输入理解和混合输入理解，具备文本和语音同时流式生成能力，多模态内容理解速度显著提升，提供了4种自然对话音色，此版本为2025年1月19日的快照版本，预计维护至下一个快照发布前的一个月左右。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30720 + +最大输出长度 + +2048 + +上下文长度 + +32768 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +0.4 + +每百万tokens + +输入：音频 + +25 + +每百万tokens + +输入：图片/视频 + +1.5 + +每百万tokens + +输出：文本（输入仅包含文本时） + +1.6 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +4.5 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +50 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen2-5-omni-7b.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen2-5-omni-7b.md new file mode 100644 index 00000000..06bd72da --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen2-5-omni-7b.md @@ -0,0 +1,185 @@ +# qwen2.5-omni-7b + +基于Qwen2.5训练的全新多模态理解生成大模型，支持文本, 图像，语音，视频输入理解和混合输入理解，具备文本和语音同时流式生成能力，多模态内容理解速度显著提升，提供了4种自然对话音色。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +30720 + +最大输出长度 + +2048 + +上下文长度 + +32768 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +0.6 + +每百万tokens + +输入：音频 + +38 + +每百万tokens + +输入：图片/视频 + +2 + +每百万tokens + +输出：文本（输入仅包含文本时） + +2.4 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +6 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +76 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +0.734 + +每百万tokens + +输入：音频 + +49.613 + +每百万tokens + +输入：图片/视频 + +2.055 + +每百万tokens + +输出：文本（输入仅包含文本时） + +2.936 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +6.165 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +99.153 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash-realtime.md new file mode 100644 index 00000000..dc5fdd36 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash-realtime.md @@ -0,0 +1,357 @@ +# qwen3.5-omni-flash-realtime + +Qwen3.5-Omni是Qwen最新一代全模态大模型，支持文本，图片，音频，音视频理解与交互。作为 Qwen3-Omni 的全面进化版本，支持60+种语言音频输入，30+语言语音输出以及可控语音对话，WebSearch和复杂FunctionCall的调用，并且具备智能语义打断的交互能力，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。该模型版本功能等同于快照模型 qwen3.5-omni-flash-realtime-2026-03-15。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +196608 + +最大输出长度 + +65536 + +上下文长度 + +262144 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +27 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +107 + +每百万tokens + +输入：文本/图片/视频 + +3.3 + +每百万tokens + +输出：文本 + +20 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +33.72 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +132.65 + +每百万tokens + +输入：文本/图片/视频 + +4.12 + +每百万tokens + +输出：文本 + +24.73 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen3.5-omni-flash-realtime-2026-03-15 + +Qwen3.5-Omni是Qwen最新一代全模态大模型，支持文本，图片，音频，音视频理解与交互。作为 Qwen3-Omni 的全面进化版本，支持60+种语言音频输入，30+语言语音输出以及可控语音对话，WebSearch和复杂FunctionCall的调用，并且具备智能语义打断的交互能力，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。该版本为2026年3月15日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +196608 + +最大输出长度 + +65536 + +上下文长度 + +262144 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +27 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +107 + +每百万tokens + +输入：文本/图片/视频 + +3.3 + +每百万tokens + +输出：文本 + +20 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +33.72 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +132.65 + +每百万tokens + +输入：文本/图片/视频 + +4.12 + +每百万tokens + +输出：文本 + +24.73 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash.md new file mode 100644 index 00000000..1c47fd1f --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-flash.md @@ -0,0 +1,357 @@ +# qwen3.5-omni-flash + +Qwen3.5-Omni是Qwen最新一代全模态大模型，支持文本，图片，音频，音视频理解与交互。作为 Qwen3-Omni 的全面进化版本， 支持超过 10 小时的音频理解及超过 400 秒的 720P（1 FPS）音视频理解与对话，并进一步拓展语言范围，支持60+种语言音频输入，30+语言语音输出，并且具备强大的结构化音视频理解能力，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态理解与交互体验。该模型版本功能等同于快照模型 qwen3.5-omni-flash-2026-03-15。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +196608 + +最大输出长度 + +65536 + +上下文长度 + +262144 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +18 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +72 + +每百万tokens + +输入：文本/图片/视频 + +2.2 + +每百万tokens + +输出：文本 + +13.3 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +22.48 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +89.18 + +每百万tokens + +输入：文本/图片/视频 + +3 + +每百万tokens + +输出：文本 + +16.49 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen3.5-omni-flash-2026-03-15 + +Qwen3.5-Omni是Qwen最新一代全模态大模型，支持文本，图片，音频，音视频理解与交互。作为 Qwen3-Omni 的全面进化版本， 支持超过 10 小时的音频理解及超过 400 秒的 720P（1 FPS）音视频理解与对话，并进一步拓展语言范围，支持60+种语言音频输入，30+语言语音输出，并且具备强大的结构化音视频理解能力，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态理解与交互体验。该版本为2026年3月15日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +196608 + +最大输出长度 + +65536 + +上下文长度 + +262144 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +18 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +72 + +每百万tokens + +输入：文本/图片/视频 + +2.2 + +每百万tokens + +输出：文本 + +13.3 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +22.48 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +89.18 + +每百万tokens + +输入：文本/图片/视频 + +3 + +每百万tokens + +输出：文本 + +16.49 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus-realtime.md new file mode 100644 index 00000000..a56c64e9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus-realtime.md @@ -0,0 +1,357 @@ +# qwen3.5-omni-plus-realtime + +Qwen3.5-Omni是Qwen最新一代全模态大模型，支持文本，图片，音频，音视频理解与交互。作为 Qwen3-Omni 的全面进化版本，支持60+种语言音频输入，30+语言语音输出以及可控语音对话，WebSearch和复杂FunctionCall的调用，并且具备智能语义打断的交互能力，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。该模型版本功能等同于快照模型 qwen3.5-omni-plus-realtime-2026-03-15。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +196608 + +最大输出长度 + +65536 + +上下文长度 + +262144 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +80 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +300 + +每百万tokens + +输入：文本/图片/视频 + +10 + +每百万tokens + +输出：文本 + +60 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +123.65 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +464.64 + +每百万tokens + +输入：文本/图片/视频 + +15.74 + +每百万tokens + +输出：文本 + +92.93 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen3.5-omni-plus-realtime-2026-03-15 + +Qwen3.5-Omni是Qwen最新一代全模态大模型，支持文本，图片，音频，音视频理解与交互。作为 Qwen3-Omni 的全面进化版本，支持60+种语言音频输入，30+语言语音输出以及可控语音对话，WebSearch和复杂FunctionCall的调用，并且具备智能语义打断的交互能力，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。该版本为2026年3月15日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +196608 + +最大输出长度 + +65536 + +上下文长度 + +262144 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +80 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +300 + +每百万tokens + +输入：文本/图片/视频 + +10 + +每百万tokens + +输出：文本 + +60 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +123.65 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +464.64 + +每百万tokens + +输入：文本/图片/视频 + +15.74 + +每百万tokens + +输出：文本 + +92.93 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus.md new file mode 100644 index 00000000..5d947b50 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-5-omni-plus.md @@ -0,0 +1,393 @@ +# qwen3.5-omni-plus + +Qwen3.5-Omni是Qwen最新一代全模态大模型，支持文本，图片，音频，音视频理解与交互。作为 Qwen3-Omni 的全面进化版本， 支持超过 10 小时的音频理解及超过 400 秒的 720P（1 FPS）音视频理解与对话，并进一步拓展语言范围，支持60+种语言音频输入，30+语言语音输出，并且具备强大的结构化音视频理解能力，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态理解与交互体验。该模型版本功能等同于快照模型 qwen3.5-omni-plus-2026-03-15。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +196608 + +最大输出长度 + +65536 + +上下文长度 + +262144 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +53 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +213 + +每百万tokens + +输入：文本/图片/视频 + +7 + +每百万tokens + +输出：文本 + +40 + +每百万tokens + +输入：音频（Batch File） + +26.5 + +每百万tokens + +输入：文本/图片/视频（Batch File） + +3.5 + +每百万tokens + +输出：文本（Batch File） + +20 + +每百万tokens + +输入：音频（Batch Chat） + +53 + +每百万tokens + +输入：文本/图片/视频（Batch Chat） + +7 + +每百万tokens + +输出：文本（Batch Chat） + +40 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +82.44 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +329.74 + +每百万tokens + +输入：文本/图片/视频 + +10.49 + +每百万tokens + +输出：文本 + +62.2 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen3.5-omni-plus-2026-03-15 + +Qwen3.5-Omni是Qwen最新一代全模态大模型，支持文本，图片，音频，音视频理解与交互。作为 Qwen3-Omni 的全面进化版本， 支持超过 10 小时的音频理解及超过 400 秒的 720P（1 FPS）音视频理解与对话，并进一步拓展语言范围，支持60+种语言音频输入，30+语言语音输出，并且具备强大的结构化音视频理解能力，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态理解与交互体验。该版本为2026年3月15日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +196608 + +最大输出长度 + +65536 + +上下文长度 + +262144 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +53 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +213 + +每百万tokens + +输入：文本/图片/视频 + +7 + +每百万tokens + +输出：文本 + +40 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +82.44 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +329.74 + +每百万tokens + +输入：文本/图片/视频 + +10.49 + +每百万tokens + +输出：文本 + +62.2 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash-realtime.md new file mode 100644 index 00000000..64516c1d --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash-realtime.md @@ -0,0 +1,561 @@ +# qwen3-omni-flash-realtime + +千问3-Omni-Flash多模态大模型的实时版，基于Thinker–Talker混合专家（MoE）架构，支持文本、图像、音频、视频的高效理解与语音生成能力，可进行119种语言文本交互和20种语言语音交互，生成类人语音实现跨语言精准沟通。模型具备强大指令跟随与系统提示定制功能，灵活适配对话风格与角色设定，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。该模型版本功能等同于快照模型 qwen3-omni-flash-realtime-2025-12-01。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +16384 + +上下文长度 + +65536 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +2.2 + +每百万tokens + +输入：音频 + +18.9 + +每百万tokens + +输入：图片/视频 + +3.9 + +每百万tokens + +输出：文本（输入仅包含文本时） + +8.3 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +15.2 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +75.1 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +3.816 + +每百万tokens + +输入：音频 + +33.54 + +每百万tokens + +输入：图片/视频 + +6.899 + +每百万tokens + +输出：文本（输入仅包含文本时） + +14.605 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +26.935 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +133.06 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen3-omni-flash-realtime-2025-12-01 + +千问3-Omni-Flash多模态大模型的实时版，基于Thinker–Talker混合专家（MoE）架构，支持文本、图像、音频、视频的高效理解与语音生成能力，可进行119种语言文本交互和20种语言语音交互，支持49种语音音色，生成类人语音实现跨语言精准沟通。模型具备强大指令跟随与系统提示定制功能，灵活适配对话风格与角色设定，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。此版本为2025年12月01日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Audio**　**Video** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +16384 + +上下文长度 + +65536 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +2.2 + +每百万tokens + +输入：音频 + +18.9 + +每百万tokens + +输入：图片/视频 + +3.9 + +每百万tokens + +输出：文本（输入仅包含文本时） + +8.3 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +15.2 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +75.1 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +### qwen3-omni-flash-realtime-2025-09-15 + +千问3-Omni-Flash多模态大模型的实时版，基于Thinker–Talker混合专家（MoE）架构，支持文本、图像、音频、视频的高效理解与语音生成能力，可进行119种语言文本交互和20种语言语音交互，生成类人语音实现跨语言精准沟通。模型具备强大指令跟随与系统提示定制功能，灵活适配对话风格与角色设定，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。此版本为2025年9月15日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +16384 + +上下文长度 + +65536 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +2.2 + +每百万tokens + +输入：音频 + +18.9 + +每百万tokens + +输入：图片/视频 + +3.9 + +每百万tokens + +输出：文本（输入仅包含文本时） + +8.3 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +15.2 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +75.1 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +3.816 + +每百万tokens + +输入：音频 + +33.54 + +每百万tokens + +输入：图片/视频 + +6.899 + +每百万tokens + +输出：文本（输入仅包含文本时） + +14.605 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +26.935 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +133.06 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash.md new file mode 100644 index 00000000..8ff3093a --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-omni/qwen3-omni-flash.md @@ -0,0 +1,747 @@ +# qwen3-omni-flash + +千问3-Omni-Flash多模态大模型，基于Thinker–Talker混合专家（MoE）架构，支持文本、图像、音频、视频的高效理解与语音生成能力，可进行119种语言文本交互和20种语言语音交互，生成类人语音实现跨语言精准沟通。模型具备强大指令跟随与系统提示定制功能，灵活适配对话风格与角色设定，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。该模型版本功能等同于快照模型 qwen3-omni-flash-2025-12-01。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +16384 + +上下文长度 + +65536 + +最大输入长度（思考模式下） + +16384 + +最大输出长度（思考模式下） + +16384 + +最大思维链长度 + +32768 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.8 + +每百万tokens + +输入：音频 + +15.8 + +每百万tokens + +输入：图片/视频 + +3.3 + +每百万tokens + +输出：文本（输入仅包含文本时） + +6.9 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +12.7 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +62.6 + +每百万tokens + +输入：文本（思考） + +1.8 + +每百万tokens + +输入：音频（思考） + +15.8 + +每百万tokens + +输入：图片/视频（思考） + +3.3 + +每百万tokens + +输出：文本（思考模式下，输入仅包含文本时） + +6.9 + +每百万tokens + +输出：文本（思考模式下，输入包含图片/音频/视频时） + +12.7 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +3.156 + +每百万tokens + +输入：音频 + +27.962 + +每百万tokens + +输入：图片/视频 + +5.725 + +每百万tokens + +输出：文本（输入仅包含文本时） + +12.183 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +22.458 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +110.896 + +每百万tokens + +输入：文本（思考） + +3.156 + +每百万tokens + +输入：音频（思考） + +27.962 + +每百万tokens + +输入：图片/视频（思考） + +5.725 + +每百万tokens + +输出：文本（思考模式下，输入仅包含文本时） + +12.183 + +每百万tokens + +输出：文本（思考模式下，输入包含图片/音频/视频时） + +22.458 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen3-omni-flash-2025-12-01 + +千问3-Omni-Flash多模态大模型，基于Thinker–Talker混合专家（MoE）架构，支持文本、图像、音频、视频的高效理解与语音生成能力，可进行119种语言文本交互和20种语言语音交互，支持49种语音音色，生成类人语音实现跨语言精准沟通。模型具备强大指令跟随与系统提示定制功能，灵活适配对话风格与角色设定，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。此版本为2025年12月01日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Audio**　**Video** + +输出模态 + +**Text**　**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +16384 + +上下文长度 + +65536 + +最大输入长度（思考模式下） + +16384 + +最大输出长度（思考模式下） + +16384 + +最大思维链长度 + +32768 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.8 + +每百万tokens + +输入：音频 + +15.8 + +每百万tokens + +输入：图片/视频 + +3.3 + +每百万tokens + +输出：文本（输入仅包含文本时） + +6.9 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +12.7 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +62.6 + +每百万tokens + +输入：文本（思考） + +1.8 + +每百万tokens + +输入：音频（思考） + +15.8 + +每百万tokens + +输入：图片/视频（思考） + +3.3 + +每百万tokens + +输出：文本（思考模式下，输入仅包含文本时） + +6.9 + +每百万tokens + +输出：文本（思考模式下，输入包含图片/音频/视频时） + +12.7 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +### qwen3-omni-flash-2025-09-15 + +Qwen3-Omni-Flash多模态大模型，基于Thinker–Talker混合专家（MoE）架构，支持文本、图像、音频、视频的高效理解与语音生成能力，可进行119种语言文本交互和20种语言语音交互，生成类人语音实现跨语言精准沟通。模型具备强大指令跟随与系统提示定制功能，灵活适配对话风格与角色设定，广泛应用于文本创作、语音助手、多媒体分析等场景，提供自然流畅的多模态交互体验。此版本为2025年9月15日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +16384 + +上下文长度 + +65536 + +最大输入长度（思考模式下） + +16384 + +最大输出长度（思考模式下） + +16384 + +最大思维链长度 + +32768 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.8 + +每百万tokens + +输入：音频 + +15.8 + +每百万tokens + +输入：图片/视频 + +3.3 + +每百万tokens + +输出：文本（输入仅包含文本时） + +6.9 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +12.7 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +62.6 + +每百万tokens + +输入：文本（思考） + +1.8 + +每百万tokens + +输入：音频（思考） + +15.8 + +每百万tokens + +输入：图片/视频（思考） + +3.3 + +每百万tokens + +输出：文本（思考模式下，输入仅包含文本时） + +6.9 + +每百万tokens + +输出：文本（思考模式下，输入包含图片/音频/视频时） + +12.7 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：文本 + +3.156 + +每百万tokens + +输入：音频 + +27.962 + +每百万tokens + +输入：图片/视频 + +5.725 + +每百万tokens + +输出：文本（输入仅包含文本时） + +12.183 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +22.458 + +每百万tokens + +输出：文本+音频（输出的文本不计费） + +110.896 + +每百万tokens + +输入：文本（思考） + +3.156 + +每百万tokens + +输入：音频（思考） + +27.962 + +每百万tokens + +输入：图片/视频（思考） + +5.725 + +每百万tokens + +输出：文本（思考模式下，输入仅包含文本时） + +12.183 + +每百万tokens + +输出：文本（思考模式下，输入包含图片/音频/视频时） + +22.458 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash-8k-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash-8k-realtime.md new file mode 100644 index 00000000..e7965377 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash-8k-realtime.md @@ -0,0 +1,213 @@ +# fun-asr-flash-8k-realtime + +通义百聆推出的新一代轻量级实时语音识别模型，依托自研的先进语音技术架构，具备强大的上下文理解能力。专为中文电话客服场景设计：覆盖多地区方言口音，在低采样率、低信噪比环境下实现低延迟、高准确率的流式转写，满足高效部署需求。该模型版本功能等同于快照模型 fun-asr-flash-8k-realtime-2026-01-28。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +## 快照版本 + +### fun-asr-flash-8k-realtime-2026-01-28 + +通义百聆推出的新一代轻量级实时语音识别模型，依托自研的先进语音技术架构，具备强大的上下文理解能力。专为中文电话客服场景设计：覆盖多地区方言口音，在低采样率、低信噪比环境下实现低延迟、高准确率的流式转写，满足高效部署需求。此版本为2026年1月28日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash.md new file mode 100644 index 00000000..f90e43ae --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-flash.md @@ -0,0 +1,269 @@ +# fun-asr-flash-2026-06-15 + +百聆2026年6月更新的大模型ASR版本，全面支持汉语传统七大方言体系（官话/吴/湘/赣/客/闽/粤），并适配 20+ 地区口音官话。针对中文古诗词的韵律、节奏与文言表达特点进行专项优化，提升对古诗词内容的识别准确率，适用于文化传承、教育讲解、有声读物等场景。优化标点预测与文本归一化能力，使输出文本更符合书面表达习惯，数字、日期、金额等信息自动转换为标准格式，增强内容的可读性与专业性。同时语种扩展至英语、日语、韩语、越南语、泰语、印尼语、马来语、菲律宾语、印地语、阿拉伯语、法语、德语、西班牙语、葡萄牙语、俄语、意大利语、荷兰语、瑞典语、丹麦语、芬兰语、挪威语、希腊语、波兰语、捷克语、匈牙利语、罗马尼亚、保加利亚语、克罗地亚语、斯洛伐克语等，共计30个语种。支持context上下文能力，可转写5分钟以内的音频。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 快照版本 + +### fun-asr-flash-2026-06-15 + +百聆2026年6月更新的大模型ASR版本，全面支持汉语传统七大方言体系（官话/吴/湘/赣/客/闽/粤），并适配 20+ 地区口音官话。针对中文古诗词的韵律、节奏与文言表达特点进行专项优化，提升对古诗词内容的识别准确率，适用于文化传承、教育讲解、有声读物等场景。优化标点预测与文本归一化能力，使输出文本更符合书面表达习惯，数字、日期、金额等信息自动转换为标准格式，增强内容的可读性与专业性。同时语种扩展至英语、日语、韩语、越南语、泰语、印尼语、马来语、菲律宾语、印地语、阿拉伯语、法语、德语、西班牙语、葡萄牙语、俄语、意大利语、荷兰语、瑞典语、丹麦语、芬兰语、挪威语、希腊语、波兰语、捷克语、匈牙利语、罗马尼亚、保加利亚语、克罗地亚语、斯洛伐克语等，共计30个语种。支持context上下文能力，可转写5分钟以内的音频。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +600 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-mtl.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-mtl.md new file mode 100644 index 00000000..b91e8935 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-mtl.md @@ -0,0 +1,237 @@ +# fun-asr-mtl + +百聆多语言语音识别大模型，支持超过31种语言，支持语种自由切换，出海用户首推，尤其东南亚出海。fun-asr为该模型的升级版本，建议切换使用fun-asr。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +100 + +## 快照版本 + +### fun-asr-mtl-2025-08-25 + +百聆多语言语音识别大模型，支持超过31种语言，支持语种自由切换，出海用户首推，尤其东南亚出海。fun-asr为该模型的升级版本，建议切换使用fun-asr。此版本为2025年8月25日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +100 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-realtime.md new file mode 100644 index 00000000..b7eece61 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr-realtime.md @@ -0,0 +1,449 @@ +# fun-asr-realtime + +通义实验室新一代端到端语音识别大模型的实时版，基于领先的自研语音技术，具备卓越的上下文感知和高精度语音转写能力。基于端到端架构，Fun-ASR 集成了创新的 RAG 技术，支持大规模热词自定义、敏感/语气词自动过滤、ITN 规范化、标点预测等多维功能，显著提升了整体识别准确率和语境贴合度。同时，Fun-ASR 支持中英文自由切换，多地区方言覆盖，具备更强的噪声鲁棒性，适应多样复杂环境。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00033 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +## 快照版本 + +### fun-asr-realtime-2026-02-28 + +通义百聆推出的新一代轻量级实时语音识别模型，依托自研的先进语音技术架构，具备强大的上下文理解能力。专为中文电话客服场景设计：覆盖多地区方言口音，在低采样率、低信噪比环境下实现低延迟、高准确率的流式转写，满足高效部署需求。此版本为2026年2月28日的快照版本，在近场VAD方面进行了优化。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00033 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +### fun-asr-realtime-2025-11-07 + +通义实验室新一代端到端语音识别大模型的实时版，基于领先的自研语音技术，具备卓越的上下文感知和高精度语音转写能力。基于端到端架构，Fun-ASR 集成了创新的 RAG 技术，支持大规模热词自定义、敏感/语气词自动过滤、ITN 规范化、标点预测等多维功能，显著提升了整体识别准确率和语境贴合度。同时，Fun-ASR 支持中英文自由切换，多地区方言覆盖，具备更强的噪声鲁棒性，适应多样复杂环境。此版本为2025年11月7日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00033 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +### fun-asr-realtime-2025-09-15 + +通义实验室新一代端到端语音识别大模型的实时版，基于领先的自研语音技术，具备卓越的上下文感知和高精度语音转写能力。基于端到端架构，Fun-ASR 集成了创新的 RAG 技术，支持大规模热词自定义、敏感/语气词自动过滤、ITN 规范化、标点预测等多维功能，显著提升了整体识别准确率和语境贴合度。同时，Fun-ASR 支持中英文自由切换，多地区方言覆盖，具备更强的噪声鲁棒性，适应多样复杂环境。此版本为2025年9月15日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00033 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr.md new file mode 100644 index 00000000..3d2164ac --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/fun-asr.md @@ -0,0 +1,403 @@ +# fun-asr + +百聆2026年4月更新的大模型ASR版本，全面支持汉语传统七大方言体系（官话/吴/湘/赣/客/闽/粤），并适配 20+ 地区口音官话。针对中文古诗词的韵律、节奏与文言表达特点进行专项优化，提升对古诗词内容的识别准确率，适用于文化传承、教育讲解、有声读物等场景。优化标点预测与文本归一化能力，使输出文本更符合书面表达习惯，数字、日期、金额等信息自动转换为标准格式，增强内容的可读性与专业性。同时语种扩展至英语、日语、韩语、越南语、泰语、印尼语、马来语、菲律宾语、印地语、阿拉伯语、法语、德语、西班牙语、葡萄牙语、俄语、意大利语、荷兰语、瑞典语、丹麦语、芬兰语、挪威语、希腊语、波兰语、捷克语、匈牙利语、罗马尼亚、保加利亚语、克罗地亚语、斯洛伐克语等，共计30个语种。此版本等同于2025年11月7日的快照版本。该模型版本功能等同于快照模型 fun-asr-2025-11-07。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 快照版本 + +### fun-asr-2025-11-07 + +百聆2026年4月更新的大模型ASR版本，全面支持汉语传统七大方言体系（官话/吴/湘/赣/客/闽/粤），并适配 20+ 地区口音官话。针对中文古诗词的韵律、节奏与文言表达特点进行专项优化，提升对古诗词内容的识别准确率，适用于文化传承、教育讲解、有声读物等场景。优化标点预测与文本归一化能力，使输出文本更符合书面表达习惯，数字、日期、金额等信息自动转换为标准格式，增强内容的可读性与专业性。同时语种扩展至英语、日语、韩语、越南语、泰语、印尼语、马来语、菲律宾语、印地语、阿拉伯语、法语、德语、西班牙语、葡萄牙语、俄语、意大利语、荷兰语、瑞典语、丹麦语、芬兰语、挪威语、希腊语、波兰语、捷克语、匈牙利语、罗马尼亚、保加利亚语、克罗地亚语、斯洛伐克语等，共计30个语种。此版本为2025年11月7日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +600 + +### fun-asr-2025-08-25 + +百聆新一代语音识别大模型，主打中文、英文、日文语音识别，多地区方言覆盖，具备更强的噪声鲁棒性，适应多样复杂环境，国内用户首推。此版本为2025年8月25日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +600 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-chat-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-chat-v1.md new file mode 100644 index 00000000..57a29e8b --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-chat-v1.md @@ -0,0 +1,105 @@ +# gummy-chat-v1 + +多语言语音转写及翻译的多模态大模型。本模型支持60秒以内的实时语音识别，适用于语音搜索、设备指令等场景。提供10个混合语种的高准确率识别服务，同时支持中英日韩互译，以其他6个语种翻译成中文或英文。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00015 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-realtime-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-realtime-v1.md new file mode 100644 index 00000000..45bcde9d --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/gummy-realtime-v1.md @@ -0,0 +1,105 @@ +# gummy-realtime-v1 + +多语言语音转写及翻译的多模态大模型。本模型提供长时间、高准确率、实时转写中/英/日/韩等10个混合语种的服务。同时支持中英日韩互译，以其他6个语种翻译成中文或英文。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00015 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash-realtime.md new file mode 100644 index 00000000..5b36abc9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash-realtime.md @@ -0,0 +1,357 @@ +# qwen3-livetranslate-flash-realtime + +Qwen3-LiveTranslate-Flash的实时版本，一款高精度、高响应、高鲁棒性的多语言实时音视频同传大模型。依托Qwen3-Omni强大的基座能力、海量多模态数据、跨语言跨模态对齐和视觉增强等技术，通义千问3-LiveTranslate-Flash 实现了离线和实时两种音视频翻译能力，能听懂19种语言，会说10种语言以及8种中文方言。该模型版本功能等同于快照模型 qwen3-livetranslate-flash-realtime-2025-09-22。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +4096 + +上下文长度 + +53248 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +64 + +每百万tokens + +输入：图片 + +8 + +每百万tokens + +输出：文本 + +64 + +每百万tokens + +输出：音频 + +240 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +73.392 + +每百万tokens + +输入：图片 + +9.541 + +每百万tokens + +输出：文本 + +73.392 + +每百万tokens + +输出：音频 + +278.891 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen3-livetranslate-flash-realtime-2025-09-22 + +千问3-LiveTranslate-Flash的实时版本，一款高精度、高响应、高鲁棒性的多语言实时音视频同传大模型。依托Qwen3-Omni强大的基座能力、海量多模态数据、跨语言跨模态对齐和视觉增强等技术，千问3-LiveTranslate-Flash 实现了离线和实时两种音视频翻译能力，能听懂19种语言，会说10种语言以及8种中文方言。此版本为2025年9月22日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Audio** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +4096 + +上下文长度 + +53248 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +64 + +每百万tokens + +输入：图片 + +8 + +每百万tokens + +输出：文本 + +64 + +每百万tokens + +输出：音频 + +240 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +73.392 + +每百万tokens + +输入：图片 + +9.541 + +每百万tokens + +输出：文本 + +73.392 + +每百万tokens + +输出：音频 + +278.891 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash.md new file mode 100644 index 00000000..fe34a355 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/model-qwen3-livetranslate-flash.md @@ -0,0 +1,289 @@ +# qwen3-livetranslate-flash + +Qwen3-LiveTranslate-Flash，一款高精度、高响应、高鲁棒性的多语言实时音视频同传大模型。依托Qwen3-Omni强大的基座能力、海量多模态数据、跨语言跨模态对齐和视觉增强等技术，Qwen3-LiveTranslate-Flash 实现了离线和实时两种音视频翻译能力，能听懂19种语言，会说10种语言以及8种中文方言。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio**　**Video** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +4096 + +上下文长度 + +53248 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +10 + +每百万tokens + +输入：图片 + +4 + +每百万tokens + +输出：文本 + +10 + +每百万tokens + +输出：音频 + +40 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +100 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +100 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen3-livetranslate-flash-2025-12-01 + +Qwen3-LiveTranslate-Flash，一款高精度、高响应、高鲁棒性的多语言实时音视频同传大模型。依托Qwen3-Omni强大的基座能力、海量多模态数据、跨语言跨模态对齐和视觉增强等技术，Qwen3-LiveTranslate-Flash 实现了离线和实时两种音视频翻译能力，能听懂19种语言，会说10种语言以及8种中文方言。此版本为2025年12月01日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio**　**Video** + +输出模态 + +**Text**　**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +4096 + +上下文长度 + +53248 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +10 + +每百万tokens + +输入：图片 + +4 + +每百万tokens + +输出：文本 + +10 + +每百万tokens + +输出：音频 + +40 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +100 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +100 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v1.md new file mode 100644 index 00000000..e74e4ad4 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v1.md @@ -0,0 +1,93 @@ +# paraformer-8k-v1 + +Paraformer语音识别提供的文件转写API，能够对常见的音频或音视频文件进行语音识别，并将结果返回给调用者。Paraformer中文语音识别模型，支持8kHz电话语音识别。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00008 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v2.md new file mode 100644 index 00000000..e00dfd56 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-8k-v2.md @@ -0,0 +1,93 @@ +# paraformer-8k-v2 + +Paraformer最新中文语音识别模型，模型结构升级，具有更好的识别效果,支持8kHz电话语音识别，仅支持中文热词。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00008 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-mtl-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-mtl-v1.md new file mode 100644 index 00000000..3f51d7ef --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-mtl-v1.md @@ -0,0 +1,93 @@ +# paraformer-mtl-v1 + +Paraformer多语言语音识别模型，支持16kHz及以上采样率的音频或视频语音识别。 支持的语种/方言包括：中文普通话、中文方言（粤语、吴语、闽南语、东北话、甘肃话、贵州话、河南话、湖北话、湖南话、宁夏话、山西话、陕西话、山东话、四川话、天津话）、英语、日语、韩语、西班牙语、印尼语、法语、德语、意大利语、马来语。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00008 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v1.md new file mode 100644 index 00000000..4122c259 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v1.md @@ -0,0 +1,93 @@ +# paraformer-realtime-8k-v1 + +Paraformer中文实时语音识别模型，支持8kHz电话客服等场景下的实时语音识别。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00024 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v2.md new file mode 100644 index 00000000..2e0ee24b --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-8k-v2.md @@ -0,0 +1,93 @@ +# paraformer-realtime-8k-v2 + +推荐使用 Paraformer最新实时语音识别模型，支持多个语种自由切换的视频直播、会议等实时场景的语音识别。可以通过language\_hints参数选择语种获得更准确的识别效果。支持8kHz电话客服等场景下的实时语音识别。 支持的语言包括：中文（含粤语等各种方言）、英文、日语、韩语。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00024 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v1.md new file mode 100644 index 00000000..7af6d727 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v1.md @@ -0,0 +1,93 @@ +# paraformer-realtime-v1 + +Paraformer中文实时语音识别模型，支持16kHz及以上采样率的视频直播、会议等实时场景下的语音识别。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00024 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v2.md new file mode 100644 index 00000000..28610811 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-realtime-v2.md @@ -0,0 +1,93 @@ +# paraformer-realtime-v2 + +推荐使用 Paraformer最新实时语音识别模型，支持多个语种自由切换的视频直播、会议等实时场景的语音识别。可以通过language\_hints参数选择语种获得更准确的识别效果。支持任意采样率。 支持的语言包括：中文（含粤语等各种方言）、英文、日语、韩语。 可支持热词。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00024 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v1.md new file mode 100644 index 00000000..299effd8 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v1.md @@ -0,0 +1,93 @@ +# paraformer-v1 + +Paraformer中英文语音识别模型，支持16kHz及以上采样率的音频或视频语音识别。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00008 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v2.md new file mode 100644 index 00000000..92200cd3 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/paraformer-v2.md @@ -0,0 +1,93 @@ +# paraformer-v2 + +推荐使用 Paraformer最新语音识别模型，支持多个语种的语音识别。可以通过language\_hints参数选择语种获得更准确的识别效果，支持任意采样率。 支持的语言包括：中文（含粤语等各种方言）、英文、日语、韩语。可支持热词。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00008 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-6.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-6.md new file mode 100644 index 00000000..73db5c7b --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-6.md @@ -0,0 +1,357 @@ +# qwen3.5-livetranslate-flash-realtime + +Qwen3.5-LiveTranslate-Flash的实时版本，一款高精度、高响应、高鲁棒性的多语言实时音视频同传大模型。依托Qwen3.5-Omni强大的基座能力、海量多模态数据、跨语言跨模态对齐和视觉增强等技术，通义千问3.5-LiveTranslate-Flash 实现了离线和实时两种音视频翻译能力，能听懂60种语言，会说29种语言。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio**　**Image** + +输出模态 + +**Audio**　**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +4096 + +上下文长度 + +53248 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +40 + +每百万tokens + +输入：图片 + +3.3 + +每百万tokens + +输出：文本 + +100 + +每百万tokens + +输出：音频 + +160 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +56.207 + +每百万tokens + +输入：图片 + +4.122 + +每百万tokens + +输出：文本 + +149.884 + +每百万tokens + +输出：音频 + +224.826 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen3.5-livetranslate-flash-realtime-2026-05-19 + +Qwen3.5-LiveTranslate-Flash的实时版本，一款高精度、高响应、高鲁棒性的多语言实时音视频同传大模型。依托Qwen3.5-Omni强大的基座能力、海量多模态数据、跨语言跨模态对齐和视觉增强等技术，通义千问3.5-LiveTranslate-Flash 实现了离线和实时两种音视频翻译能力，能听懂60种语言，会说29种语言。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio**　**Image** + +输出模态 + +**Audio**　**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +49152 + +最大输出长度 + +4096 + +上下文长度 + +53248 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +40 + +每百万tokens + +输入：图片 + +3.3 + +每百万tokens + +输出：文本 + +100 + +每百万tokens + +输出：音频 + +160 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +56.207 + +每百万tokens + +输入：图片 + +4.122 + +每百万tokens + +输出：文本 + +149.884 + +每百万tokens + +输出：音频 + +224.826 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-filetrans.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-filetrans.md new file mode 100644 index 00000000..f9bd9cfc --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-filetrans.md @@ -0,0 +1,269 @@ +# qwen3-asr-flash-filetrans + +千问3-ASR-Flash的大文件转录版本，千问3-ASR-Flash是一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，千问3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别多个语种的语音，在复杂的音频环境下能够保证精确转录。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +100 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +100 + +## 快照版本 + +### qwen3-asr-flash-filetrans-2025-11-17 + +千问3-ASR-Flash的大文件转录版本，千问3-ASR-Flash是一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，千问3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别多个语种的语音，在复杂的音频环境下能够保证精确转录。此版本为2025年11月17日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +100 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +100 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-realtime.md new file mode 100644 index 00000000..283cac5e --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-realtime.md @@ -0,0 +1,403 @@ +# qwen3-asr-flash-realtime + +千问3-ASR-Flash的实时版，一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，通义千问3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别多个语种的语音，在复杂的音频环境下能够保证精确转录。该模型版本功能等同于快照模型 qwen3-asr-flash-realtime-2025-10-27。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00033 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00066 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +## 快照版本 + +### qwen3-asr-flash-realtime-2026-02-10 + +Qwen3-ASR-Flash的实时版，一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，Qwen3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别 多个语种的语音，在复杂的音频环境下能够保证精确转录。此版本为2026年2月10日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00033 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00066 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +### qwen3-asr-flash-realtime-2025-10-27 + +千问3-ASR-Flash的实时版，一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，千问3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别 多个语种的语音，在复杂的音频环境下能够保证精确转录。此版本为2025年10月27日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00033 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00066 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +1200 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +1200 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-us.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-us.md new file mode 100644 index 00000000..efa07e23 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash-us.md @@ -0,0 +1,193 @@ +# qwen3-asr-flash-us + +Qwen3-ASR-Flash是一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，通义千问3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别 11 个语种的语音，在复杂的音频环境下能够保证精确转录。该模型版本功能等同于快照模型 qwen3-asr-flash-2025-09-08-us。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 美国（弗吉尼亚） + +部署范围：美国 + +计费项 + +价格（元） + +单位 + +语音识别 + +0.00026 + +每秒 + +## 快照版本 + +### qwen3-asr-flash-2025-09-08-us + +Qwen3-ASR-Flash是一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，Qwen3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别 11 个语种的语音，在复杂的音频环境下能够保证精确转录。此版本为2025年9月8日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 美国（弗吉尼亚） + +部署范围：美国 + +计费项 + +价格（元） + +单位 + +语音识别 + +0.00026 + +每秒 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash.md new file mode 100644 index 00000000..003cea20 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-asr-flash.md @@ -0,0 +1,403 @@ +# qwen3-asr-flash + +千问3-ASR-Flash是一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，千问3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别多个语种的语音，在复杂的音频环境下能够保证精确转录。该模型版本功能等同于快照模型 qwen3-asr-flash-2025-09-08。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +100 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +100 + +## 快照版本 + +### qwen3-asr-flash-2026-02-10 + +通义千问3-ASR-Flash，一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，通义千问3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别 多个语种的语音，在复杂的音频环境下能够保证精确转录。此版本为2026年2月10日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +100 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +100 + +### qwen3-asr-flash-2025-09-08 + +千问3-ASR-Flash是一款基于大语言模型的高精度、高智能、高鲁棒性的多语种语音识别模型。依托强大的基座模型、海量的文本与多模态数据、千万小时音频数据，千问3-ASR-Flash实现了高精度的语音识别功能，能够自动判断语种并准确识别多个语种的语音，在复杂的音频环境下能够保证精确转录。此版本为2025年9月8日的快照版本。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00022 + +每秒 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +音频时长 + +0.00026 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +100 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +100 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-omni-30b-a3b-captioner.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-omni-30b-a3b-captioner.md new file mode 100644 index 00000000..e84168ac --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/qwen3-omni-30b-a3b-captioner.md @@ -0,0 +1,153 @@ +# qwen3-omni-30b-a3b-captioner + +千问3-Omni-30b-a3b-Captioner是一款强大的音频细粒度分析模型，专为在复杂多变的音频场景中生成精准、全面的内容描述而设计，可自动解析并描述从复杂语音、环境声到音乐、影视声效等各类音频内容，能够在多声源、混合化的环境中亦保持稳定而可信的输出。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +32768 + +最大输出长度 + +32768 + +上下文长度 + +65536 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：音频 + +15.8 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +12.7 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入：音频 + +27.962 + +每百万tokens + +输出：文本（输入包含图片/音频/视频时） + +22.458 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/speech-biasing.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/speech-biasing.md new file mode 100644 index 00000000..86e2161a --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-recognition/speech-biasing.md @@ -0,0 +1,91 @@ +# speech-biasing + +热词是指用户可以预先定义的一组特定词汇或短语，这些词汇或短语在识别、翻译过程中会被赋予更高的优先级。针对您的特定业务领域，如果有部分词汇的语音识别、翻译效果不够好，可以将这些关键词或短语添加为热词进行优先识别或翻译，从而提升识别、翻译效果。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-clone-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-clone-v1.md new file mode 100644 index 00000000..a50ef4ed --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-clone-v1.md @@ -0,0 +1,93 @@ +# cosyvoice-clone-v1 + +声音复刻Cosyvoice大模型，依托先进的大模型技术进行特征提取，从而完成声音的复刻，且无需训练过程。仅需提供时长较短的音频，即可迅速生成高度相似且听感自然的定制声音。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +2 + +每万字符 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v1.md new file mode 100644 index 00000000..2b35a405 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v1.md @@ -0,0 +1,93 @@ +# cosyvoice-v1 + +CosyVoice 是通义实验室依托大规模预训练语言模型，深度融合文本理解和语音生成的新一代生成式语音合成大模型，支持文本至语音的实时流式合成。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +2 + +每万字符 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v2.md new file mode 100644 index 00000000..4682b2f9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v2.md @@ -0,0 +1,93 @@ +# cosyvoice-v2 + +cosyvoice-V2是通义实验室依托大规模预训练语言模型，在深度融合文本理解和语音生成的新一代生成式语音合成大模型，支持文本至语音的实时流式合成。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +2 + +每万字符 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-flash.md new file mode 100644 index 00000000..e8d68e1f --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-flash.md @@ -0,0 +1,105 @@ +# cosyvoice-v3.5-flash + +CosyVoice-v3.5-Flash是通义实验室CosyVoice系列的高性能语音合成大模型。对声音克隆和声音设计的语音合成效果进行全面升级，确保说话人高相似度的前提下，支持free-style指令控制，合成风格丰富多样。较之前版本大幅减少首包延迟，同时提高发音准确率，改善韵律和音质。支持跨多语种（中、英、德、法、俄、日、韩、葡、泰、印尼、越南）超自然听感实时语音合成。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-plus.md new file mode 100644 index 00000000..3cc13936 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-5-plus.md @@ -0,0 +1,105 @@ +# cosyvoice-v3.5-plus + +CosyVoice-v3.5-Plus是通义实验室CosyVoice系列的超高表现力语音合成大模型。对声音克隆和声音设计的语音合成效果进行全面升级，确保说话人高相似度的前提下，支持free-style指令控制，合成风格丰富多样。较之前版本大幅减少首包延迟，同时提高发音准确率，改善韵律和音质。支持跨多语种（中、英、德、法、俄、日、韩、葡、泰、印尼、越南）超自然听感实时语音合成。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1.5 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-flash.md new file mode 100644 index 00000000..730a99d0 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-flash.md @@ -0,0 +1,133 @@ +# cosyvoice-v3-flash + +合成能力：CosyVoice-v3-Flash是通义实验室CosyVoice系列最新版高性能的语音合成大模型，较之前版本在自然度、音质、韵律、情感表现力上有更好的表现。该模型支持文本至语音的实时流式合成。克隆能力：CosyVoice-v3-Flash也是通义实验室CosyVoice系列最新版的语音克隆大模型，较之前版本提升了发音准确性、音色相似度，并且增加了更多小语种支持（德、西、法、意、俄）。仅需提供5-20s的参考音频，即可迅速生成高度相似且听感自然的定制声音。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.9541 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-plus.md new file mode 100644 index 00000000..d5c4473a --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/cosyvoice-v3-plus.md @@ -0,0 +1,133 @@ +# cosyvoice-v3-plus + +克隆能力：CosyVoice-v3-plus是通义实验室CosyVoice系列最新版的语音克隆大模型，具有更好的音质和复刻相似度，适用于更专业的场景。仅需提供5-20s的参考音频，即可迅速生成高度相似且听感自然的定制声音。合成能力：CosyVoice-v3-plus是通义实验室CosyVoice系列最新版的语音合成大模型，具有更好的音质和表现力，适用于更专业的场景。该模型支持文本至语音的实时流式合成。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +2 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +1.9082 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/fun-music-v1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/fun-music-v1.md new file mode 100644 index 00000000..a2089a77 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/fun-music-v1.md @@ -0,0 +1,105 @@ +# fun-music-v1 + +百聆音乐生成大模型（Fun音乐大模型）支持输入开放性歌曲的创作要求或歌词，生成整首男/女声演唱的中文或英文歌曲。歌曲通俗易懂，情绪由浅入深，是人类灵感与大模型能力的完美结合。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.002 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts-realtime.md new file mode 100644 index 00000000..a5a1c8fe --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts-realtime.md @@ -0,0 +1,351 @@ +# qwen-tts-realtime + +Qwen-TTS实时模型是通义实验室“qwen系列”模型中的语音合成模型。具备双向上下文感知能力，可以低延迟高保真完成多音色、方言及长文本的双向流式生成。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +512 + +最大输出长度 + +7680 + +上下文长度 + +8192 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2.4 + +每百万tokens + +输出 + +12 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 动态更新 + +### qwen-tts-realtime-latest + +Qwen-TTS实时模型是通义实验室千问模型中语音合成利器，始终与最新快照版能力相同。具备双向上下文感知能力，可以低延迟高保真完成多音色、方言及长文本的双向流式生成。本模型是动态更新版本，模型更新不会提前通知。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +512 + +最大输出长度 + +7680 + +上下文长度 + +8192 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2.4 + +每百万tokens + +输出 + +12 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen-tts-realtime-2025-07-15 + +Qwen-TTS实时模型是通义实验室“qwen系列”模型中的语音合成利器。具备双向上下文感知能力，可以低延迟高保真完成多音色、方言及长文本的双向流式生成。此版本为2025年7月15日快照模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +512 + +最大输出长度 + +7680 + +上下文长度 + +8192 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2.4 + +每百万tokens + +输出 + +12 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts.md new file mode 100644 index 00000000..ff43d0bd --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/model-qwen-tts.md @@ -0,0 +1,467 @@ +# qwen-tts + +千问系列首个语音合成模型，支持中文、英文、中英混合输入。自适应根据输入文本调整输出语气，音色真实自然，支持流式输出。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +512 + +最大输出长度 + +7680 + +上下文长度 + +8192 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.6 + +每百万tokens + +输出：音频 + +10 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 动态更新 + +### qwen-tts-latest + +模型是动态更新版本，等同于最新版本快照模型，模型更新时不会提前通知。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +512 + +最大输出长度 + +7680 + +上下文长度 + +8192 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.6 + +每百万tokens + +输出：音频 + +10 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +## 快照版本 + +### qwen-tts-2025-05-22 + +千问系列首个语音合成模型，支持中文、英文、中英混合输入。自适应根据输入文本调整输出语气，音色真实自然，支持流式输出。本版本是2025年5月22日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +512 + +最大输出长度 + +7680 + +上下文长度 + +8192 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.6 + +每百万tokens + +输出：音频 + +10 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 + +### qwen-tts-2025-04-10 + +千问系列首个语音合成模型，支持中文、英文、中英混合输入。自适应根据输入文本调整输出语气，音色真实自然，支持流式输出。本版本是2025年4月10日快照。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +512 + +最大输出长度 + +7680 + +上下文长度 + +8192 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入：文本 + +1.6 + +每百万tokens + +输出：音频 + +10 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/music-generation-preview.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/music-generation-preview.md new file mode 100644 index 00000000..d9aa9421 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/music-generation-preview.md @@ -0,0 +1,213 @@ +# fun-music-preview + +百聆音乐生成preview版大模型（Fun音乐大模型）支持输入开放性歌曲的创作要求或歌词，生成整首男/女声演唱的中文或英文歌曲。歌曲通俗易懂，情绪由浅入深，是人类灵感与大模型能力的完美结合。本次版本为预览快照版 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.005 + +每秒 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 快照版本 + +### fun-music-preview + +百聆音乐生成preview版大模型（Fun音乐大模型）支持输入开放性歌曲的创作要求或歌词，生成整首男/女声演唱的中文或英文歌曲。歌曲通俗易懂，情绪由浅入深，是人类灵感与大模型能力的完美结合。本次版本为预览快照版 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +音频时长 + +0.005 + +每秒 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-design.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-design.md new file mode 100644 index 00000000..1be256d4 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-design.md @@ -0,0 +1,133 @@ +# qwen-voice-design + +千问voice-design模型是千问语音模型的声音设计系列模型，仅需输入简单的文字描述，即可迅速设计出符合要求的相关声音。结合qwen3-tts-vd-realtime模型使用，可设计输出10个语种的语音。且合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +声音复刻及声音设计 + +0.2 + +每次 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +声音复刻及声音设计 + +0.2 + +每次 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-enrollment.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-enrollment.md new file mode 100644 index 00000000..b66cbea6 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen-voice-enrollment.md @@ -0,0 +1,133 @@ +# qwen-voice-enrollment + +千问voice-enrollment模型是千问语音模型的声音复刻系列模型，仅需5s以上的音频，即可迅速复刻高相似度声音。结合qwen3-tts-vc-realtime模型使用，可将一个人的声音高保真复刻，输出10个语种的语音。且合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +声音复刻及声音设计 + +0.01 + +次 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.01 + +每次 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash-realtime.md new file mode 100644 index 00000000..d6d40d64 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash-realtime.md @@ -0,0 +1,403 @@ +# qwen3-tts-flash-realtime + +千问3-TTS-Flash-Realtime模型是通义实验室最新的实时语音合成大模型，不仅拥有17种高表现力的拟人音色，且能低延迟高稳定地实时合成音频；同时支持多种语言，方言，支持同一音色多语言输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型版本功能等同于快照模型 qwen3-tts-flash-realtime-2025-11-27。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.954101 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 快照版本 + +### qwen3-tts-flash-realtime-2025-11-27 + +千问3-TTS-Flash-Realtime模型是通义最新的实时语音合成大模型，不仅拥有51种高表现力的拟人音色，且能低延迟高稳定地实时合成音频；同时支持多种语言，方言，支持同一音色多语言输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2025年11月27日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.954101 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +### qwen3-tts-flash-realtime-2025-09-18 + +Qwen3-TTS-Flash-Realtime-2025-09-18模型是通义实验室最新的实时语音合成大模型，不仅拥有17种高表现力的拟人音色，且能低延迟高稳定地实时合成音频；同时支持多种语言，方言，支持同一音色多语言输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2025年9月18日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.954101 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +10 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash.md new file mode 100644 index 00000000..73c4bc3c --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-flash.md @@ -0,0 +1,403 @@ +# qwen3-tts-flash + +Qwen3-TTS-Flash模型是通义实验室最新推出的离线语音合成大模型，不仅拥有17种高表现力的拟人音色，且能低延迟高稳定地合成音频；同时支持多种语言，方言，支持同一音色多语言输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型版本功能等同于快照模型 qwen3-tts-flash-2025-11-27。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.733924 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 快照版本 + +### qwen3-tts-flash-2025-11-27 + +千问3-TTS-Flash模型是通义最新推出的离线语音合成大模型，不仅拥有51种高表现力的拟人音色，且能低延迟高稳定地合成音频；同时支持多种语言，方言，支持同一音色多语言输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2025年11月27日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.733924 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +### qwen3-tts-flash-2025-09-18 + +千问3-TTS-Flash-2025-09-18模型是通义实验室最新推出的离线语音合成大模型，不仅拥有17种高表现力的拟人音色，且能低延迟高稳定地合成音频；同时支持多种语言，方言，支持同一音色多语言输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2025年9月18日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.733924 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +10 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +10 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash-realtime.md new file mode 100644 index 00000000..e89c98a9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash-realtime.md @@ -0,0 +1,269 @@ +# qwen3-tts-instruct-flash-realtime + +千问3-TTS-Flash模型是通义实验室最新推出的实时语音合成大模型，Instruct模型可通过自然语言进行合成效果的处理，确保在不同语境下，合成情感、表达高度贴合的语音。目前支持25个音色的中英文Instruct调节。该模型等同于2026年01月22日快照版本模型。该模型版本功能等同于快照模型 qwen3-tts-instruct-flash-realtime-2026-01-22。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 快照版本 + +### qwen3-tts-instruct-flash-realtime-2026-01-22 + +千问3-TTS-Flash模型是通义实验室最新推出的实时语音合成大模型，Instruct模型可通过自然语言进行合成效果的处理，确保在不同语境下，合成情感、表达高度贴合的语音。目前支持25个音色的中英文Instruct调节。该模型为2026年01月22日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash.md new file mode 100644 index 00000000..60ac8112 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-instruct-flash.md @@ -0,0 +1,269 @@ +# qwen3-tts-instruct-flash + +Qwen3-TTS-Flash模型是通义实验室最新推出的实时语音合成大模型，Instruct模型可通过自然语言进行合成效果的处理，确保在不同语境下，合成情感、表达高度贴合的语音。目前支持25个音色的中英文Instruct调节。该模型版本功能等同于快照模型 qwen3-tts-instruct-flash-2026-01-26。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 快照版本 + +### qwen3-tts-instruct-flash-2026-01-26 + +Qwen3-TTS-Flash模型是通义实验室最新推出的实时语音合成大模型，Instruct模型可通过自然语言进行合成效果的处理，确保在不同语境下，合成情感、表达高度贴合的语音。目前支持25个音色的中英文Instruct调节。该模型为2026年01月26日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc-realtime.md new file mode 100644 index 00000000..71f9fbbc --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc-realtime.md @@ -0,0 +1,403 @@ +# qwen3-tts-vc-realtime-2026-01-15 + +千问3-TTS-Flash模型是通义最新推出的实时语音合成大模型，可对qwen-voice-enrollment服务复刻的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2026年01月15日快照版本模型。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.954101 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 快照版本 + +### qwen3-tts-vc-realtime-2026-01-15 + +千问3-TTS-Flash模型是通义最新推出的实时语音合成大模型，可对qwen-voice-enrollment服务复刻的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2026年01月15日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.954101 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +### qwen3-tts-vc-realtime-2025-11-27 + +千问3-TTS-Flash模型是通义实验室最新推出的实时语音合成大模型，可对qwen3-voice-enrollment服务复刻的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2025年11月27日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.954101 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc.md new file mode 100644 index 00000000..a6261d79 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vc.md @@ -0,0 +1,269 @@ +# qwen3-tts-vc-2026-01-22 + +Qwen3-TTS-Flash模型是通义最新推出的实时语音合成大模型，可对qwen-voice-enrollment服务复刻的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2026年01月22日快照版本模型。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 快照版本 + +### qwen3-tts-vc-2026-01-22 + +Qwen3-TTS-Flash模型是通义最新推出的实时语音合成大模型，可对qwen-voice-enrollment服务复刻的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2026年01月22日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd-realtime.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd-realtime.md new file mode 100644 index 00000000..ab514758 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd-realtime.md @@ -0,0 +1,403 @@ +# qwen3-tts-vd-realtime-2026-01-15 + +千问3-TTS-VD模型是通义实验室最新推出的实时语音合成大模型，可对qwen3-voice-design服务设计的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2026年01月15日快照版本模型。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.954101 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 快照版本 + +### qwen3-tts-vd-realtime-2026-01-15 + +千问3-TTS-VD模型是通义实验室最新推出的实时语音合成大模型，可对qwen3-voice-design服务设计的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2026年01月15日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.954101 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +### qwen3-tts-vd-realtime-2025-12-16 + +千问3-TTS-VD模型是通义实验室最新推出的实时语音合成大模型，可对qwen3-voice-design服务设计的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2025年12月16日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +1 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.954101 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd.md new file mode 100644 index 00000000..87d3a0da --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/qwen3-tts-vd.md @@ -0,0 +1,269 @@ +# qwen3-tts-vd-2026-01-26 + +Qwen3-TTS-VD模型是通义最新推出的实时语音合成大模型，可对qwen3-voice-design服务设计的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2026年01月26日快照版本模型。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 快照版本 + +### qwen3-tts-vd-2026-01-26 + +Qwen3-TTS-VD模型是通义最新推出的实时语音合成大模型，可对qwen3-voice-design服务设计的声音进行高保真实时语音合成，且同一音色支持11个语种的语音输出。该模型经过海量数据训练，合成音频可以根据文本自适应调节语气，对复杂文本合成也有较好的处理能力。该模型为2026年01月26日快照版本模型。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +语音合成 + +0.8 + +每万字符 + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +180 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +180 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/sambert.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/sambert.md new file mode 100644 index 00000000..6431df81 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/sambert.md @@ -0,0 +1,267 @@ +# Sambert语音合成 + +提供SAMBERT+NSFGAN深度神经网络算法与传统领域知识深度结合的文字转语音服务，兼具读音准确，韵律自然，声音还原度高，表现力强的特点。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +Sambert 语音合成 + +1 + +每万字符 + +## 音色列表 + +模型ID + +音色名称 + +sambert-betty-v1 + +Sambert语音合成-Betty + +sambert-brian-v1 + +Sambert语音合成-Brian + +sambert-cally-v1 + +Sambert语音合成-Cally + +sambert-camila-v1 + +Sambert语音合成-Camila + +sambert-cindy-v1 + +Sambert语音合成-Cindy + +sambert-clara-v1 + +Sambert语音合成-Clara + +sambert-donna-v1 + +Sambert语音合成-Donna + +sambert-eva-v1 + +Sambert语音合成-Eva + +sambert-hanna-v1 + +Sambert语音合成-Hanna + +sambert-indah-v1 + +Sambert语音合成-Indah + +sambert-perla-v1 + +Sambert语音合成-Perla + +sambert-waan-v1 + +Sambert语音合成-Waan + +sambert-zhichu-v1 + +Sambert语音合成-知厨 + +sambert-zhida-v1 + +Sambert语音合成-知达 + +sambert-zhide-v1 + +Sambert语音合成-知德 + +sambert-zhifei-v1 + +Sambert语音合成-知飞 + +sambert-zhigui-v1 + +Sambert语音合成-知柜 + +sambert-zhihao-v1 + +Sambert语音合成-知浩 + +sambert-zhijia-v1 + +Sambert语音合成-知佳 + +sambert-zhijing-v1 + +Sambert语音合成-知婧 + +sambert-zhilun-v1 + +Sambert语音合成-知伦 + +sambert-zhimao-v1 + +Sambert语音合成-知猫 + +sambert-zhimiao-emo-v1 + +Sambert语音合成-知妙（多情感） + +sambert-zhiming-v1 + +Sambert语音合成-知茗 + +sambert-zhimo-v1 + +Sambert语音合成-知墨 + +sambert-zhina-v1 + +Sambert语音合成-知娜 + +sambert-zhinan-v1 + +Sambert语音合成-知楠 + +sambert-zhiqi-v1 + +Sambert语音合成-知琪 + +sambert-zhiqian-v1 + +Sambert语音合成-知倩 + +sambert-zhiru-v1 + +Sambert语音合成-知茹 + +sambert-zhishu-v1 + +Sambert语音合成-知树 + +sambert-zhishuo-v1 + +Sambert语音合成-知硕 + +sambert-zhistella-v1 + +Sambert语音合成-知莎 + +sambert-zhiting-v1 + +Sambert语音合成-知婷 + +sambert-zhiwei-v1 + +Sambert语音合成-知薇 + +sambert-zhixiang-v1 + +Sambert语音合成-知祥 + +sambert-zhixiao-v1 + +Sambert语音合成-知笑 + +sambert-zhiya-v1 + +Sambert语音合成-知雅 + +sambert-zhiye-v1 + +Sambert语音合成-知晔 + +sambert-zhiying-v1 + +Sambert语音合成-知颖 + +sambert-zhiyuan-v1 + +Sambert语音合成-知媛 + +sambert-zhiyue-v1 + +Sambert语音合成-知悦 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-hd.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-hd.md new file mode 100644 index 00000000..78dfdd06 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-hd.md @@ -0,0 +1,115 @@ +# MiniMax/speech-02-hd + +MiniMax 语音大模型能够根据上下文，智能预测文本的情绪、语调等信息，并生成超自然、高保真、个性化的语音。在社交、播客、有声书、新闻资讯、教育、数字人等多种场景中展现出强大的实力。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +10000 + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +声音复刻及声音设计 + +9.9 + +每次 + +语音合成 + +3.5 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +20 + +TPM（每分钟tokens） + +20,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-turbo.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-turbo.md new file mode 100644 index 00000000..495a8092 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-02-turbo.md @@ -0,0 +1,115 @@ +# MiniMax/speech-02-turbo + +MiniMax 语音大模型能够根据上下文，智能预测文本的情绪、语调等信息，并生成超自然、高保真、个性化的语音。在社交、播客、有声书、新闻资讯、教育、数字人等多种场景中展现出强大的实力。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +10000 + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +声音复刻及声音设计 + +9.9 + +每次 + +语音合成 + +2 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +20 + +TPM（每分钟tokens） + +20,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-hd.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-hd.md new file mode 100644 index 00000000..dab06c52 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-hd.md @@ -0,0 +1,115 @@ +# MiniMax/speech-2.8-hd + +MiniMax 语音大模型能够根据上下文，智能预测文本的情绪、语调等信息，并生成超自然、高保真、个性化的语音。在社交、播客、有声书、新闻资讯、教育、数字人等多种场景中展现出强大的实力。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +10000 + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +声音复刻及声音设计 + +9.9 + +每次 + +语音合成 + +3.5 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +20 + +TPM（每分钟tokens） + +20,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-turbo.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-turbo.md new file mode 100644 index 00000000..36d321db --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/speech-2-8-turbo.md @@ -0,0 +1,115 @@ +# MiniMax/speech-2.8-turbo + +MiniMax 语音大模型能够根据上下文，智能预测文本的情绪、语调等信息，并生成超自然、高保真、个性化的语音。在社交、播客、有声书、新闻资讯、教育、数字人等多种场景中展现出强大的实力。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +10000 + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +声音复刻及声音设计 + +9.9 + +每次 + +语音合成 + +2 + +每万字符 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +20 + +TPM（每分钟tokens） + +20,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/voice-enrollment.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/voice-enrollment.md new file mode 100644 index 00000000..2b2ce542 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-speech-synthesis/voice-enrollment.md @@ -0,0 +1,103 @@ +# voice-enrollment + +大模型声音复刻服务依托先进的大模型技术进行特征提取，无需训练过程就可以完成声音的复刻。仅需提供极短的音频，即可迅速生成高度相似且听感自然的定制声音。 大模型声音设计使用FunAudioGen-VD模型，支持通过文本Prompt描述，创造声音。无需受限任何音频质量，根据目标场景对音色、语气、语调、语速、情绪等各方面表现力的需求描述，即可生成高质量语音。高度还原专业配音演员的演出水准。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Audio** + +输出模态 + +**Audio** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +— + +最大输出长度 + +— + +上下文长度 + +— + +## 模型价格 + +暂无公开定价信息。 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +600 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +600 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-ocr.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-ocr.md new file mode 100644 index 00000000..a23ee63f --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-ocr.md @@ -0,0 +1,115 @@ +# vanchin/deepseek-ocr + +DeepSeek-OCR以 “探索视觉 - 文本压缩边界” 为核心目标，从大语言模型（LLM）视角重新定义视觉编码器的功能定位，为文档识别、图像转文本等高频场景提供了兼顾精度与效率的全新解决方案。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +8192 + +最大输出长度 + +8192 + +上下文长度 + +8192 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +0.216 + +每百万tokens + +输出 + +0.216 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-1-5b.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-1-5b.md new file mode 100644 index 00000000..a1e8129c --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-1-5b.md @@ -0,0 +1,95 @@ +# deepseek-r1-distill-qwen-1.5b + +DeepSeek-R1-Distill-Qwen-1.5B是一个基于Qwen2.5-Math-1.5B的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +32768 + +最大输出长度 + +16384 + +上下文长度 + +32768 + +## 模型价格 + +暂无公开定价信息。 + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-14b.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-14b.md new file mode 100644 index 00000000..2bc285dc --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-14b.md @@ -0,0 +1,115 @@ +# deepseek-r1-distill-qwen-14b + +DeepSeek-R1-Distill-Qwen-14B是一个基于Qwen2.5-14B的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +32768 + +最大输出长度 + +16384 + +上下文长度 + +32768 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +1 + +每百万tokens + +输出 + +3 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-32b.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-32b.md new file mode 100644 index 00000000..1f337274 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-32b.md @@ -0,0 +1,115 @@ +# deepseek-r1-distill-qwen-32b + +DeepSeek-R1-Distill-Qwen-32B是一个基于Qwen2.5-32B的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +32768 + +最大输出长度 + +16384 + +上下文长度 + +32768 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2 + +每百万tokens + +输出 + +6 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-7b.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-7b.md new file mode 100644 index 00000000..82c2f6c5 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1-distill-qwen-7b.md @@ -0,0 +1,115 @@ +# deepseek-r1-distill-qwen-7b + +DeepSeek-R1-Distill-Qwen-7B是一个基于Qwen2.5-Math-7B的蒸馏大型语言模型，使用了 DeepSeek R1 的输出。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +32768 + +最大输出长度 + +16384 + +上下文长度 + +32768 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +0.5 + +每百万tokens + +输出 + +1 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1.md new file mode 100644 index 00000000..63cf7b03 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-r1.md @@ -0,0 +1,509 @@ +# deepseek-r1 + +DeepSeek-R1 在后训练阶段大规模使用了强化学习技术，在仅有极少标注数据的情况下，极大提升了模型推理能力。在数学、代码、自然语言推理等任务上，性能较高，能力较强。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +98304 + +最大输出长度 + +16384 + +上下文长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +16 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +输入（Batch File） + +2 + +每百万tokens + +输出（Batch File） + +8 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 + +## siliconflow/deepseek-r1-0528 + +DeepSeek-R1-0528 是一款强化学习（RL）驱动的推理模型，解决了模型中的重复性和可读性问题。在 RL 之前，DeepSeek-R1 引入了冷启动数据，进一步优化了推理性能。它在数学、代码和推理任务中与 OpenAI-o1 表现相当，并且通过精心设计的训练方法，提升了整体效果。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +163840 + +最大输出长度 + +32768 + +上下文长度 + +163840 + +最大输入长度（思考模式下） + +163840 + +最大输出长度（思考模式下） + +32768 + +最大思维链长度 + +65536 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +16 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +500,000 + +## vanchin/deepseek-r1 + +DeepSeek-R1 是深度求索于 2025 年 1 月开源的 6710 亿参数混合专家（MoE）推理模型，推理时仅激活 370 亿参数。作为首个通过纯强化学习（无监督微调）训练的千亿级模型，实现了链式思维（CoT）的自然涌现。模型在 RL 前加入冷启动数据解决了 R1-Zero 的重复和混语问题，在数学、代码、推理任务上达到 OpenAI o1 水平。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +98304 + +最大输出长度 + +32768 + +上下文长度 + +131072 + +最大输入长度（思考模式下） + +98304 + +最大输出长度（思考模式下） + +32768 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +16 + +每百万tokens + +输入（缓存命中） + +1.6 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 快照版本 + +### deepseek-r1-0528 + +0528为R1模型的小版本升级，相较于旧版 R1，新版在复杂推理任务中的表现有了显著提升。在数学、编程与通用逻辑等多个基准测评中取得了优异成绩。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +98304 + +最大输出长度 + +16384 + +上下文长度 + +131072 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +16 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +100,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1-terminus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1-terminus.md new file mode 100644 index 00000000..ba3b48fe --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1-terminus.md @@ -0,0 +1,257 @@ +# vanchin/deepseek-v3.1-terminus + +DeepSeek-V3.1-Terminus 是由深度求索（DeepSeek）发布的 V3.1 模型的更新版本，定位为混合智能体大语言模型。此次更新在保持模型原有能力的基础上，专注于修复用户反馈的问题并提升稳定性。它显著改善了语言一致性，减少了中英文混用和异常字符的出现。模型集成了“思考模式”（Thinking Mode）和“非思考模式”（Non-thinking Mode），用户可通过聊天模板灵活切换以适应不同任务。作为一个重要的优化，V3.1-Terminus 增强了代码智能体（Code Agent）和搜索智能体（Search Agent）的性能，使其在工具调用和执行多步复杂任务方面更加可靠 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +163840 + +最大输出长度 + +65536 + +上下文长度 + +163840 + +最大输入长度（思考模式下） + +163840 + +最大输出长度（思考模式下） + +65536 + +最大思维链长度 + +65536 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +12 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +500,000 + +## vanchin/deepseek-v3.1-terminus + +DeepSeek-V3.1-Terminus 是 DeepSeek-V3.1 的更新版本，旨在保持模型原有核心能力的同时，针对用户反馈的问题进行了修复和优化。该版本的模型结构与 DeepSeek-V3 保持一致，并在特定领域进行了显著增强。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +32768 + +最大输出长度 + +65536 + +上下文长度 + +131072 + +最大输入长度（思考模式下） + +32768 + +最大输出长度（思考模式下） + +65536 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +12 + +每百万tokens + +输入（缓存命中） + +1.6 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1.md new file mode 100644 index 00000000..7b62c749 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-1.md @@ -0,0 +1,121 @@ +# deepseek-v3.1 + +DeepSeek-V3.1为混合推理架构模型，同时支持思考模式与非思考模式，具备更高的推理效率和更强的Agent能力。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +98304 + +最大输出长度 + +65536 + +上下文长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +12 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-exp.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-exp.md new file mode 100644 index 00000000..1dcad555 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-exp.md @@ -0,0 +1,115 @@ +# deepseek-v3.2-exp + +引入了DeepSeek Sparse Attention（一种稀疏注意力机制）的实验性质版本，针对长文本的训练和推理效率进行了探索性的优化和验证。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +98304 + +最大输出长度 + +65536 + +上下文长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2 + +每百万tokens + +输出 + +3 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-think.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-think.md new file mode 100644 index 00000000..61b10aa8 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-2-think.md @@ -0,0 +1,129 @@ +# vanchin/deepseek-v3.2-think + +DeepSeek-V3.2 是一款实现了高计算效率与卓越推理及代理（Agent）性能完美协调的模型。该模型建立在 DeepSeek-V3 的基础之上，通过引入 DeepSeek 稀疏注意力（DSA）、可扩展的强化学习框架以及大规模代理任务合成流水线等关键技术突破，推动了开源大语言模型的前沿发展。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +98304 + +最大输出长度 + +65536 + +上下文长度 + +131072 + +最大输入长度（思考模式下） + +98304 + +最大输出长度（思考模式下） + +65536 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2 + +每百万tokens + +输出 + +3 + +每百万tokens + +输入（缓存命中） + +0.2 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +30 + +TPM（每分钟tokens） + +600,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-3.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-3.md new file mode 100644 index 00000000..20ddf8c9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3-3.md @@ -0,0 +1,341 @@ +# deepseek-v3.2 + +DeepSeek-V3.2是引入DeepSeek Sparse Attention（一种稀疏注意力机制）的正式版模型，也是DeepSeek推出的首个将思考融入工具使用的模型，同时支持思考模式与非思考模式的工具调用。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +98304 + +最大输出长度 + +65536 + +上下文长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2 + +每百万tokens + +输出 + +3 + +每百万tokens + +输入（缓存命中） + +0.4 + +每百万tokens + +输入（Batch File） + +1 + +每百万tokens + +输出（Batch File） + +1.5 + +每百万tokens + +显式缓存创建 + +2.5 + +每百万tokens + +显式缓存命中 + +0.2 + +每百万tokens + +输入（Batch Chat) + +2 + +每百万tokens + +输出（Batch Chat) + +3 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入 + +4.272 + +每百万tokens + +输出 + +12.815 + +每百万tokens + +输入（缓存命中） + +0.854 + +每百万tokens + +显式缓存创建 + +5.343 + +每百万tokens + +显式缓存命中 + +0.427 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +200,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +10000 + +TPM（每分钟tokens） + +1,200,000 + +## siliconflow/deepseek-v3.2 + +DeepSeek-V3.2 是一款兼具高计算效率与卓越推理和 Agent 性能的模型。其方法建立在三大关键技术突破之上：DeepSeek 稀疏注意力（DSA），一种高效的注意力机制，在保持模型性能的同时显著降低了计算复杂性，并特别针对长上下文场景进行了优化；可扩展的强化学习框架，通过该框架，模型性能可与 GPT-5 相媲美，其高算力版本在推理能力上可与 Gemini-3.0-Pro 匹敌；以及大规模 Agent 任务合成管线，旨在将推理能力整合到工具使用场景中，从而提高在复杂交互环境中的指令遵循和泛化能力。该模型在 2025 年国际数学奥林匹克（IMO）和国际信息学奥林匹克（IOI）中取得了金牌表现 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +163840 + +最大输出长度 + +65536 + +上下文长度 + +163840 + +最大输入长度（思考模式下） + +163840 + +最大输出长度（思考模式下） + +65536 + +最大思维链长度 + +65536 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2 + +每百万tokens + +输出 + +3 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +500,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3.md new file mode 100644 index 00000000..0a7f35ca --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v3.md @@ -0,0 +1,371 @@ +# deepseek-v3 + +DeepSeek-V3 为自研 MoE 模型，671B 参数，激活 37B，在 14.8T token 上进行了预训练，在长文本、代码、数学、百科、中文 能力上表现优秀。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +57344 + +最大输出长度 + +8192 + +上下文长度 + +65536 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2 + +每百万tokens + +输出 + +8 + +每百万tokens + +输入（缓存命中） + +0.4 + +每百万tokens + +输入（Batch File） + +1 + +每百万tokens + +输出（Batch File） + +4 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 + +## siliconflow/deepseek-v3-0324 + +新版 DeepSeek-V3 （DeepSeek-V3-0324）与之前的 DeepSeek-V3-1226 使用同样的 base 模型，仅改进了后训练方法。新版 V3 模型借鉴 DeepSeek-R1 模型训练过程中所使用的强化学习技术，大幅提高了在推理类任务上的表现水平，在数学、代码类相关评测集上取得了超过 GPT-4.5 的得分成绩。此外该模型在工具调用、角色扮演、问答闲聊等方面也得到了一定幅度的能力提升。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +163840 + +最大输出长度 + +163840 + +上下文长度 + +163840 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2 + +每百万tokens + +输出 + +8 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +500,000 + +## vanchin/deepseek-v3 + +DeepSeek-V3 由深度求索（DeepSeek）于 2024 年 12 月发布，是目前开源社区领先的混合专家（MoE）语言模型：总参数 671B，每个 token 仅激活 37B 参数。模型在 14.8 万亿高质量 tokens 上完成预训练，原生支持 128k 上下文。通过创新的无辅助损失负载均衡策略、多头潜在注意力（MLA）架构和 FP8 混合精度训练。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +131072 + +最大输出长度 + +16384 + +上下文长度 + +131072 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2 + +每百万tokens + +输出 + +8 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash-us.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash-us.md new file mode 100644 index 00000000..9da1cfa5 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash-us.md @@ -0,0 +1,107 @@ +# deepseek-v4-flash-us + +高效轻量化MoE模型，总参284B，激活13B，原生支持百万超长上下文能力。推理速度快、延迟低、调用成本低廉，综合能力均衡，主打高并发、轻量化任务，适合日常对话、内容创作、基础 RAG、批量文案处理等普惠刚需场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1000000 + +最大输出长度 + +393216 + +上下文长度 + +1000000 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 美国（弗吉尼亚） + +部署范围：美国 + +计费项 + +价格（元） + +单位 + +输入 + +1.499 + +每百万tokens + +输出 + +2.998 + +每百万tokens + +输入（缓存命中） + +0.3 + +每百万tokens diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash.md new file mode 100644 index 00000000..f3c4bf43 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-flash.md @@ -0,0 +1,297 @@ +# deepseek-v4-flash + +高效轻量化MoE模型，总参284B，激活13B，原生支持百万超长上下文能力。推理速度快、延迟低、调用成本低廉，综合能力均衡，主打高并发、轻量化任务，适合日常对话、内容创作、基础 RAG、批量文案处理等普惠刚需场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1000000 + +最大输出长度 + +393216 + +上下文长度 + +1000000 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +1 + +每百万tokens + +输出 + +2 + +每百万tokens + +输入（缓存命中） + +0.2 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入 + +1.499 + +每百万tokens + +输出 + +2.998 + +每百万tokens + +输入（缓存命中） + +0.3 + +每百万tokens + +## 德国（法兰克福） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +1 + +每百万tokens + +输出 + +2 + +每百万tokens + +输入（缓存命中） + +0.2 + +每百万tokens + +## 美国（弗吉尼亚） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +1 + +每百万tokens + +输出 + +2 + +每百万tokens + +输入（缓存命中） + +0.2 + +每百万tokens + +## 日本（东京） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +1 + +每百万tokens + +输出 + +2 + +每百万tokens + +输入（缓存命中） + +0.2 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +10000 + +TPM（每分钟tokens） + +1,200,000 + +## 德国（法兰克福） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 + +## 美国（弗吉尼亚） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 + +## 日本（东京） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro-us.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro-us.md new file mode 100644 index 00000000..b7a67afe --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro-us.md @@ -0,0 +1,107 @@ +# deepseek-v4-pro-us + +旗舰级 MoE 大模型，总参1.6T、激活 49B，原生支持百万级超长上下文。依托海量高质量训练数据，具备顶尖数学逻辑、复杂推理、专业代码与长文本深度解析能力，适配高阶科研、复杂办公、深度智能代理等高难度场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1000000 + +最大输出长度 + +393216 + +上下文长度 + +1000000 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 美国（弗吉尼亚） + +部署范围：美国 + +计费项 + +价格（元） + +单位 + +输入 + +17.986 + +每百万tokens + +输出 + +35.972 + +每百万tokens + +输入（缓存命中） + +1.499 + +每百万tokens diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro.md new file mode 100644 index 00000000..3ba6d7a1 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/deepseek-v4-pro.md @@ -0,0 +1,427 @@ +# deepseek-v4-pro + +旗舰级 MoE 大模型，总参1.6T、激活 49B，原生支持百万级超长上下文。依托海量高质量训练数据，具备顶尖数学逻辑、复杂推理、专业代码与长文本深度解析能力，适配高阶科研、复杂办公、深度智能代理等高难度场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1000000 + +最大输出长度 + +393216 + +上下文长度 + +1000000 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +12 + +每百万tokens + +输出 + +24 + +每百万tokens + +输入（缓存命中） + +1 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入 + +17.986 + +每百万tokens + +输出 + +35.972 + +每百万tokens + +输入（缓存命中） + +1.499 + +每百万tokens + +## 德国（法兰克福） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +12 + +每百万tokens + +输出 + +24 + +每百万tokens + +输入（缓存命中） + +1 + +每百万tokens + +## 美国（弗吉尼亚） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +12 + +每百万tokens + +输出 + +24 + +每百万tokens + +输入（缓存命中） + +1 + +每百万tokens + +## 日本（东京） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +12 + +每百万tokens + +输出 + +24 + +每百万tokens + +输入（缓存命中） + +1 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +10000 + +TPM（每分钟tokens） + +1,200,000 + +## 德国（法兰克福） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 + +## 美国（弗吉尼亚） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 + +## 日本（东京） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +15000 + +TPM（每分钟tokens） + +1,200,000 + +## vanchin/deepseek-v4-pro + +DeepSeek-V4系列是强大的混合专家（MoE）语言模型，包含DeepSeek-V4-Pro（1.6T总参数，49B激活参数）。支持高达100万（1M）token的上下文长度，是在超过32T高质量多样化token上预训练的开源模型。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1048576 + +最大输出长度 + +393216 + +上下文长度 + +1048576 + +最大输入长度（思考模式下） + +1048576 + +最大输出长度（思考模式下） + +393216 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +12 + +每百万tokens + +输出 + +24 + +每百万tokens + +输入（缓存命中） + +1 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +300,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/farui-plus.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/farui-plus.md new file mode 100644 index 00000000..549f4f76 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/farui-plus.md @@ -0,0 +1,115 @@ +# farui-plus + +通义法睿是以通义千问为基座经法律行业数据和知识专门训练的法律行业大模型产品，综合运用了模型精调、强化学习、 RAG检索增强、法律Agent技术，具有回答法律问题、推理法律适用、推荐裁判类案、辅助案情分析、生成法律文书、检索法律知识、审查合同条款等功能 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +12000 + +最大输出长度 + +2000 + +上下文长度 + +12000 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +20 + +每百万tokens + +输出 + +20 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +240 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-5-air.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-5-air.md new file mode 100644 index 00000000..0ccabaec --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-5-air.md @@ -0,0 +1,137 @@ +# glm-4.5-air + +GLM-4.5-Air采用混合专家（MoE）架构，总参数量为1060亿，激活参数120亿，相较GLM-4.5更紧凑、轻量，适用于对模型规模和资源消耗有一定限制的场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +98304 + +最大输出长度 + +16384 + +上下文长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +0.8 + +每百万tokens + +输出 + +6 + +每百万tokens + +**32k<输入<=128k** + +计费项 + +价格（元） + +单位 + +输入 + +1.2 + +每百万tokens + +输出 + +8 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-6.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-6.md new file mode 100644 index 00000000..fa746387 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-6.md @@ -0,0 +1,161 @@ +# glm-4.6 + +GLM新一代旗舰模型，核心能力较4.5全面提升。总参数量为3550 亿，激活参数320亿，上下文窗口扩展至200K。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +169984 + +最大输出长度 + +16384 + +上下文长度 + +202752 + +最大输入长度（思考模式下） + +169984 + +最大输出长度（思考模式下） + +16384 + +最大思维链长度 + +32768 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +3 + +每百万tokens + +输出 + +14 + +每百万tokens + +输入（缓存命中） + +0.6 + +每百万tokens + +**32k<输入<=200k** + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +16 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-7.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-7.md new file mode 100644 index 00000000..cbca8e9b --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4-7.md @@ -0,0 +1,149 @@ +# glm-4.7 + +智谱最新旗舰，具备更强的编程能力与更稳定的多步骤推理/执行能力。总参数355B，支持长程任务规划、编码、工具协同，问答自然、写作沉浸、创意角色扮演能力强。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +169984 + +最大输出长度 + +16384 + +上下文长度 + +202752 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +3 + +每百万tokens + +输出 + +14 + +每百万tokens + +输入（缓存命中） + +0.6 + +每百万tokens + +**32k<输入<=200k** + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +16 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4.md new file mode 100644 index 00000000..0ac548d6 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-4.md @@ -0,0 +1,137 @@ +# glm-4.5 + +GLM-4.5采用混合专家（MoE）架构，总参数量为3550 亿，激活参数320亿，在复杂推理、代码生成及智能体交互等通用能力上实现了能力融合与技术突破。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +98304 + +最大输出长度 + +16384 + +上下文长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +3 + +每百万tokens + +输出 + +14 + +每百万tokens + +**32k<输入<=128k** + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +16 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +60 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-1.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-1.md new file mode 100644 index 00000000..e068ce80 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-1.md @@ -0,0 +1,647 @@ +# glm-5.1 + +GLM-5.1是智谱AI推出的面向长程任务（Long Horizon Task）设计的模型，总参数744B，支持200K超长上下文，最大输出 128K tokens。拥有强大逻辑推理、长文本理解与代码生成能力、兼顾性能与推理效率；在多任务基准中表现优异，适用于智能交互、企业应用、开发辅助等场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +202745 + +最大输出长度 + +131072 + +上下文长度 + +202745 + +最大输入长度（思考模式下） + +169984 + +最大输出长度（思考模式下） + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +6 + +每百万tokens + +输出 + +24 + +每百万tokens + +输入（缓存命中） + +1.2 + +每百万tokens + +显式缓存创建 + +7.5 + +每百万tokens + +显式缓存命中 + +0.6 + +每百万tokens + +**32k<输入<=200k** + +计费项 + +价格（元） + +单位 + +输入 + +8 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +1.6 + +每百万tokens + +显式缓存创建 + +10 + +每百万tokens + +显式缓存命中 + +0.8 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入 + +10.492 + +每百万tokens + +输出 + +32.974 + +每百万tokens + +输入（缓存命中） + +1.948 + +每百万tokens + +## 德国（法兰克福） + +部署范围：全球 + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +6 + +每百万tokens + +输出 + +24 + +每百万tokens + +输入（缓存命中） + +1.2 + +每百万tokens + +显式缓存创建 + +7.5 + +每百万tokens + +显式缓存命中 + +0.6 + +每百万tokens + +**32k<输入<=200k** + +计费项 + +价格（元） + +单位 + +输入 + +8 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +1.6 + +每百万tokens + +显式缓存创建 + +10 + +每百万tokens + +显式缓存命中 + +0.8 + +每百万tokens + +## 美国（弗吉尼亚） + +部署范围：全球 + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +6 + +每百万tokens + +输出 + +24 + +每百万tokens + +输入（缓存命中） + +1.2 + +每百万tokens + +显式缓存创建 + +7.5 + +每百万tokens + +显式缓存命中 + +0.6 + +每百万tokens + +**32k<输入<=200k** + +计费项 + +价格（元） + +单位 + +输入 + +8 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +1.6 + +每百万tokens + +显式缓存创建 + +10 + +每百万tokens + +显式缓存命中 + +0.8 + +每百万tokens + +## 日本（东京） + +部署范围：全球 + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +6 + +每百万tokens + +输出 + +24 + +每百万tokens + +输入（缓存命中） + +1.2 + +每百万tokens + +显式缓存创建 + +7.5 + +每百万tokens + +显式缓存命中 + +0.6 + +每百万tokens + +**32k<输入<=200k** + +计费项 + +价格（元） + +单位 + +输入 + +8 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +1.6 + +每百万tokens + +显式缓存创建 + +10 + +每百万tokens + +显式缓存命中 + +0.8 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 德国（法兰克福） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 美国（弗吉尼亚） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 日本（东京） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## ZHIPU/GLM-5.1 + +GLM-5.1 是智谱最新旗舰模型，代码能力大大增强，长程任务显著提升，能够在单次任务中持续、自主地工作长达 8 小时，完成从规划、执行到迭代优化的完整闭环，交付工程级成果。 在综合能力与 Coding 能力上，GLM-5.1 整体表现对齐 Claude Opus 4.6，并在长程自主执行、复杂工程优化与真实开发场景中展现出更强的持续工作能力，是构建 Autonomous Agent 与长程 Coding Agent 的理想基座。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +204800 + +最大输出长度 + +131072 + +上下文长度 + +204800 + +最大输入长度（思考模式下） + +204800 + +最大输出长度（思考模式下） + +131072 + +最大思维链长度 + +131072 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +8 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +2 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +200 + +TPM（每分钟tokens） + +3,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-fast.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-fast.md new file mode 100644 index 00000000..f3dc2f54 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-fast.md @@ -0,0 +1,349 @@ +# glm-5.2-fast-preview + +GLM-5.2-Fast-Preview 是智谱 AI 旗舰模型 GLM-5.2 的高速版本，支持 1M 超长上下文，模型能力对齐 GLM-5.2 标准版，具备逻辑推理、长文本理解与代码生成能力。通过推理加速优化，输出 TPS 可达 GLM-5.2 标准版的 1.5～2 倍，显著提升输出速度，适用于实时对话、Agent 多轮调用、流式代码生成等对输出速度敏感的场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1048576 + +最大输出长度 + +131072 + +上下文长度 + +1048576 + +最大输入长度（思考模式下） + +1048576 + +最大输出长度（思考模式下） + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +16 + +每百万tokens + +输出 + +56 + +每百万tokens + +输入（缓存命中） + +4 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入 + +20.984 + +每百万tokens + +输出 + +65.948 + +每百万tokens + +输入（缓存命中） + +4.196 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 快照版本 + +### glm-5.2-fast-preview + +GLM-5.2-Fast-Preview 是智谱 AI 旗舰模型 GLM-5.2 的高速版本，支持 1M 超长上下文，模型能力对齐 GLM-5.2 标准版，具备逻辑推理、长文本理解与代码生成能力。通过推理加速优化，输出 TPS 可达 GLM-5.2 标准版的 1.5～2 倍，显著提升输出速度，适用于实时对话、Agent 多轮调用、流式代码生成等对输出速度敏感的场景。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1048576 + +最大输出长度 + +131072 + +上下文长度 + +1048576 + +最大输入长度（思考模式下） + +1048576 + +最大输出长度（思考模式下） + +131072 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +16 + +每百万tokens + +输出 + +56 + +每百万tokens + +输入（缓存命中） + +4 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入 + +20.984 + +每百万tokens + +输出 + +65.948 + +每百万tokens + +输入（缓存命中） + +4.196 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-us.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-us.md new file mode 100644 index 00000000..7b563335 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2-us.md @@ -0,0 +1,119 @@ +# glm-5.2-us + +GLM-5.2-US是智谱AI推出的面向长程任务（Long Horizon Task）设计的最新旗舰模型，支持1M超长上下文。拥有强大逻辑推理、长文本理解与代码生成能力、兼顾性能与推理效率；在多任务基准中表现优异，适用于智能交互、企业应用、开发辅助等场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1048576 + +最大输出长度 + +131072 + +上下文长度 + +1048576 + +最大输入长度（思考模式下） + +1048576 + +最大输出长度（思考模式下） + +131072 + +最大思维链长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 美国（弗吉尼亚） + +部署范围：美国 + +计费项 + +价格（元） + +单位 + +输入 + +10.492 + +每百万tokens + +输出 + +32.974 + +每百万tokens + +输入（缓存命中） + +2.098 + +每百万tokens diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2.md new file mode 100644 index 00000000..b764ae15 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-5-2.md @@ -0,0 +1,399 @@ +# glm-5.2 + +GLM-5.2是智谱AI推出的面向长程任务（Long Horizon Task）设计的最新旗舰模型，支持1M超长上下文。拥有强大逻辑推理、长文本理解与代码生成能力、兼顾性能与推理效率；在多任务基准中表现优异，适用于智能交互、企业应用、开发辅助等场景。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1048576 + +最大输出长度 + +131072 + +上下文长度 + +1048576 + +最大输入长度（思考模式下） + +1048576 + +最大输出长度（思考模式下） + +131072 + +最大思维链长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +8 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +2 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入 + +10.492 + +每百万tokens + +输出 + +32.974 + +每百万tokens + +输入（缓存命中） + +2.098 + +每百万tokens + +## 德国（法兰克福） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +8 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +2 + +每百万tokens + +## 美国（弗吉尼亚） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +8 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +2 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +2,000,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 德国（法兰克福） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +2,000,000 + +## 美国（弗吉尼亚） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +2,000,000 + +## ZHIPU/GLM-5.2 + +智谱原厂直供，最新旗舰模型。GLM-5.2 是智谱迄今能力最强的开源模型，支持真正可用的 1M 上下文，并在长程任务中继续保持领先。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1048576 + +最大输出长度 + +131072 + +上下文长度 + +1048576 + +最大输入长度（思考模式下） + +1048576 + +最大输出长度（思考模式下） + +131072 + +最大思维链长度 + +131072 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +8 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +2 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +200 + +TPM（每分钟tokens） + +3,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-9.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-9.md new file mode 100644 index 00000000..42af4273 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/glm-9.md @@ -0,0 +1,295 @@ +# glm-5 + +GLM-5是面向Coding与Agent场景的新一代大模型，在复杂系统工程与长程任务中达到开源 SOTA，真实编程体验逼近 Claude Opus 级别；基于 744B 新基座、异步强化学习与稀疏注意力，实现从“写代码”到“写工程”的全面升级。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +169984 + +最大输出长度 + +16384 + +上下文长度 + +202752 + +最大输入长度（思考模式下） + +169984 + +最大输出长度（思考模式下） + +16384 + +最大思维链长度 + +32768 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +18 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +**32k<输入<=200k** + +计费项 + +价格（元） + +单位 + +输入 + +6 + +每百万tokens + +输出 + +22 + +每百万tokens + +输入（缓存命中） + +1.2 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## ZHIPU/GLM-5 + +智谱新一代旗舰基座,面向AgenticEngineering,实现从代码到工程的范式跃迁,擅长复杂系统工程与长程Agent任务。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +204800 + +最大输出长度 + +131072 + +上下文长度 + +204800 + +最大输入长度（思考模式下） + +204800 + +最大输出长度（思考模式下） + +131072 + +最大思维链长度 + +131072 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +6 + +每百万tokens + +输出 + +22 + +每百万tokens + +输入（缓存命中） + +1.5 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +200 + +TPM（每分钟tokens） + +3,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-5.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-5.md new file mode 100644 index 00000000..e340883d --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-5.md @@ -0,0 +1,379 @@ +# kimi-k2.5 + +kimi-k2.5是月之暗面迄今发布最全能的模型，原生多模态架构设计，同时支持视觉与文本输入、思考与非思考模式、对话与Agent任务。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +229376 + +最大输出长度 + +16384 + +上下文长度 + +262144 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +21 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +显式缓存创建 + +5 + +每百万tokens + +显式缓存命中 + +0.4 + +每百万tokens + +## 德国（法兰克福） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +21 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +显式缓存创建 + +5 + +每百万tokens + +显式缓存命中 + +0.4 + +每百万tokens + +## 美国（弗吉尼亚） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +21 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +显式缓存创建 + +5 + +每百万tokens + +显式缓存命中 + +0.4 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 德国（法兰克福） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 美国（弗吉尼亚） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## kimi/kimi-k2.5 + +Kimi K2.5 是 Kimi 在2026年最新推出的智能模型，在 Agent、代码、视觉理解及一系列通用智能任务上取得开源 SoTA 表现。同时 Kimi K2.5 也是 Kimi 迄今最全能的模型，原生的多模态架构设计，同时支持视觉与文本输入、思考与非思考模式、对话与 Agent 任务。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +262144 + +最大输出长度 + +262144 + +上下文长度 + +262144 + +最大输入长度（思考模式下） + +262144 + +最大输出长度（思考模式下） + +262144 + +最大思维链长度 + +262144 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +21 + +每百万tokens + +输入（缓存命中） + +0.7 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +3,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-6.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-6.md new file mode 100644 index 00000000..36afdaef --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-6.md @@ -0,0 +1,267 @@ +# kimi-k2.6 + +kimi-k2.6是Kimi最新最智能的模型，具备更强更稳的长程代码编写能力，指令遵循和自我纠错能力显著提升，同时支持文本、图片与视频输入，思考与非思考模式，对话与Agent任务。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Text**　**Video** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +229376 + +最大输出长度 + +16384 + +上下文长度 + +262144 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +6.5 + +每百万tokens + +输出 + +27 + +每百万tokens + +输入（缓存命中） + +1.3 + +每百万tokens + +显式缓存创建 + +8.125 + +每百万tokens + +显式缓存命中 + +0.65 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## kimi/kimi-k2.6 + +Kimi K2.6 是 Kimi 最新最智能的模型，Kimi K2.6 的通用 Agent、代码、视觉理解等综合能力得到全面提升，其中在博士级难度的完整版人类最后的考试（Humanity’s Last Exam）、在考察模型真实软件工程能力的 SWE-Bench Pro、评估 Agent 深度检索能力的 DeepSearchQA 等基准测试中均取得行业领先的成绩，同时支持文本、图片与视频输入，思考与非思考模式，对话与 Agent 任务。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +262144 + +最大输出长度 + +262144 + +上下文长度 + +262144 + +最大输入长度（思考模式下） + +262144 + +最大输出长度（思考模式下） + +262144 + +最大思维链长度 + +262144 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +6.5 + +每百万tokens + +输出 + +27 + +每百万tokens + +输入（缓存命中） + +1.1 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +3,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code-highspeed.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code-highspeed.md new file mode 100644 index 00000000..9bf20482 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code-highspeed.md @@ -0,0 +1,133 @@ +# kimi/kimi-k2.7-code-highspeed + +K2.7 Code高速版与普通版是同一个模型，但输出速度约为普通版的 5-6 倍，常规编程场景下（取输入长度中位数）输出速度约 180 Token/s，短上下文场景可达 260 Token/s ，带来更极致的编程体验。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +262144 + +最大输出长度 + +262144 + +上下文长度 + +262144 + +最大输入长度（思考模式下） + +262144 + +最大输出长度（思考模式下） + +262144 + +最大思维链长度 + +262144 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +13 + +每百万tokens + +输出 + +54 + +每百万tokens + +输入（缓存命中） + +2.6 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +3,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code.md new file mode 100644 index 00000000..878213b1 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-7-code.md @@ -0,0 +1,423 @@ +# kimi-k2.7-code + +kimi-k2.7-code是 kimi 迄今最智能的coding模型，在长上下文中更可靠地遵循指令，能以更高的成功率完成编程任务，同时支持文本、图片与视频输入，思考模式，对话与 Agent 任务。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +229376 + +最大输出长度 + +16384 + +上下文长度 + +262144 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +6.5 + +每百万tokens + +输出 + +27 + +每百万tokens + +输入（缓存命中） + +1.3 + +每百万tokens + +显式缓存创建 + +8.125 + +每百万tokens + +显式缓存命中 + +0.65 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +计费项 + +价格（元） + +单位 + +输入 + +7.119 + +每百万tokens + +输出 + +29.977 + +每百万tokens + +输入（缓存命中） + +1.424 + +每百万tokens + +## 德国（法兰克福） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +6.5 + +每百万tokens + +输出 + +27 + +每百万tokens + +输入（缓存命中） + +1.3 + +每百万tokens + +显式缓存创建 + +8.125 + +每百万tokens + +显式缓存命中 + +0.65 + +每百万tokens + +## 美国（弗吉尼亚） + +部署范围：全球 + +计费项 + +价格（元） + +单位 + +输入 + +6.5 + +每百万tokens + +输出 + +27 + +每百万tokens + +输入（缓存命中） + +1.3 + +每百万tokens + +显式缓存创建 + +8.125 + +每百万tokens + +显式缓存命中 + +0.65 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 新加坡 + +部署范围：国际 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 德国（法兰克福） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## 美国（弗吉尼亚） + +部署范围：全球 + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## kimi/kimi-k2.7-code + +Kimi K2.7 Code 是月之暗面 Kimi发布并开源的新一代编程专用模型，定位为 Kimi 迄今最智能的 Coding 模型。Kimi K2.7 Code 是一个以编码为中心的智能体模型（coding-focused agentic model），专为长程软件工程任务优化。它擅长跨多文件重构、功能实现、长会话调试等需要可靠指令遵循和端到端完成率的复杂工作流。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +262144 + +最大输出长度 + +262144 + +上下文长度 + +262144 + +最大输入长度（思考模式下） + +262144 + +最大输出长度（思考模式下） + +262144 + +最大思维链长度 + +262144 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +6.5 + +每百万tokens + +输出 + +27 + +每百万tokens + +输入（缓存命中） + +1.3 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +3,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-thinking.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-thinking.md new file mode 100644 index 00000000..356d32d7 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k2-thinking.md @@ -0,0 +1,121 @@ +# kimi-k2-thinking + +kimi-k2-thinking模型是月之暗面提供的具有通用 Agentic能力和推理能力的思考模型，它擅长深度推理，并可通过多步工具调用，帮助解决各类难题。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +229376 + +最大输出长度 + +16384 + +上下文长度 + +262144 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +16 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k3.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k3.md new file mode 100644 index 00000000..5843285d --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/kimi-k3.md @@ -0,0 +1,133 @@ +# kimi/kimi-k3 + +Kimi K3 是 Kimi 迄今能力最强的旗舰模型，拥有 2.8 万亿参数，基于 KDA 混合线性注意力机制（Kimi Delta Attention）和注意力残差（Attention Residuals）技术构建，原生支持视觉理解，并拥有 100 万 token 上下文窗口。它是全球首个开源的 3 万亿级别模型，面向长程编程、知识工作和推理等前沿智能场景而设计。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text**　**Image**　**Video** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1048576 + +最大输出长度 + +1048576 + +上下文长度 + +1048576 + +最大输入长度（思考模式下） + +1048576 + +最大输出长度（思考模式下） + +1048576 + +最大思维链长度 + +1048576 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +20 + +每百万tokens + +输出 + +100 + +每百万tokens + +输入（缓存命中） + +2 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +3,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/mimo-v2-5-pro.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/mimo-v2-5-pro.md new file mode 100644 index 00000000..6ba047ba --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/mimo-v2-5-pro.md @@ -0,0 +1,133 @@ +# xiaomi/mimo-v2.5-pro + +MiMo-V2.5-Pro 是小米发布的最新旗舰模型。与前代模型相比，它在通用智能体能力、复杂软件工程以及长程任务等方面都有显著提升，在 ClawEval、GDPVal 和 SWE-bench Pro 等基准测试中均位列前茅。它能够独立且完全自主地完成需要人类专家耗时数天甚至数周的专业任务，涉及上千次工具调用。其高达 100 万 token 的上下文长度，非常适合集成到各种智能体框架中使用。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1048576 + +最大输出长度 + +131072 + +上下文长度 + +1048576 + +最大输入长度（思考模式下） + +1048576 + +最大输出长度（思考模式下） + +131072 + +最大思维链长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +7 + +每百万tokens + +输出 + +21 + +每百万tokens + +输入（缓存命中） + +1.4 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +100 + +TPM（每分钟tokens） + +10,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-5.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-5.md new file mode 100644 index 00000000..dfe2f3c2 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-5.md @@ -0,0 +1,255 @@ +# MiniMax-M2.5 + +MiniMax-M2.5是MiniMax推出的旗舰级开源大模型，经过数十万个真实复杂环境中的大规模强化学习训练，M2.5 在编程、工具调用和搜索、办公等生产力场景都达到或者刷新了行业的 SOTA。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +196608 + +最大输出长度 + +131072 + +上下文长度 + +204800 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2.1 + +每百万tokens + +输出 + +8.4 + +每百万tokens + +输入（缓存命中） + +0.42 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +1,000,000 + +## MiniMax/MiniMax-M2.5 + +智能体世界的SOTA，专为智能体2.0设计，将编码扩展到现实世界包括工作空间、娱乐和个人助理。模型亮点：全球SOTA开源编码与智能体模型；SWE-bench Pro和SWE-bench Verified得分高于Opus 4.6；在Excel、搜索与研究以及文档摘要方面的全球SOTA；未来工作空间的完美主力模型；闪电般快速：优化思维效率，100+ TPS，实现比 Opus 快 3 倍的速度；极致性价比，以支持始终在线的智能体。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +204800 + +最大输出长度 + +131072 + +上下文长度 + +204800 + +最大输入长度（思考模式下） + +204800 + +最大输出长度（思考模式下） + +131072 + +最大思维链长度 + +131072 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2.1 + +每百万tokens + +输出 + +8.4 + +每百万tokens + +输入（缓存命中） + +0.21 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +20,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-7.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-7.md new file mode 100644 index 00000000..4b89345b --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2-7.md @@ -0,0 +1,133 @@ +# MiniMax/MiniMax-M2.7 + +M2.7 能够自行构建复杂 Agent Harness，并基于 Agent Teams、复杂 Skills、Tool Search tool 等能力，完成高度复杂的生产力任务。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +204800 + +最大输出长度 + +131072 + +上下文长度 + +204800 + +最大输入长度（思考模式下） + +204800 + +最大输出长度（思考模式下） + +131072 + +最大思维链长度 + +131072 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2.1 + +每百万tokens + +输出 + +8.4 + +每百万tokens + +输入（缓存命中） + +0.42 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +20,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2.md new file mode 100644 index 00000000..15867b9f --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m2.md @@ -0,0 +1,239 @@ +# MiniMax-M2.1 + +MiniMax-M2.1是MiniMax推出的旗舰级开源大模型，聚焦真实世界复杂任务，以多语言编程与长链 Agent 能力为核心优势。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +172032 + +最大输出长度 + +32768 + +上下文长度 + +204800 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2.1 + +每百万tokens + +输出 + +8.4 + +每百万tokens + +输入（缓存命中） + +0.42 + +每百万tokens + +## MiniMax/MiniMax-M2.1 + +M2.1 的设计初衷在于打破“最顶级的 Agent 能力仅存在于闭源模型”的壁垒。我们在模型层面进行了针对性优化，显著提升了模型在代码生成、工具调用、复杂指令遵循及长程规划任务中的性能。从自动化进行多语言的软件开发，到执行多步骤的复杂办公工作流，MiniMax-M2.1 均表现出卓越的稳定性。我们致力于为开发者提供一个完全透明、可控且高可用的基础模型，以构建下一代自主智能体应用。 + +### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +204800 + +最大输出长度 + +131072 + +上下文长度 + +204800 + +最大输入长度（思考模式下） + +204800 + +最大输出长度（思考模式下） + +131072 + +最大思维链长度 + +131072 + +### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +2.1 + +每百万tokens + +输出 + +8.4 + +每百万tokens + +输入（缓存命中） + +0.21 + +每百万tokens + +### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +20,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m3.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m3.md new file mode 100644 index 00000000..763b226c --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/minimax-m3.md @@ -0,0 +1,125 @@ +# MiniMax/MiniMax-M3 + +MiniMax M3 凭借业界领先的 Coding 与 Agentic 能力、1M 超长上下文窗口以及原生多模态特性，可出色胜任企业级长文档理解、高质量内容生成、代码编写、Bug 修复及原生应用构建等任务；强大的 Agentic 能力端到端贯通工作流，原生多模态更带来流畅自然的图文混合交互体验。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Image**　**Text**　**Video** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +1048576 + +最大输出长度 + +— + +上下文长度 + +1048576 + +最大输入长度（思考模式下） + +1048576 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +4.2 + +每百万tokens + +输出 + +16.8 + +每百万tokens + +输入（缓存命中） + +0.84 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +500 + +TPM（每分钟tokens） + +20,000,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen-deep-research.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen-deep-research.md new file mode 100644 index 00000000..a7970b08 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen-deep-research.md @@ -0,0 +1,233 @@ +# qwen-deep-research + +千问深入研究是一款面向复杂研究任务的高级智能体系统，具备多轮推理与全局规划能力，能够运用互联网搜索等多种工具，对任务进行精细化拆解，开展推理与分析，最终为用户生成可溯源、逻辑严谨的研究型报告。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +997952 + +最大输出长度 + +32768 + +上下文长度 + +1000000 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +54 + +每百万tokens + +输出 + +163 + +每百万tokens + +## 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +TPM（每分钟tokens） + +1,200,000 + +## 快照版本 + +### qwen-deep-research-2025-12-15 + +千问深入研究是一款面向复杂研究任务的高级智能体系统，具备多轮推理与全局规划能力，能够运用互联网搜索等多种工具，对任务进行精细化拆解，开展推理与分析，最终为用户生成可溯源、逻辑严谨的研究型报告。 + +#### 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +不支持 + +Function Calling + +不支持 + +结构化输出 + +不支持 + +联网搜索 + +不支持 + +前缀续写 + +不支持 + +上下文缓存 + +不支持 + +批量推理 + +不支持 + +模型调优 + +不支持 + +#### 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +997952 + +最大输出长度 + +32768 + +上下文长度 + +1000000 + +#### 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +计费项 + +价格（元） + +单位 + +输入 + +79 + +每百万tokens + +输出 + +236 + +每百万tokens + +#### 限流 + +## 华北2（北京） + +参数 + +值 + +RPM（每分钟请求数） + +120 + +TPM（每分钟tokens） + +1,200,000 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen3-max.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen3-max.md new file mode 100644 index 00000000..c0d0f101 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/support/model-studio-model-list/model-list-text-generation/model-qwen3-max.md @@ -0,0 +1,2039 @@ +# qwen3-max + +千问3系列Max模型，相较preview版本在智能体编程与工具调用方向进行了专项升级。本次发布的正式版模型达到领域SOTA水平，适配场景更加复杂的智能体需求。该模型版本功能等同于快照模型 qwen3-max-2026-01-23。 + +## 模型能力 + +能力项 + +支持情况 + +能力项 + +支持情况 + +输入模态 + +**Text** + +输出模态 + +**Text** + +模型体验 + +支持 + +Function Calling + +支持 + +结构化输出 + +支持 + +联网搜索 + +支持 + +前缀续写 + +支持 + +上下文缓存 + +支持 + +批量推理 + +支持 + +模型调优 + +不支持 + +## 上下文限制 + +参数 + +值 + +参数 + +值 + +最大输入长度 + +258048 + +最大输出长度 + +65536 + +上下文长度 + +262144 + +最大输入长度（思考模式下） + +258048 + +最大输出长度（思考模式下） + +32768 + +最大思维链长度 + +81920 + +## 模型价格 + +本文仅展示模型调用原价，不包含限时优惠等活动信息，请前往[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)查看活动优惠。 + +## 华北2（北京） + +**输入<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +2.5 + +每百万tokens + +输出 + +10 + +每百万tokens + +输入（缓存命中） + +0.5 + +每百万tokens + +输入（Batch File） + +1.25 + +每百万tokens + +输出（Batch File） + +5 + +每百万tokens + +显式缓存创建 + +3.125 + +每百万tokens + +显式缓存命中 + +0.25 + +每百万tokens + +输入（Batch Chat) + +2.5 + +每百万tokens + +输出（Batch Chat) + +10 + +每百万tokens + +**32k<输入<=128k** + +计费项 + +价格（元） + +单位 + +输入 + +4 + +每百万tokens + +输出 + +16 + +每百万tokens + +输入（缓存命中） + +0.8 + +每百万tokens + +输入（Batch File） + +2 + +每百万tokens + +输出（Batch File） + +8 + +每百万tokens + +显式缓存创建 + +5 + +每百万tokens + +显式缓存命中 + +0.4 + +每百万tokens + +输入（Batch Chat) + +4 + +每百万tokens + +输出（Batch Chat) + +16 + +每百万tokens + +**128k<输入<=256k** + +计费项 + +价格（元） + +单位 + +输入 + +7 + +每百万tokens + +输出 + +28 + +每百万tokens + +输入（缓存命中） + +1.4 + +每百万tokens + +输入（Batch File） + +3.5 + +每百万tokens + +输出（Batch File） + +14 + +每百万tokens + +显式缓存创建 + +8.75 + +每百万tokens + +显式缓存命中 + +0.7 + +每百万tokens + +输入（Batch Chat) + +7 + +每百万tokens + +输出（Batch Chat) + +28 + +每百万tokens + +## 新加坡 + +部署范围：国际 + +**Input<=32k** + +计费项 + +价格（元） + +单位 + +输入 + +8.807 + +每百万tokens + +输出 + +44.035 + +每百万tokens + +输入（缓存命中） + +1.761 + +每百万tokens + +显式缓存创建 + +11.009 + +每百万tokens + +显式缓存命中 + +0.881 + +每百万tokens + +**32k [**费用概览**](https://bailian.console.aliyun.com/?tab=model#/costing-balance/overview)，选择**账期月份**： +登录[百炼控制台](https://bailian.console.aliyun.com/?tab=model)，单击顶部**模型**标签页，选择[**费用概览**](https://bailian.console.aliyun.com/?tab=model#/costing-balance)，选择**账期月份**： > 该页面仅展示**大模型推理**相关费用。**模型训练**和**知识库**等费用请通过[账单详情](#29f8b9b9a4lmc)查看。 @@ -31,14 +31,18 @@ 2. 选择**产品名称**为**大模型服务平台百炼**，单击**搜索**。 -3. 单击账单列表右上角的导出图标，将账单下载到本地。 +3. 单击页面顶部的**导出明细**，将账单下载到本地。 4. 打开文件，找到 实例 ID（出账粒度）列，根据下文规则进行核对。 -#### **2\. 解读关键字段** +#### **2\. 解读关键字段（模型推理账单）** -**“实例 ID（出账粒度）”字段**以英文分号 `;` 分隔，完整格式为`ApiKeyID;业务空间 ID;模型名称;输入/输出类型;调用渠道;免费额度用完即停标识`。 +**说明** + +以下字段格式说明适用于**模型推理**账单。**模型训练与调优**账单的实例ID字段格式不同（使用英文感叹号 `!` 分隔），请参见下方[训练与调优账单字段](#wi82451543-train-billing-h)。 + +**“实例 ID（出账粒度）”字段**以英文分号 `;` 分隔，完整格式为`ApiKeyID;业务空间ID;模型名称;输入/输出类型;调用渠道;免费额度用完即停标识`。 - 格式 A：标准调用（包含ApiKeyID） @@ -64,11 +68,21 @@ - 多个标签之间用英文分号 ; 隔开。 -#### **3\. 数据溯源与术语说明** +#### **3\. 训练与调优账单字段** + +**模型训练**与**模型调优**账单的**“实例 ID（出账粒度）”字段**以英文感叹号 `!` 分隔，常见格式为`业务空间ID!地域!训练任务标识`，与模型推理账单的分号 `;` 分隔格式不同。 + +**示例：**`ws-xxxxxxxxx!cn-beijing!qwen3-14b-ft-2026051510-1c3b` + +**说明** + +账单的**商品名称**字段可直接区分计费类型：训练与调优费用对应**商品名称**为`百炼大模型训练`。若无法确定对应的训练任务，可结合**账单月份**与**模型调优**页面的任务提交时间进行匹配。 + +#### **4\. 数据溯源与术语说明** - 查询 API Key：复制账单中的 `ApiKeyID`，前往[百炼API Key管理](https://bailian.console.aliyun.com/?tab=model#/api-key)页面查找对应的 Key 名称。 -- 查询业务空间：复制账单中的 `业务空间ID`，前往[百炼控制台](https://bailian.console.aliyun.com/?tab=model#/api-key)，点击左侧菜单底部的**默认业务空间**，点击**业务空间详情**，确认具体空间ID。您也可以切换到其他业务空间。 +- 查询业务空间：复制账单中的 `业务空间ID`，前往[业务空间管理](https://bailian.console.aliyun.com/?tab=globalset#/efm/business_management)页面，在列表的**业务空间ID**列确认具体空间。当前所处的业务空间显示在控制台顶部右上角的**默认业务空间**切换器中，悬停可查看其地域、业务空间ID与创建时间等详情。 - 调用渠道说明： @@ -83,13 +97,13 @@ 给**业务空间**绑定**标签**，可按部门或项目归集费用。 -1. **获取业务空间信息**：在[**业务空间管理**](https://bailian.console.aliyun.com/?tab=globalset#/efm/business_management)确定标签绑定的业务空间**Workspace ID**（示例：llm-xxx），并在[账单详情](https://usercenter2.aliyun.com/finance/expense-report/expense-detail)确定业务空间的**地域**信息。 +1. **获取业务空间信息**：在[**业务空间管理**](https://bailian.console.aliyun.com/?tab=globalset#/efm/business_management)确定标签绑定的**业务空间ID**（示例：llm-xxx），并在[账单详情](https://usercenter2.aliyun.com/finance/expense-report/expense-detail)确定业务空间的**地域**信息。 2. **绑定标签**： 1. 在[**标签管理**](https://resourcemanager.console.aliyun.com/tags#/)页面选择**资源绑定标签。** - 2. 资源选择方式选择“**输入多个资源ID**”，在产品选项卡搜索并选择“**大模型服务平台百炼:业务空间**”并选择业务空间对应地域，资源ID输入框中填写**Workspace ID**，完成后点击绑定标签按钮执行操作。 + 2. 资源选择方式选择“**输入多个资源ID**”，在产品选项卡搜索并选择“**大模型服务平台百炼:业务空间**”并选择业务空间对应地域，资源ID输入框中填写**业务空间ID**，完成后点击绑定标签按钮执行操作。 3. 在绑定标签页面中，创建标签键值或使用已创建的预置标签与业务空间绑定。当完成键值输入或选择好预置标签后，点击**确定**完成业务空间标签的绑定。 @@ -102,16 +116,14 @@ **账户可用额度 < 0** 视为欠费，可能导致模型调用等服务暂停。在[费用与成本首页](https://billing-cost.console.aliyun.com/home)悬停**账户可用额度**区域可查看，公式为：可用额度 =（现金余额 + 信控额度）-（当月未结清 + 历史未结清）。 -- **欠费影响**：按账单**商品名称**维度判定。 +- **欠费影响**：账户欠费将导致**按量付费（后付费）**模型调用等服务暂停，能否继续使用取决于计费方式。 - - 仍有**免费额度**：可继续使用，用完后停用。 - - - 仍有**节省计划**或**资源包**额度：可继续使用。 + - **免费额度、节省计划、资源包**：三者均用于抵扣按量付费费用，**欠费期间即使仍有剩余额度，也无法调用模型**，需结清欠费后恢复。 - 已购 **Coding Plan** 或 **Token Plan**：套餐额度独立于账户余额，欠费期间可继续使用，但会导致自动续费失败，到期后无法续用。 - - 以上额度均无：该商品下服务将**暂停**，需结清欠费后恢复。 - +- **代金券与余额说明**：代金券（含学生权益优惠券）不计入账户可用额度中的现金余额，后付费场景下系统会冻结当月消费金额，账户实际余额须大于当月冻结金额方可正常调用模型，详情参见[代金券说明文档](https://help.aliyun.com/zh/user-center/voucher-management)。 + - **结清欠费**：在[费用与成本](https://usercenter2.aliyun.com/home)页面单击**充值汇款**，输入金额并完成支付。 - **预防欠费**：在[高额消费预警](https://usercenter2.aliyun.com/home/alarm-threshold)页面设置消费阈值，达阈值即提醒。 @@ -121,21 +133,21 @@ 不再使用百炼时，按以下方式关停对应服务即可停止计费。 -- **停止模型推理**：停止代码中的 API 调用、关闭控制台的模型体验，即不再产生费用。为防止意外调用，可在[**API-KEY**](https://bailian.console.aliyun.com/?apiKey=1&tab=globalset#/efm/api_key)页面删除已创建的 Key。 +- **停止模型推理**：停止代码中的 API 调用、关闭控制台的模型体验，即不再产生费用。为防止意外调用，可在[**API-KEY**](https://bailian.console.aliyun.com/?tab=model#/api-key)页面删除已创建的 Key。 - **停止模型训练**：没有正在进行的训练任务时即不产生费用。 -- **取消 Coding Plan 订阅**：Coding Plan 为包月订阅产品，到期自动停止，中途不支持取消和退款。如已开启自动续费，请在[Coding Plan](https://bailian.console.aliyun.com/?tab=model#/efm/coding_plan) 页面关闭自动续费。 +- **停止 Coding Plan 计费**：Coding Plan 为包月订阅产品，到期自动停止，中途不支持取消和退款。如已开启自动续费，请在[Coding Plan](https://bailian.console.aliyun.com/?tab=model#/efm/subscription/coding-plan) 页面关闭自动续费。 - **退订 Token Plan 团队版**：在[Token Plan 控制台](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/token-plan)的**我的订阅**页面按席位退订，未产生用量消耗的席位可退订，退款原路退回支付账户。如不再续费，请关闭自动续费。 - **停止模型部署**：根据部署时的计费方式操作不同： - - 按模型调用量计费：[下线](https://bailian.console.aliyun.com/?tab=model#/efm/model_deploy)已部署的模型，或删除[API Key](https://bailian.console.aliyun.com/?tab=model#/api-key)防止意外调用。 + - **按 Token 调用计费（后付费）**：[下线](https://bailian.console.aliyun.com/?tab=model#/efm/model_deploy)已部署的模型，或删除[API Key](https://bailian.console.aliyun.com/?tab=model#/api-key)防止意外调用。 - - 按算力使用时长计费：[下线](https://bailian.console.aliyun.com/?tab=model#/efm/model_deploy)已部署的模型。 + - **按算力单元或模型单元计费（后付费）**：[下线](https://bailian.console.aliyun.com/?tab=model#/efm/model_deploy)已部署的模型。 - - 包月预付费：[下线](https://bailian.console.aliyun.com/?tab=model#/efm/model_deploy)已部署的模型，然后在[退订管理](https://usercenter2.aliyun.com/refund/refund)页面退订实例。退订时按已消费金额扣减，退回剩余金额（详见[退订说明](https://help.aliyun.com/zh/user-center/user-guide/refund-management/)）。 + - **按预置吞吐单元计费（预付费）**：[下线](https://bailian.console.aliyun.com/?tab=model#/efm/model_deploy)已部署的模型，然后在[退订管理](https://usercenter2.aliyun.com/refund/refund)页面退订实例。退订时已使用部分按 1.5 倍系数结算（详见退订说明），退回剩余金额。 ## **常见问题** @@ -161,11 +173,11 @@ **原因：**账单的“计费项”统一显示为“大模型文本消耗量”，未直接展示模型名称。 -**解决方案：**查看[账单详情](https://usercenter2.aliyun.com/finance/expense-report/expense-detail)页的**实例 ID（出账粒度）**列。字段以英文分号分隔，紧跟业务空间 ID（如 llm-xxx）之后的字段即为模型名称。例：`12xxx;llm-xxx;**qwen3.6-plus**;context_0-128k_input_token;bmp;0`表示 qwen3.6-plus 模型。 +**解决方案：**查看[账单详情](https://usercenter2.aliyun.com/finance/expense-report/expense-detail)页的**实例 ID（出账粒度）**列。字段以英文分号分隔，紧跟业务空间ID（如 llm-xxx）之后的字段即为模型名称。例：`12xxx;llm-xxx;**qwen3.6-plus**;context_0-128k_input_token;bmp;0`表示 qwen3.6-plus 模型。 **在哪里查看模型调用次数和统计？** -进入[阿里云百炼控制台](https://bailian.console.aliyun.com/?tab=model)，右上角选择目标地域，单击顶部**模型**标签页，在左侧菜单选择**用量 & 费用** > [模型用量](https://bailian.console.aliyun.com/?tab=costing-balance#/costing-balance/usage-statistics)。 +进入[阿里云百炼控制台](https://bailian.console.aliyun.com/?tab=model)，右上角选择目标地域，单击顶部**模型**标签页，选择[模型用量](https://bailian.console.aliyun.com/?tab=costing-balance#/costing-balance/usage-statistics)。 **按量付费是实时扣款吗？** @@ -190,3 +202,27 @@ 2. 检查应用代码或百炼应用配置中是否开启了 `enable_search`，如不再需要联网搜索，将该参数设为 `false` 或移除。 3. 如已停止所有调用但仍有扣费，检查是否有其他 API Key 或应用仍在运行，可在[API Key 管理](https://bailian.console.aliyun.com/?tab=model#/api-key)页面逐一排查或删除不再使用的 Key。 + + +**为什么没有主动调用 API 也会产生费用？** + +**原因：**百炼的模型部署按使用时长计费，模型完成部署即状态为**运行中**时开始收费，不依赖 API 调用。即使未主动通过 API 调用该模型，只要部署状态为**运行中**就会持续产生费用。 + +**解决方案：** + +- 前往[**模型部署**](https://bailian.console.aliyun.com/?tab=model#/efm/model_deploy)页面，下线不再使用的已部署模型，停止按时长计费。 + +- 如需防止意外调用产生推理费用，可在[**API-KEY**](https://bailian.console.aliyun.com/?tab=model#/api-key)页面删除不再使用的 Key（注意：删除后无法恢复，请谨慎操作）。 + + +**如何判断账户是否被盗用？** + +如果怀疑账户被他人盗用产生非预期费用，按以下步骤排查： + +1. 在[账单详情](https://usercenter2.aliyun.com/finance/expense-report/expense-detail)中筛选**大模型服务平台百炼**，查看**实例 ID（出账粒度）**列中的 `ApiKeyID`，确认产生费用的 API Key。 + +2. 前往[API Key 管理页面](https://bailian.console.aliyun.com/?tab=model#/api-key)，核对每个 Key 的**创建时间**，确认是否为本人创建。API Key 管理页面仅显示创建时间，不显示调用时间。 + +3. 查看调用时段分布，判断是否存在非本人操作的异常调用模式：进入[模型用量](https://bailian.console.aliyun.com/?tab=costing-balance#/costing-balance/usage-statistics)页面，按**模型**或 **API Key ID** 筛选，切换至**列表**视图查看调用时间分布。 + +4. 如发现未授权调用，立即在[API Key 管理页面](https://bailian.console.aliyun.com/?tab=model#/api-key)删除对应 API Key 并重新生成。更新所有合法调用方使用新 Key。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/model-pricing.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/model-pricing.md index 7137579d..7536077e 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/model-pricing.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/model-pricing.md @@ -28,14 +28,12 @@ **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **模式** **单次请求的输入Token数** @@ -48,7 +46,7 @@ **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3.7-max @@ -58,8 +56,6 @@ qwen3.7-max > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 0 当前能力等同于qwen3.7-max-2026-05-17 -中国内地 - 仅思考模式 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 0 当前能力等同于qwen3.6-plus-2026-04-02 -中国内地 - 0 当前能力等同于qwen3.5-plus-2026-02-15 -中国内地 - 0 [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 0 [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 0 当前能力等同于qwen3.7-flash-2026-07-15 + +> [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 + +> [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 + +非思考和思考模式 + +0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 0 当前能力等同于qwen3.5-flash-2026-02-23 @@ -2866,6 +2860,58 @@ qwen-flash-2025-07-28-us > **思维链+回答** +qwen3.7-flash + +> 当前能力等同于qwen3.7-flash-2026-07-15 + +> [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 + +国际 + +非思考和思考模式 + +0 当前能力等同于qwen3.6-flash-2026-04-16 @@ -3070,6 +3116,8 @@ qwen3.5-flash 非思考和思考模式 +0 [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 0.5元 2元 @@ -3386,8 +3424,6 @@ qwen-long qwen-long-latest -中国内地 - 0.5元 2元 @@ -3396,8 +3432,6 @@ qwen-long-latest qwen-long-2025-01-25 -中国内地 - 0.5元 2元 @@ -3410,21 +3444,19 @@ qwen-long-2025-01-25 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） **文本/图片/视频** @@ -3442,8 +3474,6 @@ qwen3.5-omni-plus > 当前能力等同于qwen3.5-omni-plus-2026-03-15 -中国内地 - 7元 53元 @@ -3456,8 +3486,6 @@ qwen3.5-omni-plus qwen3.5-omni-plus-2026-03-15 -中国内地 - 7元 53元 @@ -3472,8 +3500,6 @@ qwen3.5-omni-flash > 当前能力等同于qwen3.5-omni-flash-2026-03-15 -中国内地 - 2.2元 18元 @@ -3486,8 +3512,6 @@ qwen3.5-omni-flash qwen3.5-omni-flash-2026-03-15 -中国内地 - 2.2元 18元 @@ -3502,8 +3526,6 @@ qwen3.5-omni-flash-2026-03-15 **模型 ID（Model ID）** -**服务部署范围** - **模式** **输入单价（每百万Token）** @@ -3512,7 +3534,7 @@ qwen3.5-omni-flash-2026-03-15 **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） **文本** @@ -3536,8 +3558,6 @@ qwen3-omni-flash > 当前能力等同于qwen3-omni-flash-2025-12-01 -中国内地 - 非思考和思考模式 1.8元 @@ -3556,8 +3576,6 @@ qwen3-omni-flash qwen3-omni-flash-2025-12-01 -中国内地 - 非思考和思考模式 1.8元 @@ -3576,8 +3594,6 @@ qwen3-omni-flash-2025-12-01 qwen3-omni-flash-2025-09-15 -中国内地 - 非思考和思考模式 1.8元 @@ -3598,8 +3614,6 @@ qwen-omni-turbo > 当前能力等同于qwen-omni-turbo-2025-03-26 -中国内地 - 非思考模式 0.4元 @@ -3618,8 +3632,6 @@ qwen-omni-turbo qwen-omni-turbo-latest -中国内地 - 非思考模式 0.4元 @@ -3638,8 +3650,6 @@ qwen-omni-turbo-latest qwen-omni-turbo-2025-03-26 -中国内地 - 非思考模式 0.4元 @@ -3658,8 +3668,6 @@ qwen-omni-turbo-2025-03-26 qwen-omni-turbo-2025-01-19 -中国内地 - 非思考模式 0.4元 @@ -3898,21 +3906,19 @@ qwen-omni-turbo-2025-03-26 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） **文本/图片** @@ -3930,8 +3936,6 @@ qwen3.5-omni-plus-realtime > 当前能力等同于qwen3.5-omni-plus-realtime-2026-03-15 -中国内地 - 10元 80元 @@ -3944,8 +3948,6 @@ qwen3.5-omni-plus-realtime qwen3.5-omni-plus-realtime-2026-03-15 -中国内地 - 10元 80元 @@ -3960,8 +3962,6 @@ qwen3.5-omni-flash-realtime > 当前能力等同于qwen3.5-omni-flash-realtime-2026-03-15 -中国内地 - 3.3元 27元 @@ -3974,8 +3974,6 @@ qwen3.5-omni-flash-realtime qwen3.5-omni-flash-realtime-2026-03-15 -中国内地 - 3.3元 27元 @@ -3990,15 +3988,13 @@ qwen3.5-omni-flash-realtime-2026-03-15 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） **文本** @@ -4022,8 +4018,6 @@ qwen3-omni-flash-realtime > 当前能力等同于qwen3-omni-flash-realtime-2025-12-01 -中国内地 - 2.2元 18.9元 @@ -4040,8 +4034,6 @@ qwen3-omni-flash-realtime qwen3-omni-flash-realtime-2025-12-01 -中国内地 - 2.2元 18.9元 @@ -4058,8 +4050,6 @@ qwen3-omni-flash-realtime-2025-12-01 qwen3-omni-flash-realtime-2025-09-15 -中国内地 - 2.2元 18.9元 @@ -4078,8 +4068,6 @@ qwen-omni-turbo-realtime > 当前能力等同于qwen-omni-turbo-realtime-2025-05-08 -中国内地 - 1.6元 25元 @@ -4096,8 +4084,6 @@ qwen-omni-turbo-realtime qwen-omni-turbo-realtime-latest -中国内地 - 1.6元 25元 @@ -4114,8 +4100,6 @@ qwen-omni-turbo-realtime-latest qwen-omni-turbo-realtime-2025-05-08 -中国内地 - 1.6元 25元 @@ -4338,26 +4322,22 @@ qwen-omni-turbo-realtime-2025-05-08 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qvq-max -中国内地 - 8元 32元 @@ -4366,8 +4346,6 @@ qvq-max qvq-plus -中国内地 - 2元 5元 @@ -4400,14 +4378,12 @@ qvq-max **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **模式** **单次请求的输入Token数** @@ -4420,7 +4396,7 @@ qvq-max **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-vl-plus @@ -4430,8 +4406,6 @@ qwen3-vl-plus > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 无阶梯计价 1.6元 @@ -4624,8 +4584,6 @@ qwen-vl-plus > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 无阶梯计价 0.8元 @@ -5280,26 +5238,22 @@ qwen3-vl-plus-2025-09-23 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3.5-ocr -中国内地 - 0.5元 2元 @@ -5312,8 +5266,6 @@ qwen-vl-ocr > [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 0.3元 0.5元 @@ -5324,8 +5276,6 @@ qwen-vl-ocr-latest > [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 0.3元 0.5元 @@ -5334,8 +5284,6 @@ qwen-vl-ocr-latest qwen-vl-ocr-2025-11-20 -中国内地 - 0.3元 0.5元 @@ -5344,8 +5292,6 @@ qwen-vl-ocr-2025-11-20 qwen-vl-ocr-2025-08-28 -中国内地 - 5元 5元 @@ -5354,8 +5300,6 @@ qwen-vl-ocr-2025-08-28 qwen-vl-ocr-2025-04-13 -中国内地 - 5元 5元 @@ -5364,8 +5308,6 @@ qwen-vl-ocr-2025-04-13 qwen-vl-ocr-2024-10-28 -中国内地 - 5元 5元 @@ -5466,20 +5408,16 @@ qwen-vl-ocr-2025-11-20 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen-audio-turbo -中国内地 - 目前仅供免费体验。 > 免费额度用完后不可调用，推荐使用[全模态（Qwen-Omni）](https://help.aliyun.com/zh/model-studio/qwen-omni)作为替代模型 @@ -5488,8 +5426,6 @@ qwen-audio-turbo qwen-audio-turbo-latest -中国内地 - ### 千问数学模型 计费规则：按输入Token和输出Token计费。 @@ -5498,20 +5434,16 @@ qwen-audio-turbo-latest **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen-math-plus -中国内地 - 4元 12元 @@ -5520,8 +5452,6 @@ qwen-math-plus qwen-math-turbo -中国内地 - 2元 6元 @@ -5534,14 +5464,12 @@ qwen-math-turbo **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **单次请求的输入Token数** **输入单价（每百万Token）** @@ -5550,7 +5478,7 @@ qwen-math-turbo **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-coder-plus @@ -5558,8 +5486,6 @@ qwen3-coder-plus > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 0 当前能力等同于qwen3-coder-flash-2025-07-28 -中国内地 - 0 免费额度用完后不可调用，推荐使用[全模态（Qwen-Omni）](https://help.aliyun.com/zh/model-studio/qwen-omni)作为替代模型。 @@ -8260,22 +8076,18 @@ qwen2-audio-instruct qwen-audio-chat -中国内地 - ### **Qwen-Coder** 计费规则：按输入Token和输出Token计费。 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **单次请求的输入Token数** **输入单价（每百万Token）** @@ -8284,12 +8096,10 @@ qwen-audio-chat **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-coder-next -中国内地 - 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 12元 24元 @@ -8614,8 +8416,6 @@ deepseek-v4-flash > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 1元 2元 @@ -8626,8 +8426,6 @@ deepseek-v3.2 > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 2元 3元 @@ -8636,8 +8434,6 @@ deepseek-v3.2 deepseek-v3.2-exp -中国内地 - 2元 3元 @@ -8646,8 +8442,6 @@ deepseek-v3.2-exp deepseek-v3.1 -中国内地 - 4元 12元 @@ -8658,8 +8452,6 @@ deepseek-r1 > [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 4元 16元 @@ -8668,8 +8460,6 @@ deepseek-r1 deepseek-r1-0528 -中国内地 - 4元 16元 @@ -8680,8 +8470,6 @@ deepseek-v3 > [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 2元 8元 @@ -8690,14 +8478,10 @@ deepseek-v3 deepseek-r1-distill-qwen-1.5b -中国内地 - 限时免费 deepseek-r1-distill-qwen-7b -中国内地 - 0.5元 1元 @@ -8706,8 +8490,6 @@ deepseek-r1-distill-qwen-7b deepseek-r1-distill-qwen-14b -中国内地 - 1元 3元 @@ -8716,8 +8498,6 @@ deepseek-r1-distill-qwen-14b deepseek-r1-distill-qwen-32b -中国内地 - 2元 6元 @@ -8726,16 +8506,12 @@ deepseek-r1-distill-qwen-32b deepseek-r1-distill-llama-8b -中国内地 - 已下线 > 该模型已下线，推荐使用[深度思考](https://help.aliyun.com/zh/model-studio/deep-thinking)、[DeepSeek-阿里云](https://help.aliyun.com/zh/model-studio/deepseek-api)、[Kimi-阿里云](https://help.aliyun.com/zh/model-studio/kimi-api)作为替代模型。 deepseek-r1-distill-llama-70b -中国内地 - 目前仅供免费体验 > 免费额度用完后不可调用，推荐使用[深度思考](https://help.aliyun.com/zh/model-studio/deep-thinking)、[DeepSeek-阿里云](https://help.aliyun.com/zh/model-studio/deepseek-api)、[Kimi-阿里云](https://help.aliyun.com/zh/model-studio/kimi-api)作为替代模型 @@ -8926,8 +8702,6 @@ deepseek-v4-flash **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** @@ -8938,8 +8712,6 @@ deepseek-v4-flash siliconflow/deepseek-v3.2 -中国内地 - 2元 3元 @@ -8948,24 +8720,18 @@ siliconflow/deepseek-v3.2 siliconflow/deepseek-v3.1-terminus -中国内地 - 4元 12元 siliconflow/deepseek-r1-0528 -中国内地 - 4元 16元 siliconflow/deepseek-v3-0324 -中国内地 - 2元 8元 @@ -8976,8 +8742,6 @@ siliconflow/deepseek-v3-0324 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** @@ -8990,8 +8754,6 @@ vanchin/deepseek-v3.2-think > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 2元 3元 @@ -9002,8 +8764,6 @@ vanchin/deepseek-v3.1-terminus > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 4元 12元 @@ -9012,8 +8772,6 @@ vanchin/deepseek-r1 > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 4元 16元 @@ -9022,16 +8780,12 @@ vanchin/deepseek-v3 > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 2元 8元 vanchin/deepseek-ocr -中国内地 - 0.216元 0.216元 @@ -9042,14 +8796,12 @@ vanchin/deepseek-ocr **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **模式** **输入单价（每百万Token）** @@ -9058,12 +8810,10 @@ vanchin/deepseek-ocr **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） kimi-k2.7-code -中国内地 - 仅思考模式 6.5元 @@ -9074,8 +8824,6 @@ kimi-k2.7-code kimi-k2.6 -中国内地 - 非思考和思考模式 6.5元 @@ -9086,8 +8834,6 @@ kimi-k2.6 kimi-k2.5 -中国内地 - 非思考和思考模式 4元 @@ -9098,8 +8844,6 @@ kimi-k2.5 kimi-k2-thinking -中国内地 - 仅思考模式 4元 @@ -9110,8 +8854,6 @@ kimi-k2-thinking Moonshot-Kimi-K2-Instruct -中国内地 - 非思考模式 4元 @@ -9236,8 +8978,6 @@ kimi-k2.7-code **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** @@ -9246,24 +8986,28 @@ kimi-k2.7-code **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -kimi/kimi-k2.7-code-highspeed +kimi/kimi-k3 > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 +20元 + +100元 + +无 + +kimi/kimi-k2.7-code-highspeed + +> [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 13元 54元 -无 - kimi/kimi-k2.7-code > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 6.5元 27元 @@ -9272,8 +9016,6 @@ kimi/kimi-k2.6 > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 6.5元 27元 @@ -9282,8 +9024,6 @@ kimi/kimi-k2.5 > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 4元 21元 @@ -9294,14 +9034,12 @@ kimi/kimi-k2.5 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **模式** **单次请求的输入Token数** @@ -9314,12 +9052,10 @@ kimi/kimi-k2.5 **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） glm-5.2 -中国内地 - 非思考和思考模式 不区分阶梯 @@ -9332,8 +9068,6 @@ glm-5.2 glm-5.2-fast-preview -中国内地 - 非思考和思考模式 不区分阶梯 @@ -9346,8 +9080,6 @@ glm-5.2-fast-preview glm-5.1 -中国内地 - 非思考和思考模式 0 [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 非思考和思考模式 4.2元 @@ -9804,8 +9508,6 @@ MiniMax/MiniMax-M2.7 > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 仅思考模式 2.1元 @@ -9816,8 +9518,6 @@ MiniMax/MiniMax-M2.5 > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 仅思考模式 2.1元 @@ -9828,8 +9528,6 @@ MiniMax/MiniMax-M2.1 > [上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)享有折扣 -中国内地 - 仅思考模式 2.1元 @@ -9844,8 +9542,6 @@ MiniMax/MiniMax-M2.1 **模型 ID（Model ID）** -**服务部署范围** - **输入Token数量** **输入单价（每百万Token）** @@ -9858,8 +9554,6 @@ MiniMax/MiniMax-M2.1 xiaomi/mimo-v2.5-pro -中国内地 - 0 当前能力等同于qwen-image-2.0-pro-2026-04-22 -中国内地 - 0.5元/张 100张 qwen-image-2.0-pro-2026-06-22 -中国内地 - 0.5元/张 100张 qwen-image-2.0-pro-2026-04-22 -中国内地 - 0.5元/张 100张 qwen-image-2.0-pro-2026-03-03 -中国内地 - 0.5元/张 100张 @@ -9980,16 +9664,12 @@ qwen-image-2.0 > 当前能力等同于qwen-image-2.0-2026-03-03 -中国内地 - 0.2元/张 100张 qwen-image-2.0-2026-03-03 -中国内地 - 0.2元/张 100张 @@ -9998,16 +9678,12 @@ qwen-image-max > 当前能力等同于qwen-image-max-2025-12-30 -中国内地 - 0.5元/张 100张 qwen-image-max-2025-12-30 -中国内地 - 0.5元/张 100张 @@ -10016,24 +9692,18 @@ qwen-image-plus > 当前能力等同于qwen-image -中国内地 - 0.2元/张 100张 qwen-image-plus-2026-01-09 -中国内地 - 0.2元/张 100张 qwen-image -中国内地 - 0.25元/张 100张 @@ -10046,6 +9716,12 @@ qwen-image **输出单价** +qwen-image-3.0-pro + +国际 + +限时免费 + qwen-image-2.0-pro > 当前能力等同于qwen-image-2.0-pro-2026-04-22 @@ -10126,50 +9802,44 @@ qwen-image **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） + +qwen-image-3.0-pro + +限时免费 qwen-image-2.0-pro > 当前能力等同于qwen-image-2.0-pro-2026-04-22 -中国内地 - 0.5元/张 100张 qwen-image-2.0-pro-2026-06-22 -中国内地 - 0.5元/张 100张 qwen-image-2.0-pro-2026-04-22 -中国内地 - 0.5元/张 100张 qwen-image-2.0-pro-2026-03-03 -中国内地 - 0.5元/张 100张 @@ -10178,16 +9848,12 @@ qwen-image-2.0 > 当前能力等同于qwen-image-2.0-2026-03-03 -中国内地 - 0.2元/张 100张 qwen-image-2.0-2026-03-03 -中国内地 - 0.2元/张 100张 @@ -10196,16 +9862,12 @@ qwen-image-edit-max > 当前能力等同于qwen-image-edit-max-2026-01-16 -中国内地 - 0.5元/张 100张 qwen-image-edit-max-2026-01-16 -中国内地 - 0.5元/张 100张 @@ -10214,32 +9876,24 @@ qwen-image-edit-plus > 当前能力等同于qwen-image-edit-plus-2025-10-30 -中国内地 - 0.2元/张 100张 qwen-image-edit-plus-2025-12-15 -中国内地 - 0.2元/张 100张 qwen-image-edit-plus-2025-10-30 -中国内地 - 0.2元/张 100张 qwen-image-edit -中国内地 - 0.3元/张 100张 @@ -10252,6 +9906,12 @@ qwen-image-edit **输出单价** +qwen-image-3.0-pro + +国际 + +限时免费 + qwen-image-2.0-pro > 当前能力等同于qwen-image-2.0-pro-2026-04-22 @@ -10340,18 +10000,14 @@ qwen-image-edit **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen-mt-image -中国内地 - 0.003元/张 100张 @@ -10362,24 +10018,20 @@ qwen-mt-image **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） z-image-turbo -中国内地 - 关闭提示词改写（`prompt_extend=false`）：0.1元/张 开启提示词改写（`prompt_extend=true`）：0.2元/张 @@ -10408,80 +10060,62 @@ z-image-turbo **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.6-t2i -中国内地 - 0.20元/张 50张 wan2.5-t2i-preview -中国内地 - 0.20元/张 50张 wan2.2-t2i-plus -中国内地 - 0.20元/张 100张 wan2.2-t2i-flash -中国内地 - 0.14元/张 100张 wanx2.1-t2i-plus -中国内地 - 0.20元/张 500张 wanx2.1-t2i-turbo -中国内地 - 0.14元/张 500张 wanx2.0-t2i-turbo -中国内地 - 0.04元/张 500张 wanx-v1 -中国内地 - 0.16元/张 500张 @@ -10564,40 +10198,32 @@ wan2.6-t2i **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.7-image-pro -中国内地 - 0.50元/张 50张 wan2.7-image -中国内地 - 0.20元/张 50张 wan2.6-image -中国内地 - 0.20元/张 50张 @@ -10662,32 +10288,26 @@ wan2.6-image **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.5-i2i-preview -中国内地 - 0.20元/张 50张 wanx2.1-imageedit -中国内地 - 0.14元/张 500张 @@ -10714,18 +10334,14 @@ wan2.5-i2i-preview **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wanx-sketch-to-image-lite -中国内地 - 0.06元/张 500张 @@ -10738,18 +10354,14 @@ wanx-sketch-to-image-lite **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wanx-x-painting -中国内地 - 目前仅供免费体验。 > 免费额度用完后不可调用，推荐参考[图像编辑-千问](https://help.aliyun.com/zh/model-studio/qwen-image-edit-guide)或[图像编辑-万相2.1](https://help.aliyun.com/zh/model-studio/wanx-image-edit)获取替代方案。 @@ -10764,18 +10376,14 @@ wanx-x-painting **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wanx-style-repaint-v1 -中国内地 - 0.12元/张 500张 @@ -10788,18 +10396,14 @@ wanx-style-repaint-v1 **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wanx-background-generation-v2 -中国内地 - 0.08元/张 500张 @@ -10812,18 +10416,14 @@ wanx-background-generation-v2 **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） image-out-painting -中国内地 - 0.18元/张 500张 @@ -10836,18 +10436,14 @@ image-out-painting **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） image-instance-segmentation -中国内地 - 目前仅供免费体验。 > 免费额度用完后不可调用。 @@ -10862,18 +10458,14 @@ image-instance-segmentation **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） image-erase-completion -中国内地 - 目前仅供免费体验。 > 免费额度用完后不可调用，推荐参考[图像编辑-千问](https://help.aliyun.com/zh/model-studio/qwen-image-edit-guide)或[图像编辑-万相2.1](https://help.aliyun.com/zh/model-studio/wanx-image-edit)获取替代方案。 @@ -10888,18 +10480,14 @@ image-erase-completion **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wanx-virtualmodel -中国内地 - 目前仅供免费体验。 > 免费额度用完后不可调用，推荐参考[图像编辑-千问](https://help.aliyun.com/zh/model-studio/qwen-image-edit-guide)或[图像编辑-万相2.1](https://help.aliyun.com/zh/model-studio/wanx-image-edit)获取替代方案。 @@ -10908,8 +10496,6 @@ wanx-virtualmodel virtualmodel-v2 -中国内地 - ### **鞋靴模特** > 仅输出计费，计费规则请参见[图像生成](#26310bc5cf4do)。 @@ -10918,18 +10504,14 @@ virtualmodel-v2 **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） shoemodel-v1 -中国内地 - 目前仅供免费体验。 > 免费额度用完后不可调用。 @@ -10944,18 +10526,14 @@ shoemodel-v1 **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wanx-poster-generation-v1 -中国内地 - 目前仅供免费体验。 > 免费额度用完后不可调用，推荐参考[图像编辑-千问](https://help.aliyun.com/zh/model-studio/qwen-image-edit-guide)或[图像编辑-万相2.1](https://help.aliyun.com/zh/model-studio/wanx-image-edit)获取替代方案。 @@ -10975,39 +10553,31 @@ wanx-poster-generation-v1 **模型 ID（Model ID）** -**服务部署范围** - **单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) facechain-facedetect -中国内地 - 限时免费 限时免费 facechain-finetune -中国内地 - 2.5元/次 50次 -有效期：申请通过后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） facechain-generation -中国内地 - 0.18元/张 500张 -有效期：申请通过后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） ### **创意文字生成-WordArt锦书** @@ -11017,26 +10587,20 @@ facechain-generation **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wordart-texture -中国内地 - 0.08元/张 500张 wordart-semantic -中国内地 - 0.24元/张 ### **AI试衣-OutfitAnyone** @@ -11054,42 +10618,30 @@ wordart-semantic **模型 ID（Model ID）** -**服务部署范围** - **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） aitryon -中国内地 - 400张 aitryon-plus -中国内地 - 400张 aitryon-parsing-v1 -中国内地 - 400张 aitryon-refiner -中国内地 - 100张 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **单价** **折扣** @@ -11098,8 +10650,6 @@ aitryon-refiner aitryon -中国内地 - 0.20元/张 无 @@ -11108,8 +10658,6 @@ aitryon aitryon-plus -中国内地 - 0.50元/张 无 @@ -11118,8 +10666,6 @@ aitryon-plus aitryon-parsing-v1 -中国内地 - 0.004元/张 无 @@ -11128,8 +10674,6 @@ aitryon-parsing-v1 aitryon-refiner -中国内地 - 0.30元/张 无 @@ -11182,8 +10726,6 @@ aitryon-refiner **模型 ID（Model ID）** -**服务部署范围** - **输出图像分辨率** **输出单价** @@ -11192,8 +10734,6 @@ aitryon-refiner kling/kling-v3-image-generation -中国内地 - 1K 0.2元/张 @@ -11206,8 +10746,6 @@ kling/kling-v3-image-generation kling/kling-v3-omni-image-generation -中国内地 - 1K 0.2元/张 @@ -11228,8 +10766,6 @@ kling/kling-v3-omni-image-generation **模型 ID（Model ID）** -**服务部署范围** - **输出图像分辨率** **输出单价** @@ -11238,8 +10774,6 @@ kling/kling-v3-omni-image-generation vidu/vidu-image\_reference2image -中国内地 - 1K 0.625元/张 @@ -11256,8 +10790,6 @@ vidu/vidu-image\_reference2image vidu/viduq3-fast\_reference2image -中国内地 - 1K 0.46875元/张 @@ -11272,8 +10804,6 @@ vidu/viduq3-fast\_reference2image vidu/viduq2-pro\_reference2image -中国内地 - 1K 0.9375元/张 @@ -11288,8 +10818,6 @@ vidu/viduq2-pro\_reference2image vidu/viduq2-fast\_reference2image -中国内地 - 1K 0.28125元/张 @@ -11302,26 +10830,20 @@ vidu/viduq2-fast\_reference2image **模型 ID（Model ID）** -**服务部署范围** - **输出单价（每秒）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） fun-music-preview -中国内地 - 0.005元 1,000秒 fun-music-v1 -中国内地 - 0.002元 ## **语音合成（文本转语音）** @@ -11332,32 +10854,26 @@ fun-music-v1 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen-audio-3.0-tts-plus -中国内地 - 1.4元 1万字符 qwen-audio-3.0-tts-flash -中国内地 - 1元 1万字符 @@ -11386,7 +10902,7 @@ qwen-audio-3.0-tts-flash **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） @@ -11396,22 +10912,18 @@ qwen-audio-3.0-tts-flash **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-tts-instruct-flash > 当前能力等同于qwen3-tts-instruct-flash-2026-01-26 -中国内地 - 0.8元 不计费 @@ -11420,8 +10932,6 @@ qwen3-tts-instruct-flash qwen3-tts-instruct-flash-2026-01-26 -中国内地 - 0.8元 不计费 @@ -11434,20 +10944,16 @@ qwen3-tts-instruct-flash-2026-01-26 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-tts-vd-2026-01-26 -中国内地 - 0.8元 不计费 @@ -11460,20 +10966,16 @@ qwen3-tts-vd-2026-01-26 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-tts-vc-2026-01-22 -中国内地 - 0.8元 不计费 @@ -11486,22 +10988,18 @@ qwen3-tts-vc-2026-01-22 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-tts-flash > 当前能力等同于qwen3-tts-flash-2025-11-27 -中国内地 - 0.8元 不计费 @@ -11510,8 +11008,6 @@ qwen3-tts-flash qwen3-tts-flash-2025-11-27 -中国内地 - 0.8元 不计费 @@ -11520,8 +11016,6 @@ qwen3-tts-flash-2025-11-27 qwen3-tts-flash-2025-09-18 -中国内地 - 0.8元 不计费 @@ -11534,20 +11028,16 @@ qwen3-tts-flash-2025-09-18 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen-tts-flash -中国内地 - 1.6元 10元 @@ -11556,8 +11046,6 @@ qwen-tts-flash qwen-tts-latest -中国内地 - 1.6元 10元 @@ -11566,8 +11054,6 @@ qwen-tts-latest qwen-tts-2025-05-22 -中国内地 - 1.6元 10元 @@ -11576,8 +11062,6 @@ qwen-tts-2025-05-22 qwen-tts-2025-04-10 -中国内地 - 1.6元 10元 @@ -11676,7 +11160,7 @@ qwen3-tts-flash-2025-09-18 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） @@ -11686,22 +11170,18 @@ qwen3-tts-flash-2025-09-18 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-tts-instruct-flash-realtime > 当前能力等同于qwen3-tts-instruct-flash-realtime-2026-01-22 -中国内地 - 1元 不计费 @@ -11710,8 +11190,6 @@ qwen3-tts-instruct-flash-realtime qwen3-tts-instruct-flash-realtime-2026-01-22 -中国内地 - 1元 不计费 @@ -11724,20 +11202,16 @@ qwen3-tts-instruct-flash-realtime-2026-01-22 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-tts-vd-realtime-2026-01-15 -中国内地 - 1元 不计费 @@ -11746,8 +11220,6 @@ qwen3-tts-vd-realtime-2026-01-15 qwen3-tts-vd-realtime-2025-12-16 -中国内地 - 1元 不计费 @@ -11760,20 +11232,16 @@ qwen3-tts-vd-realtime-2025-12-16 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-tts-vc-realtime-2026-01-15 -中国内地 - 1元 不计费 @@ -11782,8 +11250,6 @@ qwen3-tts-vc-realtime-2026-01-15 qwen3-tts-vc-realtime-2025-11-27 -中国内地 - 1万字符 #### 千问3-TTS-Flash-Realtime @@ -11792,20 +11258,16 @@ qwen3-tts-vc-realtime-2025-11-27 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-tts-flash-realtime -中国内地 - 1元 不计费 @@ -11814,8 +11276,6 @@ qwen3-tts-flash-realtime qwen3-tts-flash-realtime-2025-11-27 -中国内地 - 1元 不计费 @@ -11824,8 +11284,6 @@ qwen3-tts-flash-realtime-2025-11-27 qwen3-tts-flash-realtime-2025-09-18 -中国内地 - 1元 不计费 @@ -11838,20 +11296,16 @@ qwen3-tts-flash-realtime-2025-09-18 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen-tts-realtime -中国内地 - 2.4元 12元 @@ -11860,8 +11314,6 @@ qwen-tts-realtime qwen-tts-realtime-latest -中国内地 - 2.4元 12元 @@ -11870,8 +11322,6 @@ qwen-tts-realtime-latest qwen-tts-realtime-2025-07-15 -中国内地 - 2.4元 12元 @@ -11982,24 +11432,20 @@ qwen3-tts-flash-realtime-2025-09-18 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **单价（每个音色）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen-voice-enrollment -中国内地 - 0.01元 1000个音色/账号 @@ -12024,24 +11470,20 @@ qwen-voice-enrollment **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **单价（每个音色）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen-voice-design -中国内地 - 0.2元 10个音色/账号 @@ -12064,7 +11506,7 @@ qwen-voice-design **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） @@ -12072,58 +11514,44 @@ qwen-voice-design **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） cosyvoice-v3.5-plus -中国内地 - 1.5元 1万字符 cosyvoice-v3.5-flash -中国内地 - 0.8元 1万字符 cosyvoice-v3-plus -中国内地 - 2元 1万字符 cosyvoice-v3-flash -中国内地 - 1元 1万字符 cosyvoice-v2 -中国内地 - 2元 1万字符 cosyvoice-v1 -中国内地 - 2元 1万字符 @@ -12158,16 +11586,12 @@ cosyvoice-v3-flash **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每万字符）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) 参见[模型列表](https://help.aliyun.com/zh/model-studio/sambert-java-sdk#57d33631f7doi) -中国内地 - 1元 每主账号每模型每月3万字符。 @@ -12180,8 +11604,6 @@ cosyvoice-v3-flash **模型名称** -**服务部署范围** - **语音合成单价（每万字符）** [复刻一个音色](https://help.aliyun.com/zh/model-studio/mini-clone-api) @@ -12190,8 +11612,6 @@ cosyvoice-v3-flash MiniMax/speech-2.8-hd -中国内地 - 3.5元 9.9元 @@ -12202,20 +11622,14 @@ MiniMax/speech-2.8-hd MiniMax/speech-02-hd -中国内地 - 3.5元 MiniMax/speech-2.8-turbo -中国内地 - 2元 MiniMax/speech-02-turbo -中国内地 - 2元 ## **语音识别（语音转文本）与翻译（语音转成指定语种的文本）** @@ -12226,21 +11640,19 @@ MiniMax/speech-02-turbo **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） **输入：音频** @@ -12252,8 +11664,6 @@ MiniMax/speech-02-turbo qwen3.5-livetranslate-flash-realtime -中国内地 - 40元 3.3元 @@ -12266,8 +11676,6 @@ qwen3.5-livetranslate-flash-realtime qwen3.5-livetranslate-flash-realtime-2026-05-19 -中国内地 - 40元 3.3元 @@ -12282,8 +11690,6 @@ qwen3-livetranslate-flash-realtime > 当前能力等同于qwen3-livetranslate-flash-realtime-2025-09-22 -中国内地 - 64元 8元 @@ -12296,8 +11702,6 @@ qwen3-livetranslate-flash-realtime qwen3-livetranslate-flash-realtime-2025-09-22 -中国内地 - 64元 8元 @@ -12382,21 +11786,19 @@ qwen3-livetranslate-flash-realtime-2025-09-22 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） **输入：音频** @@ -12408,8 +11810,6 @@ qwen3-livetranslate-flash-realtime-2025-09-22 qwen3-livetranslate-flash -中国内地 - 10元 4元 @@ -12422,8 +11822,6 @@ qwen3-livetranslate-flash qwen3-livetranslate-flash-2025-12-01 -中国内地 - 10元 4元 @@ -12480,7 +11878,7 @@ qwen3-livetranslate-flash-2025-12-01 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） @@ -12488,20 +11886,16 @@ qwen3-livetranslate-flash-2025-12-01 **模型 ID（Model ID）** -**服务部署范围** - **输入单价** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-asr-flash-filetrans -中国内地 - 0.00022元/秒 不计费 @@ -12510,28 +11904,20 @@ qwen3-asr-flash-filetrans qwen3-asr-flash-filetrans-2025-11-17 -中国内地 - 36,000秒（10小时） qwen3-asr-flash > 当前能力等同于qwen3-asr-flash-2025-09-08 -中国内地 - 36,000秒（10小时） qwen3-asr-flash-2026-02-10 -中国内地 - 36,000秒（10小时） qwen3-asr-flash-2025-09-08 -中国内地 - 36,000秒（10小时） #### 美国（弗吉尼亚） @@ -12612,40 +11998,32 @@ qwen3-asr-flash-2025-09-08 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-asr-flash-realtime > 当前能力等同于qwen3-asr-flash-realtime-2025-10-27 -中国内地 - 0.00033元/秒 36,000秒（10小时） qwen3-asr-flash-realtime-2026-02-10 -中国内地 - 36,000秒（10小时） qwen3-asr-flash-realtime-2025-10-27 -中国内地 - 36,000秒（10小时） #### 新加坡 @@ -12680,58 +12058,44 @@ qwen3-asr-flash-realtime-2025-10-27 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） fun-asr > 当前能力等同于fun-asr-2025-11-07 -中国内地 - 0.00022元/秒 36,000秒（10小时） fun-asr-2025-11-07 -中国内地 - 36,000秒（10小时） fun-asr-2025-08-25 -中国内地 - 36,000秒（10小时） fun-asr-mtl -中国内地 - 36,000秒（10小时） fun-asr-mtl-2025-08-25 -中国内地 - 36,000秒（10小时） fun-asr-flash-2026-06-15 -中国内地 - 0.00022元/秒 36,000秒（10小时） @@ -12782,66 +12146,48 @@ fun-asr-flash-2026-06-15 **模型 ID（Model ID）** -**服务部署范围** - **输入单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） fun-asr-realtime -中国内地 - 0.00033元/秒 36,000秒（10小时） fun-asr-realtime-2026-02-28 -中国内地 - 36,000秒（10小时） fun-asr-realtime-2025-11-07 -中国内地 - 36,000秒（10小时） fun-asr-realtime-2025-09-15 -中国内地 - 36,000秒（10小时） fun-asr-mtl-realtime -中国内地 - 36,000秒（10小时） fun-asr-mtl-realtime-2025-12-10 -中国内地 - 36,000秒（10小时） fun-asr-flash-8k-realtime > 当前能力等同于fun-asr-flash-8k-realtime-2026-01-28 -中国内地 - 0.00022元/秒 36,000秒（10小时） fun-asr-flash-8k-realtime-2026-01-28 -中国内地 - 36,000秒（10小时） #### 新加坡 @@ -12872,16 +12218,12 @@ fun-asr-realtime-2025-11-07 **模型 ID（Model ID）** -**服务部署范围** - **输入单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) paraformer-v2 -中国内地 - 0.00008元/秒 36,000秒（10小时） @@ -12892,20 +12234,12 @@ paraformer-v2 paraformer-8k-v2 -中国内地 - paraformer-v1 -中国内地 - paraformer-8k-v1 -中国内地 - paraformer-mtl-v1 -中国内地 - #### **实时语音识别** 计费规则：按输入音频的秒数计费，输出不计费。 @@ -12914,16 +12248,12 @@ paraformer-mtl-v1 **模型 ID（Model ID）** -**服务部署范围** - **输入单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) paraformer-realtime-v2 -中国内地 - 0.00024元/秒 36,000秒（10小时） @@ -12934,16 +12264,10 @@ paraformer-realtime-v2 paraformer-realtime-v1 -中国内地 - paraformer-realtime-8k-v2 -中国内地 - paraformer-realtime-8k-v1 -中国内地 - ## **语音对话** ### **实时语音对话** @@ -12965,21 +12289,19 @@ paraformer-realtime-8k-v1 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） **文本** @@ -12991,8 +12313,6 @@ paraformer-realtime-8k-v1 qwen-audio-3.0-realtime-plus -中国内地 - 5元 40元 @@ -13005,8 +12325,6 @@ qwen-audio-3.0-realtime-plus qwen-audio-3.0-realtime-flash -中国内地 - 3元 30元 @@ -13042,26 +12360,22 @@ qwen-audio-3.0-realtime-flash **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） happyhorse-1.1-t2v -中国内地 - 720P 原价0.9元/秒 限时6折 @@ -13074,8 +12388,6 @@ happyhorse-1.1-t2v happyhorse-1.0-t2v -中国内地 - 720P 原价0.9元/秒 限时8折 @@ -13188,32 +12500,50 @@ happyhorse-1.0-t2v 原价1.6元/秒 限时8折 +#### 日本（东京） + +**模型 ID（Model ID）** + +**服务部署范围** + +**输出视频分辨率** + +**输出单价** + +happyhorse-1.1-t2v + +全球 + +720P + +原价0.9元/秒 限时6折 + +1080P + +原价1.2元/秒 限时6折 + ### **HappyHorse-图生视频-基于首帧** > 仅输出计费，计费规则请参见[视频生成](#d809366847gza)。 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） happyhorse-1.1-i2v -中国内地 - 720P 原价0.9元/秒 限时6折 @@ -13226,8 +12556,6 @@ happyhorse-1.1-i2v happyhorse-1.0-i2v -中国内地 - 720P 原价0.9元/秒 限时8折 @@ -13340,32 +12668,50 @@ happyhorse-1.0-i2v 原价1.6元/秒 限时8折 +#### 日本（东京） + +**模型 ID（Model ID）** + +**服务部署范围** + +**输出视频分辨率** + +**输出单价** + +happyhorse-1.1-i2v + +全球 + +720P + +原价0.9元/秒 限时6折 + +1080P + +原价1.2元/秒 限时6折 + ### **HappyHorse-参考生视频** > 仅输出计费，计费规则请参见[视频生成](#d809366847gza)。 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） happyhorse-1.1-r2v -中国内地 - 720P 原价0.9元/秒 限时6折 @@ -13378,8 +12724,6 @@ happyhorse-1.1-r2v happyhorse-1.0-r2v -中国内地 - 720P 原价0.9元/秒 限时8折 @@ -13492,11 +12836,33 @@ happyhorse-1.0-r2v 原价1.6元/秒 限时8折 +#### 日本（东京） + +**模型 ID（Model ID）** + +**服务部署范围** + +**输出视频分辨率** + +**输出单价** + +happyhorse-1.1-r2v + +全球 + +720P + +原价0.9元/秒 限时6折 + +1080P + +原价1.2元/秒 限时6折 + ### **HappyHorse-视频编辑** **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） @@ -13504,20 +12870,16 @@ happyhorse-1.0-r2v **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输入和输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） happyhorse-1.0-video-edit -中国内地 - 720P 原价0.9元/秒 限时8折 @@ -13600,32 +12962,52 @@ happyhorse-1.0-video-edit 原价1.6元/秒 限时8折 +#### 日本（东京） + +计费规则：输入视频和输出视频均计费，按**视频秒数**计费，失败不计费也不占用免费额度。 + +**模型 ID（Model ID）** + +**服务部署范围** + +**输出视频分辨率** + +**输入和输出单价** + +happyhorse-1.0-video-edit + +全球 + +720P + +原价0.9元/秒 限时8折 + +1080P + +原价1.6元/秒 限时8折 + ### **万相-文生视频** > 仅输出计费，计费规则请参见[视频生成](#d809366847gza)。 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.7-t2v-2026-06-12 -中国内地 - 720P 0.6元/秒 @@ -13638,8 +13020,6 @@ wan2.7-t2v-2026-06-12 wan2.7-t2v-2026-04-25 -中国内地 - 720P 0.6元/秒 @@ -13652,8 +13032,6 @@ wan2.7-t2v-2026-04-25 wan2.7-t2v -中国内地 - 720P 0.6元/秒 @@ -13666,8 +13044,6 @@ wan2.7-t2v wan2.6-t2v -中国内地 - 720P 0.6元/秒 @@ -13680,8 +13056,6 @@ wan2.6-t2v wan2.5-t2v-preview -中国内地 - 480P 0.3元/秒 @@ -13698,8 +13072,6 @@ wan2.5-t2v-preview wan2.2-t2v-plus -中国内地 - 480P 0.14元/秒 @@ -13712,8 +13084,6 @@ wan2.2-t2v-plus wanx2.1-t2v-turbo -中国内地 - 480P 0.24元/秒 @@ -13726,8 +13096,6 @@ wanx2.1-t2v-turbo wanx2.1-t2v-plus -中国内地 - 720P 0.70元/秒 @@ -13902,14 +13270,12 @@ wan2.6-t2v **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出视频类型** **输出视频分辨率** @@ -13918,12 +13284,10 @@ wan2.6-t2v **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.7-i2v-2026-04-25 -中国内地 - 有声视频 720P @@ -13938,8 +13302,6 @@ wan2.7-i2v-2026-04-25 wan2.7-i2v -中国内地 - 有声视频 720P @@ -13998,14 +13360,12 @@ wan2.7-i2v **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出视频类型** **输出视频分辨率** @@ -14014,12 +13374,10 @@ wan2.7-i2v **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.6-i2v-flash -中国内地 - 有声视频 `audio=true` @@ -14048,8 +13406,6 @@ wan2.6-i2v-flash wan2.6-i2v -中国内地 - 有声视频 720P @@ -14064,8 +13420,6 @@ wan2.6-i2v wan2.5-i2v-preview -中国内地 - 有声视频 480P @@ -14084,8 +13438,6 @@ wan2.5-i2v-preview wan2.2-i2v-flash -中国内地 - 无声视频 480P @@ -14104,8 +13456,6 @@ wan2.2-i2v-flash wan2.2-i2v-plus -中国内地 - 无声视频 480P @@ -14120,8 +13470,6 @@ wan2.2-i2v-plus wanx2.1-i2v-turbo -中国内地 - 无声视频 480P @@ -14136,8 +13484,6 @@ wanx2.1-i2v-turbo wanx2.1-i2v-plus -中国内地 - 无声视频 720P @@ -14342,26 +13688,22 @@ wan2.6-i2v **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.2-kf2v-flash -中国内地 - 480P 0.10元/秒 @@ -14378,8 +13720,6 @@ wan2.2-kf2v-flash wanx2.1-kf2v-plus -中国内地 - 720P 0.70元/秒 @@ -14419,8 +13759,6 @@ wan2.1-kf2v-plus **模型 ID（Model ID）** -**服务部署范围** - **输出视频类型** **输出视频分辨率** @@ -14429,12 +13767,10 @@ wan2.1-kf2v-plus **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.7-r2v-2026-06-12 -中国内地 - 有声视频 720P @@ -14449,8 +13785,6 @@ wan2.7-r2v-2026-06-12 wan2.7-r2v -中国内地 - 有声视频 720P @@ -14465,8 +13799,6 @@ wan2.7-r2v wan2.6-r2v-flash -中国内地 - 有声视频 `audio=true` @@ -14495,8 +13827,6 @@ wan2.6-r2v-flash wan2.6-r2v -中国内地 - 有声视频 720P @@ -14647,7 +13977,7 @@ wan2.6-r2v **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） @@ -14655,20 +13985,16 @@ wan2.6-r2v **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输入和输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.7-videoedit -中国内地 - 720P 0.6元/秒 @@ -14683,20 +14009,16 @@ wan2.7-videoedit **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wanx2.1-vace-plus -中国内地 - 720P 0.70元/秒 @@ -14756,26 +14078,20 @@ wan2.1-vace-plus **模型 ID（Model ID）** -**服务部署范围** - **单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.2-s2v-detect -中国内地 - 输入图像：0.004元/张 200张 wan2.2-s2v -中国内地 - 输出视频： - 480P：0.5元/秒 @@ -14791,33 +14107,29 @@ wan2.2-s2v **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出视频模式** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.2-animate-move -中国内地 - 标准模式`wan-std` 0.4元/秒 50秒 -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） 专业模式`wan-pro` @@ -14851,33 +14163,29 @@ wan2.2-animate-move **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输出视频模式** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） wan2.2-animate-mix -中国内地 - 标准模式`wan-std` 0.6元/秒 50秒 -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） 专业模式`wan-pro` @@ -14918,34 +14226,26 @@ wan2.2-animate-mix **模型 ID（Model ID）** -**服务部署范围** - **单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） animate-anyone-detect-gen2 -中国内地 - 输入图像：0.004元/张 200张 animate-anyone-template-gen2 -中国内地 - 输出视频：0.08元/秒 1800秒（30分钟） animate-anyone-gen2 -中国内地 - 输出视频：0.08元/秒 1800秒（30分钟） @@ -14961,26 +14261,20 @@ animate-anyone-gen2 **模型 ID（Model ID）** -**服务部署范围** - **单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） emo-detect-v1 -中国内地 - 输入图像：0.004元/张 200张 emo-v1 -中国内地 - 输出视频： - 1:1画幅视频：0.08元/秒 @@ -15001,26 +14295,20 @@ emo-v1 **模型 ID（Model ID）** -**服务部署范围** - **单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） liveportrait-detect -中国内地 - 输入图像：0.004元/张 200张 liveportrait -中国内地 - 输出视频：0.02元/秒 1800秒（30分钟） @@ -15036,26 +14324,20 @@ liveportrait **模型 ID（Model ID）** -**服务部署范围** - **单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） emoji-detect-v1 -中国内地 - 输入图像：0.004元/张 200张 emoji-v1 -中国内地 - 输出视频：0.08元/秒 1800秒（30分钟） @@ -15068,18 +14350,14 @@ emoji-v1 **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） videoretalk -中国内地 - 0.08元/秒 1800秒（30分钟） @@ -15092,20 +14370,16 @@ videoretalk **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） video-style-transform -中国内地 - 540P 0.2元/秒 @@ -15126,8 +14400,6 @@ video-style-transform **模型 ID（Model ID）** -**服务部署范围** - **输出视频类型** **输出视频分辨率** @@ -15138,8 +14410,6 @@ video-style-transform pixverse/pixverse-c1-t2v -中国内地 - 有声视频 `audio=true` @@ -15184,8 +14454,6 @@ pixverse/pixverse-c1-t2v pixverse/pixverse-v6-t2v -中国内地 - 有声视频 `audio=true` @@ -15230,8 +14498,6 @@ pixverse/pixverse-v6-t2v pixverse/pixverse-v5.6-t2v -中国内地 - 有声视频 `audio=true` @@ -15276,8 +14542,6 @@ pixverse/pixverse-v5.6-t2v pixverse/pixverse-v5.6-it2v -中国内地 - 有声视频 `audio=true` @@ -15328,8 +14592,6 @@ pixverse/pixverse-v5.6-it2v **模型 ID（Model ID）** -**服务部署范围** - **输出视频类型** **输出视频分辨率** @@ -15340,8 +14602,6 @@ pixverse/pixverse-v5.6-it2v pixverse/pixverse-c1-it2v -中国内地 - 有声视频 `audio=true` @@ -15386,8 +14646,6 @@ pixverse/pixverse-c1-it2v pixverse/pixverse-v6-it2v -中国内地 - 有声视频 `audio=true` @@ -15432,8 +14690,6 @@ pixverse/pixverse-v6-it2v pixverse/pixverse-v5.6-it2v -中国内地 - 有声视频 `audio=true` @@ -15484,8 +14740,6 @@ pixverse/pixverse-v5.6-it2v **模型 ID（Model ID）** -**服务部署范围** - **输出视频类型** **输出视频分辨率** @@ -15496,8 +14750,6 @@ pixverse/pixverse-v5.6-it2v pixverse/pixverse-c1-kf2v -中国内地 - 有声视频 `audio=true` @@ -15542,8 +14794,6 @@ pixverse/pixverse-c1-kf2v pixverse/pixverse-v6-kf2v -中国内地 - 有声视频 `audio=true` @@ -15588,8 +14838,6 @@ pixverse/pixverse-v6-kf2v pixverse/pixverse-v5.6-kf2v -中国内地 - 有声视频 `audio=true` @@ -15640,8 +14888,6 @@ pixverse/pixverse-v5.6-kf2v **模型 ID（Model ID）** -**服务部署范围** - **输出视频类型** **输出视频分辨率** @@ -15652,8 +14898,6 @@ pixverse/pixverse-v5.6-kf2v pixverse/pixverse-c1-r2v -中国内地 - 有声视频 `audio=true` @@ -15698,8 +14942,6 @@ pixverse/pixverse-c1-r2v pixverse/pixverse-v5.6-r2v -中国内地 - 有声视频 `audio=true` @@ -15750,16 +14992,12 @@ pixverse/pixverse-v5.6-r2v **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[**（注）**](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) pixverse/pixverse-lipsync -中国内地 - 0.12元/秒 无免费额度 @@ -15772,8 +15010,6 @@ pixverse/pixverse-lipsync **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** @@ -15782,8 +15018,6 @@ pixverse/pixverse-lipsync pixverse/pixverse-motioncontrol -中国内地 - 360P 0.27元/秒 @@ -15806,16 +15040,12 @@ pixverse/pixverse-motioncontrol **模型 ID（Model ID）** -**服务部署范围** - **输出单价** **免费额度**[**（注）**](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) pixverse/pixverse-upscale -中国内地 - 0.15元/秒 无免费额度 @@ -15828,8 +15058,6 @@ pixverse/pixverse-upscale **模型 ID（Model ID）** -**服务部署范围** - **输出视频类型** **输出视频分辨率** @@ -15840,8 +15068,6 @@ pixverse/pixverse-upscale kling/kling-v3-video-generation -中国内地 - 无声视频 720P @@ -15866,8 +15092,6 @@ kling/kling-v3-video-generation kling/kling-v3-omni-video-generation -中国内地 - 无声视频（无参考视频） 720P @@ -15908,8 +15132,6 @@ kling/kling-v3-omni-video-generation **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** @@ -15918,8 +15140,6 @@ kling/kling-v3-omni-video-generation vidu/viduq3-pro\_text2video -中国内地 - 540P 0.3125元/秒 @@ -15936,8 +15156,6 @@ vidu/viduq3-pro\_text2video vidu/viduq3-turbo\_text2video -中国内地 - 540P 0.25元/秒 @@ -15954,8 +15172,6 @@ vidu/viduq3-turbo\_text2video vidu/viduq2\_text2video -中国内地 - 540P 0.1125元/秒 @@ -15978,8 +15194,6 @@ vidu/viduq2\_text2video **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** @@ -15988,8 +15202,6 @@ vidu/viduq2\_text2video vidu/viduq3-pro-fast\_img2video -中国内地 - 720P 0.375元/秒 @@ -16002,8 +15214,6 @@ vidu/viduq3-pro-fast\_img2video vidu/viduq3-pro\_img2video -中国内地 - 540P 0.3125元/秒 @@ -16020,8 +15230,6 @@ vidu/viduq3-pro\_img2video vidu/viduq3-turbo\_img2video -中国内地 - 540P 0.25元/秒 @@ -16038,8 +15246,6 @@ vidu/viduq3-turbo\_img2video vidu/viduq2-pro\_img2video -中国内地 - 540P 0.15625元/秒 @@ -16056,8 +15262,6 @@ vidu/viduq2-pro\_img2video vidu/viduq2-turbo\_img2video -中国内地 - 540P 0.0875元/秒 @@ -16074,8 +15278,6 @@ vidu/viduq2-turbo\_img2video vidu/viduq2-pro-fast\_img2video -中国内地 - 720P 0.1元/秒 @@ -16094,8 +15296,6 @@ vidu/viduq2-pro-fast\_img2video **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** @@ -16104,8 +15304,6 @@ vidu/viduq2-pro-fast\_img2video vidu/viduq3-pro\_start-end2video -中国内地 - 540P 0.3125元/秒 @@ -16122,8 +15320,6 @@ vidu/viduq3-pro\_start-end2video vidu/viduq3-turbo\_start-end2video -中国内地 - 540P 0.25元/秒 @@ -16140,8 +15336,6 @@ vidu/viduq3-turbo\_start-end2video vidu/viduq2-pro\_start-end2video -中国内地 - 540P 0.15625元/秒 @@ -16158,8 +15352,6 @@ vidu/viduq2-pro\_start-end2video vidu/viduq2-turbo\_start-end2video -中国内地 - 540P 0.0875元/秒 @@ -16182,8 +15374,6 @@ vidu/viduq2-turbo\_start-end2video **模型 ID（Model ID）** -**服务部署范围** - **输出视频分辨率** **输出单价** @@ -16192,8 +15382,6 @@ vidu/viduq2-turbo\_start-end2video vidu/viduq3-ad\_reference2video -中国内地 - 720P 0.75元/秒 @@ -16206,8 +15394,6 @@ vidu/viduq3-ad\_reference2video vidu/viduq3-drama\_reference2video -中国内地 - 1080P 0.875元/秒 @@ -16216,8 +15402,6 @@ vidu/viduq3-drama\_reference2video vidu/viduq3-mix\_reference2video -中国内地 - 720P 0.78125元/秒 @@ -16230,8 +15414,6 @@ vidu/viduq3-mix\_reference2video vidu/viduq3\_reference2video -中国内地 - 540P 0.3125元/秒 @@ -16248,8 +15430,6 @@ vidu/viduq3\_reference2video vidu/viduq3-turbo\_reference2video -中国内地 - 540P 0.15625元/秒 @@ -16266,8 +15446,6 @@ vidu/viduq3-turbo\_reference2video vidu/viduq2-pro\_reference2video -中国内地 - 540P 0.25元/秒 @@ -16284,8 +15462,6 @@ vidu/viduq2-pro\_reference2video vidu/viduq2\_reference2video -中国内地 - 540P 0.21875元/秒 @@ -16310,8 +15486,6 @@ vidu/viduq2\_reference2video **模型 ID（Model ID）** -**服务部署范围** - **3D任务类型** **输出规格** @@ -16320,8 +15494,6 @@ vidu/viduq2\_reference2video Tripo/Tripo-H3.1 -中国内地 - 文生3D 标准版+无贴图 @@ -16376,8 +15548,6 @@ Tripo/Tripo-H3.1 Tripo/Tripo-P1.0 -中国内地 - 文生3D 无贴图 @@ -16416,18 +15586,14 @@ Tripo/Tripo-P1.0 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3.7-text-embedding -中国内地 - 0.5元 100万Token @@ -16436,8 +15602,6 @@ text-embedding-v4 > [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 0.5元 100万Token @@ -16446,8 +15610,6 @@ text-embedding-v3 > [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 0.5元 50万Token @@ -16456,8 +15618,6 @@ text-embedding-v2 > [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 0.7元 50万Token @@ -16466,24 +15626,18 @@ text-embedding-v1 > [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)半价 -中国内地 - 0.7元 50万Token text-embedding-async-v2 -中国内地 - 0.7元 2000万Token text-embedding-async-v1 -中国内地 - 0.7元 2000万Token @@ -16516,13 +15670,11 @@ text-embedding-v3 **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） **文本** @@ -16530,8 +15682,6 @@ text-embedding-v3 qwen3-vl-embedding -中国内地 - 0.7元 1.8元 @@ -16540,14 +15690,10 @@ qwen3-vl-embedding qwen2.5-vl-embedding -中国内地 - 100万Token tongyi-embedding-vision-plus -中国内地 - 0.5元 0.5元 @@ -16556,8 +15702,6 @@ tongyi-embedding-vision-plus tongyi-embedding-vision-flash -中国内地 - 0.15元 0.15元 @@ -16566,8 +15710,6 @@ tongyi-embedding-vision-flash multimodal-embedding-v1 -中国内地 - 0.7元 0.9元 @@ -16582,24 +15724,20 @@ multimodal-embedding-v1 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen3-vl-rerank -中国内地 - 文本输入：0.7元 图片输入：1.8元 @@ -16608,16 +15746,12 @@ qwen3-vl-rerank qwen3-rerank -中国内地 - 文本输入：0.5元 100万Token gte-rerank-v2 -中国内地 - 文本输入：0.8元 100万Token @@ -16646,8 +15780,6 @@ qwen3-rerank **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** @@ -16656,8 +15788,6 @@ qwen3-rerank farui-plus -中国内地 - 20元 20元 @@ -16672,20 +15802,16 @@ farui-plus **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） tongyi-intent-detect-v3 -中国内地 - 0.4元 1元 @@ -16698,28 +15824,24 @@ tongyi-intent-detect-v3 **说明** -以下模型仅在中国内地服务部署范围下有免费额度，其他服务部署范围下均无免费额度。 +以下模型仅在华北2（北京）下有免费额度，其他地域均无免费额度。 #### 华北2（北京） **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） qwen-plus-character > [Session Cache](https://help.aliyun.com/zh/model-studio/role-play#6034f997cde74)享有折扣 -中国内地 - 0.8元 2元 @@ -16730,8 +15852,6 @@ qwen-flash-character > [Session Cache](https://help.aliyun.com/zh/model-studio/role-play#6034f997cde74)享有折扣 -中国内地 - 0.25元 1.5元 @@ -16742,8 +15862,6 @@ qwen-flash-character-2026-02-26 > [Session Cache](https://help.aliyun.com/zh/model-studio/role-play#6034f997cde74)享有折扣 -中国内地 - 0.18元 1.5元 @@ -16850,20 +15968,16 @@ qwen-plus-character **模型 ID（Model ID）** -**服务部署范围** - **输入单价（每百万Token）** **输出单价（每百万Token）** **免费额度**[（注）](https://help.aliyun.com/zh/model-studio/new-free-quota#591f3dfedfyzj) -有效期：阿里云百炼开通后90天内 +有效期：自开通百炼/模型发布/申请通过之日起90天内（以较晚者为准） gui-plus -中国内地 - 1.5元 4.5元 @@ -16872,8 +15986,6 @@ gui-plus gui-plus-2026-02-26 -中国内地 - ## 错误码 如果模型调用失败并返回报错信息，请参见[错误码](https://help.aliyun.com/zh/model-studio/error-code)进行解决。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/model-training-and-deployment-billing.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/model-training-and-deployment-billing.md index 8bc376aa..9f893367 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/model-training-and-deployment-billing.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/model-training-and-deployment-billing.md @@ -474,9 +474,9 @@ wan2.2-kf2v-flash - 后付费时，如果账户欠费，部署的资源将继续保留并计费 24 小时，在这 24 小时内服务仍可正常使用。超过 24 小时后系统停止计费，模型部署进入欠费状态，底层资源将被删除，但模型部署任务仍会保留。补足欠费后，系统将重新分配资源并恢复使用（恢复后继续产生费用）。如果您不希望继续产生费用，可删除模型部署任务，删除成功后将不再计费。 -当模型输入超过最长输入 Token 或 超出购买的 TPM 量时，相关调用将自动切换为当前模型的按量付费模式。此时，推理性能可能下降，将受业务空间中当前快照模型的公共流量的管控，[费用](https://help.aliyun.com/zh/model-studio/model-pricing)按模型调用（按量付费）标准计收。 +当模型输入超过最长输入 Token 时，相关调用将自动切换为当前模型的按量付费模式；超出购买的 TPM 量时，按创建时选择的溢出策略处理（「自动溢出」切换为按量付费，「仅使用 PTU 容量」返回 429）。此时，推理性能可能下降，将受业务空间中当前快照模型的公共流量的管控，[费用](https://help.aliyun.com/zh/model-studio/model-pricing)按模型调用（按量付费）标准计收。 -- 此时，调用 API 返回 Header 将包含：`x-dashscope-ptu-overflow:true`。 +- 此时（仅「自动溢出」策略下），调用 API 返回 Header 将包含：`x-dashscope-ptu-overflow:true`。 - TPM 统计请前往：[模型监控（北京）](https://bailian.console.aliyun.com/?tab=model#/model-telemetry)。 @@ -820,15 +820,33 @@ MU2 x 8 ¥240,288 +MU3 x 8 + +¥1,096 + +¥527,752 + 千问3.6-35B-A3B qwen3.6-35b-a3b -MU8 x 1 +MU1 x 8 -¥47 +¥432 -¥22,400 +¥208,944 + +MU2 x 8 + +¥504 + +¥240,288 + +MU3 x 8 + +¥1,096 + +¥527,752 MU9 x 1 @@ -846,6 +864,12 @@ MU1 x 2 ¥52,236 +MU3 x 8 + +¥1,096 + +¥527,752 + 千问3.6-Plus-2026-04-02 qwen3.6-plus-2026-04-02 @@ -866,12 +890,6 @@ PD分离模式：¥417,888 qwen3.5-397b-a17b -MU2 x 8 - -¥504 - -¥240,288 - MU3 x 8 MU3 x 16（PD分离模式） @@ -884,6 +902,12 @@ PD分离模式：¥2,192 PD分离模式：¥1,055,504 +MU6 x 16 + +¥400 + +¥193,424 + 千问3.5-122B-A10B qwen3.5-122b-a10b @@ -894,11 +918,11 @@ MU1 x 4 ¥104,472 -MU2 x 8 +MU6 x 16 -¥504 +¥400 -¥240,288 +¥193,424 千问3.5-35B-A3B @@ -916,10 +940,40 @@ MU2 x 8 ¥240,288 +MU3 x 8 + +¥1,096 + +¥527,752 + +MU9 x 1 + +¥51 + +¥24,600 + 千问3.5-27B qwen3.5-27b +MU2 x 8 + +¥504 + +¥240,288 + +MU3 x 8 + +¥1,096 + +¥527,752 + +MU8 x 1 + +¥47 + +¥22,400 + MU9 x 1 ¥51 @@ -936,6 +990,12 @@ MU1 x 2 ¥52,236 +MU2 x 2 + +¥126 + +¥60,072 + MU8 x 1 ¥47 @@ -962,12 +1022,24 @@ MU1 x 2 qwen3.5-plus-2026-02-15 +MU1 x 8 + MU1 x 16（PD分离模式） +¥432 + PD分离模式：¥864 +¥208,944 + PD分离模式：¥417,888 +MU2 x 8 + +¥504 + +¥240,288 + MU3 x 8 MU3 x 16（PD分离模式） @@ -996,41 +1068,25 @@ MU2 x 8 ¥240,288 -千问3-Next-80B-A3B-Instruct - -qwen3-next-80b-a3b-instruct - -MU1 x 2 - -¥108 - -¥52,236 - 千问3-32B qwen3-32b -MU1 x 4 - -¥216 - -¥104,472 - -MU6 x 4 +MU6 x 16 -¥100 +¥400 -¥48,356 +¥193,424 -千问3-30B-A3B +千问3-30B-A3B-Thinking-2507 -qwen3-30b-a3b +qwen3-30b-a3b-thinking-2507 -MU9 x 2 +MU1 x 2 -¥102 +¥108 -¥49,200 +¥52,236 千问3-8B @@ -1080,12 +1136,6 @@ MU1 x 2 ¥52,236 -MU5 x 1 - -¥21 - -¥10,139 - 千问3-Embedding-0.6B qwen3-embedding-0.6b @@ -1154,6 +1204,16 @@ MU5 x 1 ¥10,139 +千问2.5-开源版-72B + +qwen2.5-72b-instruct + +MU1 x 8 + +¥432 + +¥208,944 + 千问2.5-开源版-32B qwen2.5-32b-instruct @@ -1190,26 +1250,6 @@ MU5 x 1 ¥10,139 -千问2.5-开源版-3B - -qwen2.5-3b-instruct - -MU5 x 1 - -¥21 - -¥10,139 - -千问-Flash-2025-07-28 - -qwen-flash-2025-07-28 - -MU1 x 4 - -¥216 - -¥104,472 - 千问-Plus-2025-07-28 qwen-plus-2025-07-28 @@ -1266,6 +1306,12 @@ GLM-5.1 glm-5.1 +MU2 x 8 + +¥504 + +¥240,288 + MU3 x 16（PD分离模式） PD分离模式：¥2,192 @@ -1298,6 +1344,16 @@ PD分离模式：¥800 PD分离模式：¥386,848 +GLM-4.7-Flash + +glm-4.7-flash + +MU3 x 16（PD分离模式） + +PD分离模式：¥2,192 + +PD分离模式：¥1,055,504 + ###### DeepSeek **模型名称** @@ -1318,11 +1374,11 @@ DeepSeek-v4-Flash deepseek-v4-flash -MU1 x 8 +MU3 x 8 -¥432 +¥1,096 -¥208,944 +¥527,752 DeepSeek-v3.2 @@ -1350,16 +1406,6 @@ PD分离模式：¥480,576 **最小计费：天** -MiniMax-M2.5 - -MiniMax-M2.5 - -MU1 x 16（PD分离模式） - -PD分离模式：¥864 - -PD分离模式：¥417,888 - Kimi-K2.5 kimi-k2.5 @@ -1402,15 +1448,21 @@ MU2 x 8 **最小计费：天** -千问3-VL-235B-A22B-Instruct +千问3-VL-32B-Instruct -qwen3-vl-235b-a22b-instruct +qwen3-vl-32b-instruct -MU1 x 4 +MU2 x 8 -¥216 +¥504 -¥104,472 +¥240,288 + +MU3 x 8 + +¥1,096 + +¥527,752 千问3-VL-8B-Instruct @@ -1422,6 +1474,12 @@ MU1 x 2 ¥52,236 +MU5 x 1 + +¥21 + +¥10,139 + 千问3-VL-4B-Instruct qwen3-vl-4b-instruct @@ -1442,6 +1500,16 @@ MU5 x 1 ¥10,139 +千问3-VL-Embedding-2B + +qwen3-vl-embedding-2b + +MU5 x 1 + +¥21 + +¥10,139 + 千问3-VL-Flash-2025-10-15 qwen3-vl-flash-2025-10-15 @@ -1472,16 +1540,6 @@ MU6 x 4 ¥48,356 -千问VL-OCR-2025-11-20 - -qwen-vl-ocr-2025-11-20 - -MU6 x 4 - -¥100 - -¥48,356 - ###### 千问 Omni **模型名称** @@ -1562,6 +1620,14 @@ MU5 **元/千Token** +千问3.5-27B 邀测中 + +qwen3.5-27b + +¥0.0006 + +¥0.0048 + 千问3-32B qwen3-32b @@ -1592,6 +1658,16 @@ qwen3-8b 思考模式：¥0.005 +千问3-4B-Instruct-2507 + +qwen3-4b-instruct-2507 + +¥0.0003 + +非思考模式：¥0.0012 + +思考模式：¥0.003 + 千问2.5-开源版-72B qwen2.5-72b-instruct @@ -1624,6 +1700,14 @@ qwen2.5-7b-instruct ¥0.001 +千问2-开源版-7B + +qwen2-7b-instruct + +¥0.001 + +¥0.002 + ##### 千问VL **基础模型** @@ -1670,6 +1754,14 @@ qwen2.5-vl-7b-instruct ¥0.005 +千问2.5-VL-3B-Instruct + +qwen2.5-vl-3b-instruct + +¥0.0012 + +¥0.0036 + ### **图像生成模型-万相** 经过SFT-LoRA高效微调的万相图像生成模型，部署免费，调用按微调的基础模型的标准调用价格计费。模型训练和部署流程请参见[图像生成模型调优](https://help.aliyun.com/zh/model-studio/wan-image-generation-finetune-guide)。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/new-free-quota.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/new-free-quota.md index 6e121bcc..85b1cbde 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/new-free-quota.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/new-free-quota.md @@ -4,17 +4,17 @@ **说明** -仅华北2（北京）地域且服务部署范围为[中国内地](https://help.aliyun.com/zh/model-studio/regions/#080da663a75xh)的模型享有免费额度，其他地域和部署范围无免费额度。 +仅华北2（北京）地域模型享有免费额度，其他地域无免费额度。 ## 规则说明 ### 有效期 -免费额度的有效期为 30～90 天，从开通阿里云百炼或模型申请通过之日起计算。额度到期或耗尽后，继续调用模型推理服务将[产生计费](https://help.aliyun.com/zh/model-studio/billing-for-model-studio)。 +免费额度的有效期为 90 天，从开通阿里云百炼、模型发布或模型申请通过之日起计算（以较晚者为准）。额度到期或耗尽后，继续调用模型推理服务将[产生计费](https://help.aliyun.com/zh/model-studio/billing-for-model-studio)。 **重要** -自**2025年9月8日11点**起，首次开通阿里云百炼的用户，获赠的新人免费额度有效期调整为 90 天，在此之前已开通的用户不受影响，详情参考[阿里云百炼新人免费额度有效期调整通知](https://help.aliyun.com/zh/model-studio/new-free-quota-validity-adjustment)。 +**2025年9月8日11点**前已开通阿里云百炼的用户，免费额度有效期可能不足90天；在此之后开通的用户有效期为90天。详情参考[阿里云百炼新人免费额度有效期调整通知](https://help.aliyun.com/zh/model-studio/new-free-quota-validity-adjustment)。 免费额度过期后自动失效，不支持补发、延期或重置： @@ -55,7 +55,7 @@ ## 获取免费额度 -访问[阿里云百炼-中国内地版](https://bailian.console.aliyun.com/#/model-market)，阅读并同意协议后，系统将自动开通阿里云百炼并发放**免费推理额度**。 +访问[阿里云百炼-华北2（北京）地域](https://bailian.console.aliyun.com/#/model-market)，阅读并同意协议后，系统将自动开通阿里云百炼并发放**免费推理额度**。 > 如果未弹出服务协议，表示您已经开通过阿里云百炼且获得免费额度。 @@ -212,9 +212,12 @@ 免费额度列显示**无免费额度**或**免费额度**区域不显示，可能由以下原因之一导致： -- **免费额度已到期或耗尽**：免费额度的有效期为 30～90 天，从开通阿里云百炼或模型申请通过之日起计算，到期或耗尽后将不再显示，继续调用模型将产生计费。 -- **该模型所在地域或服务部署范围不享有免费额度**：仅华北2（北京）地域且服务部署范围为中国内地的模型、以及仅新加坡地域且服务部署范围为国际的模型享有免费额度，其他地域和部署范围无免费额度。 +- **免费额度已到期或耗尽**：免费额度的有效期为 90 天，从开通阿里云百炼、模型发布或模型申请通过之日起计算（以较晚者为准），到期或耗尽后将不再显示，继续调用模型将产生计费。 + +- **该模型所在地域或服务部署范围不享有免费额度**：仅华北2（北京）地域的模型以及仅新加坡地域的模型享有免费额度，其他地域和部署范围无免费额度。 + - **该模型本身不提供免费额度**：部分模型不参与新人免费额度发放。 + ### 使用哪种 API Key 才能消耗免费额度？ diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/savings-plan-and-resource-package.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/savings-plan-and-resource-package.md index 5ce9dfe5..a1867096 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/savings-plan-and-resource-package.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/test-1/savings-plan-and-resource-package.md @@ -80,7 +80,7 @@ AI 通用型节省计划是针对大模型按量付费使用场景设计的折 - C 类：qwen3.6-max-preview、DeepSeek、Kimi、GLM、MiniMax、HappyHorse - > 三方直供模型不支持抵扣，详情参见[三方直供模型支持抵扣 AI 通用型节省计划吗？](#85a29cab67489) + > 三方直供模型不支持抵扣，其中 DeepSeek、Kimi、GLM 已有阿里云直供版本可抵扣，MiniMax 暂无阿里云直供版本。详情参见[三方直供模型支持抵扣 AI 通用型节省计划吗？](#85a29cab67489) **每月承诺消费金额范围** @@ -802,7 +802,7 @@ ASR模型按秒计费，TTS模型按字符计费，请前往[百炼控制台](ht - 根据[退订规则](https://help.aliyun.com/zh/user-center/cancel-subscription/)，预付费商品未发生使用的部分，可按未使用额度费用[申请退款](https://billing-cost.console.aliyun.com/refund/refund?commodityType=RESOURCE_PLANS&refundType=NOREASON_REFUND)；已使用的部分则无法退款。 -**使用限制**：资源包按模型名称严格匹配，**跨版本或子型号不通用**，请以资源包购买页标注的适用模型为准。例如，qwen-plus 资源包不支持抵扣 qwen-max 或 qwen-turbo 的调用费用；若需要同时覆盖多个模型版本的调用费用，建议选择 AI 通用型节省计划。 +**使用限制**：资源包按模型名称严格匹配，**跨版本或子型号不通用**，请以资源包购买页标注的适用模型为准。例如，qwen-plus 资源包不支持抵扣 qwen-max 的调用费用；若需要同时覆盖多个模型版本的调用费用，建议选择 AI 通用型节省计划。 ### **大语言模型推理资源包** @@ -812,48 +812,36 @@ ASR模型按秒计费，TTS模型按字符计费，请前往[百炼控制台](ht [大语言模型推理资源包 qwen-max](https://common-buy.aliyun.com/?commodityCode=sfm_llminference2_dp_cn#/buy) -[大语言模型推理资源包 qwen-turbo](https://common-buy.aliyun.com/?commodityCode=sfm_llminference3_dp_cn#/buy) - **适用地域** 华北2（北京） 华北2（北京） -华北2（北京） - **适用模型** qwen-plus 及 qwen-plus-latest的实时推理服务（[非思考模式](https://help.aliyun.com/zh/model-studio/deep-thinking)） qwen-max的实时推理服务（[非思考模式](https://help.aliyun.com/zh/model-studio/deep-thinking)） -qwen-turbo的实时推理服务（[非思考模式](https://help.aliyun.com/zh/model-studio/deep-thinking)） - **包含输入和输出总Tokens** 1,200万/1.1亿 1,800万/3,900万/3.9亿/11.7亿/19.5亿 -3,500万/3.5亿/17.5亿/35亿 - **价格（元）** 11.66/114.4 57.6/125/1250/3750/6250 -11.45/114.45/572.25/1144.5 - **有效期** 自购买日起生效，有效期可选 3 个月、6 个月或 1 年。 自购买之日起有效期为 1 年。 -自购买之日起有效期为 1 年。 - **使用限制** - **qwen-plus**、**qwen-plus-latest** @@ -866,7 +854,7 @@ qwen-turbo的实时推理服务（[非思考模式](https://help.aliyun.com/zh/m - [Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)、[上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)、[模型调优](https://help.aliyun.com/zh/model-studio/model-training-overview)、[模型部署](https://help.aliyun.com/zh/model-studio/model-deployment-introduction)产生的费用。 -- **qwen-max**、**qwen-turbo** +- **qwen-max** - 仅支持抵扣实时推理产生的费用（[非思考模式](https://help.aliyun.com/zh/model-studio/deep-thinking)，包含输入和输出），不支持抵扣[Batch调用](https://help.aliyun.com/zh/model-studio/batch-interfaces-compatible-with-openai/)、[上下文缓存](https://help.aliyun.com/zh/model-studio/context-cache)、[模型调优](https://help.aliyun.com/zh/model-studio/model-training-overview)、[模型部署](https://help.aliyun.com/zh/model-studio/model-deployment-introduction)产生的费用。 @@ -895,7 +883,7 @@ qwen-turbo的实时推理服务（[非思考模式](https://help.aliyun.com/zh/m **图像编辑**：qwen-image-edit-plus -**资源包容量 (生成图片张数)** +**资源包容量（生成图片张数）** 80/400 @@ -962,7 +950,7 @@ qwen-turbo的实时推理服务（[非思考模式](https://help.aliyun.com/zh/m ### **三方直供模型支持抵扣 AI 通用型节省计划吗？** -[C 类模型](#ho1f5x10wuun0)中，阿里直供的模型支持抵扣，三方直供的模型不支持抵扣。可以在[百炼模型广场](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)中通过模型卡片右上角标识（如"阿里直供"或"三方直供"标签）判断。 +[C 类模型](#ho1f5x10wuun0)中，阿里直供的模型支持抵扣，三方直供的模型不支持抵扣。目前，DeepSeek、Kimi、GLM 有阿里云直供版本，可通过 AI 通用型节省计划抵扣；MiniMax 暂无阿里云直供版本，暂不支持通过节省计划抵扣。可以在[百炼模型广场](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)中通过模型卡片右上角标识（如"阿里直供"或"三方直供"标签）查看最新的直供模型列表。 ### **购买节省计划后如何使用？** diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/add-vision-skill.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/add-vision-skill.md index 23651b2d..f20936f9 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/add-vision-skill.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/add-vision-skill.md @@ -1,6 +1,6 @@ # 添加视觉理解能力 -百炼 Coding Plan 中的部分模型（qwen3.6-plus、qwen3.5-plus、kimi-k2.5）原生支持视觉理解，可直接处理图片输入。对于 glm-5、MiniMax-M2.5 等纯文本模型，可通过添加本地 Skill 使其获得视觉能力。 +百炼 Coding Plan 中的部分模型（qwen3.7-plus、qwen3.6-plus、qwen3.5-plus、kimi-k2.5）原生支持视觉理解，可直接处理图片输入。对于 glm-5、MiniMax-M2.5 等纯文本模型，可通过添加本地 Skill 使其获得视觉能力。 **说明** @@ -21,6 +21,8 @@ **说明** +- qwen3.7-plus + - qwen3.6-plus - qwen3.5-plus @@ -51,7 +53,7 @@ ## 方法 1：直接使用视觉模型（推荐） -qwen3.6-plus、qwen3.5-plus 和 kimi-k2.5 具备视觉理解能力。如果经常需要处理图片，直接切换到这些模型是最简单、推荐的做法。 +qwen3.7-plus、qwen3.6-plus、qwen3.5-plus 和 kimi-k2.5 具备视觉理解能力。如果经常需要处理图片，直接切换到这些模型是最简单、推荐的做法。 **工具** @@ -59,15 +61,15 @@ qwen3.6-plus、qwen3.5-plus 和 kimi-k2.5 具备视觉理解能力。如果经 Claude Code -`/model qwen3.6-plus`或`/model qwen3.5-plus`或 `/model kimi-k2.5` +`/model qwen3.7-plus`或`/model qwen3.6-plus`或`/model qwen3.5-plus`或 `/model kimi-k2.5` OpenCode -`/models`→ 搜索并选择`qwen3.6-plus`或`qwen3.5-plus`或`kimi-k2.5` +`/models`→ 搜索并选择`qwen3.7-plus`或`qwen3.6-plus`或`qwen3.5-plus`或`kimi-k2.5` Qwen Code -`/model`→ 选择`qwen3.6-plus`或`qwen3.5-plus`或`kimi-k2.5` +`/model`→ 选择`qwen3.7-plus`或`qwen3.6-plus`或`qwen3.5-plus`或`kimi-k2.5` 更多编程工具中的模型切换方式请参考[接入客户端/开发工具](https://help.aliyun.com/zh/model-studio/use-chat-client-or-development-tool/)。切换后可直接在对话中引用图片路径，或拖拽/粘贴图片。 @@ -91,9 +93,9 @@ Qwen Code --- name: image-analyzer description: 帮助没有视觉能力的模型进行图像理解。当需要分析图像内容、提取图片中的信息、文字、界面元素，或理解截图、图表、架构图等任何视觉内容时，使用此技能，传入图片路径即可获得描述信息。 - model: qwen3.6-plus + model: qwen3.7-plus --- - qwen3.6-plus具有视觉理解能力，请直接使用qwen3.6-plus模型进行图片理解。 + qwen3.7-plus具有视觉理解能力，请直接使用qwen3.7-plus模型进行图片理解。 ``` 创建完成后的目录结构如下： @@ -111,7 +113,7 @@ Qwen Code 2. 下载[aliyun.png](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260225/hxwnny/aliyun.png)到项目目录下，并提问：`请加载image-analyzer skill，描述一下 aliyun.png banner位置是什么信息。`可收到如下回复： - ![image.png](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/5928202771/p1054884.png) + aliyun.png 为阿里云官网首页截图，banner 区域标题为 **Coding Plan 已支持 Qwen3.5**，正文介绍阿里云百炼支持 Qwen3.5、Kimi-k2.5、GLM-4.7 等模型，新客首月仅 7.9 元，页面提供**立即订阅**和**在线咨询**入口。 ### OpenCode @@ -128,13 +130,13 @@ Qwen Code **说明** - model 字段必须使用 OpenCode 配置文件中定义的 provider 和模型名称。参考 [OpenCode](https://help.aliyun.com/zh/model-studio/opencode) 文档的配置示例，应为`bailian-coding-plan/qwen3.6-plus`。 + model 字段必须使用 OpenCode 配置文件中定义的 provider 和模型名称。参考 [OpenCode](https://help.aliyun.com/zh/model-studio/opencode) 文档的配置示例，应为`bailian-coding-plan/qwen3.7-plus`。 ``` --- description: Analyzes images using a vision-capable model. Use this agent when the user needs to understand image content, extract information from screenshots, diagrams, UI mockups, or any visual content. Invoke with @image-analyzer followed by the image path and your question. mode: subagent - model: bailian-coding-plan/qwen3.6-plus + model: bailian-coding-plan/qwen3.7-plus tools: write: false edit: false @@ -156,7 +158,18 @@ Qwen Code 2. 下载[aliyun.png](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260225/hxwnny/aliyun.png)到项目目录下，通过`@`唤起`image-analyzer`并提问：`@image-analyzer，描述一下 aliyun.png banner位置是什么信息。`可收到如下回复： - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6472262771/p1055847.png) + ``` + Banner位置包含： + 左侧： + - 汉堡菜单图标 + - 阿里云橙色logo和"阿里云"文字 + - 导航菜单：大模型、产品、解决方案、权益、定价、云市场、伙伴、服务、了解阿里云 + 右侧： + - 搜索框(显示"大模型") + - 图标：蓝色圆圈、地球、耳机 + - 链接：文档、备案、控制台 + Build · glm-5 · 37.0s + ``` ## **常见问题** @@ -181,6 +194,24 @@ Qwen Code "apiKey": "sk-sp-xxx" }, "models": { + "qwen3.7-plus": { + "name": "Qwen3.7 Plus", + "modalities": { + "input": [ + "text", + "image" + ], + "output": [ + "text" + ] + }, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 1024 + } + } + }, "qwen3.6-plus": { "name": "Qwen3.6 Plus", "modalities": { @@ -259,6 +290,15 @@ Qwen Code "apiKey": "YOUR_API_KEY", "api": "openai-completions", "models": [ + { + "id": "qwen3.7-plus", + "name": "qwen3.7-plus", + "reasoning": false, + "input": ["text", "image"], + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "contextWindow": 1000000, + "maxTokens": 65536 + }, { "id": "qwen3.6-plus", "name": "qwen3.6-plus", @@ -296,6 +336,7 @@ Qwen Code "primary": "bailian/qwen3.6-plus" }, "models": { + "bailian/qwen3.7-plus": {}, "bailian/qwen3.6-plus": {}, "bailian/qwen3.5-plus": {}, "bailian/kimi-k2.5": {} diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan.md index 1ab71393..57403f95 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan.md @@ -50,11 +50,11 @@ Coding Plan 整合了千问、GLM、Kimi 、MiniMax顶级模型，并兼容主 - 每月**90,000** 次请求 -- **限时优惠：**活动已结束，当前价格以下单页为准。 +- **限时优惠**：活动已结束，当前价格以下单页为准。 - **限量抢购**：名额有限、先到先得。每日 09:30:00（UTC+08:00）补充，可前往[Coding Plan 页面](https://www.aliyun.com/benefit/scene/codingplan)抢购。 -- **额度消耗：**单次提问将按实际“模型调用次数”扣除额度。简单任务约消耗 5-10 次，复杂任务约 10-30+ 次，实际消耗受任务难度、上下文及工具使用影响。在[Coding Plan 页面](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan)可以查看用量。 +- **额度消耗：**单次提问将按实际“模型调用次数”扣除额度。简单任务约消耗 5-10 次，复杂任务约 10-30+ 次，实际消耗受任务难度、上下文及工具使用影响。在[Coding Plan 页面](https://bailian.console.aliyun.com/cn-beijing/?tab=plan#/efm/subscription/coding-plan)可以查看用量。 - **额度恢复**： @@ -84,7 +84,7 @@ Coding Plan 整合了千问、GLM、Kimi 、MiniMax顶级模型，并兼容主 您需要获取并配置套餐专属的 API Key 和 Base URL，才能正确使用并抵扣套餐额度。 -- **API Key**：在[Coding Plan 页面](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan)，获取Coding Plan 专属 API Key（格式为`sk-sp-xxxxx`）。 +- **API Key**：在[Coding Plan 页面](https://bailian.console.aliyun.com/cn-beijing/?tab=plan#/efm/subscription/coding-plan)，获取Coding Plan 专属 API Key（格式为`sk-sp-xxxxx`）。 - **Base URL**：后续需在 AI 工具中配置以下其中一个Base URL（因工具而异），具体操作请参见对应的AI工具文档。 @@ -99,33 +99,33 @@ Coding Plan 专属的 API Key 和 Base URL 与百炼按量计费的 API Key（`s ### **步骤三：接入AI工具** - [**OpenClaw**开源、自托管个人 AI 助手](https://help.aliyun.com/zh/model-studio/openclaw) +[**OpenClaw**开源、自托管个人 AI 助手](https://help.aliyun.com/zh/model-studio/openclaw) - [**Hermes Agent**开源 AI 代理框架，内置自学习循环](https://help.aliyun.com/zh/model-studio/hermes-agent) +[**Hermes Agent**开源 AI 代理框架，内置自学习循环](https://help.aliyun.com/zh/model-studio/hermes-agent) - [**Claude Code**AI 终端编码助手，支持自然语言编程](https://help.aliyun.com/zh/model-studio/claude-code) +[**Claude Code**AI 终端编码助手，支持自然语言编程](https://help.aliyun.com/zh/model-studio/claude-code) - [**OpenCode**开源 AI 编程代理工具](https://help.aliyun.com/zh/model-studio/opencode) +[**OpenCode**开源 AI 编程代理工具](https://help.aliyun.com/zh/model-studio/opencode) - [**Cursor**AI 原生代码编辑器](https://help.aliyun.com/zh/model-studio/cursor) +[**Cursor**AI 原生代码编辑器](https://help.aliyun.com/zh/model-studio/cursor) - [**Codex**OpenAI 推出的命令行编程工具](https://help.aliyun.com/zh/model-studio/codex) +[**Codex**OpenAI 推出的命令行编程工具](https://help.aliyun.com/zh/model-studio/codex) - [**Qwen Code**开源命令行 AI 编码工具](https://help.aliyun.com/zh/model-studio/qwen-code) +[**Qwen Code**开源命令行 AI 编码工具](https://help.aliyun.com/zh/model-studio/qwen-code) - [**QwenPaw**开源个人 AI 助手，支持本地与云端部署](https://help.aliyun.com/zh/model-studio/qwenpaw) +[**QwenPaw**开源个人 AI 助手，支持本地与云端部署](https://help.aliyun.com/zh/model-studio/qwenpaw) - [**Cherry Studio**多模型桌面客户端](https://help.aliyun.com/zh/model-studio/cherry-studio) +[**Cherry Studio**多模型桌面客户端](https://help.aliyun.com/zh/model-studio/cherry-studio) - [**Chatbox**跨平台 AI 桌面客户端](https://help.aliyun.com/zh/model-studio/chatbox) +[**Chatbox**跨平台 AI 桌面客户端](https://help.aliyun.com/zh/model-studio/chatbox) - [**Cline**VS Code 扩展，智能代码补全和调试](https://help.aliyun.com/zh/model-studio/cline) +[**Cline**VS Code 扩展，智能代码补全和调试](https://help.aliyun.com/zh/model-studio/cline) - [**Qoder**面向真实软件开发的 Agentic 编码平台](https://help.aliyun.com/zh/model-studio/qoder-agent) +[**Qoder**面向真实软件开发的 Agentic 编码平台](https://help.aliyun.com/zh/model-studio/qoder-agent) - [**Lingma**阿里云推出的智能编码辅助工具](https://help.aliyun.com/zh/model-studio/lingma-agent) +[**Lingma**阿里云智能编码助手，提供独立 IDE](https://help.aliyun.com/zh/model-studio/lingma-agent) - [**Kilo CLI**轻量高性能命令行编程工具](https://help.aliyun.com/zh/model-studio/kilo-cli) +[**Kilo CLI**轻量高性能命令行编程工具](https://help.aliyun.com/zh/model-studio/kilo-cli) [··· **更多工具**其他编程工具](https://help.aliyun.com/zh/model-studio/more-tools) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/add-vision-skill.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/add-vision-skill.md new file mode 100644 index 00000000..dad44bd5 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/add-vision-skill.md @@ -0,0 +1,329 @@ +# 添加视觉理解能力 + +Token Plan 支持的部分模型（qwen3.7-plus 等）原生支持视觉理解，可直接处理图片输入。对于 glm-5、MiniMax-M2.5 等纯文本模型，可通过添加本地 Skill 使其获得视觉能力。 + +**说明** + +运行图片理解 Skill 会消耗 Token Plan Credits，无其他收费项。 + +## 前提条件 + +1. 已订阅 [Token Plan](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription)。 + +2. 已在 AI 工具中完成接入配置，且能正常对话，详情请参见[接入客户端/开发工具](https://help.aliyun.com/zh/model-studio/use-chat-client-or-development-tool/)。 + + +## 视觉支持情况 + +**模型** + +**是否支持视觉** + +**说明** + +- qwen3.8-max-preview + +- qwen3.7-plus + +- qwen3.6-plus + +- kimi-k2.5 等 + + +是 + +无需额外配置，可直接传入图片 + +- qwen3-max-2026-01-23 + +- qwen3-coder-next + +- qwen3-coder-plus + +- glm-5 + +- glm-4.7 + +- MiniMax-M2.5 + + +否 + +需通过 Skill 或 Agent 辅助模型获得视觉能力 + +## 方法 1：直接使用视觉模型（推荐） + +qwen3.7-plus 等模型具备视觉理解能力。如果经常需要处理图片，直接切换到这些模型是最简单、推荐的做法。 + +**工具** + +**模型切换方式** + +Claude Code + +`/model qwen3.7-plus`或`/model qwen3.6-plus`或`/model qwen3.5-plus`或 `/model kimi-k2.5` + +OpenCode + +`/models`→ 搜索并选择`qwen3.7-plus`或`qwen3.6-plus`或`qwen3.5-plus`或`kimi-k2.5` + +Qwen Code + +`/model`→ 选择`qwen3.7-plus`或`qwen3.6-plus`或`qwen3.5-plus`或`kimi-k2.5` + +更多编程工具中的模型切换方式请参考[接入客户端/开发工具](https://help.aliyun.com/zh/model-studio/use-chat-client-or-development-tool/)。切换后可直接在对话中引用图片路径，或拖拽/粘贴图片。 + +## 方法 2：通过 Skill 或 Agent 添加视觉能力 + +如需使用 glm-5、MiniMax-M2.5 等不支持视觉的模型处理图片，可通过配置 Skill 或 Agent 实现。 + +## Claude Code + +1. **添加 Skill** + + 在项目目录下的 `.claude` 文件夹中新建 `skills/image-analyzer` 目录： + + ``` + mkdir -p .claude/skills/image-analyzer + ``` + + 在该目录下创建 `SKILL.md` 文件，并写入以下内容： + + ``` + --- + name: image-analyzer + description: 帮助没有视觉能力的模型进行图像理解。当需要分析图像内容、提取图片中的信息、文字、界面元素，或理解截图、图表、架构图等任何视觉内容时，使用此技能，传入图片路径即可获得描述信息。 + model: qwen3.7-plus + --- + qwen3.7-plus具有视觉理解能力，请直接使用qwen3.7-plus模型进行图片理解。 + ``` + + 创建完成后的目录结构如下： + + ``` + .claude/ + └── skills/ + └── image-analyzer/ + └── SKILL.md + ``` + +2. **开始使用** + + 1. 在项目目录下运行`claude`启动 Claude Code，并运行`/model glm-5`切换到`glm-5`模型。 + + 2. 下载[aliyun.png](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260225/hxwnny/aliyun.png)到项目目录下，并提问：`请加载image-analyzer skill，描述一下 aliyun.png banner位置是什么信息。`可收到如下回复： + + aliyun.png 为阿里云官网首页截图，banner 区域标题为 **Coding Plan 已支持 Qwen3.5**，正文介绍阿里云百炼支持 Qwen3.5、Kimi-k2.5、GLM-4.7 等模型，新客首月仅 7.9 元，页面提供**立即订阅**和**在线咨询**入口。 + + +### OpenCode + +1. **添加 Agent** + + 在项目目录下的 `.opencode` 文件夹中新建 `agents` 目录： + + ``` + mkdir -p .opencode/agents + ``` + + 在该目录下创建`image-analyzer.md`文件，并写入以下内容： + + **说明** + + model 字段必须使用 OpenCode 配置文件中定义的 provider 和模型名称。参考 [OpenCode](https://help.aliyun.com/zh/model-studio/opencode) 文档的配置示例，应为`bailian-token-plan/qwen3.7-plus`。 + + ``` + --- + description: Analyzes images using a vision-capable model. Use this agent when the user needs to understand image content, extract information from screenshots, diagrams, UI mockups, or any visual content. Invoke with @image-analyzer followed by the image path and your question. + mode: subagent + model: bailian-token-plan/qwen3.7-plus + tools: + write: false + edit: false + --- + You have vision capabilities. Analyze the provided image and return a clear, structured description focused on what the user is asking about. + ``` + + 创建完成后的目录结构如下： + + ``` + .opencode/ + └── agents/ + └── image-analyzer.md + ``` + +2. **开始使用** + + 1. 在项目目录下运行`opencode`启动 OpenCode，并切换到`glm-5`模型。 + + 2. 下载[aliyun.png](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260225/hxwnny/aliyun.png)到项目目录下，通过`@`唤起`image-analyzer`并提问：`@image-analyzer，描述一下 aliyun.png banner位置是什么信息。`可收到如下回复： + + ``` + Banner位置包含： + 左侧： + - 汉堡菜单图标 + - 阿里云橙色logo和"阿里云"文字 + - 导航菜单：大模型、产品、解决方案、权益、定价、云市场、伙伴、服务、了解阿里云 + 右侧： + - 搜索框(显示"大模型") + - 图标：蓝色圆圈、地球、耳机 + - 链接：文档、备案、控制台 + Build · glm-5 · 37.0s + ``` + + +## **常见问题** + +### **OpenCode + 视觉理解模型为什么无法理解图片？** + +**原因**：OpenCode 默认不启用模型的视觉能力，需要在配置文件中显式声明 `modalities` 参数。 + +**解决方案**：在 OpenCode 配置文件的模型定义中添加 `modalities` 字段，将 `input` 设为 `["text", "image"]`，如下所示： + +> 将sk-sp-xxx替换为Token Plan API Key。 + +``` +{ + "$schema": "https://opencode.ai/config.json", + "provider": { + "bailian-token-plan": { + "npm": "@ai-sdk/anthropic", + "name": "Model Studio Token Plan", + "options": { + "baseURL": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "apiKey": "sk-sp-xxx" + }, + "models": { + "qwen3.6-plus": { + "name": "Qwen3.6 Plus", + "modalities": { + "input": [ + "text", + "image" + ], + "output": [ + "text" + ] + }, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 1024 + } + } + }, + "qwen3.5-plus": { + "name": "Qwen3.5 Plus", + "modalities": { + "input": [ + "text", + "image" + ], + "output": [ + "text" + ] + }, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 1024 + } + } + }, + "kimi-k2.5": { + "name": "Kimi K2.5", + "modalities": { + "input": [ + "text", + "image" + ], + "output": [ + "text" + ] + }, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 1024 + } + } + } + } + } + } +} +``` + +### **OpenClaw + 视觉理解模型为什么无法理解图片？** + +**原因**：OpenClaw 需要通过配置文件中的 input 字段来判断模型是否支持视觉能力。 + +**解决方案**： + +1. 在`~/.openclaw/openclaw.json`配置文件中，确保模型定义包含`"input": ["text", "image"]`字段。 + + ``` + { + "models": { + "mode": "merge", + "providers": { + "bailian": { + "baseUrl": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "apiKey": "YOUR_API_KEY", + "api": "openai-completions", + "models": [ + { + "id": "qwen3.6-plus", + "name": "qwen3.6-plus", + "reasoning": false, + "input": ["text", "image"], + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "contextWindow": 1000000, + "maxTokens": 65536 + }, + { + "id": "qwen3.5-plus", + "name": "qwen3.5-plus", + "reasoning": false, + "input": ["text", "image"], + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "contextWindow": 1000000, + "maxTokens": 65536 + }, + { + "id": "kimi-k2.5", + "name": "kimi-k2.5", + "reasoning": false, + "input": ["text", "image"], + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "contextWindow": 262144, + "maxTokens": 32768 + } + ] + } + } + }, + "agents": { + "defaults": { + "model": { + "primary": "bailian/qwen3.6-plus" + }, + "models": { + "bailian/qwen3.6-plus": {}, + "bailian/qwen3.5-plus": {}, + "bailian/kimi-k2.5": {} + } + } + }, + "gateway": { + "mode": "local" + } + } + ``` + +2. 修改配置后，需要清除 OpenClaw 的模型缓存并重启，否则旧的配置仍会生效。 + + ``` + rm ~/.openclaw/agents/main/agent/models.json + openclaw gateway restart + ``` diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-harness-tool.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-harness-tool.md new file mode 100644 index 00000000..88c71bd8 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-harness-tool.md @@ -0,0 +1,79 @@ +# 接入 Harness 工具 + +Token Plan 支持的部分 Qwen 模型内置 Harness 工具，可为 AI 编程工具扩展联网搜索、代码解释器、网页抓取等能力。 + +**说明** + +适用于 Token Plan，不适用于 Coding Plan。 + +## **工具概览** + +**工具** + +**说明** + +联网搜索 + +检索互联网信息，结合搜索结果生成回答 + +代码解释器 + +在沙箱环境中编写与运行 Python 代码，用于数学计算、数据分析等场景 + +网页抓取 + +访问指定 URL 并提取内容，为大模型提供所需信息 + +以图搜图 + +根据输入图片从互联网搜索视觉相似的图片，适用于以图找同款、视觉内容溯源等场景 + +文搜图 + +根据文本描述从互联网搜索相关图片，适用于可视化问答、配图推荐等场景 + +## **支持的模型和工具** + +### **个人版** + +**模型** + +**支持的工具** + +qwen3.8-max-preview + +联网搜索、代码解释器、网页抓取、以图搜图、文搜图 + +qwen3.7-max + +联网搜索、代码解释器、网页抓取 + +qwen3.7-plus + +联网搜索、代码解释器、网页抓取、以图搜图、文搜图 + +### **团队版** + +**模型** + +**支持的工具** + +qwen3.8-max-preview + +联网搜索、代码解释器、网页抓取、以图搜图、文搜图 + +qwen3.7-max + +联网搜索、代码解释器、网页抓取 + +qwen3.7-plus + +联网搜索、代码解释器、网页抓取、以图搜图、文搜图 + +## **费用说明** + +Harness 工具按成功调用次数计费，费用从套餐 Credits 中抵扣。 + +## **使用方式** + +将 AI 编程工具的模型切换为上述支持 Harness 的 Qwen 模型，在对话中直接提问即可。模型会根据问题自动调用相应的内置工具，无需额外配置。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-multimodal-gen.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-multimodal-gen.md index 4daf85ea..55d418ac 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-multimodal-gen.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-multimodal-gen.md @@ -1,14 +1,6 @@ # 接入多模态生成模型 -图像生成模型需通过工具的扩展机制（Skill、Slash Command 或 Agent）接入。 - -## **前提：获取套餐专属凭证** - -在控制台「我的订阅」打开 Token Plan 套餐详情页，接入信息卡片展示套餐专属 API Key（以 `sk-sp-` 为前缀，掩码显示），支持生成、重置与复制 API Key。 - -**说明** - -套餐详情页「可使用模型」以文本、编程模型为主；图像生成模型不在该列表展示，需通过 `multimodal-generation` API 调用。 +Token Plan 中的图像生成、视频生成模型需通过工具的扩展机制（Skill、Slash Command 或 Agent）接入。 ## **示例：在 Claude Code 中接入图像生成模型** @@ -27,7 +19,7 @@ ## 步骤 -1. 从用户需求中提取 prompt（图片描述）、model、size（默认 1024*1024）。若用户明确指定了模型（如“模型=wan2.7-image”或“用 wan2.7-image 画”），必须严格使用用户指定的模型名，不要回退到默认模型；仅当用户未指定模型时才使用默认 qwen-image-2.0。常用图像生成模型有 qwen-image-2.0、qwen-image-2.0-pro、wan2.7-image、wan2.7-image-pro、z-image-turbo 等，完整列表以百炼模型列表为准。 +1. 从用户需求中提取 prompt（图片描述）、model、size（默认 1024*1024）。若用户明确指定了模型（如“模型=wan2.7-image”或“用 wan2.7-image 画”），必须严格使用用户指定的模型名，不要回退到默认模型；仅当用户未指定模型时才使用默认 qwen-image-2.0。常用图像生成模型有 qwen-image-2.0、qwen-image-2.0-pro、wan2.7-image、wan2.7-image-pro 等，完整列表以百炼模型列表为准。 2. 调用 API 生成图片（使用 Bash 工具执行 curl）： @@ -55,9 +47,73 @@ curl -s -X POST "https://token-plan.cn-beijing.maas.aliyuncs.com/api/v1/services 在 Claude Code 中输入 `/text-to-image 画一只猫`。如需使用默认模型以外的图像生成模型，在指令中写明模型名即可，例如 `/text-to-image 用 wan2.7-image 画一只猫`。 +## **示例：在 Claude Code 中接入视频生成模型** + +以 Claude Code 为例，通过 Slash Command 接入视频生成模型。视频生成为异步接口，流程为"提交任务 → 轮询状态 → 下载视频"。 + +### **步骤一：创建 Slash Command** + +将套餐专属 API Key（以 `sk-sp-` 为前缀）配置为环境变量 `$ANTHROPIC_AUTH_TOKEN`，供后续 curl 鉴权使用。 + +在项目根目录创建 `.claude/commands/text-to-video.md`，写入以下内容： + +``` +调用 Token Plan 文生视频 API，根据描述生成视频并自动下载到本地。 + +用户需求：$ARGUMENTS + +## 步骤 + +1. 从用户需求中提取 prompt（视频描述）、model（默认 happyhorse-1.1-t2v）、resolution（默认 720P）、ratio（默认 16:9）、duration（默认 5 秒）。若用户明确指定了模型（如"模型=happyhorse-1.0-t2v"），必须严格使用用户指定的模型名。 + +2. 使用 Bash 工具执行以下脚本，一次性完成提交任务、等待完成、下载视频： + +```bash +#!/bin/bash +set -e + +TASK_RESPONSE=$(curl -s -X POST "https://token-plan.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis" \ + -H "X-DashScope-Async: enable" \ + -H "Authorization: Bearer $ANTHROPIC_AUTH_TOKEN" \ + -H "Content-Type: application/json" \ + -d '{ + "model": "", + "input": {"prompt": ""}, + "parameters": {"resolution": "", "ratio": "", "duration": } + }') + +TASK_ID=$(echo "$TASK_RESPONSE" | grep -o '"task_id":"[^"]*"' | head -1 | cut -d'"' -f4) +if [ -z "$TASK_ID" ]; then echo "提交失败: $TASK_RESPONSE"; exit 1; fi +echo "任务已提交，ID: $TASK_ID，等待生成..." + +while true; do + sleep 15 + STATUS_RESPONSE=$(curl -s "https://token-plan.cn-beijing.maas.aliyuncs.com/api/v1/tasks/$TASK_ID" \ + -H "Authorization: Bearer $ANTHROPIC_AUTH_TOKEN") + STATUS=$(echo "$STATUS_RESPONSE" | grep -o '"task_status":"[^"]*"' | cut -d'"' -f4) + if [ "$STATUS" = "SUCCEEDED" ]; then + VIDEO_URL=$(echo "$STATUS_RESPONSE" | grep -o '"video_url":"[^"]*"' | cut -d'"' -f4) + OUTPUT="generated_$(date +%Y%m%d_%H%M%S).mp4" + curl -s -o "$OUTPUT" "$VIDEO_URL" + echo "视频已下载: $(pwd)/$OUTPUT" + exit 0 + elif [ "$STATUS" = "FAILED" ]; then + echo "生成失败: $STATUS_RESPONSE"; exit 1 + fi + echo "生成中..." +done +``` + +3. 向用户展示生成的视频文件路径。 +``` + +### **步骤二：生成视频** + +在 Claude Code 中输入 `/text-to-video 一只白色的猫在阳台上晒太阳`。如需使用其他视频生成模型，在指令中写明模型名即可，例如 `/text-to-video 用 happyhorse-1.1-r2v 生成一只猫跳跃的视频`。 + ## **其他工具** -控制台套餐详情页「快速接入 AI 编程工具」入口提供 Qwen Code、Qoder、OpenClaw、Claude Code、OpenCode 等工具的接入文档。不同工具的扩展机制和配置文件路径如下表所示，该表为支持扩展机制的主流 AI 编程工具示例。将上述 Claude Code 示例中的配置内容保存到对应路径即可。 +不同工具的扩展机制和配置文件路径如下表所示，该表为支持扩展机制的主流 AI 编程工具示例。将上述 Claude Code 示例中的配置内容保存到对应路径即可。 工具 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-tool.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-tool.md deleted file mode 100644 index aaf9db77..00000000 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-tool.md +++ /dev/null @@ -1,253 +0,0 @@ -# 工具调用 - -Token Plan 团队版支持通过模型内置工具和 MCP 服务两种方式为 AI 编程工具扩展能力，如联网搜索、代码解释器、网页抓取等。 - -## **工具概览** - -Token Plan 团队版提供两种方式接入工具： - -- **模型内置工具**：qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash 模型的 Responses API 内置了联网搜索、代码解释器、网页抓取、以图搜图、文搜图五种工具。启用后，模型会在需要时自动调用相应工具。 - -- **MCP 服务**：其他模型（如 deepseek-v3.2、glm-5 等）可通过百炼 MCP 广场的 MCP 服务获取工具能力。本文以联网搜索 MCP 为例说明接入方式，其他 MCP 服务的接入方式类似。 - - -## **费用说明** - -### **模型内置工具** - -qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash 模型内置工具的费用可通过 Token Plan 团队版抵扣，内置工具不额外收费，产生的 token 消耗统一从套餐 Credits 中抵扣。具体价格以[控制台模型详情页](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/detail/qwen3.6-plus)为准。 - -### **MCP 服务** - -百炼 MCP 广场提供联网搜索、代码解释器、网页抓取等 MCP 服务。联网搜索 MCP 全部用户前 2000 次调用免费，免费额度用尽后按 29 元/千次计费；其他 MCP 服务部分**限时免费**，每月提供一定免费额度。具体价格以[MCP 广场](https://bailian.console.aliyun.com/cn-beijing/?tab=app#/mcp-market)各服务详情页为准。 - -## **使用方式** - -### **使用 qwen3.7-max / qwen3.7-plus / qwen3.6-plus / qwen3.6-flash 模型内置工具** - -将 AI 工具的模型设置为 `qwen3.7-max`、`qwen3.7-plus`、`qwen3.6-plus` 或 `qwen3.6-flash`，在对话中直接提问即可。模型会根据问题自动调用相应的内置工具： - -**工具** - -**说明** - -联网搜索 - -检索互联网信息，结合搜索结果生成回答 - -代码解释器 - -调用模型时启用内置的 Python 代码解释器，可使模型在沙箱环境里编写与运行 Python 代码，以解决数学计算、数据分析等复杂问题。 - -网页抓取 - -网页抓取工具可以访问指定 URL 并提取内容，为大模型提供所需信息。 - -以图搜图 - -图搜图工具使模型能够根据输入图片从互联网搜索视觉相似的图片，并基于搜索结果进行分析和推理，适用于以图找同款、视觉内容溯源等场景。 - -文搜图 - -文搜图工具使模型能够根据文本描述从互联网搜索相关图片，并基于图片内容进行描述和推理，适用于可视化问答、配图推荐等场景。 - -### **通过 MCP 服务接入工具** - -其他模型可通过百炼 MCP 广场的 MCP 服务获取工具能力。以下以联网搜索 MCP 为例说明接入方式。 - -#### **前提条件** - -已获取[百炼 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。此处的 API Key 为百炼通用 API Key（格式为 sk-xxx），用于调用 MCP 服务，与 Token Plan 团队版专属 API Key（格式为 sk-sp-xxx）不同。 - -#### **开通 MCP 服务** - -1. 进入百炼的[MCP 广场](https://bailian.console.aliyun.com/cn-beijing/?tab=app#/mcp-market)，找到需要的 MCP 服务（如联网搜索）。 - -2. 点击**立即开通**，确认开通。 - -3. 开通成功后，获取以下配置信息： - - - **Streamable HTTP Endpoint**：MCP 服务的连接地址。 - - - **API Key**：即百炼 API Key，是 MCP 服务的鉴权密钥。 - - -#### **接入工具** - -将 MCP 服务添加到 AI 编程工具中。以下以联网搜索 MCP 为例，示例中的 `YOUR_API_KEY` 需替换为百炼 API Key。接入其他 MCP 服务时，将 Endpoint 地址替换为对应服务的地址即可。 - -## OpenClaw - -1. 在终端执行如下命令安装 MCPorter。 - - ``` - npm install -g mcporter - ``` - -2. 在终端执行如下命令启用 MCPorter。 - - ``` - openclaw config set skills.entries.mcporter.enabled true - ``` - -3. 在 `~/.openclaw/workspace` 目录下，执行如下命令添加联网搜索 MCP。 - - ``` - mcporter config add WebSearch https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp --transport http --header "Authorization=Bearer YOUR_API_KEY" - ``` - -4. 执行如下命令确认 MCP 已安装。 - - ``` - mcporter list - ``` - -5. 执行如下命令使配置生效。 - - ``` - openclaw gateway restart - ``` - -6. 发送提问 `用 mcporter 搜索阿里云的新闻` 即可看到搜索结果。 - - -## OpenCode - -1. 在配置文件 `~/.config/opencode/opencode.json` 中写入 MCP 配置信息。 - - ``` - { - "mcp": { - "WebSearch": { - "type": "remote", - "httpUrl": "https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp", - "headers": { - "Authorization": "Bearer YOUR_API_KEY" - } - } - } - } - ``` - - 若 `opencode.json` 中已有其他配置（如 provider），将 mcp 字段合并到现有配置中即可。 - -2. 在终端执行以下命令进入 OpenCode。 - - ``` - opencode - ``` - -3. 在对话框执行 `/mcps` 确认 `websearch` 状态为 Enabled。确认后按 Esc 退出。 - -4. 发送提问 `用 websearch MCP 搜索阿里云的新闻` 即可看到搜索结果。 - - > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 - - -## Claude Code - -1. 在终端执行以下命令添加联网搜索 MCP 服务。 - - ``` - claude mcp add WebSearch https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp -t http -H "Authorization: Bearer YOUR_API_KEY" - ``` - - 终端返回 `Added SSE MCP server xx` 即表示添加成功。 - -2. 执行以下命令进入 Claude Code。 - - ``` - claude - ``` - -3. 在对话框执行 `/mcp` 命令，确认 `websearch` 的状态为 connected。 - -4. 按 Esc 退出 MCP 列表后，发送提问 `用 websearch MCP 搜索阿里云的新闻` 即可看到搜索结果。 - - > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 - - -## Qwen Code - -1. 在终端执行以下命令添加联网搜索 MCP。 - - ``` - qwen mcp add WebSearch \ - -t http \ - "https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp" \ - -H "Authorization: Bearer YOUR_API_KEY" - ``` - -2. 在终端执行以下命令进入 Qwen Code。 - - ``` - qwen - ``` - -3. 在对话框执行 `/mcp` 命令确认 MCP 连接状态。 - -4. 发送提问 `用 websearch MCP 搜索阿里云的新闻` 即可看到搜索结果。 - - > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 - - -## Kilo CLI - -1. 在配置文件 `~/.config/kilo/opencode.json` 中写入 MCP 配置信息。 - - ``` - { - "mcp": { - "websearch": { - "type": "remote", - "url": "https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp", - "headers": { - "Authorization": "Bearer YOUR_API_KEY" - } - } - } - } - ``` - - 若 `opencode.json` 中已有其他配置（如 provider），将 mcp 字段合并到现有配置中即可。 - -2. 在终端执行以下命令查看 MCP 状态。Connected 即表示连接成功。 - - ``` - kilocode mcp list - ``` - -3. 在终端执行以下命令进入 Kilo CLI。 - - ``` - kilo - ``` - -4. 发送提问 `用 websearch MCP 搜索阿里云的新闻` 即可看到搜索结果。 - - > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 - - -## Kilo Code IDE 插件 - -1. 打开 Kilo Code IDE 插件，配置联网搜索 MCP 信息。 - - ``` - { - "mcpServers": { - "websearch": { - "type": "streamable-http", - "url": "https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp", - "headers": { - "Authorization": "Bearer YOUR_API_KEY" - } - } - } - } - ``` - - 当联网搜索的 MCP 状态显示为绿色时，表示添加成功。 - -2. 返回对话界面，发送提问 `用 websearch MCP 搜索阿里云的新闻` 即可看到搜索结果。 - - > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/web-search-for-coding-plan.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/web-search-mcp.md similarity index 51% rename from skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/web-search-for-coding-plan.md rename to skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/web-search-mcp.md index 3208f230..5fd78221 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/coding-plan-guide/web-search-for-coding-plan.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-best-practice/web-search-mcp.md @@ -1,6 +1,6 @@ # 联网搜索 -在 Coding Plan 支持的编程工具中添加联网搜索工具，使模型能够检索实时信息。 +在 Token Plan 支持的编程工具中添加联网搜索工具，使模型能够检索实时信息。 ## 适用范围 @@ -8,11 +8,11 @@ ## 前提条件 -1. 已订阅 [Coding Plan](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan)，详情请参见[快速开始](https://help.aliyun.com/zh/model-studio/coding-plan-quickstart)。 +1. 已订阅 [Token Plan](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription)，详情请参见[快速开始](https://help.aliyun.com/zh/model-studio/coding-plan-quickstart)。 -2. 已在 Coding Plan 工具（如 Claude Code、Qwen Code）中完成接入配置，且能正常对话，详情请参见[接入客户端/开发工具](https://help.aliyun.com/zh/model-studio/use-chat-client-or-development-tool/)。 +2. 已在 Token Plan 支持的工具（如 Claude Code、Qwen Code）中完成接入配置，且能正常对话，详情请参见[接入客户端/开发工具](https://help.aliyun.com/zh/model-studio/use-chat-client-or-development-tool/)。 -3. 已获取[百炼 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。此处的 API Key 为百炼通用 API Key（格式为 sk-xxx），用于调用 MCP 服务，与 Coding Plan 专属 API Key（格式为 sk-sp-xxx）不同。 +3. 已获取[百炼 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。此处的 API Key 为百炼通用 API Key（格式为 sk-xxx），用于调用 MCP 服务，与 Token Plan 专属 API Key（格式为 sk-sp-xxx）不同。 ## 开通或升级联网搜索 MCP @@ -27,6 +27,8 @@ - 联网搜索 MCP 全部用户前 2000 次调用免费，免费额度用尽后按 29 元/千次计费。如果使用第三方 MCP 服务，该 MCP 服务可能收费，以 MCP 服务的介绍信息为准。 + - 部分MCP服务支持**个人FC资源部署**，按实际调用时长和次数计费，适用于需要专属资源、指定资源地域等场景。 + 3. 开通成功后，可以获取以下配置信息： 1. **Streamable HTTP Endpoint**：MCP 服务的连接地址。联网搜索 MCP 的连接地址为`https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp`。 @@ -38,10 +40,12 @@ 1. 进入百炼的[MCP广场](https://bailian.console.aliyun.com/cn-beijing/?tab=app#/mcp-market)，找到**联网搜索** MCP 服务。 -2. 单击右侧**取消开通**，再单击**立即开通**。 +2. 单击右侧**取消开通**，再单击**立即开通** > **确认开通**。 - 联网搜索 MCP 全部用户前 2000 次调用免费，免费额度用尽后按 29 元/千次计费。如果使用第三方 MCP 服务，该 MCP 服务可能收费，以 MCP 服务的介绍信息为准。 + - 部分MCP服务支持**个人FC资源部署**，按实际调用时长和次数计费，适用于需要专属资源、指定资源地域等场景。 + 3. 重新开通成功后，即完成协议升级，可以获取以下配置信息： 1. **Streamable HTTP Endpoint**：MCP 服务的连接地址。联网搜索 MCP 的连接地址为`https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp`。 @@ -78,8 +82,11 @@ ``` mcporter list ``` - - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9753812771/p1055340.png) + ``` + mcporter 0.7.3 — Listing 1 server(s) (per-server timeout: 30s) + - websearch (1 tool, 2.4s) + [ok] Listed 1 server (1 healthy). + ``` 5. 在终端执行如下命令使配置生效。 @@ -89,7 +96,7 @@ 6. 发送提问`用 mcporter 搜索阿里云的新闻`即可看到搜索结果。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9753812771/p1055129.png) + 系统返回包含 4 条阿里云相关新闻的卡片，每条新闻显示标题、来源及发布时间，底部提示是否需要抓取某篇新闻的详细内容。 ## OpenCode @@ -120,13 +127,27 @@ 3. 在对话框执行`/mcps`确认`websearch`状态是否为 Enabled。确认后按 Esc 退出。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/4753812771/p1054875.png) - 4. 发送提问`用websearch MCP搜索阿里云的新闻`即可看到搜索结果。 > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9228202771/p1054876.png) + ``` + 已为您搜索到最新的阿里云相关新闻： + 主要新闻摘要 + 1. 阿里云2026年战略目标 + 阿里云目标在2026年拿下中国AI云市场增量的80%，重点发力MaaS（模型即服务）和GPU算力服务。 + 目前阿里云上调用大模型API的客户中，有七成同时使用其GPU算力服务。 + 2. 阿里云服务器2026年优惠活动 + - 最便宜服务器：38元/年（2核2G、200M峰值带宽） + - 99元/年 ECS服务器（新老用户同享，续费同价） + - 199元/年企业专享服务器（2核4G、5M带宽） + 3. AI自动化工具生态 + 阿里云推出OpenClaw（原Clawdbot）AI + 自动化代理工具部署方案，支持一键部署到轻量应用服务器，可与阿里云百炼大模型无缝联动。 + 4. 企业智能客服 + 瓴羊Quick Service智能客服平台基于通义千问大模型，提供大模型驱动的语义理解、多轮对话、 + 知识库智能生成等功能。 + ``` ## Claude Code @@ -155,17 +176,54 @@ 3. 在对话框执行`/mcp`命令，确认`websearch`的状态为 connected。首次添加可能需要等待状态从 connecting 变成 connected。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9276802771/p1054987.png) - 若连接状态显示 failed，请选中该 MCP 并选择 Reconnect 重连。若重试 1-2 次仍失败，请核实配置。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0613342771/p1055684.png) + ``` + /mcp + WebSearch MCP Server + Status: [ok] connected + Auth: [ok] authenticated + URL: https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp + xxx + Capabilities: tools + Tools: 1 tools + 1. View tools + 2. Re-authenticate + 3. Clear authentication + > 4. Reconnect + 5. Disable + ``` 4. 按 Esc 退出 MCP 列表后，发送提问`用websearch MCP搜索阿里云的新闻`即可看到搜索结果。 > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9276802771/p1054990.png) + ``` + > 用websearch MCP搜索阿里云的新闻 + * websearch – bailian_web_search (MCP)(query: "阿里云 新闻", count: 10) + └ { + "status": 0, + "pages": [ + … +51 lines (ctrl+o to expand) + ] + } + * 搜索结果已返回，以下是阿里云的最新新闻摘要： + 阿里云近期新闻 + 1. 编程模型订阅服务推出（2026 年 2 月 25 日） + 阿里云百炼推出 Coding Plan 编程模型订阅服务，包含四款开源模型 API： + - Qwen3.5、GLM-5、MiniMax M2.5、Kimi K2.5 + - 用户可在 Qwen Code、Claude Code、Cline 等 AI 工具上无缝切换使用 + - 降低开发者模型选型与接入成本 + 2. 千问 3.5 新模型开源（2026 年 2 月 25 日） + 阿里发布三款中型千问 3.5 新模型： + - Qwen3.5-35B-A3B、Qwen3.5-122B-A10B、Qwen3.5-27B + - 基于 Qwen3.5-35B-A3B 的托管模型 Qwen3.5-Flash 已上线阿里云百炼 + - 每百万 Token 输入低至 0.2 元 + 3. 市场表现亮眼 + - 千问（Qwen）大模型调用量占比跃升至 32.1%，排名中国企业级大模型第一 + - 2026 财年第二季度收入 398.24 亿元，同比增长 34% + - AI 相关产品收入连续第九个季度实现三位数增长 + ``` ## Qwen Code @@ -197,13 +255,80 @@ 3. 在`Qwen Code`对话框执行`/mcp`命令以确认 MCP 连接状态。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/4753812771/p1054878.png) + ``` + > /mcp + Configured MCP servers: + * WebSearch – Ready (1 tool) + Tools: + – mcp__WebSearch__bailian_web_search + [tip] Tips: + – Use /mcp desc to show server and tool descriptions + – Use /mcp schema to show tool parameter schemas + – Use /mcp nodesc to hide descriptions + – Use /mcp auth to authenticate with OAuth-enabled servers + – Press Ctrl+T to toggle tool descriptions on/off + ``` 4. 发送提问`用 websearch MCP 搜索阿里云的新闻`即可看到搜索结果。 > 为了避免和其他工具混淆，建议提问时明确提及`websearch MCP`。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9228202771/p1054879.png) + ``` + * 搜索结果已返回，我需要整理并总结其中与阿里云相关的主要新闻信息，以清晰的方式呈现给用户。 + * 已为您搜索到阿里云相关新闻，以下是主要内容： + 阿里云最新动态 + 1. 阿里云千问大模型表现亮眼 + - 2025 年下半年，阿里云千问（Qwen）在中国企业级大模型市场占比跃升至32.1%，几乎翻倍（上半年为 + 17.7%），位居第一 + - 字节豆包占比 21.3% 排名第二，DeepSeek 占比 18.4% 排名第三 + 2. AI 战略投入 + - 阿里豪掷3800 亿元押注 AI 基础设施 + - 2026 年，阿里云将 AI 相关收入目标增速上调至200% + - 2026 年 2 月推出30 亿元"春节请客计划"，联合淘宝、飞猪、盒马等以免单形式发放福利 + 3. Coding Plan 服务更新 + - 阿里云 Coding Plan 订阅服务新增支持 Qwen 3.5-Plus、GLM-4.7、Kimi-K2.5 等编程模型 + - 新用户首月首购享两折优惠 + 4. 技术进展 + - 除夕夜开源的千问 Qwen3.5-Plus 采用混合注意力机制和稀疏 MoE 架构 + - 总参数 3970 亿，激活 170 亿，推理速度达主流模型的 2 倍 + 需要我为您搜索更具体的阿里云新闻主题吗？ + > 输入您的消息或 @ 文件路径 + 按 ? 查看快捷键 1.8% context used + ``` + + +## Cline + +1. 点击MCP Servers，进入Configure页面，点击Configure MCP Servers。 + + ![2026-02-26\_13-58-09](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9276802771/p1054977.jpg) + +2. 在配置文件`cline_mcp_settings.json`中写入 MCP 配置信息，并将`YOUR_API_KEY`替换为上一步骤获取的 API Key。配置完成后，保存文件。 + + ``` + { + "mcpServers": { + "websearch": { + "type": "http", + "url": "https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp", + "headers": { + "Authorization": "Bearer YOUR_API_KEY" + }, + "disabled": false + } + } + } + ``` + +3. 配置完成后，在Configure页面下可以查看MCP的相关信息。 + + ![2026-02-26\_14-05-44](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9276802771/p1054984.jpg) + +4. 新开对话，发送提问`用websearch MCP搜索阿里云的新闻`即可看到搜索结果。 + + > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 + + ![2026-02-26\_14-10-05](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9276802771/p1054989.jpg) ## Kilo CLI @@ -231,8 +356,14 @@ ``` kilocode mcp list ``` - - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/0248612771/p1055115.png) + ``` + MCP Servers + │ + * [ok] websearch connected + │ https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp + │ + └ 1 server(s) + ``` 3. 在终端中执行以下命令进入 Kilo CLI。 @@ -244,14 +375,31 @@ > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/9228202771/p1054891.png) + ``` + 以下是阿里云的最新新闻要点： + 阿里云千问大模型市场份额跃升第一 + - 市场份额翻倍：根据沙利文2026年2月报告，阿里云千问（Qwen）占比跃升至*32.1%*，相较2025年上半年的17.7%几乎翻倍，超越字节豆包（21.3%）和DeepSeek(18.4%)，成为企业级大模型市场第一 + - 日均调用量激增：2025年下半年中国企业级大模型日均调用量达37.0万亿tokens，较上半年增长263% + AI基础设施投资战略 + - 3800亿元投资：阿里宣布豪掷3800亿元押注AI基础设施 + - 收入目标：2026年将AI相关收入目标增速上调至200% + - 市场目标：阿里云目标拿下中国AI云市场增量的80% + 千问模型技术进展 + - 开源Qwen3.5-Plus：除夕夜开源，采用混合注意力机制和稀疏MoE架构 + - 高效能设计：总参数3970亿，仅激活170亿，推理速度达到主流模型的2倍 + - 春节免单计划：千问启动30亿元"春节请客计划"，联合淘宝、飞猪、盒马等生态业务发放福利 + 其他动态 + - Coding Plan上新：支持千问3.5、GLM-4.7、Kimi-K2.5等编程模型，新用户首月两折优惠 + - 云服务器优惠：轻量应用服务器38元/年起，200M带宽不限流量 + Code · kimi-k2.5 · 49.8s + ``` ## Kilo Code IDE 插件 1. 打开Kilo Code IDE插件配置联网搜索 MCP 信息，并将`YOUR_API_KEY`替换为上一步骤获取的 API Key。配置完成后，保存文件。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3100902771/p1055026.png) + 依次单击左上角齿轮图标，在侧边栏选择 **Agent Behaviour**，单击 **MCP Servers** 页签，然后单击底部 **Edit Global MCP** 按钮，打开 `mcp_settings.json` 配置文件，将以下 JSON 内容粘贴并保存。 ``` { @@ -269,14 +417,10 @@ 当联网搜索的MCP状态显示为绿色时，表示添加成功。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3100902771/p1055021.png) - 2. 返回对话界面，发送提问`用websearch MCP搜索阿里云的新闻`即可看到搜索结果。 > 为了避免和其他工具混淆，建议提问时明确提及 websearch MCP。 - ![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3100902771/p1055023.png) - ## 常见问题 @@ -292,7 +436,7 @@ - 如果使用的 URL 为 `https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/sse`，说明开通的是旧版 SSE 协议，请将协议[升级至Streamable HTTP](#da7cd47fa8pcd)。 -3. **API Key 错误**：请确认使用了有效的百炼通用 API Key（格式为 sk-xxx，非 Coding Plan 专属 API Key），并已在命令中正确替换 YOUR\_API\_KEY。 +3. **API Key 错误**：请确认使用了有效的百炼通用 API Key（格式为 sk-xxx，非 Token Plan 专属 API Key），并已在命令中正确替换 YOUR\_API\_KEY。 4. **免费额度用尽**：联网搜索 MCP 全部用户前 2000 次调用免费，免费额度用尽后按 29 元/千次计费，请确认账户余额充足。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-faq.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-faq.md deleted file mode 100644 index 76b285c1..00000000 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-faq.md +++ /dev/null @@ -1,292 +0,0 @@ -# 常见问题 - -Token Plan 团队版常见问题汇总，涵盖购买、使用、计量和性能相关的问题解答。 - -## **Token Plan 团队版和 Coding Plan 有什么区别？** - -**Token Plan 团队版** - -**Coding Plan** - -适用场景 - -一人公司/团队/企业日常办公 - -个人开发场景 - -支持的模型 - -文本生成、图像生成模型 - -文本生成模型 - -计费方式 - -按 Token 消耗抵扣 Credits - -按模型调用次数 - -使用频次 - -无每 5 小时/每周限额 - -每 5 小时/每周限额 - -API Key 和 Base URL - -在[管理后台](https://tokenplan-enterprise.bailian.aliyunportal.com)生成专属 API Key，Base URL 详见[快速开始](https://help.aliyun.com/zh/model-studio/token-plan-quickstart) - -在[Coding Plan 页面](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/coding-plan)获取专属 API Key 和专属 Base URL - -高峰期性能 - -多租户隔离 - -高峰期间可能排队 - -数据安全 - -承诺不使用数据训练模型 - -用户数据授权 - -## **接入与调用** - -### **如何在编程工具中使用图像生成模型？** - -图像生成模型使用独立的接口，无法通过文本模型的 Base URL 直接调用。需要通过工具的 Skill 或扩展机制接入，具体配置方法请参见[接入多模态生成模型](https://help.aliyun.com/zh/model-studio/token-plan-multimodal-gen)。 - -### **常见报错及解决方案** - -**报错信息** - -**可能原因** - -**解决方案** - -**401 InvalidApiKey: No API-key provided.** - -请求头中未携带 API Key（`Authorization: Bearer` 或 `x-api-key` 均未传）。 - -在管理后台生成 API Key，并在工具中完成配置。 - -**401 InvalidApiKey: Invalid API-key provided.** - -1. 误用了百炼通用 API Key（sk-xxx 格式）或 Coding Plan 的 API Key - -2. Token Plan 团队版订阅过期 - -3. API Key 复制不完整或包含空格 - - -1. 确认使用的是 Token Plan 专属 API Key，确保完整且无空格。 - -2. 确认订阅是否过期。 - -3. 如仍报错，重置 API Key，重置后使用新 Key 配置。 - - -**404 model 'xxx' not found or not supported** - -**400 Model not exist.** - -1. 模型名称拼写错误或大小写错误 - -2. 模型 ID 不在套餐支持列表中 - - -1. 确认模型名称区分大小写，与套餐支持的模型 ID 一致。 - -2. 检查所选套餐是否包含该模型。 - - -**401 invalid access token or token expired** - -误用了 Coding Plan 或其他套餐的 Base URL - -Anthropic 兼容端点：`https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` - -OpenAI 兼容端点：`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` - -**401 Incorrect API key provided** - -误用了百炼通用 Base URL（dashscope.aliyuncs.com） - -Anthropic 兼容端点：`https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` - -OpenAI 兼容端点：`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` - -**400 InvalidParameter: Range of input length should be \[1, xxx\]** - -输入内容（含对话历史、代码上下文等）超出模型的最大上下文长度 - -新建会话清空历史，或使用工具自带的上下文压缩命令（如 Claude Code 的 `/compact`、Qwen Code 的 `/clear`）。也可切换上下文窗口更大的模型。 - -**400 InvalidParameter: url error, please check url！** - -Base URL 路径与协议不匹配。例如把 OpenAI 兼容路径配在 Anthropic 端点上，或反之。 - -按工具实际使用的协议选择对应的端点： - -- Anthropic 兼容协议（Claude Code 等）：以 `/apps/anthropic` 结尾。 - -- OpenAI 兼容协议（Cursor、Qwen Code 等）：以 `/compatible-mode/v1` 结尾。 - - -**400 InvalidParameter: Range of max\_tokens should be \[1, xxxx\]** - -请求中的 `max_tokens`（或工具配置中的最大输出长度）超出当前模型支持的最大输出 Token 数。 - -将 `max_tokens` 调整为不超过报错信息中提示的上限值。 - -**400 invalid\_parameter\_error: The thinking\_budget parameter must be a positive integer and not greater than xxxxx** - -工具配置中的思维链长度（如 `thinking_budget`、`budgetTokens`）超过当前模型支持的上限。各模型上限不同，以报错中的数值为准。 - -将思维链长度调整为不超过报错提示的上限值，或在不支持思考模式的模型上移除该配置项。 - -**400 data\_inspection\_failed: Input text data may contain inappropriate content.** - -输入或输出命中平台内容安全策略。 - -修改输入内容后重新提交。如多次触发，调整提示词避免敏感话题。 - -**429 API-Key Requests rate limit exceeded, please try again later.** - -短时间内请求过于密集，触发模型调用限流。 - -等待一分钟后重试；如频繁触发请降低请求频率，并确认 API Key 未被他人共享使用。 - -**429 Throttling.AllocationQuota: Allocated quota exceeded, please increase your quota limit.** - -**insufficient\_quota: You exceeded your current quota, please check your plan and billing details.** - -该报错可能由以下两种原因触发： - -**套餐额度已用尽**：坐席额度和共享用量包均已耗尽。 - -**触发模型调用限流**：即使套餐额度充足，每秒或每分钟消耗的 Token 数（TPS/TPM）超过模型限流阈值也会触发。限流按主账号维度计算，账号下所有 RAM 子账号、业务空间和 API Key 的调用量合并计算；即使每分钟总调用量未超限，短时间内的请求激增也可能触发。 - -**额度已用尽**：可加购坐席（加购后需将新坐席分配给成员后再使用）、加购共享用量包，或等待下一计费周期额度自动重置。 - -**触发限流**：等待约一分钟后重试，并采用平滑请求策略（如匀速调度、指数退避或请求队列缓冲）避免瞬时高峰。 - -**Connection error** - -Base URL 域名拼写错误或网络连接异常 - -检查 Base URL 域名拼写及网络连接。 - -## **产品功能相关** - -### **Token Plan 团队版的 API Key 能与其他套餐或普通 API 混用吗？** - -不能。Token Plan 团队版、Coding Plan 和百炼按量计费三者的 API Key 和 Base URL 互不相通，请勿混用。误用其他 API Key 不会抵扣 Token Plan 团队版的套餐额度。 - -### **能在多个工具中使用同一订阅吗？** - -可以。同一 API Key 可在全部兼容的 AI 编程和智能体工具中使用，额度共享消耗。每个成员持有独立的 API Key，不可共享给其他成员。 - -### **有哪些使用限制？** - -仅限在兼容的 AI 编程和智能体工具中交互式使用，不可用于自动化脚本或应用后端。违规使用可能导致订阅暂停或 API Key 封禁。 - -### **团队管理入口在哪里？** - -阿里云主账号或 RAM 用户登录[Token Plan 控制台](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/token-plan)后，在左侧菜单进入**我的订阅**，通过订阅卡片的**设置**、**用量分析**、**分配座席**入口进入成员管理与设置面板；也可点击**进入管理平台**跳转独立管理平台。通过 SSO 或钉钉加入的成员，通过管理员分发的**管理平台地址**（形如 tokenplan-enterprise.bailian.aliyunportal.com）登录管理平台。详见[访问入口](https://help.aliyun.com/zh/model-studio/token-plan-team#tp05-enter)。 - -### **成员如何获取 API Key？** - -管理员在管理后台创建成员账号并分配席位后，为成员生成 API Key。成员无法自行生成，需联系管理员获取。详见[团队管理](https://help.aliyun.com/zh/model-studio/token-plan-team)。 - -### **回收席位、修改角色、移出组织有什么区别？** - -这三个操作都在成员管理页面执行，但作用范围不同： - -- **回收席位**：撤销成员的席位使用权并将席位释放回席位池。成员失去使用权，但仍留在组织中，可被重新分配席位。 - -- **修改角色**：变更成员的权限（如管理员/普通成员），不影响席位分配和组织归属。 - -- **移出组织**：将成员从团队完全移除，席位自动回收至席位池，成员从成员列表消失。 - - -### **为什么 API Key 只能查看一次，丢失后如何处理？** - -为避免团队间 API Key 混用导致计费混乱，API Key 仅在首次生成或重置时显示，后续无法再次查看或复制。若 API Key 丢失，在**成员管理**页面找到对应成员，点击**重置**生成新 Key，原 Key 立即失效，需在工具中重新配置。 - -## **购买相关** - -### **可以同时购买多个套餐吗？** - -每个阿里云账号限购一个订阅，同一订阅下每种坐席类型均可购买多个。共享用量包可叠加购买，单次最多 1000 个。 - -### **可以单独购买共享用量包吗？** - -不可以。共享用量包是 Token Plan 团队版的附加商品，需先订阅 Token Plan 团队版坐席套餐后，才能购买共享用量包。 - -### **套餐是否支持退订？** - -支持按席位退订。在控制台**Token Plan 订阅详情页**的订阅明细中，点击对应席位的**退订**，已有用量消耗的席位不可退订。也可勾选多个席位后点击**批量退订**。退款原路退回支付账户，预计 1-3 个工作日到账。详见[订阅管理](https://help.aliyun.com/zh/model-studio/token-plan-overview#tp01-sub-mgmt)。 - -### **阿里云账号欠费是否影响 Token Plan 团队版的使用？** - -Token Plan 团队版为预付费订阅产品，只要套餐额度未用尽且订阅仍在有效期内，阿里云账号欠费不影响 Token Plan 团队版的正常使用。 - -### **如何加购坐席？** - -在 Token Plan 订阅详情页的订阅明细中，点击**加购座席**，选择坐席类型和数量后提交订单。加购后需将新坐席分配给成员才能使用。详见[团队管理](https://help.aliyun.com/zh/model-studio/token-plan-team)。 - -### **如何关闭或开启自动续费？** - -在 Token Plan 订阅详情页的订阅明细中，点击**关闭自动续费**或**开启自动续费**。关闭后订阅到期不自动续费，需手动续费。 - -### **套餐变更与退订重购有哪些限制？** - -Token Plan 团队版与 Coding Plan 是两个独立的订阅计划，不支持相互转换： - -- **套餐互转限制**：不支持将已购买的 Token Plan 团队版更换为 Coding Plan，也不支持将 Coding Plan 直接转换为 Token Plan 团队版（即使补差价也不行）。可以同时订阅这两个计划，各自独立计费。 - -- **退订重购注意事项**：退款并重新购买后，API Key 和 Base URL 会发生变更，需在工具中重新配置新的专属 API Key 才能正常使用。 - - -## **计量相关** - -### **Credits 抵扣规则是什么？** - -Token Plan 团队版实际消耗取决于每次请求中输入 Token、缓存 Token 和输出 Token 的组合。优先从坐席额度抵扣，坐席额度用尽后从共享用量包抵扣，全部用尽后服务暂停至下一计费周期或购买共享用量包补充额度。 - -### **如何查看用量？** - -在[Token Plan 订阅详情页](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/token-plan)可查看套餐和共享用量包的用量详情。管理员还可在[管理后台](https://tokenplan-enterprise.bailian.aliyunportal.com)的用量分析页面查看全部成员的消耗明细。 - -### **用量如何重置？** - -坐席额度在每个订阅月到期时重置，未用完的额度不累积到下月。共享用量包额度购买后有效期为 1 个月，到期后需重新购买，不随座席额度按月重置。 - -### **超出限额之后怎么办？** - -坐席额度用尽后自动从共享用量包抵扣；全部额度用尽后服务暂停。可通过以下方式恢复： - -- 购买共享用量包补充额度。 - -- 等待下一计费周期额度自动重置。 - - -### **续费后为什么 Credits 没有增加？** - -Token Plan 团队版的坐席额度按订阅周期计算，每个订阅月到期时自动重置。续费（续订）仅延长订阅有效期或预定下一计费周期的额度，**不会叠加补充至当前计费周期**。 - -若当前周期额度已用尽且需立即恢复服务，续订下月额度无法即时补充，可通过以下方式恢复： - -- 购买共享用量包补充额度。 - -- 升级至更高规格的坐席。 - -- 加购坐席（加购后需将新坐席分配给成员方可使用）。 - - -## **数据安全** - -### **数据安全如何保障？** - -Token Plan 团队版承诺不使用对话数据训练模型，传输过程采用 HTTPS 加密，并基于多租户隔离架构保障企业级数据隔离。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-overview.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-overview.md index 8abf5759..9043d571 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-overview.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-overview.md @@ -1,299 +1,187 @@ -# Token Plan（团队版）概述 +# Token Plan 概述 -Token Plan 团队版是阿里云百炼推出的 AI 大模型订阅服务，以 Credits 统一计量，支持文本生成与图像生成模型，兼容主流 AI 编程与智能体工具，提供团队管理后台、数据安全保障，调用平稳运行。 +Token Plan 是阿里云百炼推出的 AI 大模型订阅服务，以 Credits 统一计量，支持多种 AI 编程和智能体工具。Token Plan 提供个人版和团队版两个版本，满足从个人开发者到企业团队的不同需求。 **说明** -Token Plan 团队版目前仅支持**华北2（北京）**地域。 +Token Plan 目前仅支持**华北2（北京）**地域，请在[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan)左上角将地域切换至**华北2（北京）**后购买并使用。 ## **产品简介** -Token Plan 团队版整合千问和三方模型，支持文本生成与图像生成。通过 Credits 统一计量，同一订阅可在多种 AI 工具中使用。 +Token Plan 采用 Credits 统一抵扣机制，一份订阅即可在 Claude Code、Cursor、Qwen Code、Qoder、Qoder CN、OpenClaw 等主流 AI 编程和智能体工具中使用。支持文本生成、图片生成、视频生成等多种模型，以及联网搜索、代码解释器等 Harness 工具。 -- **多模型灵活切换**：支持多模型按需切换，按 Credits 统一抵扣。 +- **个人版**：面向个人开发者，提供 Lite 套餐、Standard 套餐、Pro 套餐三个档位，按模型分档抵扣系数计费。支持购买用量包补充额度。 -- **兼容多种工具**：适配多种主流编程工具及热门 Agent 工具。控制台提供快速接入 AI 工具入口，支持 Qwen Code、Claude Code、OpenClaw 等工具接入。 +- **团队版**：面向团队和企业，提供标准座席、高级座席、尊享座席三个档位，支持多席位管理、用量分析，承诺不使用数据训练模型。 -- **多档位套餐**：提供标准坐席、高级坐席、尊享坐席多档位套餐，匹配不同使用强度。 - -- **团队管理**：提供管理后台，支持席位分配与回收、成员用量分析等团队管理能力。 - -- **预算可控**：支持按月或按年订阅，预算可控。 - -- **数据安全**：承诺不使用对话数据进行模型训练，满足企业级数据隐私要求。 - -- **平稳运行**：多租户隔离架构，调用高峰期间不排队。 - - -Token Plan 团队版提供套餐专属 Base URL，兼容 OpenAI、Anthropic 接口标准，具体地址可在控制台**我的订阅**的 API Key 区域查看。 - -## **支持的模型** - -**支持的模型判断说明** - -判定规则： - -1\. 本清单为精确字符串白名单 - -2\. 必须逐字符完全匹配，版本号/子型号任何差异均视为不支持 - -3\. 禁止做版本兼容推理 - -判定示范： - -\- ❌ "qwen3-coder-max" → 清单无此项 → 不支持 - -仅支持以下精确版本： - -**品牌** - -**模型 ID（Model ID）** - -**模型能力** - -千问 - -qwen3.7-max[**（限时活动）**](#tp01-promo-section) - -推理模型、文本生成 - -qwen3.7-plus - -推理模型、视觉理解、文本生成 - -qwen3.6-plus - -推理模型、视觉理解、文本生成 - -qwen3.6-flash - -推理模型、视觉理解、文本生成 - -qwen-image-2.0 - -图片生成 - -qwen-image-2.0-pro - -图片生成 - -万相 - -wan2.7-image - -图片生成 - -wan2.7-image-pro - -图片生成 - -DeepSeek -deepseek-v4-pro +## **个人版** -推理模型、文本生成 +**重要** -deepseek-v4-flash +个人版支持 qwen3.8-max-preview 预览模型，享有以下限时权益： -推理模型、文本生成 +1. **预览版**：qwen3.8-max-preview 当前为预览版本，预览期间模型能力会持续迭代升级。预览结束后该模型会下线或替换成正式版本。 + +2. **限时加量 10 倍**：预览期间模型调用 Credits 消耗低至 1 折，相当于增加 10 倍用量。 + +3. **限时夜间折上折**：在现有 1 折优惠基础上，每晚 22:00 - 次日 08:00 期间调用模型，Credits 消耗再享 2 折（即原标准的 0.2 折）。 + -deepseek-v3.2 +阿里云百炼有权根据运营情况对活动进行变更或调整，包括不限于活动内容和有效期等，请以页面最新内容或阿里云通知为准。 -推理模型、文本生成 +**Lite 套餐** -月之暗面 +**Standard 套餐** -kimi-k2.7-code +**Pro 套餐** -推理模型、视觉理解、文本生成 +**用量包** -kimi-k2.6 +**定价** -推理模型、视觉理解、文本生成 +原价 60 元/月 +限时 **39 元/月** -kimi-k2.5 +原价 180 元/月 +限时 **139 元/月** -推理模型、视觉理解、文本生成 +原价 600 元/月 +限时 **499 元/月** -智谱 AI +100 元/个/月 +20,000 Credits/个 -glm-5.2 +**5 小时限额** -推理模型、文本生成 +700 Credits -glm-5.1 +3,000 Credits -推理模型、文本生成 +12,000 Credits -glm-5 +无限制 -推理模型、文本生成 +**每 7 天限额** -MiniMax +2,500 Credits -MiniMax-M2.5 +10,000 Credits -推理模型、文本生成 +40,000 Credits -## **套餐与定价** +无限制 -前往 [Token Plan 团队版购买页面](https://common-buy.aliyun.com/token-plan/)选择坐席类型、数量和订阅周期，完成订阅。主账号和 RAM 账号均可订阅。 +**并发 Agent** -订阅周期支持按月购买、按年购买、连续包月包年。 +1-2 个 -### **限时活动** +3-4 个 -即日起至 2026 年 7 月 22 日 23:59（UTC+8），qwen3.7-max 模型 Credits 消耗减半，同时支持隐式缓存。 +6-8 个 -### **Token Plan 团队版** +\- -提供标准坐席、高级坐席、尊享坐席三个档位，匹配不同使用强度。 +**购买条件** -席位（坐席）是 Token Plan 团队版的最小订阅单位，代表一个团队成员的使用名额。管理员在[团队管理](https://help.aliyun.com/zh/model-studio/token-plan-team)中将席位分配给成员后，系统自动为该成员生成专属的 API Key。每个席位绑定一个成员、对应一个 API Key，不可共享。 +\- -**坐席类型** +需有效订阅，最多 5 个 -**价格** +**模型** -**额度** +支持文本生成、图像生成、视频生成等多种模型（[查看完整列表](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview#tpp01-models)） -**适用场景** +**Harness 工具** -标准坐席 +支持多种 Harness 工具，以 Credits 统一抵扣（[查看详情](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview#tpp01-harness)） -¥198/坐席/月 +## **团队版** -25,000 Credits/坐席/月 +**重要** -轻度使用 AI 辅助的团队成员 +团队版支持 qwen3.8-max-preview 预览模型，享有以下限时权益： -高级坐席 +1. **预览版**：qwen3.8-max-preview 当前为预览版本，预览期间模型能力会持续迭代升级。预览结束后该模型会下线或替换成正式版本。 + +2. **限时加量 10 倍**：预览期间模型调用 Credits 消耗低至 1 折，相当于增加 10 倍用量。 + -¥698/坐席/月 +阿里云百炼有权根据运营情况对活动进行变更或调整，包括不限于活动内容和有效期等，请以页面最新内容或阿里云通知为准。 -100,000 Credits/坐席/月 +**标准座席 Standard** -日常高频使用 AI 编程或办公的团队成员 +**高级座席 Pro** -尊享坐席 +**尊享座席 Max** -¥1,398/坐席/月 +**共享用量包 Extra Bundle** -250,000 Credits/坐席/月 +**定价** -重度依赖 AI 的核心开发者或高强度使用者 +原价 198 元/座席/月 +限时 **150 元/座席/月** -### Token Plan 团队版 - 共享用量包 +原价 698 元/座席/月 +限时 **550 元/座席/月** -跨坐席共享的弹性用量包，当个别坐席用量超出套餐额度时，可从共享用量包中抵扣。每个共享用量包有效期为 1 个月，到期未使用的额度自动清零。持有多个共享用量包时，优先抵扣最近到期的用量包。 +**1,398 元/座席/月** -**档位** +5,000 元/个/月 -**价格** +**每月总额度** -**额度** +25,000 Credits/座席/月 -Token Plan 团队版 - 共享用量包 +100,000 Credits/座席/月 -¥5,000/个 +250,000 Credits/座席/月 625,000 Credits/个 -## **订阅管理** - -在[Token Plan 控制台](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/token-plan)的**我的订阅**页面管理订阅： - -- **加购席位**：点击**加购座席**，新加席位与现有订阅统一到期，费用和 Credits 额度均按剩余时长折算。 - -- **升级席位**：在**订阅明细**中点击席位的**升级**，将低档位席位升级为高档位，按差价补缴费用。升级当月的 Credits 额度按**高低档位差值**、依本计费周期的剩余时长折算后补充到当月可用额度；自下一计费周期起按新档位的完整额度发放。实际到账额度以控制台**我的订阅**页面为准。 - -- **退订席位**：在**订阅明细**中点击席位的**退订**，按席位维度退订；已有用量消耗的席位不可退订。退款原路退回支付账户，预计 1-3 个工作日到账。 - -- **批量操作**：勾选多个席位后点击**批量升级**或**批量退订**。 - -- **续费**：点击**续费**按钮，续费周期与订阅时一致（按月订阅则按月续费，按年订阅则按年续费），到期前完成可避免服务中断。 - -- **自动续费**：点击**开启自动续费**，确认后次日生效，到期前 9 天系统按订阅周期自动扣款续费。如需关闭，点击**关闭自动续费**并确认，关闭后停止到期前自动扣款续费，订阅到期后自动停订。 - -- **加购共享用量包**：在**共享用量包**区域点击**前往购买**。 - - -## **Credits 计费机制** - -### **计费说明** - -单次请求消耗的 Credits **并非固定值**，由模型类型、Token 用量、思考模式及工具调用等动态决定。其中 Token 用量会随多轮对话累积的上下文（历史消息、代码、工具返回、检索内容等）持续增长，且部分模型按**上下文长度阶梯计费**（上下文越长，单价档位可能越高），因此同一模型在不同请求下的消耗可能相差较大。实际消耗以[控制台订阅页用量明细](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan)为准。 - -### **计算示例** +**5 小时限额** -以 qwen3.6-plus 为例，预估单次请求消耗明细如下（不同模型的单价不同，实际以账单为准）： +无限制 -**Token 类型** +**7 天限额** -**数量** +无限制 -**消耗 Credits** +**模型** -输入 tokens +支持文本生成、图像生成、视频生成等多种模型（[查看完整列表](https://help.aliyun.com/zh/model-studio/token-plan-team-overview#tpt01-models)） -8,349 +**Harness 工具** -1.67 +支持多种 Harness 工具，以 Credits 统一抵扣（[查看详情](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview#tpp01-harness)） -缓存 tokens +**团队管理** -40,794 +支持多席位管理和用量分析（[团队管理](https://help.aliyun.com/zh/model-studio/token-plan-team-management)） -0.82 +## **套餐限额** -输出 tokens +### **个人版** -573 +个人版采用 5 小时和 7 天两层固定窗口限额，限额单位为 Credits： -0.69 - -**合计** - -**约 3.18 Credits** - -**说明** - -上表**仅为单次请求的示例**，并不代表每次请求都固定消耗约 3 Credits。在 AI 编程、智能体等**多轮对话**场景中，每次请求都会携带累积的上下文（历史对话、代码、工具返回等），随着对话进行输入 Token 持续增多，单次消耗的 Credits 也会**相应上升**；若模型按上下文长度阶梯计费，长上下文可能进入更高价位档，消耗速度进一步加快。 - -如需控制消耗，建议：任务切换或话题变更时**及时开启新会话**、清理无关历史，以缩短上下文；对长文档、大代码库按需拆分输入；并在上述控制台订阅页用量明细中关注各模型的实时消耗趋势。 - -**说明** - -上表以 qwen3.6-plus 展示 Token 类型分布，其中**“缓存 tokens”一行为通用 Token 分类示意**。**Token Plan 团队版中，隐式缓存的命中取决于所选模型是否支持**——目前仅 qwen3.7-max 在限时活动期间支持隐式缓存（详见上文[限时活动](#tp01-promo-section)章节）；使用其他模型时，输入 Token 不会进入隐式缓存抵扣，此时“缓存 tokens”一行不适用。 - -### **抵扣顺序** - -1. 优先从坐席套餐的月度额度中抵扣。 - -2. 坐席额度用尽后，从共享用量包中抵扣。持有多个共享用量包时，优先抵扣最近到期的用量包。 +- **5 小时限额**：自首次调用起开启 5 小时计时窗口，窗口期内累计消耗达到限额后暂停服务，需等待满 5 小时后额度重置。 -3. 全部额度用尽后，服务将暂停至下一计费周期或购买共享用量包补充额度。 +- **7 天限额**：自首次调用起开启 7 天计时窗口，窗口期内累计消耗达到限额后暂停服务，需等待满 7 天后额度重置。 -## **查看额度消耗情况** +任一层限额触顶即暂停服务，可购买用量包补充额度继续使用，或等待对应窗口周期结束后额度重置。窗口期内未用完的额度不结转至下一周期。 -**通过控制台**：登录[Token Plan 控制台](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/token-plan)，在**我的订阅**页面查看总额度使用百分比、重置时间、团队席位分配情况，以及各席位与共享用量包的状态和到期时间。 +### **团队版** -**通过团队管理平台**：在**用量分析**页面，可查看近 1、7、30 天的 Credits 消耗趋势、各模型用量，以及每个成员的消耗明细。详见[团队管理](https://help.aliyun.com/zh/model-studio/token-plan-team)。 +团队版采用月度总额度制，无 5 小时和 7 天窗口限额。每个座席的月度额度在计费周期内可用，到期未使用的额度不结转。超出月度总额度后调用将被阻断，可购买共享用量包补充额度。 -## **使用细则** +## **常见问题** -1. **使用范围**：仅限在兼容的 AI 编程和智能体工具中交互式使用，不可用于自动化脚本或应用后端。违规使用可能导致订阅暂停或 API Key 封禁。 - -2. **数据安全**：Token Plan 团队版不会使用对话数据训练模型。 - -3. **账号规范**：API Key 仅限已分配席位的成员本人使用，不可共享或公开泄露。 - -4. **退订与退款**：在控制台**我的订阅**页面按席位退订，已有用量消耗的席位不可退订。退款原路退回支付账户，预计 1-3 个工作日到账。 - -5. **服务地域**：Token Plan 团队版目前仅在特定地域提供服务，如需从海外调用，请确认符合当地法律法规要求。关于百炼支持的地域和服务部署范围，请参见[选择地域和服务部署范围](https://help.aliyun.com/zh/model-studio/regions/)。 - +### **个人版和团队版可以同时购买吗？** + +可以。同一阿里云账号可以同时持有个人版和团队版，各自独立计费。 -## 错误码 +### **关于 Coding Plan** -如果模型调用失败并返回报错信息，请参见[错误码](https://help.aliyun.com/zh/model-studio/error-code)进行解决。 +Coding Plan 和 Token Plan 是两个独立的订阅产品，两者之间无法迁移或升级。Coding Plan Lite 已于 2026 年 3 月 20 日停止新购，2026 年 4 月 13 日停止续费和升级；Coding Plan Pro 为限量抢购，库存售罄后不再补充。推荐使用 **Token Plan**，支持更多模型和 Harness 工具。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-faq.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-faq.md new file mode 100644 index 00000000..2e703bf7 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-faq.md @@ -0,0 +1,186 @@ +# 常见问题 + +Token Plan 个人版的额度、购买、订阅和接入常见问题。 + +## **额度与限额** + +### 5 小时限额和 7 天限额是什么意思？ + +Token Plan 个人版采用 5 小时和每 7 天两层固定窗口限额，限额单位为 Credits。两层限额独立计算，任一层触顶即暂停服务，需等待对应窗口周期结束后额度重置。窗口期内未用完的额度不结转至下一周期。 + +- **5 小时限额**：自首次调用起开启 5 小时计时窗口，窗口期内累计消耗达到限额后暂停服务，需等待满 5 小时后额度重置。 + + 例如：Standard 套餐的 5 小时限额为 3,000 Credits。您在 7 月 20 日 10:00 首次调用，系统开启窗口（10:00 ~ 15:00）。10:00 消耗 2,000 Credits，11:00 消耗 1,000 Credits，累计 3,000 Credits 触顶暂停。需等到 15:00 窗口结束，额度重置为 3,000 Credits，服务恢复。 + +- **7 天限额**：自首次调用起开启 7 天计时窗口，窗口期内累计消耗达到限额后暂停服务，需等待满 7 天后额度重置。 + + 例如：Standard 套餐的 7 天限额为 10,000 Credits。您在 7 月 20 日首次调用，系统开启窗口（7 月 20 日 ~ 7 月 27 日）。7 月 20 日消耗 4,000 Credits，7 月 22 日消耗 6,000 Credits，累计 10,000 Credits 触顶暂停。需等到 7 月 27 日窗口结束，额度重置为 10,000 Credits，服务恢复。 + + +各档位限额如下： + +**档位** + +**5 小时限额** + +**7 天限额** + +Lite 套餐 + +700 Credits + +2,500 Credits + +Standard 套餐 + +3,000 Credits + +10,000 Credits + +Pro 套餐 + +12,000 Credits + +40,000 Credits + +### 5 小时限额到了但7 天限额还有余量，还能继续使用吗？ + +不能。任一层限额触顶即暂停服务，可购买用量包补充额度、使用额度重置功能，或等待对应窗口周期结束后额度自动重置。 + +### 7 天限额是固定日期重置吗？ + +不是。7 天限额采用固定窗口机制，自首次调用起计时 7 天，到期后额度重置。重置时间取决于您首次调用的时间，而非固定的日历日期（如每周一）。 + +### 额度用完了怎么办？ + +限额用完后调用会被阻断，不会按量计费。恢复方式： + +- 等待额度释放。 + +- 升级套餐。 + +- 购买用量包，获得不受 5 小时和 7 天限额约束的额外额度。 + + +### 用量包和套餐是什么关系？需要先买套餐吗？ + +用量包是套餐的补充，提供不受套餐限额约束的额外 Credits，需先订阅有效套餐后才能购买，最多同时持有 5 个。 + +### 用量包的额度有 5 小时/周限制吗？ + +没有。用量包额度不受套餐的 5 小时和 7 天窗口限额约束，购买后即可使用。 + +### 用量包有效期多久？ + +用量包有效期为 1 个月，到期后未使用的额度自动作废，不支持退款。 + +## **接入报错** + +### 常见报错及解决方案 + +**报错信息** + +**可能原因** + +**解决方案** + +401 InvalidApiKey: No API-key provided. + +请求头中未携带 API Key + +生成 API Key 并在工具中完成配置 + +401 InvalidApiKey: Invalid API-key provided. + +误用了按量计费的 API Key 或 Coding Plan 的 Key；订阅过期；Key 复制不完整 + +确认使用 Token Plan 个人版专属 API Key，确保完整且无空格 + +404 model 'xxx' not found or not supported + +模型名称拼写错误或不在支持列表 + +确认模型名称区分大小写，与套餐支持的模型 ID 一致。 + +401 invalid access token or token expired + +误用了 Coding Plan 或其他计费模式的 Base URL + +使用 Token Plan 个人版专属 Base URL + +401 Incorrect API key provided + +误用了百炼通用 Base URL（dashscope.aliyuncs.com） + +使用 Token Plan 个人版专属 Base URL + +429 Requests rate limit exceeded + +短时间内请求过于密集 + +等待一分钟后重试，降低请求频率 + +429 Allocated quota exceeded + +5 小时或7 天限额用尽 + +等待窗口释放额度 + +## **并发与性能** + +### 最多支持多少个 Agent 并发？ + +并发能力与套餐档位相关： + +**档位** + +**建议并发** + +Lite 套餐 + +可同时支持 1-2 个 Agent 并发运行 + +Standard 套餐 + +可同时支持 3-4 个 Agent 并发运行 + +Pro 套餐 + +可同时支持 6-8 个 Agent 并发运行 + +### 高峰期响应会变慢吗？ + +高峰期可能出现排队等待。如需更稳定的吞吐，可升级到更高档位或使用团队版。 + +## **使用规则** + +### "禁止 API 生产自动化调用"具体是什么意思？ + +Token Plan 个人版仅供个人通过官方指定工具（如 Cursor、Claude Code、Windsurf 等）进行交互式开发。不允许将 API Key 用于生产环境的自动化服务、批量脚本或后台定时任务等非交互场景。 + +### 多人共用一个账号可以吗？ + +不可以。Token Plan 个人版限单人使用，不允许多人共用同一账号或 API Key。如需多人协作，请使用 Token Plan 团队版。 + +## **购买与订阅** + +### RAM 用户可以使用 Token Plan 吗？ + +可以，需由主账号完成以下授权： + +1. 在 [RAM 控制台](https://ram.console.aliyun.com/)为该 RAM 用户授予 `AliyunTokenPlanReadOnlyAccess`（只读）或 `AliyunTokenPlanFullAccess`（管理）系统策略，同时授予 `AliyunBSSReadOnlyAccess` 系统策略。 + +2. 在百炼控制台[账号管理](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)页面，为该 RAM 用户分配管理员或订阅套餐权限。 + + +### 可以升配吗？升配后额度怎么算？ + +支持从低档位升级到更高档位。升级按剩余时长补缴差价，升级后每 5 小时限额和每 7 天限额立即提升至新档位对应额度。 + +### 可以降配吗？ + +不支持降配。如需更换为更低档位，可在订阅到期后重新购买。 + +### 自动续费怎么取消？ + +登录[百炼控制台 Token Plan](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan) 页面，在订阅管理中关闭自动续费。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-overview.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-overview.md new file mode 100644 index 00000000..6d7e420e --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-overview.md @@ -0,0 +1,257 @@ +# 概述 + +[Token Plan 个人版](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)是面向个人开发者的 AI 大模型订阅服务，以 Credits 统一计量，支持文本、多模态模型及 Harness 工具，适配主流 AI 编程和智能体工具。 + +**说明** + +Token Plan 个人版目前仅支持**华北2（北京）**地域。 + +## **核心特性** + +- **Credits 统一计量**：通过 Credits 统一抵扣不同模型和 Harness 工具的费用。 + +- **多模态模型支持**：覆盖文本生成、推理、视觉理解、图片生成、语音合成、语音识别、视频生成等能力。 + +- **Harness 工具集成**：支持联网搜索、文搜图、图搜图、网页抓取、代码解释器。 + +- **用量包**：超出限额后可购买用量包，不受限制继续使用。 + +- **兼容多种工具**：适配 Claude Code、Cursor、Qwen Code、Qoder、Qoder CN、OpenClaw 等主流 AI 编程和智能体工具。 + + +## **套餐档位与定价** + +**Lite 套餐** + +**Standard 套餐** + +**Pro 套餐** + +**用量包** + +**定价** + +原价 60 元/月 +限时 **39 元/月** + +原价 180 元/月 +限时 **139 元/月** + +原价 600 元/月 +限时 **499 元/月** + +100 元/个/月 +20,000 Credits/个 + +**每 7 天限额** + +2,500 Credits + +10,000 Credits + +40,000 Credits + +无限制 + +**每 5 小时限额** + +700 Credits + +3,000 Credits + +12,000 Credits + +无限制 + +**并发 Agent** + +1-2 个 + +3-4 个 + +6-8 个 + +\- + +**权益** + +文本、视觉等多模态模型 + +联网搜索等 Harness 工具 + +适配主流工具并持续扩展 + +享受 Lite 套餐所有权益 + +4x Lite 套餐用量 + +享受 Standard 套餐所有权益 + +16x Lite 套餐用量 + +更高的并发上限 + +不受 5 小时和 7 天限额约束 + +需先订阅套餐后购买 + +最多同时持有 5 个 + +- **每 5 小时限额**：自首次调用起开启 5 小时计时窗口，窗口期内累计消耗达到限额后暂停服务，需等待满 5 小时后额度重置。 + +- **每 7 天限额**：自首次调用起开启 7 天计时窗口，窗口期内累计消耗达到限额后暂停服务，需等待满 7 天后额度重置。 + + +任一层限额触顶即暂停服务，可购买用量包补充额度继续使用，或等待对应窗口周期结束后额度重置。窗口期内未用完的额度不结转至下一周期。 + +## **Credits 计费机制** + +### **计费说明** + +单次消耗的 Credits 由模型类型、Token 用量、思考模式及工具调用等动态决定，实际消耗以[控制台订阅页](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/personal)用量详情为准。 + +### **抵扣顺序** + +1. 每次调用消耗的 Credits 同时计入每 5 小时限额和每 7 天限额。 + +2. 任一层限额（5 小时或 7 天）触顶后，服务暂停，可购买用量包补充额度、使用额度重置功能，或等待对应窗口周期结束后额度自动重置。 + + +### **额度重置** + +可自主使用额度重置功能，将 5 小时/7 天限额归零重新计算。额度重置后，当前窗口内已消耗的 Credits 清零，限额从零开始重新累计。 + +## **支持的模型** + +**重要** + +个人版支持 qwen3.8-max-preview 预览模型，享有以下限时权益： + +1. **预览版**：qwen3.8-max-preview 当前为预览版本，**预览期间模型能力会持续迭代升级**。预览结束后该模型会下线或替换成正式版本。 + +2. **限时加量 10 倍**：限时活动期间，模型调用 Credits 消耗低至 1 折，相当于增加 10 倍用量。 + +3. **限时夜间折上折**：在现有 1 折优惠基础上，每晚 22:00 - 次日 08:00 期间调用模型，Credits 消耗再享 2 折（即原标准的 0.2 折）。 + + +阿里云百炼有权根据运营情况对活动进行变更或调整，包括不限于活动内容和有效期等，请以页面最新内容或阿里云通知为准。 + +**品牌** + +**模型 ID（Model ID）** + +**模型能力** + +千问 + +qwen3.8-max-preview + +推理模型、视觉理解、文本生成 + +qwen3.7-max + +推理模型、文本生成 + +qwen3.7-plus + +推理模型、视觉理解、文本生成 + +qwen3.6-flash + +推理模型、视觉理解、文本生成 + +智谱 AI + +glm-5.2 + +推理模型、文本生成 + +DeepSeek + +deepseek-v4-pro + +推理模型、文本生成 + +万相 + +wan2.7-image + +图片生成 + +wan2.7-image-pro + +图片生成 + +HappyHorse + +happyhorse-1.1-i2v + +视频生成 + +happyhorse-1.1-t2v + +视频生成 + +happyhorse-1.1-r2v + +视频生成 + +## **支持的 Harness 工具** + +**工具能力** + +**工具名称** + +联网搜索 + +web\_search + +文搜图 + +t2i\_search + +图搜图 + +i2i\_search + +网页抓取 + +web\_extractor + +代码解释器 + +code\_interpreter + +## **订阅管理** + +### **升级** + +支持从低档位升级到更高档位。升级按剩余时长补缴差价，升级后每 5 小时限额和每 7 天限额立即提升至新档位对应额度。 + +### **续费** + +续费支持切换续费周期，也可一次续费多个周期。 + +- **手动续费**：支持选择不同续费时长，可续费多个周期。 + +- **自动续费**：开启后到期前系统自动扣款续费。 + + +续费仅延长订阅有效期，不会叠加补充至当前计费周期的额度。 + +### **其他** + +- 个人版暂不支持退订。 + +- 订阅到期后重新购买，API Key 会发生变更，需在工具中重新配置。新 API Key 可在控制台[**我的订阅**](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/personal)页面的 API Key 区域获取。 + + +## **订阅前须知** + +1. **严禁 API 调用**：仅限在编程工具和智能体工具（如 Claude Code、Cursor、Qwen Code、Qoder、Qoder CN、OpenClaw 等）中使用，禁止以 API 调用的形式用于自动化脚本、自定义应用程序后端或任何非交互式批量调用场景。将套餐 API Key 用于允许范围之外的调用将被视为违规或滥用，可能会导致订阅被暂停或 API Key 被封禁。 + +2. **数据使用授权**：使用 Token Plan 个人版期间，模型输入以及模型生成的内容将用于服务改进与模型优化。停止使用 Token Plan 个人版服务可终止后续数据授权，但终止授权的范围不涵盖已授权使用的数据。详细条款请参见[阿里云百炼服务协议](https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html)第 5.2 条。 + +3. **账号使用规范**：套餐为订阅人专享使用，禁止共享。账号共享可能导致订阅权益受限。 + +4. **购买限制**：同一实名认证主体限购一份，可同时购买个人版和团队版。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-quick-start.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-quick-start.md new file mode 100644 index 00000000..10116c46 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-quick-start.md @@ -0,0 +1,81 @@ +# 快速开始 + +三步完成 Token Plan 个人版订阅和接入：选择套餐、获取 API Key、配置 AI 工具。 + +## **步骤一：订阅 Token Plan 个人版** + +访问 [Token Plan 个人版购买页面](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)，选择套餐档位和订阅周期，完成订阅。 + +购买须知： + +- **RAM 用户授权**：RAM 用户使用 Token Plan 前，需由主账号完成以下授权： + + 1. 在 [RAM 控制台](https://ram.console.aliyun.com/)为该 RAM 用户授予 `AliyunTokenPlanReadOnlyAccess`（只读）或 `AliyunTokenPlanFullAccess`（管理）系统策略，同时授予 `AliyunBSSReadOnlyAccess` 系统策略。 + + 2. 在百炼控制台[账号管理](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)页面，为该 RAM 用户分配管理员或订阅套餐权限。 + + +## **步骤二：获取 API Key 和 Base URL** + +- **API Key**：订阅完成后，在 Token Plan 控制台的**我的订阅**页面生成 API Key。API Key 仅在生成时完整显示一次，请立即复制并妥善保存。 + +- **Base URL**：根据 AI 工具支持的协议，选择对应的 Base URL。 + + +**协议** + +**Base URL** + +OpenAI 兼容 + +`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` + +Anthropic 兼容 + +`https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` + +**重要** + +Token Plan 的 API Key 以 `sk-sp-` 开头，与百炼通用 API Key（`sk-` 开头）格式不同，两者不可混用。Token Plan、Coding Plan 和按量付费的 API Key 与 Base URL 完全隔离，必须配套使用。 + +## **步骤三：接入 AI 工具** + +将 API Key 和 Base URL 配置到 AI 工具中，即可开始使用。 + +[**OpenClaw**开源、自托管个人 AI 助手](https://help.aliyun.com/zh/model-studio/openclaw) + +[**Hermes Agent**开源 AI 代理框架，内置自学习循环](https://help.aliyun.com/zh/model-studio/hermes-agent) + +[**Claude Code**AI 终端编码助手，支持自然语言编程](https://help.aliyun.com/zh/model-studio/claude-code) + +[**OpenCode**开源 AI 编程代理工具](https://help.aliyun.com/zh/model-studio/opencode) + +[**Cursor**AI 原生代码编辑器](https://help.aliyun.com/zh/model-studio/cursor) + +[**Codex**OpenAI 推出的命令行编程工具](https://help.aliyun.com/zh/model-studio/codex) + +[**Qwen Code**开源命令行 AI 编码工具](https://help.aliyun.com/zh/model-studio/qwen-code) + +[**QwenPaw**开源个人 AI 助手，支持本地与云端部署](https://help.aliyun.com/zh/model-studio/qwenpaw) + +[**Cherry Studio**多模型桌面客户端](https://help.aliyun.com/zh/model-studio/cherry-studio) + +[**Chatbox**跨平台 AI 桌面客户端](https://help.aliyun.com/zh/model-studio/chatbox) + +[**Cline**VS Code 扩展，智能代码补全和调试](https://help.aliyun.com/zh/model-studio/cline) + +[**Qoder**面向真实软件开发的 Agentic 编码平台](https://help.aliyun.com/zh/model-studio/qoder-agent) + +[**Lingma**阿里云智能编码助手，提供独立 IDE](https://help.aliyun.com/zh/model-studio/lingma-agent) + +[**Kilo CLI**轻量高性能命令行编程工具](https://help.aliyun.com/zh/model-studio/kilo-cli) + +[··· **更多工具**其他编程工具](https://help.aliyun.com/zh/model-studio/more-tools) + +## **可选：接入多模态生成模型** + +Token Plan 个人版支持多模态生成模型（wan2.7-image、happyhorse-1.1-t2v 等）。多模态生成模型使用独立的接口，需要通过 AI 工具的 Skill 或扩展机制接入，详见[接入多模态生成模型](https://help.aliyun.com/zh/model-studio/token-plan-multimodal-gen)。 + +## **可选：接入 Harness 工具** + +通过 Harness 工具调用，模型可以在对话中调用联网搜索、文搜图、图搜图、网页抓取、代码解释器等扩展能力。当前仅 qwen3.7、qwen3.8 支持原生工具调用，通过 Responses API 直接调用，Harness 工具按抵扣系数消耗 Credits。详见[接入 Harness 工具](https://help.aliyun.com/zh/model-studio/token-plan-harness-tool)。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-faq.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-faq.md new file mode 100644 index 00000000..c24ef0ee --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-faq.md @@ -0,0 +1,211 @@ +# 常见问题 + +Token Plan 团队版的常见问题解答，包括产品选择、Credits 计费与额度、模型与工具兼容性、使用限制、购买续费与退订等。 + +## **产品定位与套餐选择** + +### **个人版和团队版有什么区别？我该买哪个？** + +**对比项** + +**个人版** + +**团队版** + +适用场景 + +个人开发者 + +团队/企业 + +额度机制 + +5 小时 + 7 天固定窗口限额 + +固定月额度 + +团队管理 + +不支持 + +席位分配与回收、成员用量分析、SSO 接入 + +数据安全 + +数据使用遵循服务协议 + +承诺不使用对话数据训练模型 + +高峰期性能 + +高峰期可能出现等待 + +多租户隔离，高峰期不排队 + +个人开发者、日常使用 AI 编程工具，选择个人版即可；团队需要多人协作、统一管理席位和用量、对数据安全有更高要求，选择团队版。 + +### **个人版和团队版能同时买吗？额度是分开算的还是共享的？** + +可以同时购买。同一阿里云账号可以同时持有个人版和团队版，各自独立计费，额度不共享。 + +## **Credits 计费与额度** + +### **团队版的额度机制是怎样的？** + +团队版采用固定月额度，无 5 小时/7 天窗口限制。各坐席类型的月度额度： + +- **标准坐席**：25,000 Credits/坐席/月 + +- **高级坐席**：100,000 Credits/坐席/月 + +- **尊享坐席**：250,000 Credits/坐席/月 + + +坐席额度在每个订阅月到期时重置，未用完的额度不累积到下月。 + +### **坐席额度用完了怎么办？** + +坐席额度用尽后调用会被阻断，不会按量计费。恢复方式： + +- 等待下一个订阅月额度自动重置。 + +- 购买共享用量包（625,000 Credits/个，有效期 1 个月），团队内全部成员共享使用。 + + +## **模型与工具兼容性** + +### **Token Plan 团队版支持哪些模型？** + +团队版支持文本生成、推理、视觉理解、图片生成和视频生成模型。具体模型列表和抵扣系数请参见控制台的模型列表页面。 + +### **支持 Cursor / Claude Code / Cline 等第三方工具吗？** + +支持。Token Plan 兼容 OpenAI 和 Anthropic 协议，任何支持自定义 Base URL 和 API Key 的工具均可接入，包括 Cursor、Claude Code、Qwen Code、Qoder、Qoder CN、Cline、OpenClaw、Cherry Studio、Chatbox 等。具体配置方法请参见快速开始中的接入 AI 工具部分。 + +### **团队版的 API Key 能用在个人版上吗？** + +不能。个人版和团队版各自生成独立的 API Key，不可混用。系统会根据 API Key 自动识别对应的套餐。 + +### **Harness 工具是什么？** + +Harness 工具是模型内置的扩展能力，包括联网搜索、文搜图、图搜图、网页抓取、代码解释器等。团队版支持 Harness 工具，调用时按工具抵扣系数消耗 Credits。当前仅 qwen3.7 和 qwen3.8 系列模型支持原生 Harness 工具调用。 + +## **使用限制** + +### **模型调用限流是怎么计算的？** + +每秒或每分钟消耗的 Token 数（TPS/TPM）超过模型限流阈值时会触发限流。限流按主账号维度计算，账号下全部 RAM 子账号、业务空间和 API Key 的调用量合并计算。 + +### **高峰期性能如何？** + +团队版基于多租户隔离架构，调用高峰期间不排队。 + +## **购买、续费与退订** + +### **支持 RAM 子账号购买吗？** + +支持。RAM 子账号使用前，需主账号完成以下授权： + +1. 在 RAM 控制台为该 RAM 用户授予 `AliyunTokenPlanReadOnlyAccess`（只读）或 `AliyunTokenPlanFullAccess`（管理）系统策略。 + +2. 在百炼控制台账号管理页面，为该 RAM 用户分配管理员或订阅套餐权限。 + + +### **可以升配吗？升配后额度怎么算？** + +支持坐席升配。升配后立即生效，限额按新坐席类型执行。升配需补缴差价（按剩余天数折算）。 + +### **可以降配吗？** + +不支持降配。如需使用更低坐席类型，可在当前订阅到期后重新订阅。 + +### **自动续费怎么取消？** + +登录[费用中心 > 续费管理](https://usercenter2.aliyun.com/finance/renew-manage)，找到 Token Plan 团队版订单，关闭自动续费。 + +### **续费时可以更换订阅时长吗？** + +不可以。续费仅支持按原订阅时长续费。如需更换订阅时长，可在订阅到期后重新购买。 + +### **限时优惠的计费规则是什么？** + +限时优惠仅适用于包月订阅的新购、续费和自动续费，包年订阅和升级坐席不参与。 + +加购坐席时按剩余时长折算费用，实际收费取折算金额与限时价中的较低值。 + +**示例**：标准坐席原价 ¥198/月，限时价 ¥150/月。 + +- 加购时按剩余时长折算为 ¥99，低于限时价，实际收取 ¥99。 + +- 加购时按剩余时长折算为 ¥165，高于限时价，实际收取 ¥150。 + + +## **与现有产品的关系** + +### **已有个人版，再买团队版会冲突吗？** + +不冲突。两者可以同时持有，各自独立计费。使用时根据 API Key 自动匹配对应套餐。 + +## **接入报错** + +### **常见报错及解决方案** + +**报错信息** + +**可能原因** + +**解决方案** + +401 InvalidApiKey: No API-key provided. + +请求头中未携带 API Key + +生成 API Key 并在工具中完成配置 + +401 InvalidApiKey: Invalid API-key provided. + +误用了按量计费的 API Key 或 Coding Plan 的 Key；订阅过期；Key 复制不完整 + +确认使用 Token Plan 团队版专属 API Key，确保完整且无空格 + +404 model 'xxx' not found or not supported + +模型名称拼写错误或不在支持列表 + +确认模型名称区分大小写，与套餐支持的模型 ID 一致 + +401 invalid access token or token expired + +误用了 Coding Plan 或其他计费模式的 Base URL + +使用 Token Plan 团队版专属 Base URL + +401 Incorrect API key provided + +误用了百炼通用 Base URL（dashscope.aliyuncs.com） + +使用 Token Plan 团队版专属 Base URL + +400 Range of input length should be \[1, xxx\] + +输入内容超出模型最大上下文长度 + +新建会话清空历史，或使用工具的上下文压缩命令 + +400 调用图像/视频生成模型报错（如 qwen-image-2.0、wan2.7-image） + +图像、视频生成模型使用独立接口，无法通过文本模型的 Base URL 直接调用 + +通过工具的 Skill、Slash Command 或 Agent 扩展机制接入，详见[接入多模态生成模型](https://help.aliyun.com/zh/document_detail/6546109.html) + +429 Requests rate limit exceeded + +短时间内请求过于密集，触发模型限流 + +等待一分钟后重试，降低请求频率 + +429 Allocated quota exceeded + +坐席月度额度用尽 + +购买共享用量包或等待下月额度重置 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-management.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-management.md new file mode 100644 index 00000000..bc5280b9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-management.md @@ -0,0 +1,183 @@ +# 团队管理 + +在 Token Plan 控制台或管理平台中添加和管理团队成员、分配和回收席位、监控 Credits 用量。 + +## **访问入口** + +**阿里云主账号或 RAM 用户**：登录 Token Plan 控制台，在**我的订阅**页面进行团队管理：在**团队版**卡片点击**设置**修改组织名称、登录方式（SSO/钉钉）等；在**订阅明细**区域点击**分配座席**进入成员管理。 + +**说明** + +RAM 用户使用 Token Plan 前，需由主账号完成以下授权： + +1. 在 RAM 控制台为该 RAM 用户授予以下系统策略：`AliyunTokenPlanReadOnlyAccess`（只读）或 `AliyunTokenPlanFullAccess`（管理），用于使用 Token Plan，权限范围按实际需要选择；同时授予 `AliyunBSSFullAccess`（费用中心权限），用于查看**我的订阅**页面中主账号已订阅的 Token Plan 套餐与用量。缺少费用中心权限时，RAM 用户登录后**我的订阅**页面将显示为空。 + +2. 在百炼控制台**账号管理**页面，为该 RAM 用户分配**管理员**或**订阅套餐**权限。 + + +通过 **SSO 或钉钉**加入的成员：通过管理员分发的**管理平台地址**登录管理平台。管理平台地址可在**设置**页面的**基本信息**区域获取。 + +## **角色与权限** + +**角色** + +**权限** + +所有者 + +添加或移除成员、分配或回收席位、修改成员角色、查看全部成员和模型的用量 + +管理员 + +权限范围与所有者相同。管理员由所有者授予，可被移除或降级。 + +成员 + +使用管理员分配的 API Key 和 Base URL 调用模型 + +## **成员管理** + +### **添加成员** + +- **手动添加**（不能登录管理平台，仅供 API 调用）：在**成员管理**页面点**添加成员**，在弹窗中填用户名（仅支持英文字母、数字、下划线）和角色，可选择同时分配席位。在该成员的操作列点**分配席位**选择席位版本，分配后系统自动生成 API Key，连同 Base URL 发给成员即可调用模型。 + +- **SSO 或钉钉登录**（成员可登录管理平台，自管席位和 API Key）：完成 SAML 或钉钉接入配置后，成员从登录页对应入口登录即自动加入。 + + +### **修改成员角色** + +在**成员管理**页面，找到目标成员，点击操作列的**修改角色**，在弹窗中选择新角色（管理员或成员）后保存。所有者角色不可修改。 + +### **重置 API Key** + +在**成员管理**页面，找到目标成员，点击操作列的**重置**。重置后原 API Key 立即失效，新 API Key 需重新发给成员。 + +### **移出成员** + +在**成员管理**页面，找到目标成员，点击操作列的**移出组织**。移出后席位自动回收、API Key 立即失效。 + +## **SAML 接入** + +通过标准 SAML 2.0 对接企业 IdP，对应登录页的**SSO**入口。配置后，成员用 IdP 账号登录管理平台即自动加入组织。 + +### **SAML 配置** + +**前提条件**：组织内有成员时无法编辑 SSO 配置，需先全部移出。 + +1. 登录 Token Plan 控制台，在**我的订阅**页面的**团队版**卡片点击**设置**，找到**SSO 配置**区域。 + +2. 从企业 IdP 获取 IdP 信息（IdP Entity ID、IdP SSO URL、IdP Certificate）。点击**编辑**，填入自定义的 SP Entity ID 和上述 IdP 信息，保存。 + +3. 保存后系统自动生成 ACS URL。将 SP Entity ID 和 ACS URL 填入企业 IdP 的 SSO 应用配置中。 + +4. 在**基本信息**区域复制**管理平台地址**，分享给团队成员。成员访问该地址，在登录页选择 SSO 登录方式即可加入组织。 + + +**SP 信息**（百炼侧） + +**参数** + +**说明** + +SP Entity ID + +百炼在 SSO 流程中的唯一标识，自定义填写，需同步填入企业 IdP 的 SSO 应用配置。 + +ACS URL + +IdP 认证成功后回传响应的地址，保存 SSO 配置后由百炼自动生成，需填入企业 IdP。 + +SP Certificate + +百炼侧 SAML 签名证书，保存 SSO 配置后由系统自动生成，用于企业 IdP 验证百炼侧响应签名，需同步配置到企业 IdP 信任链。 + +**IdP 信息**（企业侧，需从企业 IdP 获取后填入） + +**参数** + +**说明** + +IdP Entity ID + +企业身份提供商的唯一标识。 + +IdP SSO URL + +企业 IdP 的登录入口地址，成员登录时跳转到此地址进行认证。 + +IdP Certificate + +企业 IdP 的签名证书，百炼用来验证响应确实来自该企业。 + +### **配置示例：阿里云 IDaaS** + +**前提条件**：已开通阿里云 IDaaS EIAM 实例。 + +1. **在 IDaaS 中创建 SAML 应用**：登录 IDaaS 实例管理平台，进入**应用管理**，点击**添加应用**，选择**标准 SAML 2.0**应用模板，填写应用名称后创建。 + +2. **从 IDaaS 获取 IdP 信息并填入百炼**：在 IDaaS 应用的**单点登录**页面底部的**应用配置信息**区域，复制 IdP 信息，填入 Token Plan 控制台的 SSO 配置区域。 + +3. **将 SP Entity ID 和 ACS URL 填入 IDaaS**：保存后百炼 SSO 配置区域显示自动生成的 ACS URL，在 IDaaS 应用的**登录访问** > **单点登录**页面填入对应参数。 + +4. **在 IDaaS 中创建账户并授权**：在 IDaaS 的**账户管理**中为团队成员创建账户，在 SAML 应用的**登录访问** > **应用授权**页面添加授权。 + +5. **分享管理平台地址给成员**：在**设置**页面的**基本信息**区域复制**管理平台地址**，成员访问该地址选择 SSO 登录即可加入组织。 + + +## **钉钉接入** + +使用钉钉作为身份系统的企业可直接接入，对应登录页的**钉钉**入口。配置后，成员用钉钉账号登录管理平台即自动加入组织。 + +**前提条件**：已在钉钉中创建企业，并将待登录的成员加入企业。 + +1. **创建钉钉企业内部应用**：登录钉钉开发者后台，进入**应用开发** > **企业内部应用** > **钉钉应用**，点击**创建应用**。 + +2. **获取应用凭证**：在应用详情页的**基础信息** > **凭证与基础信息**，记录 Client ID 和 Client Secret。 + +3. **配置回调域名**：在**开发配置** > **安全设置**，在**重定向 URL（回调域名）**中填入 `https://account-enterprise.bailian.aliyunportal.com/api/v1/auth/dingtalk/callback`。 + +4. **开通通讯录读权限**：在应用的**权限管理**页面，开通**通讯录个人信息读权限**。 + +5. **发布应用**：在**应用发布** > **版本管理与发布**页面，创建新版本并发布。 + +6. **在 Token Plan 管理平台填入凭证**：在 Token Plan 控制台**我的订阅**页面的**团队版**卡片点击**设置**，在**SSO 配置**区域切换到**钉钉**选项卡，填入配置名称、钉钉应用 AppKey 和 AppSecret。 + +7. **把管理平台地址分享给成员**：在**设置**页面的**基本信息**区域复制**管理平台地址**，成员访问该地址选择钉钉登录即可加入组织。 + + +## **席位操作** + +### **查看席位状态** + +在控制台**我的订阅**页面的**团队座席**区域查看各档位席位的已分配数量和总数；在**订阅明细**区域可查看每个席位的状态和到期时间。 + +### **分配席位** + +1. 在**成员管理**页面，找到目标成员，点击操作列的**分配席位**。 + +2. 在弹窗中选择席位版本，点击**确定**。 + + +分配后系统自动生成 API Key，连同 Base URL 发给成员即可调用模型。 + +### **回收席位** + +在**成员管理**页面，找到目标成员，点击操作列的**回收席位**，在确认弹窗中点**确定**。回收后席位转为未分配，原成员将无法使用席位的 Credits；重新分配后系统会为新成员生成新的 API Key。 + +### **加购席位** + +在 Token Plan 控制台的**我的订阅**页面，点击**加购座席**，选择席位档位和数量后提交订单。新加席位与现有订阅统一到期，费用按剩余时长折算。 + +### **升级席位** + +在 Token Plan 控制台的**我的订阅**页面，在**订阅明细**中找到目标席位，点击**升级**，选择更高档位后提交订单。需要批量操作时，勾选多个席位后点击**批量升级**。升级按剩余时长补缴差价。 + +## **用量分析** + +用量分析可从 Token Plan 控制台的我的订阅页面或管理平台进入。在**用量分析**页面，所有者可查看： + +- **用量趋势**：近 1、7、30 天的 Credits 消耗趋势图。 + +- **模型用量**：组织内各模型的 Credits 消耗。 + +- **成员用量**：各成员的 Credits 消耗。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-overview.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-overview.md new file mode 100644 index 00000000..4b4358d0 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-overview.md @@ -0,0 +1,314 @@ +# 概述 + +Token Plan 团队版是阿里云百炼推出的 AI 大模型订阅服务，以 Credits 统一计量，支持文本生成、图像生成与视频生成模型，兼容主流 AI 编程与智能体工具，提供团队管理后台、数据安全保障，调用平稳运行。 + +**说明** + +Token Plan 团队版目前仅支持**华北2（北京）**地域。 + +## **产品简介** + +Token Plan 团队版整合千问和三方模型，支持文本生成、图像生成与视频生成。通过 Credits 统一计量，同一订阅可在多种 AI 工具中使用。 + +- **多模型灵活切换**：支持多模型按需切换，按 Credits 统一抵扣。 + +- **兼容多种工具**：适配多种主流编程工具及热门 Agent 工具。控制台提供快速接入 AI 工具入口，支持 Qwen Code、Claude Code、Qoder、Qoder CN、OpenClaw 等工具接入。 + +- **多档位套餐**：提供标准坐席、高级坐席、尊享坐席多档位套餐，匹配不同使用强度。 + +- **团队管理**：提供管理后台，支持席位分配与回收、成员用量分析等团队管理能力。 + +- **预算可控**：支持按月或按年订阅，预算可控。 + +- **数据安全**：承诺不使用对话数据进行模型训练，满足企业级数据隐私要求。 + +- **平稳运行**：多租户隔离架构，调用高峰期间不排队。 + + +Token Plan 团队版提供套餐专属 Base URL，兼容 OpenAI、Anthropic 接口标准，具体地址可在控制台**我的订阅**的 API Key 区域查看。 + +## **支持的模型** + +**重要** + +团队版支持 qwen3.8-max-preview 预览模型，享有以下限时权益： + +1. **预览版**：qwen3.8-max-preview 当前为预览版本，**预览期间模型能力会持续迭代升级**。预览结束后该模型会下线或替换成正式版本。 + +2. **限时加量 10 倍**：限时活动期间，模型调用 Credits 消耗低至 1 折，相当于增加 10 倍用量。 + + +阿里云百炼有权根据运营情况对活动进行变更或调整，包括不限于活动内容和有效期等，请以页面最新内容或阿里云通知为准。 + +**品牌** + +**模型 ID（Model ID）** + +**模型能力** + +千问 + +qwen3.8-max-preview + +推理模型、视觉理解、文本生成 + +qwen3.7-max + +推理模型、文本生成 + +qwen3.7-plus + +推理模型、视觉理解、文本生成 + +qwen3.6-plus + +推理模型、视觉理解、文本生成 + +qwen3.6-flash + +推理模型、视觉理解、文本生成 + +qwen-image-2.0 + +图片生成 + +qwen-image-2.0-pro + +图片生成 + +万相 + +wan2.7-image + +图片生成 + +wan2.7-image-pro + +图片生成 + +DeepSeek + +deepseek-v4-pro + +推理模型、文本生成 + +deepseek-v4-flash + +推理模型、文本生成 + +deepseek-v3.2 + +推理模型、文本生成 + +月之暗面 + +kimi-k2.7-code + +推理模型、视觉理解、文本生成 + +kimi-k2.6 + +推理模型、视觉理解、文本生成 + +kimi-k2.5 + +推理模型、视觉理解、文本生成 + +智谱 AI + +glm-5.2 + +推理模型、文本生成 + +glm-5.1 + +推理模型、文本生成 + +glm-5 + +推理模型、文本生成 + +MiniMax + +MiniMax-M2.5 + +推理模型、文本生成 + +HappyHorse + +happyhorse-1.1-i2v + +视频生成 + +happyhorse-1.1-t2v + +视频生成 + +happyhorse-1.1-r2v + +视频生成 + +## **套餐与定价** + +前往 Token Plan 团队版购买页面选择坐席类型、数量和订阅周期，完成订阅。主账号和 RAM 账号均可订阅。订阅周期支持按月购买、按年购买、连续包月包年。 + +### **限时活动** + +即日起至 2026 年 7 月 22 日 23:59（UTC+8），qwen3.7-max 模型 Credits 消耗减半，同时支持隐式缓存。 + +### **Token Plan 团队版** + +提供标准坐席、高级坐席、尊享坐席三个档位，匹配不同使用强度。 + +席位（坐席）是 Token Plan 团队版的最小订阅单位，代表一个团队成员的使用名额。管理员在团队管理中将席位分配给成员后，系统自动为该成员生成专属的 API Key。每个席位绑定一个成员、对应一个 API Key，不可共享。 + +**坐席类型** + +**价格** + +**额度** + +**适用场景** + +标准坐席 + +原价 ¥198/坐席/月 +限时 **¥150/坐席/月** + +25,000 Credits/坐席/月 + +轻度使用 AI 辅助的团队成员 + +高级坐席 + +原价 ¥698/坐席/月 +限时 **¥550/坐席/月** + +100,000 Credits/坐席/月 + +日常高频使用 AI 编程或办公的团队成员 + +尊享坐席 + +¥1,398/坐席/月 + +250,000 Credits/坐席/月 + +重度依赖 AI 的核心开发者或高强度使用者 + +**重要** + +限时优惠仅适用于包月订阅的新购、续费和自动续费，包年订阅和升级坐席不参与。加购坐席时按剩余时长折算费用，实际收费取折算金额与限时价中的较低值。 + +### **Token Plan 团队版 - 共享用量包** + +跨坐席共享的弹性用量包，当个别坐席用量超出套餐额度时，可从共享用量包中抵扣。每个共享用量包有效期为 1 个月，到期未使用的额度自动清零。持有多个共享用量包时，优先抵扣最近到期的用量包。 + +**档位** + +**价格** + +**额度** + +Token Plan 团队版 - 共享用量包 + +¥5,000/个 + +625,000 Credits/个 + +## **Credits 计费机制** + +### **计费说明** + +单次消耗的 Credits 由模型类型、Token 用量、思考模式及工具调用等动态决定，实际消耗以控制台订阅页用量明细为准。 + +### **计算示例** + +以 qwen3.6-plus 为例，预估单次请求消耗明细如下（不同模型的单价不同，实际以账单为准）： + +**Token 类型** + +**数量** + +**消耗 Credits** + +输入 tokens + +8,349 + +1.67 + +缓存 tokens + +40,794 + +0.82 + +输出 tokens + +573 + +0.69 + +**合计** + +**约 3.18 Credits** + +### **抵扣顺序** + +1. 优先从坐席套餐的月度额度中抵扣。 + +2. 坐席额度用尽后，从共享用量包中抵扣。持有多个共享用量包时，优先抵扣最近到期的用量包。 + +3. 全部额度用尽后，服务将暂停至下一计费周期或购买共享用量包补充额度。 + + +## **查看额度消耗情况** + +**通过控制台**：登录 Token Plan 控制台，在**我的订阅**页面查看总额度使用百分比、重置时间、团队席位分配情况，以及各席位与共享用量包的状态和到期时间。 + +**通过团队管理平台**：在**用量分析**页面，可查看近 1、7、30 天的 Credits 消耗趋势、各模型用量，以及每个成员的消耗明细。详见团队管理。 + +## **订阅管理** + +### **加购席位** + +在 Token Plan 控制台的**我的订阅**页面，点击**加购座席**，选择席位档位和数量后提交订单。新加席位与现有订阅统一到期，费用和 Credits 额度均按剩余时长折算。 + +### **升级席位** + +在**订阅明细**中找到目标席位，点击**升级**，选择更高档位后提交订单。需要批量操作时，勾选多个席位后点击**批量升级**。升级按剩余时长补缴差价。 + +### **续费** + +- 点击**续费**按钮，续费周期与订阅时一致（按月订阅则按月续费，按年订阅则按年续费），到期前完成可避免服务中断。 + +- **自动续费**：点击**开启自动续费**，确认后次日生效，到期前 9 天系统按订阅周期自动扣款续费。如需关闭，点击**关闭自动续费**并确认。 + + +### **退订席位** + +在**订阅明细**中点击席位的**退订**，按席位维度退订；已有用量消耗的席位不可退订。也可勾选多个席位后点击**批量退订**。退款原路退回支付账户，预计 1-3 个工作日到账。 + +### **加购共享用量包** + +在**共享用量包**区域点击**前往购买**。 + +### **注意事项** + +- 退订重购后，API Key 和 Base URL 会发生变更，需在工具中重新配置。新 API Key 和 Base URL 均可在控制台[**我的订阅**](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/enterprise)页面的 API Key 区域获取。 + +- 续费仅延长订阅有效期，不会叠加补充至当前计费周期的额度。 + +- 团队版与个人版可同时持有，各自独立计费。 + + +## **使用细则** + +1. **使用范围**：仅限在兼容的 AI 编程和智能体工具中交互式使用，不可用于自动化脚本或应用后端。违规使用可能导致订阅暂停或 API Key 封禁。 + +2. **数据安全**：Token Plan 团队版不会使用对话数据训练模型。 + +3. **账号规范**：API Key 仅限已分配席位的成员本人使用，不可共享或公开泄露。 + +4. **服务地域**：Token Plan 团队版目前仅在特定地域提供服务，如需从海外调用，请确认符合当地法律法规要求。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-quickstart.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-quickstart.md similarity index 64% rename from skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-quickstart.md rename to skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-quickstart.md index 4486ad53..8aa9ac0d 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-quickstart.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-quickstart.md @@ -4,16 +4,18 @@ ## **步骤一：订阅 Token Plan 团队版** -访问 [Token Plan 团队版购买页面](https://common-buy.aliyun.com/token-plan/)，选择坐席类型、数量和订阅周期（按月或按年）并完成订阅，主账号和 RAM 账号均可订阅。 +访问 [Token Plan 团队版购买页面](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)，选择套餐档位和订阅周期并完成订阅，主账号和 RAM 账号均可订阅。 -购买须知： - -- RAM 子账号订阅前，需主账号在 RAM 控制台授予 AliyunBailianFullAccess 权限。 +- **RAM 用户授权**：RAM 用户使用 Token Plan 前，需由主账号完成以下授权： + 1. 在 [RAM 控制台](https://ram.console.aliyun.com/)为该 RAM 用户授予 `AliyunTokenPlanReadOnlyAccess`（只读）或 `AliyunTokenPlanFullAccess`（管理）系统策略，同时授予 `AliyunBSSReadOnlyAccess` 系统策略。 + + 2. 在百炼控制台[账号管理](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)页面，为该 RAM 用户分配管理员或订阅套餐权限。 + ## **步骤二：获取 API Key 和 Base URL** -- **API Key**：在 Token Plan 控制台的[成员管理页面](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)或管理平台创建成员账号，分配席位后，为成员生成 API Key。详见[团队管理](https://help.aliyun.com/zh/model-studio/token-plan-team)。 +- **API Key**：在 Token Plan 控制台的成员管理页面或管理平台创建成员账号，分配席位后，为成员生成 API Key。详见团队管理。 API Key 使用须知： @@ -53,6 +55,8 @@ Token Plan、Coding Plan 和按量付费的 API Key 与 Base URL 完全隔离， ## **步骤三：接入 AI 工具** +将 API Key 和 Base URL 配置到 AI 工具中，即可开始使用。 + [**OpenClaw**开源、自托管个人 AI 助手](https://help.aliyun.com/zh/model-studio/openclaw) [**Hermes Agent**开源 AI 代理框架，内置自学习循环](https://help.aliyun.com/zh/model-studio/hermes-agent) @@ -83,17 +87,10 @@ Token Plan、Coding Plan 和按量付费的 API Key 与 Base URL 完全隔离， [··· **更多工具**其他编程工具](https://help.aliyun.com/zh/model-studio/more-tools) -## **可选：接入图像生成模型** +## **可选：接入多模态生成模型** -Token Plan 团队版支持图像生成模型（qwen-image-2.0、wan2.7-image 等）。图像生成模型使用独立的接口，需要通过 AI 工具的 Skill 或扩展机制接入。具体配置方法请参见[接入多模态生成模型](https://help.aliyun.com/zh/model-studio/token-plan-multimodal-gen)。 +Token Plan 团队版支持多模态生成模型（wan2.7-image、happyhorse-1.1-t2v 等）。多模态生成模型使用独立的接口，需要通过 AI 工具的 Skill 或扩展机制接入，详见[接入多模态生成模型](https://help.aliyun.com/zh/model-studio/token-plan-multimodal-gen)。 -## **可选：工具调用** - -通过接入工具调用，模型可以在对话中调用联网搜索、代码解释器等扩展能力。 - -- qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash：内置联网搜索、代码解释器、网页抓取、以图搜图、文搜图 5 个工具，通过 Responses API 直接调用。内置工具不额外收费，产生的 token 消耗统一从套餐 Credits 中抵扣。 - -- 其他模型：通过 MCP 服务接入工具。 - +## **可选：接入 Harness 工具** -详细说明请参见[工具调用](https://help.aliyun.com/zh/model-studio/token-plan-tool)。 +通过 Harness 工具调用，模型可以在对话中调用联网搜索、文搜图、图搜图、网页抓取、代码解释器等扩展能力。当前仅 qwen3.7、qwen3.8 支持原生工具调用，通过 Responses API 直接调用，Harness 工具按抵扣系数消耗 Credits。详见[接入 Harness 工具](https://help.aliyun.com/zh/model-studio/token-plan-harness-tool)。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team.md deleted file mode 100644 index 9e83ccff..00000000 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/token-plan-guide/token-plan-team.md +++ /dev/null @@ -1,272 +0,0 @@ -# 团队管理 - -在 Token Plan 控制台或管理平台中添加和管理团队成员、分配和回收席位、监控 Credits 用量。 - -## **访问入口** - -**阿里云主账号或 RAM 用户**：登录[Token Plan 控制台](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/token-plan)，在**我的订阅**页面进行团队管理：在**团队版**卡片点击**设置**修改组织名称、登录方式（SSO/钉钉）等；在**订阅明细**区域点击**分配座席**进入成员管理。 - -**说明** - -RAM 用户使用 Token Plan 前，需由主账号完成以下授权： - -1. 在[RAM 控制台](https://ram.console.aliyun.com/users)为该 RAM 用户授予 `AliyunTokenPlanReadOnlyAccess`（只读）或 `AliyunTokenPlanFullAccess`（管理）系统策略，具体权限范围按实际需要选择。 - -2. 在百炼控制台[**账号管理**](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/user_management/user_management)页面，为该 RAM 用户分配**管理员**或**订阅套餐**权限。 - - -通过 **SSO 或钉钉**加入的成员：通过管理员分发的**管理平台地址**登录管理平台。管理平台地址可在**设置**页面的**基本信息**区域获取。 - -## **角色与权限** - -**角色** - -**权限** - -所有者 - -添加或移除成员、分配或回收席位、修改成员角色、查看全部成员和模型的用量 - -管理员 - -权限范围与所有者相同。管理员由所有者授予，可被移除或降级。 - -成员 - -使用管理员分配的 API Key 和 Base URL 调用模型 - -## **成员管理** - -### **添加成员** - -- **手动添加**（不能登录管理平台，仅供 API 调用）：在**成员管理**页面点**添加成员**，在弹窗中填用户名（仅支持英文字母、数字、下划线）和角色，可选择同时分配席位（默认不分配，需加购可点弹窗内加购更多席位）。在该成员的操作列点**分配席位**选择席位版本，分配后系统自动生成 API Key，连同 Base URL 发给成员即可调用模型。 - -- **SSO 或钉钉登录**（成员可登录管理平台，自管席位和 API Key）：完成 [SAML](#tp05-sso) 或[钉钉](#tp05-dingtalk)接入配置后，成员从登录页对应入口登录即自动加入。 - - -### **修改成员角色** - -在**成员管理**页面，找到目标成员，点击操作列的**修改角色**，在弹窗中选择新角色（管理员或成员）后保存。所有者角色不可修改。 - -### **重置 API Key** - -在**成员管理**页面，找到目标成员，点击操作列的**重置**。重置后原 API Key 立即失效，新 API Key 需重新发给成员。 - -### **移出成员** - -在**成员管理**页面，找到目标成员，点击操作列的**移出组织**。移出后席位自动回收、API Key 立即失效。 - -## **SAML 接入** - -通过标准 SAML 2.0 对接企业 IdP，对应登录页的**SSO**入口。配置后，成员用 IdP 账号登录管理平台即自动加入组织。 - -### **SAML 配置** - -**前提条件**：组织内有成员时无法编辑 SSO 配置，需先全部移出。 - -1. 登录 Token Plan 控制台，在**我的订阅**页面的**团队版**卡片点击**设置**，找到**SSO 配置**区域。 - -2. 从企业 IdP 获取 IdP 信息（IdP Entity ID、IdP SSO URL、IdP Certificate）。点击**编辑**，填入自定义的 SP Entity ID 和上述 IdP 信息，保存。 - -3. 保存后系统自动生成 ACS URL。将 SP Entity ID 和 ACS URL 填入企业 IdP 的 SSO 应用配置中。 - -4. 在**基本信息**区域复制**管理平台地址**，分享给团队成员。成员访问该地址，在登录页选择 SSO 登录方式即可加入组织。 - - -**参数说明** - -**SP 信息**（百炼侧） - -**参数** - -**说明** - -SP Entity ID - -百炼在 SSO 流程中的唯一标识，自定义填写，需同步填入企业 IdP 的 SSO 应用配置。 - -ACS URL - -IdP 认证成功后回传响应的地址，保存 SSO 配置后由百炼自动生成，需填入企业 IdP。 - -SP Certificate - -百炼侧 SAML 签名证书，保存 SSO 配置后由系统自动生成，用于企业 IdP 验证百炼侧响应签名，需同步配置到企业 IdP 信任链。 - -**IdP 信息**（企业侧，需从企业 IdP 获取后填入） - -**参数** - -**说明** - -IdP Entity ID - -企业身份提供商的唯一标识（Entity ID）。 - -IdP SSO URL - -企业 IdP 的登录入口地址，成员登录时跳转到此地址进行认证。 - -IdP Certificate - -企业 IdP 的签名证书，百炼用来验证响应确实来自该企业。 - -### **配置示例：**[**阿里云 IDaaS**](https://www.aliyun.com/product/idaas) - -**前提条件**：已开通阿里云 IDaaS EIAM 实例。如未开通，登录[IDaaS 控制台](https://yundun.console.aliyun.com/?p=idaas)，在 EIAM 云身份服务实例列表页点击**免费创建实例**。 - -1. **在 IDaaS 中创建 SAML 应用** - - 登录 IDaaS 实例管理平台，进入**应用管理**，点击**添加应用**，选择**标准 SAML 2.0**应用模板，填写应用名称后创建。 - -2. **从 IDaaS 获取 IdP 信息并填入百炼** - - 在 IDaaS 应用的**单点登录**页面底部的**应用配置信息**区域，复制 IdP 信息。然后在 Token Plan 控制台**我的订阅**页面的**团队版**卡片点击**设置**，在 **SSO 配置**区域，点击**编辑**，填入自定义的 SP Entity ID 和以下对应的 IdP 信息后保存： - - **IDaaS 应用配置信息** - - **填入百炼的字段** - - **说明** - - IdP 唯一标识（IdP Entity ID） - - IdP Entity ID - - IDaaS 在应用中的唯一标识 - - IdP SSO 地址（IdP Sign-in URL） - - IdP SSO URL - - 成员 SSO 登录时跳转到的认证地址 - - 公钥证书（Certificate） - - IdP Certificate - - 点击“复制证书内容”获取完整证书 - -3. **将 SP Entity ID 和 ACS URL 填入 IDaaS** - - 保存后，百炼 SSO 配置区域显示自动生成的 ACS URL。在 IDaaS 应用的**登录访问** > **单点登录**页面，填入以下对应参数： - - **百炼 SP 信息** - - **填入 IDaaS 的字段** - - **说明** - - SP Entity ID - - 应用唯一标识（SP Entity ID） - - 填入在百炼中自定义的 SP Entity ID - - ACS URL - - 单点登录地址（ACS URL） - - 填入百炼自动生成的 ACS URL - -4. **在 IDaaS 中创建账户并授权** - - 在 IDaaS 实例管理平台的**账户管理**中，为需要使用 Token Plan 的团队成员创建 IDaaS 账户。 - - 然后在 SAML 应用的**登录访问** > **应用授权**页面，点击**添加授权**，选择需要通过 SSO 登录的账户、组或组织机构。后续添加用户重复此步骤即可，无需重新配置 SSO 参数。 - -5. **分享管理平台地址给成员** - - 在**设置**页面的**基本信息**区域复制**管理平台地址**，分享给已授权的成员。成员访问该地址，在登录页选择 SSO 登录方式即可加入组织。 - - -## **钉钉接入** - -使用钉钉作为身份系统的企业可直接接入，对应登录页的**钉钉**入口。配置后，成员用钉钉账号登录管理平台即自动加入组织。 - -**前提条件**：已在钉钉中创建企业，并将待登录的成员加入企业。 - -1. **创建钉钉企业内部应用** - - 登录[钉钉开发者后台](https://open-dev.dingtalk.com/)，进入**应用开发** > **企业内部应用** > **钉钉应用**，点击**创建应用**。填写应用名称和描述后保存。 - -2. **获取应用凭证** - - 在应用详情页，进入**基础信息** > **凭证与基础信息**，记录 Client ID 和 Client Secret（即钉钉应用 AppKey 与 AppSecret），后续在 Token Plan 管理平台中使用。 - -3. **配置回调域名** - - 在应用详情页，进入**开发配置** > **安全设置**，在**重定向 URL（回调域名）**中填入 `https://account-enterprise.bailian.aliyunportal.com/api/v1/auth/dingtalk/callback`，保存。 - -4. **开通通讯录读权限** - - 进入应用的**权限管理**页面，开通**通讯录个人信息读权限**。该权限用于在登录时识别成员的钉钉账号。 - -5. **发布应用** - - 进入**应用发布** > **版本管理与发布**页面，创建新版本并发布。 - -6. **在 Token Plan 管理平台填入凭证** - - 在 Token Plan 控制台**我的订阅**页面的**团队版**卡片点击**设置**，在**SSO 配置**区域，切换到**钉钉**选项卡，填入配置名称、**钉钉应用 AppKey**和**钉钉应用 AppSecret**（步骤 2 中获取），保存。 - -7. **把管理平台地址分享给成员** - - 在**设置**页面的**基本信息**区域复制**管理平台地址**，分享给团队成员。成员访问该地址，在登录页选择钉钉登录方式即可加入组织。 - - -## **席位操作** - -### **查看席位状态** - -在控制台**我的订阅**页面的**团队座席**区域查看各档位席位的已分配数量和总数；在**订阅明细**区域可查看每个席位的状态和到期时间。 - -### **分配席位** - -1. 在**成员管理**页面，找到目标成员，点击操作列的**分配席位**。 - -2. 在弹窗中选择席位版本，点击**确定**。 - - -分配后系统自动生成 API Key，连同 Base URL 发给成员即可调用模型。 - -### **回收席位** - -在**成员管理**页面，找到目标成员，点击操作列的**回收席位**，在确认弹窗中点**确定**。回收后席位转为未分配，原成员将无法使用席位的 Credits；重新分配后系统会为新成员生成新的 API Key。 - -### **加购席位** - -在[Token Plan 控制台](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/token-plan)的**我的订阅**页面，点击**加购座席**，选择席位档位和数量后提交订单。 - -新加席位与现有订阅统一到期，费用按剩余时长折算。例如订阅周期剩余 15 天，加购一个标准席位，则只需支付 15 天对应的费用。 - -### **升级席位** - -在[Token Plan 控制台](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/token-plan)的**我的订阅**页面，在**订阅明细**中找到目标席位，点击**升级**，选择更高档位后提交订单。需要批量操作时，勾选多个席位后点击**批量升级**。 - -升级按剩余时长补缴差价。例如将标准席位升级为高级席位，只需补缴剩余时长内两个档位的差价。 - -## **用量分析** - -用量分析可从 Token Plan 控制台的我的订阅页面或管理平台进入。在**用量分析**页面，所有者可查看： - -- **用量趋势**：近 1、7、30 天的 Credits 消耗趋势图。 - -- **模型用量**：组织内各模型的 Credits 消耗。 - -- **成员用量**：各成员的 Credits 消耗。 - - -## **常见问题** - -### **席位到期后 API Key 是否仍可使用** - -席位到期后，对应的 API Key 将无法调用模型。续订套餐并重新分配席位后即可恢复使用。 - -### **回收席位后能否分配给其他成员** - -可以。回收后席位转为未分配，重新分配给其他成员后系统会生成新的 API Key。 - -### **能否为同一成员更换席位** - -每个成员同一时间只能持有一个席位。如需更换，先回收当前席位，再分配新的席位即可。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/model-training-best-practices.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/model-training-best-practices.md index 9c756531..727165e7 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/model-training-best-practices.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/model-training-best-practices.md @@ -36,7 +36,7 @@ 在阿里云百炼，**完成调优的模型必须部署后才能调用和评测**。因此，您需要首先完成模型部署，方可继续评测模型。 -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/3946933871/CAEQSBiBgICa8texgxkiIGU5OGM2ZmE4ZjJjNDRhMWY5MmMwYmQwZmIzZWIyODM04430647_20240626141408.264.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/6543415871/CAEQSBiBgICa8texgxkiIGU5OGM2ZmE4ZjJjNDRhMWY5MmMwYmQwZmIzZWIyODM04430647_20240626141408.264.svg) ## **前提条件** @@ -149,7 +149,7 @@ 2. **数据集类型**：包含**训练集**和**评测集**。每次新增数据集时，您只能选择一种类型的数据集。**训练集**由一轮或多轮的Prompt+Completion数据组成，而**评测集**仅包含Prompt数据。您需要分别创建这两种数据集。 - 3. **存储位置**：默认为**平台存储**。 + 3. **存储位置**：默认为**平台OSS存储**。 4. **导入方式**：默认为**本地上传**。 @@ -241,11 +241,11 @@ ### **新建训练任务** -在[模型调优](https://bailian.console.aliyun.com/#/efm/model_manager)页面，点击**训练新模型**，即可设置训练参数。具体操作如下： +在[模型调优](https://bailian.console.aliyun.com/#/efm/model_manager)页面，点击**创建训练任务**，即可设置训练参数。具体操作如下： 1. 前往**[模型调优](https://bailian.console.aliyun.com/#/efm/model_manager)**页面。这里展示了所有调优任务。 -2. 点击**训练新模型**，阿里云百炼将会引导您配置训练参数： +2. 点击**创建训练任务**，阿里云百炼将会引导您配置训练参数： 1. **选择模型训练方式**：阿里云百炼支持**SFT微调训练**、**DPO偏好训练**和**CPT继续预训练**三种方式，以下是选择建议： diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api-by-vanchin.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api-by-vanchin.md index f13ea9b6..c4e229b1 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api-by-vanchin.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api-by-vanchin.md @@ -31,7 +31,8 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -92,7 +93,8 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); let reasoningContent = ''; // 完整思考过程 @@ -161,7 +163,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -222,7 +225,8 @@ import os client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -240,7 +244,8 @@ import OpenAI from "openai"; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: "https://dashscope.aliyuncs.com/compatible-mode/v1", + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", }); const completion = await openai.chat.completions.create({ @@ -254,7 +259,8 @@ console.log(completion.choices[0].message.content); ## **curl** ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -280,7 +286,8 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -326,7 +333,8 @@ import process from 'process'; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); async function main() { @@ -373,7 +381,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api.md index fe1e76b9..84527826 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api.md @@ -4,19 +4,19 @@ **重要** -deepseek-v3、deepseek-v3.1、deepseek-v3.2、deepseek-v3.2-exp、deepseek-r1、deepseek-r1-0528、deepseek-r1-distill-qwen-7b/14b/32b 将于**2026年7月9日**下架。推荐转用：[qwen3.7-plus](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.7-plus)、[qwen3.7-max](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.7-max)、[qwen3.6-flash](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.6-flash)。 +deepseek-v3、deepseek-v3.1、deepseek-v3.2、deepseek-v3.2-exp、deepseek-r1、deepseek-r1-0528、deepseek-r1-distill-qwen-7b/14b/32b 将于**2026年10月10日**下架。推荐转用：[qwen3.7-plus](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.7-plus)、[qwen3.7-max](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.7-max)、[qwen3.6-flash](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/detail/qwen3.6-flash)。 ## **服务接入地址** -不同地域的服务接入地址不同，请根据您选择的地域配置对应的 Base URL。各地域可调用的模型及限流不同，请参见[限流](https://help.aliyun.com/zh/model-studio/rate-limit)文档。 +不同地域的服务接入地址不同，请根据您选择的地域配置对应的 Base URL（调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)）。各地域可调用的模型及限流不同，请参见[限流](https://help.aliyun.com/zh/model-studio/rate-limit)文档。 ## **OpenAI兼容** ## 华北2（北京） -SDK 调用配置的`base_url`：`https://dashscope.aliyuncs.com/compatible-mode/v1` +SDK 调用配置的`base_url`：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -HTTP 请求地址：`POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions` +HTTP 请求地址：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions` ## 美国（弗吉尼亚） @@ -30,31 +30,25 @@ SDK 调用配置的`base_url`：`https://{WorkspaceId}.ap-southeast-1.maas.aliyu HTTP 请求地址：`POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions` -调用时请将`WorkspaceId`替换为真实的 Workspace ID。 - ## 德国（法兰克福） SDK 调用配置的`base_url`：`https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/compatible-mode/v1` HTTP 请求地址：`POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions` -调用时请将`WorkspaceId`替换为真实的 Workspace ID。 - ## 日本（东京） SDK 调用配置的`base_url`：`https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1` HTTP 请求地址：`POST https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions` -调用时请将`WorkspaceId`替换为真实的 Workspace ID。 - ## **DashScope** ## 华北2（北京） -HTTP 请求地址为`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation` +HTTP 请求地址为`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation` -SDK 调用无需配置 `base_url`。 +SDK 调用配置的`base_url`：`dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"` ## 美国（弗吉尼亚） @@ -68,24 +62,18 @@ HTTP 请求地址为`POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com SDK 调用配置的`base_url`：`dashscope.base_http_api_url = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1"` -调用时请将`WorkspaceId`替换为真实的 Workspace ID。 - ## 德国（法兰克福） HTTP 请求地址为`POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation` SDK 调用配置的`base_url`：`dashscope.base_http_api_url = "https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1"` -调用时请将`WorkspaceId`替换为真实的 Workspace ID。 - ## 日本（东京） HTTP 请求地址为`POST https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation` SDK 调用配置的`base_url`：`dashscope.base_http_api_url = "https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/api/v1"` -调用时请将`WorkspaceId`替换为真实的 Workspace ID。 - ## **快速开始** deepseek-v4-pro 是 DeepSeek 系列最新模型，在编程、数学和通用任务方面表现出色。您可以通过`enable_thinking`参数在思考与非思考模式之间切换。以下示例展示如何调用思考模式的 deepseek-v4-pro 模型。 @@ -109,6 +97,7 @@ import os client = OpenAI( # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) messages = [{"role": "user", "content": "你是谁"}] @@ -174,6 +163,7 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); let reasoningContent = ''; // 完整思考过程 @@ -253,6 +243,7 @@ Request ID: chatcmpl-a1b2c3d4-e5f6-7890-abcd-ef1234567890 ## **curl** ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -280,8 +271,9 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/ ``` import os +import dashscope from dashscope import Generation -# 以下为华北2（北京）地域的配置，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的配置不同。 +# 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" # 初始化请求参数 messages = [{"role": "user", "content": "你是谁？"}] @@ -357,8 +349,10 @@ import io.reactivex.Flowable; import java.lang.System; import java.util.Arrays; public class Main { - // 以下为华北2（北京）地域的配置，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的配置不同。 + static { + // 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; + } private static StringBuilder reasoningContent = new StringBuilder(); private static StringBuilder finalContent = new StringBuilder(); private static boolean isFirstPrint = true; @@ -434,6 +428,7 @@ public class Main { ## **curl** ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation" \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -469,6 +464,7 @@ import os client = anthropic.Anthropic( # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic", ) @@ -496,6 +492,7 @@ for event in message: ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic/v1/messages \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -514,11 +511,9 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic/v ## **推理强度（reasoning\_effort）** -deepseek-v4-pro 和 deepseek-v4-flash 默认开启思考模式。通过`reasoning_effort`参数可以调整推理强度，可选值为`high`和`max`，默认为`high`。 - -**说明** +deepseek-v4-pro 和 deepseek-v4-flash 默认开启思考模式。通过`reasoning_effort`参数可以调整推理强度，可选值为`low`、`medium`、`high`、`xhigh`和`max`，默认为`high`。 -设为`low`或`medium`时会映射为`high`，设为`xhigh`时会映射为`max`。 +其中，`low`和`medium`的效果等同于`high`；`xhigh`的效果等同于`max`。 ## **OpenAI兼容** @@ -529,6 +524,7 @@ from openai import OpenAI import os client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -545,6 +541,7 @@ print(completion.choices[0].message.content) import OpenAI from "openai"; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", }); const completion = await openai.chat.completions.create({ @@ -558,6 +555,7 @@ console.log(completion.choices[0].message.content); ## **curl** ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -572,8 +570,9 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/ ``` import os +import dashscope from dashscope import Generation -# 以下为华北2（北京）地域的配置，调用时请将WorkspaceId替换为真实的业务空间ID，各地域的配置不同。 +# 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" response = Generation.call( api_key=os.getenv("DASHSCOPE_API_KEY"), diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm-zhipu.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm-zhipu.md index 155b701e..7c153915 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm-zhipu.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm-zhipu.md @@ -45,6 +45,7 @@ import os client = OpenAI( # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) @@ -135,6 +136,7 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); @@ -229,6 +231,7 @@ main(); ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -291,6 +294,7 @@ import os client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) @@ -348,7 +352,8 @@ import process from 'process'; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); const tools = [ @@ -416,6 +421,7 @@ main(); ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -474,6 +480,7 @@ import os client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) @@ -502,11 +509,12 @@ print(completion.usage.prompt_tokens) # true 时少于 false ## **curl** ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ - "model": "glm-5.2", + "model": "ZHIPU/GLM-5.2", "messages": [ {"role": "user", "content": "请计算 15 * 23 是多少？"}, {"role": "assistant", "content": "15 乘以 23 等于 345。", "reasoning_content": "15 * 23 = 345"}, diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm.md index 0b7948dc..e5181358 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/glm.md @@ -14,9 +14,9 @@ glm-4.6、glm-4.7 将于**2026年7月9日**下架。推荐转用：[qwen3.7-plus ## 华北2（北京） -SDK 调用配置的`base_url`：`https://dashscope.aliyuncs.com/compatible-mode/v1` +SDK 调用配置的`base_url`：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -HTTP 请求地址：`POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions` +HTTP 请求地址：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions` ## 美国（弗吉尼亚） @@ -30,23 +30,42 @@ SDK 调用配置的`base_url`：`https://{WorkspaceId}.eu-central-1.maas.aliyunc HTTP 请求地址：`POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## 新加坡 SDK 调用配置的`base_url`：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` HTTP 请求地址：`POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions` -调用时请将`WorkspaceId`替换为真实的 [Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## DashScope ## 华北2（北京） -HTTP 请求地址：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation` +HTTP 请求地址：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation` + +SDK 调用配置的`base_url`： -SDK 调用无需配置 `base_url`。 +## **Python** + +``` +dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' +``` + +## **Java** + +- **方式一：** + + ``` + import com.alibaba.dashscope.protocol.Protocol; + Generation gen = new Generation(Protocol.HTTP.getValue(), "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"); + ``` + +- **方式二：** + + ``` + import com.alibaba.dashscope.utils.Constants; + Constants.baseHttpApiUrl="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; + ``` + ## 美国（弗吉尼亚） @@ -81,22 +100,16 @@ dashscope.base_http_api_url = 'https://dashscope-us.aliyuncs.com/api/v1' HTTP 请求地址：`POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation` -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - SDK 调用配置的`base_url`： ## **Python** -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ``` dashscope.base_http_api_url = 'https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/api/v1' ``` ## **Java** -调用时请将`WorkspaceId`替换为真实的[Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - - **方式一：** ``` @@ -118,7 +131,7 @@ HTTP 请求地址为`POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com SDK 调用配置的`base_url`：`dashscope.base_http_api_url = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1"` -调用时请将`WorkspaceId`替换为真实的 [Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **快速开始** @@ -144,7 +157,8 @@ import os client = OpenAI( # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) messages = [{"role": "user", "content": "你是谁"}] @@ -233,7 +247,8 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); let reasoningContent = ''; // 完整思考过程 @@ -325,7 +340,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -525,7 +541,8 @@ public class Main { ## curl ``` -curl -X POST "https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation" \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation" \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -H "X-DashScope-SSE: enable" \ @@ -560,7 +577,8 @@ import os client = anthropic.Anthropic( # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/apps/anthropic", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic", ) message = client.messages.create( @@ -587,7 +605,8 @@ for event in message: ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/apps/anthropic/v1/messages \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic/v1/messages \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -H "anthropic-version: 2023-06-01" \ @@ -645,7 +664,8 @@ import os client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) tools = [ @@ -702,7 +722,8 @@ import process from 'process'; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); const tools = [ @@ -774,7 +795,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -867,7 +889,8 @@ for chunk in completion: ## curl ``` -curl -X POST "https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation" \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation" \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -H "X-DashScope-SSE: enable" \ @@ -950,7 +973,8 @@ from openai import OpenAI import os client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( model="glm-5.2", @@ -966,7 +990,8 @@ print(completion.choices[0].message.content) import OpenAI from "openai"; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: "https://dashscope.aliyuncs.com/compatible-mode/v1", + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", }); const completion = await openai.chat.completions.create({ model: "glm-5.2", @@ -979,7 +1004,8 @@ console.log(completion.choices[0].message.content); ## **curl** ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -1029,7 +1055,8 @@ import os client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) # 多轮对话，assistant 消息中携带 reasoning_content（历史思考过程） @@ -1057,7 +1084,8 @@ print(completion.usage.prompt_tokens) # true 时少于 false ## **curl** ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api-by-moonshot-ai.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api-by-moonshot-ai.md index 23cc13c0..d942d163 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api-by-moonshot-ai.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api-by-moonshot-ai.md @@ -24,22 +24,28 @@ - 如果通过SDK调用，需要[安装SDK](https://help.aliyun.com/zh/model-studio/install-sdk#8833b9274f4v8) -kimi/kimi-k2.7-code-highspeed、kimi/kimi-k2.7-code、kimi/kimi-k2.6、kimi/kimi-k2.5 均支持输入文本、图像或视频。kimi/kimi-k2.7-code-highspeed 与 kimi/kimi-k2.7-code 功能完全一致，速度提升5~6倍。kimi/kimi-k2.7-code-highspeed、kimi/kimi-k2.7-code 为仅思考模型（`enable_thinking` 始终为 true，无法设置为 false）。kimi/kimi-k2.6、kimi/kimi-k2.5 可通过 `enable_thinking` 参数控制思考模式，默认开启思考模式： +kimi 系列模型均支持输入文本、图像或视频： -- **思考模式**（`enable_thinking: true`）：模型会输出详细的推理过程（`reasoning_content`） +1. kimi/kimi-k2.7-code-highspeed 与 kimi/kimi-k2.7-code 功能完全一致，速度提升5~6倍。 -- **非思考模式**（`enable_thinking: false` 或不设置）：直接输出结果，不包含推理过程 +2. kimi/kimi-k3、kimi/kimi-k2.7-code-highspeed、kimi/kimi-k2.7-code 为仅思考模型。 +3. kimi/kimi-k2.6、kimi/kimi-k2.5 可通过 `enable_thinking` 参数控制思考模式，默认开启思考模式： + + - **思考模式**（`reasoning_effort: "max"`）：模型会输出详细的推理过程（`reasoning_content`） + + - **非思考模式**（`enable_thinking: false` 或不设置）：直接输出结果，不包含推理过程 + -kimi/kimi-k2.7-code-highspeed、kimi/kimi-k2.7-code、kimi/kimi-k2.6 支持通过 `preserve_thinking` 参数在多轮对话中传递思考过程，详情请参见[传递思考过程](https://help.aliyun.com/zh/model-studio/deep-thinking#jln7docdq5et5)。 +除 kimi/kimi-k2.5外，其他模型均支持通过 `preserve_thinking` 参数在多轮对话中传递思考过程，详情请参见[传递思考过程](https://help.aliyun.com/zh/model-studio/deep-thinking#jln7docdq5et5)。 -以下示例演示如何调用思考模式的 kimi-k2.6 模型进行文本生成。 +以下示例演示如何调用思考模式的 kimi/kimi-k3 模型进行文本生成。 ## OpenAI兼容 **说明** -`enable_thinking`非 OpenAI 标准参数，OpenAI Python SDK 通过 `extra_body`传入，Node.js SDK 作为顶层参数传入。 +`reasoning_effort`非 OpenAI 标准参数，OpenAI Python SDK 通过 `extra_body`传入，Node.js SDK 作为顶层参数传入。 ## Python @@ -49,14 +55,15 @@ import os client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( - model="kimi/kimi-k2.6", + model="kimi/kimi-k3", messages=[{"role": "user", "content": "1+1等于多少？"}], - # 通过 extra_body 设置 enable_thinking 开启思考模式 - extra_body={"enable_thinking": True} + # 通过 extra_body 设置 reasoning_effort 开启思考模式 + extra_body={"reasoning_effort": "max"} ) msg = completion.choices[0].message @@ -100,7 +107,8 @@ import process from 'process'; const client = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: "https://dashscope.aliyuncs.com/compatible-mode/v1", + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", }); const messages = [ @@ -109,9 +117,9 @@ const messages = [ ]; const response = await client.chat.completions.create({ - model: "kimi/kimi-k2.6", + model: "kimi/kimi-k3", messages, - extra_body: { enable_thinking: true }, + extra_body: { reasoning_effort: "max" }, }); const msg = response.choices[0].message; @@ -152,11 +160,12 @@ console.log(msg.content); ## curl ``` -curl --location 'https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions' \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions' \ --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ --header 'Content-Type: application/json' \ --data '{ - "model": "kimi/kimi-k2.6", + "model": "kimi/kimi-k3", "messages":[ { "role": "system", @@ -167,13 +176,13 @@ curl --location 'https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completi "content": "1+1等于多少？" } ], - "enable_thinking": true + "reasoning_effort": "max" }' ``` ## **多模态调用示例** -kimi/kimi-k2.7-code-highspeed、kimi/kimi-k2.7-code、kimi/kimi-k2.6、kimi/kimi-k2.5不仅支持纯文本对话，还具备强大的多模态理解能力。本章节将介绍如何让模型理解图像和视频内容。 +Kimi 系列模型不仅支持纯文本对话，还具备强大的多模态理解能力。本章节将介绍如何让模型理解图像和视频内容。 **重要** @@ -193,12 +202,13 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) # 单图传入示例（开启思考模式） completion = client.chat.completions.create( - model="kimi/kimi-k2.6", + model="kimi/kimi-k3", messages=[ { "role": "user", @@ -213,7 +223,7 @@ completion = client.chat.completions.create( ] } ], - extra_body={"enable_thinking":True} # 开启思考模式 + extra_body={"reasoning_effort":"max"} # 开启思考模式 ) # 输出思考过程 @@ -227,7 +237,7 @@ print(completion.choices[0].message.content) # 多图传入示例（开启思考模式，取消注释使用） # completion = client.chat.completions.create( -# model="kimi/kimi-k2.6", +# model="kimi/kimi-k3", # messages=[ # { # "role": "user", @@ -244,7 +254,7 @@ print(completion.choices[0].message.content) # ] # } # ], -# extra_body={"enable_thinking":True} +# extra_body={"reasoning_effort":"max"} # ) # # # 输出思考过程和回复 @@ -261,12 +271,13 @@ import process from 'process'; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); // 单图传入示例（开启思考模式） const completion = await openai.chat.completions.create({ - model: 'kimi/kimi-k2.6', + model: 'kimi/kimi-k3', messages: [ { role: 'user', @@ -281,7 +292,7 @@ const completion = await openai.chat.completions.create({ ] } ], - enable_thinking: true // 开启思考模式 + reasoning_effort: "max" // 开启思考模式 }); // 输出思考过程 @@ -313,7 +324,7 @@ console.log(completion.choices[0].message.content); // ] // } // ], -// enable_thinking: true +// reasoning_effort: "max" // }); // // // 输出思考过程和回复 @@ -328,11 +339,12 @@ console.log(completion.choices[0].message.content); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ - "model": "kimi/kimi-k2.6", + "model": "kimi/kimi-k3", "messages": [ { "role": "user", @@ -350,15 +362,16 @@ curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions ] } ], - "enable_thinking": true + "reasoning_effort": "max" }' # 多图输入示例（取消注释使用） -# curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +# curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ # -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ # -H "Content-Type: application/json" \ # -d '{ -# "model": "kimi/kimi-k2.6", +# "model": "kimi/kimi-k3", # "messages": [ # { # "role": "user", @@ -382,7 +395,7 @@ curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions # ] # } # ], -# "enable_thinking": true +# "reasoning_effort": "max" # }' ``` @@ -400,11 +413,12 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( - model="kimi/kimi-k2.6", + model="kimi/kimi-k3", messages=[ { "role": "user", @@ -435,12 +449,13 @@ import OpenAI from "openai"; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: "https://dashscope.aliyuncs.com/compatible-mode/v1" + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" }); async function main() { const response = await openai.chat.completions.create({ - model: "kimi/kimi-k2.6", + model: "kimi/kimi-k3", messages: [ { role: "user", @@ -496,11 +511,12 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ -d '{ - "model": "kimi/kimi-k2.6", + "model": "kimi/kimi-k3", "messages": [ { "role": "user", @@ -560,7 +576,7 @@ curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions [结构化输出](https://help.aliyun.com/zh/model-studio/qwen-structured-output) -kimi/kimi-k2.7-code-highspeed +kimi/kimi-k3 支持 @@ -574,22 +590,34 @@ kimi/kimi-k2.7-code-highspeed 支持 +kimi/kimi-k2.7-code-highspeed + kimi/kimi-k2.7-code kimi/kimi-k2.6 kimi/kimi-k2.5 -- kimi/kimi-k2.7-code-highspeed、kimi/kimi-k2.7-code、kimi/kimi-k2.6、kimi/kimi-k2.5支持上下文缓存（隐式缓存，自动开启），kimi/kimi-k2.7-code-highspeed命中缓存的输入Token按输入价格的20.0%计费，kimi/kimi-k2.7-code命中缓存的输入Token按输入价格的20.0%计费，kimi/kimi-k2.6命中缓存的输入Token按输入价格的16.9%计费，kimi/kimi-k2.5命中缓存的输入Token按输入价格的17.5%计费。 +以上模型支持上下文缓存（隐式缓存，自动开启）： + +- kimi/kimi-k3命中缓存的输入Token按输入价格的10%计费 -- 在思考模式下，使用 kimi/kimi-k2.7-code-highspeed、kimi/kimi-k2.7-code、kimi/kimi-k2.6、kimi/kimi-k2.5 进行工具调用时：必须在每轮 assistant 消息中保留 `reasoning_content` 字段，`tool_choice` 也仅支持 `"auto"`（默认）和 `"none"`），否则会报错。 +- kimi/kimi-k2.7-code-highspeed、kimi/kimi-k2.7-code 命中缓存的输入Token按输入价格的20.0%计费， + +- kimi/kimi-k2.6命中缓存的输入Token按输入价格的16.9%计费 + +- kimi/kimi-k2.5命中缓存的输入Token按输入价格的17.5%计费。 ## **参数默认值** **模型** -**stream\_options** +**tool\_choice** + +**preserve\_thinking** + +**reasoning\_effort** **temperature** @@ -599,33 +627,33 @@ kimi/kimi-k2.5 **presence\_penalty** -**tool\_choice** +**stream\_options** -**top\_k** +kimi/kimi-k3 -**preserve\_thinking** +auto -kimi/kimi-k2.7-code-highspeed +默认关闭 -仅支持设置为`true` +max -1.0 +1 0.95 0.0 -0.0 +\- -auto +仅支持设置为`true` -\- +kimi/kimi-k2.7-code-highspeed -默认开启 +auto -kimi/kimi-k2.7-code +默认开启 -仅支持设置为`true` +\- 1.0 @@ -635,33 +663,37 @@ kimi/kimi-k2.7-code 0.0 +仅支持设置为`true` + +kimi/kimi-k2.7-code + auto +默认开启 + \- -默认开启 +1.0 -kimi/kimi-k2.6 +0.95 + +0.0 + +0.0 仅支持设置为`true` +kimi/kimi-k2.6 + +思考模式/非思考模式：auto + +默认关闭 + +\- + 思考模式：1.0 非思考模式：0.6 - - - - - - - - - - - - - - 思考模式/非思考模式：0.95 @@ -669,27 +701,25 @@ kimi/kimi-k2.6 思考模式/非思考模式：0.0 -思考模式/非思考模式：auto - -\- - -默认关闭 +仅支持设置为`true` kimi/kimi-k2.5 \- +\- + - `stream_options`仅支持设置为`true`，`temperature`、`top_p`、`repetition_penalty`、`presence_penalty`不支持设置为其他值； -- 在思考模式下，不支持强制调用某个工具，`tool_choice`仅支持设置为`auto`（默认值）和`none`。 +- kimi/kimi-k3 支持 `reasoning_effort` 参数，唯一支持值为 `max`。 + +- 在思考模式下，使用 Kimi 模型进行工具调用时：必须在每轮 assistant 消息中保留 `reasoning_content` 字段；对于`tool_choice`参数，`kimi-k3` 支持 `auto` / `none` / `required` 三档；其他模型不支持 `required`，传入会报错。`kimi-k3` 支持动态加载工具，详细用法请参见[动态加载工具](https://platform.kimi.com/docs/guide/use-dynamic-tool-loading)。 - ”-”表示没有默认值，也不支持设置。 ## **模型列表与计费** -kimi/kimi-k2.7-code-highspeed、kimi/kimi-k2.7-code 为仅思考模型（`enable_thinking` 始终为 true，无法设置为 false）。kimi/kimi-k2.7-code-highspeed 与 kimi/kimi-k2.7-code 功能完全一致，速度提升5~6倍。kimi/kimi-k2.6、kimi/kimi-k2.5属于混合思考模型，通过`enable_thinking`参数控制是否开启思考模式（注意：均无法通过`thinking_budget`限制思考长度）。 - 模型上下文长度与价格信息请参见[百炼控制台](https://bailian.console.aliyun.com/cn-beijing?tab=model#/model-market/all)。 按照模型的输入与输出 Token 计费。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api.md index 676f6ad3..e453bbed 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api.md @@ -32,16 +32,12 @@ SDK 调用配置的`base_url`：`https://{WorkspaceId}.eu-central-1.maas.aliyunc HTTP 请求地址：`POST https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions` -调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## 新加坡 SDK 调用配置的`base_url`：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` HTTP 请求地址：`POST https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1/chat/completions` -调用时请将`{WorkspaceId}`替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 - ## 日本（东京） SDK 调用配置的`base_url`：`https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1` @@ -61,6 +57,7 @@ SDK调用配置的`base_url`： ## **Python代码** ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1' ``` @@ -70,6 +67,7 @@ dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.co ``` import com.alibaba.dashscope.protocol.Protocol; + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 Generation gen = new Generation(Protocol.HTTP.getValue(), "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"); ``` @@ -77,6 +75,7 @@ dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.co ``` import com.alibaba.dashscope.utils.Constants; + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 Constants.baseHttpApiUrl="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; ``` @@ -101,7 +100,7 @@ dashscope.base_http_api_url = 'https://dashscope-us.aliyuncs.com/api/v1' ``` import com.alibaba.dashscope.protocol.Protocol; - Generation gen = new Generation(Protocol.HTTP.getValue(), “https://dashscope-us.aliyuncs.com/api/v1"); + Generation gen = new Generation(Protocol.HTTP.getValue(), "https://dashscope-us.aliyuncs.com/api/v1"); ``` - **方式二：** @@ -154,6 +153,7 @@ SDK调用配置的`base_url`： ## **Python代码** ``` +# 以下为新加坡地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 dashscope.base_http_api_url = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1' ``` @@ -163,6 +163,7 @@ dashscope.base_http_api_url = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyunc ``` import com.alibaba.dashscope.protocol.Protocol; + // 以下为新加坡地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 Generation gen = new Generation(Protocol.HTTP.getValue(), "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1"); ``` @@ -170,6 +171,7 @@ dashscope.base_http_api_url = 'https://{WorkspaceId}.ap-southeast-1.maas.aliyunc ``` import com.alibaba.dashscope.utils.Constants; + // 以下为新加坡地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 Constants.baseHttpApiUrl="https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1"; ``` @@ -223,6 +225,7 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) @@ -291,6 +294,7 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); @@ -375,6 +379,7 @@ main(); ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -456,18 +461,20 @@ for chunk in completion: message = chunk.output.choices[0].message # 只收集思考内容 - if message.reasoning_content: + reasoning_chunk = message.get("reasoning_content") + if reasoning_chunk: if not is_answering: - print(message.reasoning_content, end="", flush=True) - reasoning_content += message.reasoning_content + print(reasoning_chunk, end="", flush=True) + reasoning_content += reasoning_chunk - # 收到 content，开始进行回复 - if message.content: + # 收到 content，开始进行回复（content 为列表，需提取其中的文本） + if message.get("content"): + text = message.content[0].get("text", "") if not is_answering: print("\n" + "=" * 20 + "完整回复" + "=" * 20 + "\n") is_answering = True - print(message.content, end="", flush=True) - answer_content += message.content + print(text, end="", flush=True) + answer_content += text # 循环结束后，reasoning_content 和 answer_content 变量中包含了完整的内容 # 您可以在这里根据需要进行后续处理 @@ -572,6 +579,7 @@ public class Main { ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation" \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -630,6 +638,7 @@ import os client = anthropic.Anthropic( # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic", ) @@ -657,6 +666,7 @@ for event in message: ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic/v1/messages \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -702,6 +712,7 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) @@ -770,6 +781,7 @@ import process from 'process'; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); @@ -835,6 +847,7 @@ console.log(completion.choices[0].message.content); ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -861,6 +874,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/ }' # 多图输入示例（取消注释使用） +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 # curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ # -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ # -H "Content-Type: application/json" \ @@ -956,7 +970,7 @@ print(response.output.choices[0].message.content[0]["text"]) ## Java ``` -// dashscope SDK的版本 >= 2.19.4 +// dashscope SDK的版本 >= 2.22.24 import com.alibaba.dashscope.aigc.multimodalconversation.MultiModalConversation; import com.alibaba.dashscope.aigc.multimodalconversation.MultiModalConversationParam; import com.alibaba.dashscope.aigc.multimodalconversation.MultiModalConversationResult; @@ -1047,6 +1061,7 @@ public class KimiK26MultiModalExample { ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation" \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ @@ -1073,6 +1088,7 @@ curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services }' # 多图输入示例（取消注释使用） +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 # curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation" \ # -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ # -H "Content-Type: application/json" \ @@ -1131,6 +1147,7 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) @@ -1167,6 +1184,7 @@ import OpenAI from "openai"; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" }); @@ -1203,6 +1221,7 @@ main(); ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -1237,6 +1256,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/ import dashscope import os +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" messages = [ @@ -1314,6 +1334,7 @@ public class Main { ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -1342,6 +1363,7 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) @@ -1369,6 +1391,7 @@ import OpenAI from "openai"; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" }); @@ -1404,6 +1427,7 @@ main(); ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -1427,6 +1451,7 @@ curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/ import os import dashscope +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" messages = [{"role": "user", @@ -1466,6 +1491,7 @@ import com.alibaba.dashscope.exception.UploadFileException; import com.alibaba.dashscope.utils.Constants; public class Main { + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 static {Constants.baseHttpApiUrl="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1";} private static final String MODEL_NAME = "kimi-k2.6"; @@ -1504,6 +1530,7 @@ public class Main { ## curl ``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ @@ -1559,6 +1586,7 @@ base64_image = encode_image("xxx/eagle.png") client = OpenAI( api_key=os.getenv('DASHSCOPE_API_KEY'), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -1602,6 +1630,7 @@ import { readFileSync } from 'fs'; const openai = new OpenAI( { apiKey: process.env.DASHSCOPE_API_KEY, + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" } ); @@ -1655,6 +1684,7 @@ import os import dashscope from dashscope import MultiModalConversation +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" # 编码函数： 将本地文件转换为 Base64 编码的字符串 @@ -1715,6 +1745,7 @@ import com.alibaba.dashscope.utils.Constants; public class Main { + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 static {Constants.baseHttpApiUrl="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1";} private static String encodeToBase64(String imagePath) throws IOException { @@ -1826,6 +1857,7 @@ import os from dashscope import MultiModalConversation import dashscope +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" # 将xxx/eagle.png替换为你本地图像的绝对路径 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/mimo.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/mimo.md index 7de583d8..349b678a 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/mimo.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/mimo.md @@ -30,7 +30,8 @@ import os client = OpenAI( # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) messages = [{"role": "user", "content": "1+1 等于多少？"}] @@ -99,7 +100,8 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); let reasoningContent = ''; // 完整思考过程 @@ -179,7 +181,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api-by-minimax.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api-by-minimax.md index 38765019..9790a912 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api-by-minimax.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api-by-minimax.md @@ -31,7 +31,8 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -91,7 +92,8 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); let reasoningContent = ''; // 完整思考过程 @@ -159,7 +161,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -236,7 +239,8 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) # 单图传入示例（自适应思考模式） @@ -304,7 +308,8 @@ import process from 'process'; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); // 单图传入示例（自适应思考模式） @@ -371,7 +376,8 @@ console.log(completion.choices[0].message.content); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -397,7 +403,8 @@ curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions }' # 多图输入示例（取消注释使用） -# curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +# curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ # -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ # -H "Content-Type: application/json" \ # -d '{ @@ -441,7 +448,8 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -476,7 +484,8 @@ import OpenAI from "openai"; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: "https://dashscope.aliyuncs.com/compatible-mode/v1" + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" }); async function main() { @@ -513,7 +522,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ -d '{ diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api.md index 2fc5120d..3a5396cd 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/minimax-api.md @@ -8,7 +8,7 @@ MiniMax-M2.1 将于**2026年7月9日**下架。推荐转用：[qwen3.7-plus](htt **重要** -本文档仅适用于中国内地地域。如需使用模型，需从中国内地地域[获取API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。 +本文档描述的功能仅在华北2（北京）地域可用，如需使用模型，需从华北2（北京）地域[获取API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。 ## **快速开始** @@ -26,7 +26,8 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -83,7 +84,8 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); let reasoningContent = ''; // 完整思考过程 @@ -148,7 +150,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -205,7 +208,10 @@ curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions ``` import os +import dashscope from dashscope import Generation +# 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 +dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" # 初始化请求参数 messages = [{"role": "user", "content": "你是谁？"}] @@ -276,6 +282,7 @@ import com.alibaba.dashscope.common.Role; import com.alibaba.dashscope.exception.ApiException; import com.alibaba.dashscope.exception.InputRequiredException; import com.alibaba.dashscope.exception.NoApiKeyException; +import com.alibaba.dashscope.utils.Constants; import io.reactivex.Flowable; import java.lang.System; import java.util.Arrays; @@ -283,6 +290,10 @@ import org.slf4j.Logger; import org.slf4j.LoggerFactory; public class Main { + static { + // 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; + } private static final Logger logger = LoggerFactory.getLogger(Main.class); private static StringBuilder reasoningContent = new StringBuilder(); private static StringBuilder finalContent = new StringBuilder(); @@ -365,7 +376,8 @@ public class Main { ## curl ``` -curl -X POST "https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation" \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation" \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -415,6 +427,64 @@ curl -X POST "https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generatio } ``` +## Anthropic兼容 + +## Python + +### **示例代码** + +``` +import anthropic +import os + +client = anthropic.Anthropic( + # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" + api_key=os.getenv("DASHSCOPE_API_KEY"), + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic", +) + +message = client.messages.create( + model="MiniMax-M2.5", + max_tokens=1024, + messages=[ + {"role": "user", "content": "你是谁"} + ], + stream=True, +) + +for event in message: + if event.type == "content_block_delta": + if hasattr(event.delta, "thinking"): + print(event.delta.thinking, end="", flush=True) + if hasattr(event.delta, "text"): + print(event.delta.text, end="", flush=True) +``` + +## HTTP + +### **示例代码** + +## curl + +``` +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic/v1/messages \ +-H "Authorization: Bearer $DASHSCOPE_API_KEY" \ +-H "Content-Type: application/json" \ +-H "anthropic-version: 2023-06-01" \ +-d '{ + "model": "MiniMax-M2.5", + "max_tokens": 1024, + "messages": [ + { + "role": "user", + "content": "你是谁" + } + ] +}' +``` + ## **其它功能** **模型** diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/siliconflow-deepseek-api.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/siliconflow-deepseek-api.md index 6d5c5d09..bf122dc0 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/siliconflow-deepseek-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/siliconflow-deepseek-api.md @@ -41,7 +41,8 @@ import os client = OpenAI( # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) messages = [{"role": "user", "content": "你是谁"}] @@ -120,7 +121,8 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); let reasoningContent = ''; // 完整思考过程 @@ -215,7 +217,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -242,8 +245,12 @@ curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions ``` import os +import dashscope from dashscope import Generation +# 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 +dashscope.base_http_api_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1" + # 初始化请求参数 messages = [{"role": "user", "content": "你是谁？"}] @@ -334,8 +341,14 @@ import com.alibaba.dashscope.exception.NoApiKeyException; import io.reactivex.Flowable; import java.lang.System; import java.util.Arrays; +import com.alibaba.dashscope.utils.Constants; public class Main { + static { + // 以下为华北2（北京）地域的配置，调用时请将 {WorkspaceId} 替换为真实的业务空间ID，各地域的配置不同。 + Constants.baseHttpApiUrl = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1"; + } + private static StringBuilder reasoningContent = new StringBuilder(); private static StringBuilder finalContent = new StringBuilder(); private static boolean isFirstPrint = true; @@ -406,7 +419,8 @@ DeepSeek-V3，一个由深度求索公司创造的智能助手！我可以帮助 ## curl ``` -curl -X POST "https://dashscope.aliyuncs.com/api/v1/services/aigc/text-generation/generation" \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation" \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -H "X-DashScope-SSE: enable" \ diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/stepfun.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/stepfun.md index f02e6efc..861138b5 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/stepfun.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-cases/third-party-model-integration-tutorial/stepfun.md @@ -24,7 +24,8 @@ import os client = OpenAI( # 如果没有配置环境变量，请用阿里云百炼API Key替换：api_key="sk-xxx" api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) messages = [{"role": "user", "content": "9.9和9.11哪个大？"}] @@ -98,7 +99,8 @@ import process from 'process'; const openai = new OpenAI({ // 如果没有配置环境变量，请用阿里云百炼API Key替换：apiKey: "sk-xxx" apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); let reasoningContent = ''; // 完整思考过程 @@ -181,7 +183,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -216,7 +219,8 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -256,7 +260,8 @@ import process from 'process'; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1' + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1' }); const completion = await openai.chat.completions.create({ @@ -294,7 +299,8 @@ console.log(completion.choices[0].message.content); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H "Content-Type: application/json" \ -d '{ @@ -332,7 +338,8 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), - base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + # 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( @@ -366,7 +373,8 @@ import OpenAI from "openai"; const openai = new OpenAI({ apiKey: process.env.DASHSCOPE_API_KEY, - baseURL: "https://dashscope.aliyuncs.com/compatible-mode/v1" + // 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 + baseURL: "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" }); async function main() { @@ -402,7 +410,8 @@ main(); ## curl ``` -curl -X POST https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions \ +# 以下为华北2（北京）地域的URL。请将 {WorkspaceId} 替换为您的百炼业务空间ID，各地域的URL不同。 +curl -X POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ -H 'Content-Type: application/json' \ -d '{ diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/chatbox.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/chatbox.md index 51c73192..6d4b8347 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/chatbox.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/chatbox.md @@ -1,6 +1,6 @@ # Chatbox -Chatbox 是一款跨平台 AI 客户端应用，可以通过Token Plan 团队版、Coding Plan或按量计费接入阿里云百炼。 +Chatbox 是一款跨平台 AI 客户端应用，可以通过 Token Plan 个人版、Token Plan 团队版、Coding Plan 或按量计费接入阿里云百炼。 ## **下载安装 Chatbox** @@ -12,6 +12,8 @@ Chatbox 是一款跨平台 AI 客户端应用，可以通过Token Plan 团队版 百炼提供三种计费方案，根据需要选择： +- **Token Plan 个人版**：按 token 消耗抵扣个人 Credits。 + - **Token Plan 团队版**：按坐席订阅，按 token 消耗抵扣 Credits。 - **Coding Plan**：固定月费订阅，按模型调用次数计量。 @@ -19,6 +21,24 @@ Chatbox 是一款跨平台 AI 客户端应用，可以通过Token Plan 团队版 - **按量计费**：按实际调用量后付费。 +### Token Plan 个人版 + +**配置项** + +**说明** + +**API 密钥** + +填入 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/personal)。 + +**API 主机** + +`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` + +**模型** + +填入 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)，如 `qwen3.8-max-preview`。 + ### Token Plan 团队版 **配置项** @@ -27,7 +47,7 @@ Chatbox 是一款跨平台 AI 客户端应用，可以通过Token Plan 团队版 **API 密钥** -填入 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)。 +填入 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/enterprise)。 **API 主机** @@ -71,11 +91,11 @@ Chatbox 是一款跨平台 AI 客户端应用，可以通过Token Plan 团队版 **API 主机** -根据模型部署地域，填入对应 URL： +根据模型部署地域，填入对应 URL（请将 URL 中的 `{WorkspaceId}` 替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)）： -- 华北2（北京）：`https://dashscope.aliyuncs.com/compatible-mode/v1` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` - 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1` @@ -100,4 +120,6 @@ Chatbox 是一款跨平台 AI 客户端应用，可以通过Token Plan 团队版 - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cherry-studio.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cherry-studio.md index af83a91a..a3880109 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cherry-studio.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cherry-studio.md @@ -1,6 +1,6 @@ # Cherry Studio -Cherry Studio 是一款开源 AI 桌面客户端，可以通过 Token Plan 团队版、Coding Plan或按量计费接入阿里云百炼。 +Cherry Studio 是一款开源 AI 桌面客户端，可以通过 Token Plan 个人版、Token Plan 团队版、Coding Plan 或按量计费接入阿里云百炼。 ## **安装 Cherry Studio** @@ -12,6 +12,8 @@ Cherry Studio 是一款开源 AI 桌面客户端，可以通过 Token Plan 团 百炼提供三种计费方案，根据需要选择： +- **Token Plan 个人版**：按 token 消耗抵扣个人 Credits。 + - **Token Plan 团队版**：按坐席订阅，按 token 消耗抵扣 Credits。 - **Coding Plan**：固定月费订阅，按模型调用次数计量。 @@ -19,6 +21,24 @@ Cherry Studio 是一款开源 AI 桌面客户端，可以通过 Token Plan 团 - **按量计费**：按实际调用量后付费。 +### Token Plan 个人版 + +**配置项** + +**说明** + +**API 密钥** + +填入 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/personal)。 + +**API 地址** + +`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` + +**模型** + +可用模型请参考 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)。 + ### Token Plan 团队版 **配置项** @@ -27,7 +47,7 @@ Cherry Studio 是一款开源 AI 桌面客户端，可以通过 Token Plan 团 **API 密钥** -填入 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)。 +填入 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/enterprise)。 **API 地址** @@ -67,11 +87,11 @@ Cherry Studio 是一款开源 AI 桌面客户端，可以通过 Token Plan 团 **API 地址** -根据地域，填入对应 URL： +根据地域，填入对应 URL（请将 URL 中的 `{WorkspaceId}` 替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)）： -- 华北2（北京）：`https://dashscope.aliyuncs.com/compatible-mode/v1` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` - 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1` @@ -96,7 +116,9 @@ Cherry Studio 是一款开源 AI 桌面客户端，可以通过 Token Plan 团 - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) ### 报错 The value of the enable\_thinking parameter is restricted to True @@ -109,7 +131,7 @@ Cherry Studio 是一款开源 AI 桌面客户端，可以通过 Token Plan 团 可能的原因： -- **地域不匹配**：免费额度仅适用于华北2（北京）地域（[中国内地服务部署范围](https://help.aliyun.com/zh/model-studio/regions/#080da663a75xh)）的模型。使用其他地域和部署范围的模型会产生费用。请检查**API 地址**是否与目标地域匹配，详情请参见[新人免费额度](https://help.aliyun.com/zh/model-studio/new-free-quota)。 +- **地域不匹配**：免费额度仅适用于华北2（北京）地域的模型。使用其他地域的模型会产生费用。请检查**API 地址**是否与目标地域匹配，详情请参见[新人免费额度](https://help.aliyun.com/zh/model-studio/new-free-quota)。 - **额度按模型独立计算**：各模型的免费额度相互独立，不可跨模型共享。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/claude-code.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/claude-code.md index 00895664..36af72c4 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/claude-code.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/claude-code.md @@ -1,6 +1,6 @@ # Claude Code -Claude Code 是 Anthropic 推出的命令行 AI 编程助手。通过阿里云百炼，可以使用按量计费、Coding Plan 或 Token Plan 团队版接入 Claude Code。 +Claude Code 是 Anthropic 推出的命令行 AI 编程助手。通过阿里云百炼，可以使用按量计费、Coding Plan、Token Plan 个人版或 Token Plan 团队版接入 Claude Code。 ## **安装 Claude Code** @@ -47,6 +47,36 @@ npm install -g @anthropic-ai/claude-code 新建 `~/.claude/settings.json`（Windows 路径：`C:\Users\<用户名>\.claude\settings.json`），写入对应套餐的配置。 +### Token Plan 个人版 + +将 YOUR\_API\_KEY 替换为 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)。可用模型：qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-flash、glm-5.2、deepseek-v4-pro。完整说明参见 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)。 + +``` +{ + "env": { + "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY", + "ANTHROPIC_BASE_URL": "https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic", + "ANTHROPIC_MODEL": "qwen3.8-max-preview", + "ANTHROPIC_DEFAULT_HAIKU_MODEL": "qwen3.6-flash", + "ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3.8-max-preview", + "ANTHROPIC_DEFAULT_OPUS_MODEL": "qwen3.8-max-preview", + "CLAUDE_CODE_SUBAGENT_MODEL": "qwen3.7-max", + "CLAUDE_CODE_MAX_CONTEXT_TOKENS": "983616" + } +} +``` + +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- **thinking**：始终开启，不支持关闭。 + +- **temperature**：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- **reasoning\_effort**：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Token Plan 团队版 将 YOUR\_API\_KEY 替换为 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)。可用模型参见 Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-overview)。 @@ -56,15 +86,27 @@ npm install -g @anthropic-ai/claude-code "env": { "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY", "ANTHROPIC_BASE_URL": "https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic", - "ANTHROPIC_MODEL": "qwen3.7-max", + "ANTHROPIC_MODEL": "qwen3.8-max-preview", "ANTHROPIC_DEFAULT_HAIKU_MODEL": "qwen3.6-flash", - "ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3.7-max", - "ANTHROPIC_DEFAULT_OPUS_MODEL": "qwen3.7-max", - "CLAUDE_CODE_SUBAGENT_MODEL": "qwen3.7-max" + "ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3.8-max-preview", + "ANTHROPIC_DEFAULT_OPUS_MODEL": "qwen3.8-max-preview", + "CLAUDE_CODE_SUBAGENT_MODEL": "qwen3.7-max", + "CLAUDE_CODE_MAX_CONTEXT_TOKENS": "983616" } } ``` +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- **thinking**：始终开启，不支持关闭。 + +- **temperature**：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- **reasoning\_effort**：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Coding Plan 将 YOUR\_API\_KEY 替换为 Coding Plan 专属 [API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan)。可用模型参见 Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-plan)。 @@ -87,18 +129,18 @@ npm install -g @anthropic-ai/claude-code 将 YOUR\_API\_KEY 替换为[阿里云百炼API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。可用模型参见[Anthropic 兼容 API](https://help.aliyun.com/zh/model-studio/anthropic-api-messages#07833dedefft7)。 -`ANTHROPIC_BASE_URL` 按地域设置，API Key 需与所选地域对应： +`ANTHROPIC_BASE_URL` 按地域设置，API Key 需与所选地域对应，并将`WorkspaceId`替换为真实的 [Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)： -- 华北2（北京）：`https://dashscope.aliyuncs.com/apps/anthropic` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/apps/anthropic`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/apps/anthropic` ``` { "env": { "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY", - "ANTHROPIC_BASE_URL": "https://dashscope.aliyuncs.com/apps/anthropic", + "ANTHROPIC_BASE_URL": "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic", "ANTHROPIC_MODEL": "qwen3.7-max", "ANTHROPIC_DEFAULT_HAIKU_MODEL": "qwen3.6-flash", "ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3.7-max", @@ -122,7 +164,7 @@ Claude Code 默认使用 200K 上下文窗口。如果需要处理大型代码 { "env": { "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY", - "ANTHROPIC_BASE_URL": "https://dashscope.aliyuncs.com/apps/anthropic", + "ANTHROPIC_BASE_URL": "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic", "ANTHROPIC_MODEL": "qwen3.7-plus", "CLAUDE_CODE_MAX_CONTEXT_TOKENS": "1000000" } @@ -137,7 +179,7 @@ Claude Code 默认使用 200K 上下文窗口。如果需要处理大型代码 { "env": { "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY", - "ANTHROPIC_BASE_URL": "https://dashscope.aliyuncs.com/apps/anthropic", + "ANTHROPIC_BASE_URL": "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic", "ANTHROPIC_MODEL": "qwen3.7-plus[1m]", "ANTHROPIC_DEFAULT_SONNET_MODEL": "qwen3.7-plus[1m]", "ANTHROPIC_DEFAULT_OPUS_MODEL": "qwen3.7-plus[1m]", @@ -169,6 +211,14 @@ Claude Code 默认使用 200K 上下文窗口。如果需要处理大型代码 **配置信息** + Token Plan 个人版 + + 供应商名称：百炼-Token Plan 个人版 + + API Key：[控制台获取](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview) + + 请求地址：`https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` + Token Plan 团队版 供应商名称：百炼-Token Plan @@ -191,7 +241,7 @@ Claude Code 默认使用 200K 上下文窗口。如果需要处理大型代码 API Key：[百炼API Key](https://help.aliyun.com/zh/model-studio/get-api-key) - 请求地址：`https://dashscope.aliyuncs.com/apps/anthropic` + 请求地址：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic` 2. 展开**高级选项**配置模型映射，将主模型与 Haiku、Sonnet、Opus 默认模型设置为对应套餐[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-overview)。映射关系按需选择，示例如下： @@ -208,35 +258,23 @@ Claude Code 默认使用 200K 上下文窗口。如果需要处理大型代码 ### 接入 Claude Code 桌面版 +Claude Code 桌面版（Claude Desktop）与 Claude Code CLI 是两个独立入口，在 CC Switch 中分别对应 **Claude Code** 与 **Claude Desktop** 面板。桌面版通过 CC Switch 本地网关访问百炼：网关地址与鉴权令牌均由 CC Switch 自动写入桌面版配置，**无需在桌面版中手动填写百炼 API Key**——百炼 API Key 只在 CC Switch 供应商配置中填写，由本地路由转发时自动注入。 + +**重要** + +请勿在桌面版的第三方推理配置中手动填写百炼 API Key。桌面版对 CC Switch 本地网关（地址 `http://127.0.0.1:15721/claude-desktop`）的鉴权令牌由 CC Switch 自动生成并写入，手动填入百炼 API Key 会因令牌不匹配导致鉴权失败。桌面版第三方配置写入目前仅支持 macOS、Windows。 + 1. 从 [Claude 下载页](https://claude.ai/download)安装 Claude Code 桌面版。 -2. 顶部菜单 **Help** → **Troubleshooting** → **Enable Developer Mode**，重启后顶部出现 **Developer** 菜单。 - -3. **Developer** → **Configure Third-Party Inference**，**Connection** 选 **Gateway**，按下表填写后点击 **Apply locally**： - - **字段** - - **填写** - - Gateway base URL - - CC Switch 路由监听地址，默认 `http://127.0.0.1:15721`，如已修改则与下一步保持一致。 +2. 在 CC Switch 左侧应用切换器切换到 **Claude Desktop** 面板。若未显示该入口，前往**设置 → 通用 → 应用可见性**确认 Claude Desktop 未被隐藏。 - Gateway API key +3. 添加百炼供应商：若已在 **Claude Code** 面板配置过百炼供应商，可点击**将 Claude Code 中已有的供应商导入**一键复用；也可点击右上角 **+** 新增。由于百炼模型 ID（如 `qwen3.7-max`）不是 Claude Desktop 识别的 `claude-sonnet-* / claude-opus-* / claude-haiku-*` 三档角色 ID，需开启**需要模型映射**，为 Sonnet、Opus、Haiku 三档分别填写实际请求的百炼模型（如 Sonnet → qwen3.7-max）。 - 百炼API Key +4. 开启本地路由：前往**设置 → 路由 → 本地路由**，打开**在主页面显示本地路由开关**；回到 Claude Desktop 面板，打开 **Claude Desktop 本地路由**开关，监听地址默认 `127.0.0.1:15721`。 - Gateway auth scheme +5. 在供应商卡片点击**启用**，CC Switch 会自动将第三方推理配置写入 Claude Code 桌面版。 - bearer - - Model list - - Model ID须为 Anthropic 风格，如 `claude-opus-4.7`。实际调用模型由 CC Switch 路由决定，对应关系即供应商[高级选项中的模型映射](#ccswitch-add-li2)。Display name 仅影响下拉显示。 - -4. CC Switch 左上角设置 → **路由**，开启**路由总开关**，监听地址默认 `127.0.0.1:15721`，如需修改请同步上一步。 - -5. 在桌面版模型下拉中选择已配置的模型ID即可使用。 +6. 保持 CC Switch 运行，**完全退出并重启** Claude Code 桌面版后生效，在模型菜单中选择已配置的模型即可使用。 ## **Claude Code IDE 插件** @@ -295,7 +333,9 @@ Claude Code 默认使用 200K 上下文窗口。如果需要处理大型代码 - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) ### 启动 Claude Code 后，界面显示"Unable to connect to Anthropic services. Failed to connect to api.anthropic.com: ERR\_BAD\_REQUEST" diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cline.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cline.md index 0d220bf9..8126ee3f 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cline.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cline.md @@ -1,6 +1,6 @@ # Cline -Cline 是一款 VSCode 智能编程插件，可以通过 Token Plan 团队版、Coding Plan或按量计费接入阿里云百炼。 +Cline 是一款 VSCode 智能编程插件，可以通过 Token Plan 个人版、Token Plan 团队版、Coding Plan 或按量计费接入阿里云百炼。 ## **安装 Cline** @@ -15,6 +15,8 @@ Cline 是一款 VSCode 智能编程插件，可以通过 Token Plan 团队版、 阿里云百炼提供三种计费方案，根据需要选择： +- **Token Plan 个人版**：按 token 消耗抵扣个人 Credits。 + - **Token Plan 团队版**：按坐席订阅，按 token 消耗抵扣 Credits。 - **Coding Plan**：固定月费订阅，按模型调用次数计量。 @@ -22,6 +24,28 @@ Cline 是一款 VSCode 智能编程插件，可以通过 Token Plan 团队版、 - **按量计费**：按实际调用量后付费。 +### Token Plan 个人版 + +**配置项** + +**说明** + +API Provider + +选择 **OpenAI Compatible**。 + +Base URL + +`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` + +API Key + +填入 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)。 + +Model ID + +填入 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)，如 `qwen3.8-max-preview`。 + ### Token Plan 团队版 **配置项** @@ -42,7 +66,7 @@ API Key Model ID -填入 Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-overview)，如 `qwen3.7-max`。 +填入 Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-overview)，如 `qwen3.8-max-preview`。 ### Coding Plan @@ -78,11 +102,11 @@ API Provider Base URL -根据地域，填入对应 URL： +根据地域，填入对应 URL（请将 URL 中的 `{WorkspaceId}` 替换为真实的[业务空间ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)）： -- 华北2（北京）：`https://dashscope.aliyuncs.com/compatible-mode/v1` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` - 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1` @@ -133,7 +157,9 @@ Model ID - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) ### 报错 401 Incorrect API key provided diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/codex.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/codex.md index 76a25932..eaae7c03 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/codex.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/codex.md @@ -1,6 +1,6 @@ # Codex -Codex 是 OpenAI 推出的终端 AI 编程助手。可通过 Token Plan 团队版、Coding Plan 或按量计费接入阿里云百炼。 +Codex 是 OpenAI 推出的终端 AI 编程助手。可通过 Token Plan 个人版、Token Plan 团队版、Coding Plan 或按量计费接入阿里云百炼。 ## **安装 Codex** @@ -21,19 +21,199 @@ Codex 是 OpenAI 推出的终端 AI 编程助手。可通过 Token Plan 团队 ## **配置接入凭证** -接入需要编辑配置文件`~/.codex/config.toml`并配置环境变量`OPENAI_API_KEY`。根据所选计费方案替换对应值，阿里云百炼提供三种计费方案： +接入需要编辑配置文件`~/.codex/config.toml`并配置环境变量`OPENAI_API_KEY`。根据所选计费方案替换对应值，阿里云百炼提供以下计费方案： + +### 配置模型元数据 + +使用自定义模型（如 qwen3.8-max-preview）时，需要配置模型元数据文件，使 Codex 正确识别模型的上下文窗口、推理深度等参数。 + +1. 新建文件 `~/.codex/model-catalog.local.json`，写入以下内容： + + ``` + { + "models": [ + { + "slug": "qwen3.8-max-preview", + "display_name": "qwen3.8-max-preview", + "description": "DashScope model: qwen3.8-max-preview", + "default_reasoning_level": "xhigh", + "supported_reasoning_levels": [ + { + "effort": "low", + "description": "Fast responses with lighter reasoning" + }, + { + "effort": "medium", + "description": "Greater reasoning depth for complex problems" + }, + { + "effort": "xhigh", + "description": "Extra high reasoning depth for complex problems" + } + ], + "context_window": 983616, + "effective_context_window_percent": 95, + "supports_parallel_tool_calls": false, + "supports_image_detail_original": true, + "input_modalities": ["text", "image"], + "shell_type": "default", + "visibility": "list", + "supported_in_api": true, + "priority": 1, + "base_instructions": "", + "support_verbosity": false, + "supports_reasoning_summaries": false, + "experimental_supported_tools": [], + "truncation_policy": { + "mode": "bytes", + "limit": 10000 + } + } + ] + } + ``` + +2. 在 `~/.codex/config.toml` 中添加以下配置，指向元数据文件： + + ``` + model_catalog_json = "~/.codex/model-catalog.local.json" + ``` + + +### Token Plan 个人版 + +`model`请选择[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)，可用模型包括 qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-flash、glm-5.2、deepseek-v4-pro。将`OPENAI_API_KEY`环境变量设置为 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)。 + +#### Responses API（qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash） + +qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-plus 和 qwen3.6-flash 支持 Responses API，可使用最新版 Codex。 + +``` +model_provider = "Model_Studio_Token_Plan_Personal" +model = "qwen3.8-max-preview" +[model_providers.Model_Studio_Token_Plan_Personal] +name = "Model_Studio_Token_Plan_Personal" +base_url = "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" +env_key = "OPENAI_API_KEY" +wire_api = "responses" +``` + +#### Chat/Completions API（其他模型） + +其他模型需通过 Chat/Completions API 接入，需安装旧版本 Codex，如 0.80.0： + +``` +npm install -g @openai/codex@0.80.0 +``` +``` +model_provider = "Model_Studio_Token_Plan_Personal" +model = "glm-5" +[model_providers.Model_Studio_Token_Plan_Personal] +name = "Model_Studio_Token_Plan_Personal" +base_url = "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" +env_key = "OPENAI_API_KEY" +wire_api = "chat" +``` + +#### 配置环境变量 + +将`OPENAI_API_KEY`环境变量设置为 Token Plan 个人版专属 API Key。 + +## macOS + +1. 在终端中执行以下命令，查看默认 Shell 类型。 + + ``` + echo $SHELL + ``` + +2. 根据 Shell 类型设置环境变量： + + ## Zsh + + ``` + # 将 YOUR_API_KEY 替换为 Token Plan 个人版 API Key + echo 'export OPENAI_API_KEY="YOUR_API_KEY"' >> ~/.zshrc + ``` + + ## Bash + + ``` + # 将 YOUR_API_KEY 替换为 Token Plan 个人版 API Key + echo 'export OPENAI_API_KEY="YOUR_API_KEY"' >> ~/.bash_profile + ``` + +3. 执行以下命令使环境变量生效。 + + ## Zsh + + ``` + source ~/.zshrc + ``` + + ## Bash + + ``` + source ~/.bash_profile + ``` + + +## Windows + +## CMD + +1. 在 CMD 中运行以下命令，设置环境变量。 + + ``` + REM 将 YOUR_API_KEY 替换为 Token Plan 个人版 API Key + setx OPENAI_API_KEY "YOUR_API_KEY" + ``` + +2. 打开一个新的 CMD 窗口，运行以下命令检查环境变量是否生效。 + + ``` + echo %OPENAI_API_KEY% + ``` + + +## PowerShell + +1. 在 PowerShell 中运行以下命令，设置环境变量。 + + ``` + # 将 YOUR_API_KEY 替换为 Token Plan 个人版 API Key + [Environment]::SetEnvironmentVariable("OPENAI_API_KEY", "YOUR_API_KEY", [EnvironmentVariableTarget]::User) + ``` + +2. 打开一个新的 PowerShell 窗口，运行以下命令检查环境变量是否生效。 + + ``` + echo $env:OPENAI_API_KEY + ``` + + +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + ### Token Plan 团队版 `model`请选择[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-overview)。将`OPENAI_API_KEY`环境变量设置为 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)。 -#### Responses API（qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash） +#### Responses API（qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash） -qwen3.7-max、qwen3.7-plus、qwen3.6-plus 和 qwen3.6-flash 支持 Responses API，可使用最新版 Codex。 +qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-plus 和 qwen3.6-flash 支持 Responses API，可使用最新版 Codex。 ``` model_provider = "Model_Studio_Token_Plan" -model = "qwen3.7-max" +model = "qwen3.8-max-preview" [model_providers.Model_Studio_Token_Plan] name = "Model_Studio_Token_Plan" base_url = "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" @@ -135,6 +315,17 @@ wire_api = "chat" ``` +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Coding Plan `model`请选择[支持的模型](https://help.aliyun.com/zh/model-studio/coding-plan-overview)。将`OPENAI_API_KEY`环境变量设置为 Coding Plan 专属 [API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan)。 @@ -237,11 +428,11 @@ wire_api = "chat" 将`OPENAI_API_KEY`环境变量设置为[百炼 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。可用模型参见[支持的模型](https://help.aliyun.com/zh/model-studio/compatibility-of-openai-with-dashscope#7f9c78ae99pwz)。 -根据地域设置`base_url`，API Key 须与所选地域对应： +根据地域设置`base_url`，API Key 须与所选地域对应，请将 URL 中的 `{WorkspaceId}` 替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)： -- 华北2（北京）：`https://dashscope.aliyuncs.com/compatible-mode/v1` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` 按量计费支持 Responses API 和 Chat/Completions API 两种接入方式，请根据使用的模型选择： @@ -255,7 +446,7 @@ model_provider = "Model_Studio" model = "qwen3.7-max" [model_providers.Model_Studio] name = "Model_Studio" -base_url = "https://dashscope.aliyuncs.com/compatible-mode/v1" +base_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" env_key = "OPENAI_API_KEY" wire_api = "responses" ``` @@ -272,7 +463,7 @@ model_provider = "Model_Studio" model = "qwen3.6-plus" [model_providers.Model_Studio] name = "Model_Studio" -base_url = "https://dashscope.aliyuncs.com/compatible-mode/v1" +base_url = "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1" env_key = "OPENAI_API_KEY" wire_api = "chat" ``` @@ -368,11 +559,11 @@ codex ### **第三方工具提示“不支持国内模型”或“检查被拒 / Bad request (400)”怎么办？** -**原因**：部分第三方管理工具（如 CC-Switch）在切换供应商时会发起“健康检查/连接测试”探测请求，该探测请求的格式与 Codex 实际调用的请求格式不同，百炼网关可能因此返回 400 Bad request 并提示“检查被拒”，工具据此显示“不支持国内模型”。此提示仅代表健康检查探测未通过，**并不代表百炼不支持国内模型，也不影响 Codex 的实际使用。** +**原因**：部分第三方管理工具（如 CC-Switch）在切换供应商时会发起“健康检查/连接测试”探测请求，该探测请求的格式与 Codex 实际调用的请求格式不同，百炼网关可能因此返回 400 Bad request 并提示“检查被拒”，工具据此显示“不支持国内模型”。此提示仅代表健康检查探测未通过，**并不代表百炼不支持中国内地模型，也不影响 Codex 的实际使用。** -**说明**：百炼支持通过 Codex 使用 qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash、glm-5 等国内模型，配置方式详见上文[配置接入凭证](#cdx-config)。 +**说明**：百炼支持通过 Codex 使用 qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash、glm-5 等中国内地模型，配置方式详见上文[配置接入凭证](#cdx-config)。 -**解决方案**：建议参照上文配置接入凭证，直接在`~/.codex/config.toml`中完成配置，无需依赖第三方工具的健康检查结果；配置完成后参照[验证配置](#cdx-verify)启动 Codex，若能正常进入对话界面即表示可正常使用国内模型。 +**解决方案**：建议参照上文配置接入凭证，直接在`~/.codex/config.toml`中完成配置，无需依赖第三方工具的健康检查结果；配置完成后参照[验证配置](#cdx-verify)启动 Codex，若能正常进入对话界面即表示可正常使用中国内地模型。 ### **报错 wire\_api 配置问题怎么办？** @@ -394,7 +585,7 @@ codex **原因**： -- 误用了其他方案的 API Key（Token Plan 团队版、Coding Plan 和按量计费的 API Key 互不相通） +- 误用了其他方案的 API Key（Token Plan 个人版、Token Plan 团队版、Coding Plan 和按量计费的 API Key 互不相通） - 订阅过期 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cursor.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cursor.md index f903bf87..61f5ca45 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cursor.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/cursor.md @@ -1,6 +1,6 @@ # Cursor -Cursor 是一款 AI 编程 IDE，可以通过按量计费、Coding Plan 或 Token Plan 团队版接入阿里云百炼。 +Cursor 是一款 AI 编程 IDE，可以通过按量计费、Coding Plan、Token Plan 个人版或 Token Plan 团队版接入阿里云百炼。 ## **安装 Cursor** @@ -12,6 +12,8 @@ Cursor 是一款 AI 编程 IDE，可以通过按量计费、Coding Plan 或 Toke 阿里云百炼提供三种计费方案，根据需要选择： +- **Token Plan 个人版**：按 token 消耗抵扣个人 Credits。 + - **Token Plan 团队版**：按坐席订阅，按 token 消耗抵扣 Credits。 - **Coding Plan**：固定月费订阅，按模型调用次数计量。 @@ -19,6 +21,20 @@ Cursor 是一款 AI 编程 IDE，可以通过按量计费、Coding Plan 或 Toke - **按量计费**：按实际调用量后付费。 +### Token Plan 个人版 + +**API Key** + +Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview) + +**Base URL** + +`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` + +**可用模型** + +Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview) + ### Token Plan 团队版 **API Key** @@ -59,11 +75,11 @@ Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-pla **Base URL** -根据地域，填入对应 URL： +根据地域，填入对应 URL，请将 URL 中的 `{WorkspaceId}` 替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)： -- 华北2（北京）：`https://dashscope.aliyuncs.com/compatible-mode/v1` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` - 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1` @@ -110,7 +126,9 @@ Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-pla - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) ### 在 Cursor 中无法调用已添加的模型 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/dify.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/dify.md index 68960a12..e0558cb7 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/dify.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/dify.md @@ -1,6 +1,6 @@ # Dify -Dify 是一个开源的大模型应用开发平台，您可以基于阿里云百炼提供的模型 API 来构建大模型应用。 +Dify 是一个开源的大模型应用开发平台，您可以基于阿里云百炼提供的模型 API（按量付费或 Token Plan 个人版）来构建大模型应用。 ## **前提条件** @@ -36,7 +36,7 @@ Dify 是一个开源的大模型应用开发平台，您可以基于阿里云百 单击**通义千问**卡片中的显示模型，打开您需要使用的模型开关。 -> 若插件内暂未包含最新版千问模型，可尝试安装 **OpenAI-API-compatible** 插件，在插件设置中的 **API endpoint URL** 填入`https://dashscope.aliyuncs.com/compatible-mode/v1`（华北2（北京）地域）或`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`（新加坡地域）。其中`WorkspaceId`需替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 +> 若插件内暂未包含最新版千问模型，可尝试安装 **OpenAI-API-compatible** 插件，在插件设置中的 **API endpoint URL** 填入`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1`（华北2（北京）地域）或`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`（新加坡地域）。其中`WorkspaceId`需替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 ## **2\. 开始使用** @@ -101,6 +101,23 @@ Dify 具有多种大模型应用类型，请选择您使用的类型进行参考 > Embedding 模型暂时无法选择multimodal-embedding-v1模型，敬请关注后续动态。 +## **Token Plan 个人版** + +如果您已订阅 Token Plan 个人版，可通过 OpenAI-API-compatible 插件接入，无需使用千问插件。 + +1. 在 [Dify 市场](https://cloud.dify.ai/plugins?category=discover)安装 **OpenAI-API-compatible** 插件。 + +2. 在插件设置中配置以下信息： + + - **API endpoint URL**：`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` + + - **API Key**：填入 Token Plan 个人版专属 API Key。[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview) + +3. 在模型列表中选择 Token Plan 个人版支持的模型，即可开始使用。支持的模型请参考[Token Plan 个人版支持模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)。 + + +常见问题请参考[Token Plan 常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq)。 + ## **常见问题** ### **Q1：在千问插件内配置 API Key 报错？** diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/first-call-to-image-and-video-api.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/first-call-to-image-and-video-api.md index b4f0ca4f..c614a907 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/first-call-to-image-and-video-api.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/first-call-to-image-and-video-api.md @@ -22,7 +22,7 @@ Postman 和 cURL仅适用于快速测试与功能验证。对于生产环境 **HTTP调用示例（文生图）** -![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/4178421871/CAEQZRiBgICJ0seH1xkiIDgzYWE2MTBkZjkzODRkNDA5NzczNTE0NjBiMGE1Y2Nm5274221_20250627113930.173.svg) +![image](https://help-static-aliyun-doc.aliyuncs.com/assets/img/zh-CN/4507315871/CAEQZRiBgICJ0seH1xkiIDgzYWE2MTBkZjkzODRkNDA5NzczNTE0NjBiMGE1Y2Nm5274221_20250627113930.173.svg) ## **方式一：使用Postman发送请求（推荐）** @@ -147,7 +147,6 @@ curl -X POST https://dashscope.aliyuncs.com/api/v1/services/aigc/text2image/imag - （可选）点击页面右侧的 `**Beautify**`，可以格式化JSON格式，使其更易阅读。 - 5. 点击**Send**发送请求，并获取 `task_id`。有效期 24 小时，过期后无法查询，请及时获取结果。 ``` @@ -189,7 +188,6 @@ curl -X GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id} \ 5. 点击**Send**发送请求。 - 2. 检查返回结果。重复发送此请求（轮询），直到 task\_status 变为 SUCCEEDED，获取图像的URL。图像URL有效期为**24小时**，请及时下载。 ``` diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/hermes-agent.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/hermes-agent.md index 247ca4b6..f96c40c4 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/hermes-agent.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/hermes-agent.md @@ -1,6 +1,6 @@ # Hermes Agent -Hermes Agent 是一款终端 AI 编程工具，可以通过按量计费、Coding Plan 或 Token Plan 团队版接入阿里云百炼。 +Hermes Agent 是一款终端 AI 编程工具，可以通过按量计费、Coding Plan、Token Plan 个人版或 Token Plan 团队版接入阿里云百炼。 ## **安装 Hermes Agent** @@ -31,6 +31,8 @@ Hermes Agent 是一款终端 AI 编程工具，可以通过按量计费、Coding 通过 `hermes config set` 命令配置接入参数，根据所选方案填入对应的 Base URL 和 API Key： +- **Token Plan 个人版**：个人订阅，按 token 消耗抵扣 Credits。 + - **Token Plan 团队版**：按坐席订阅，按 token 消耗抵扣 Credits。 - **Coding Plan**：固定月费订阅，按模型调用次数计量。 @@ -40,10 +42,46 @@ Hermes Agent 是一款终端 AI 编程工具，可以通过按量计费、Coding **说明** -本文示例均使用 **Anthropic 兼容协议**：Base URL 以 `/apps/anthropic` 结尾，并将 `api_mode` 设为 `anthropic_messages`。Hermes Agent 同样支持 **OpenAI 兼容协议**：将 Base URL 结尾的 `/apps/anthropic` 替换为 `/compatible-mode/v1`，并删除 `api_mode` 配置项即可。例如按量计费（华北2·北京）的 OpenAI 兼容 Base URL 为 `https://dashscope.aliyuncs.com/compatible-mode/v1`。 +本文示例均使用 **Anthropic 兼容协议**：Base URL 以 `/apps/anthropic` 结尾，并将 `api_mode` 设为 `anthropic_messages`。Hermes Agent 同样支持 **OpenAI 兼容协议**：将 Base URL 结尾的 `/apps/anthropic` 替换为 `/compatible-mode/v1`，并删除 `api_mode` 配置项即可。例如按量计费（华北2·北京）的 OpenAI 兼容 Base URL 为 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1`。 除命令行版外，Hermes Agent 还提供桌面版（Hermes Desktop）。可从 [Hermes 官网](https://hermes-agent.nousresearch.com/) 下载安装包，或在命令行版安装完成后运行 `hermes desktop` 启动。桌面版与命令行版共用同一份 `~/.hermes/config.yaml` 配置文件，接入参数与本文一致；在桌面版中以自定义端点（Custom Endpoint）方式接入时，请使用上述 OpenAI 兼容 Base URL。 +### Token Plan 个人版 + +将 `YOUR_API_KEY` 替换为 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)。可用模型：qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-flash、glm-5.2、deepseek-v4-pro，完整列表请参考 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)。 + +``` +hermes config set model.provider custom +hermes config set model.base_url https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic +hermes config set model.api_mode anthropic_messages +hermes config set model.api_key YOUR_API_KEY +hermes config set model.default qwen3.8-max-preview +``` + +以上命令将配置写入 `~/.hermes/config.yaml`。也可以直接编辑该文件，写入以下内容： + +config.yaml 配置示例 + +``` +model: + default: qwen3.8-max-preview + provider: custom + base_url: https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic + api_mode: anthropic_messages + api_key: YOUR_API_KEY +``` + +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Token Plan 团队版 将 `YOUR_API_KEY` 替换为 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)。可用模型请参考 Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-overview)。 @@ -53,7 +91,7 @@ hermes config set model.provider custom hermes config set model.base_url https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic hermes config set model.api_mode anthropic_messages hermes config set model.api_key YOUR_API_KEY -hermes config set model.default qwen3.7-max +hermes config set model.default qwen3.8-max-preview ``` 以上命令将配置写入 `~/.hermes/config.yaml`。也可以直接编辑该文件，写入以下内容： @@ -62,13 +100,24 @@ config.yaml 配置示例 ``` model: - default: qwen3.7-max + default: qwen3.8-max-preview provider: custom base_url: https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic api_mode: anthropic_messages api_key: YOUR_API_KEY ``` +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Coding Plan 将 `YOUR_API_KEY` 替换为 Coding Plan 专属 [API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan)。可用模型请参考 Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-plan)。 @@ -98,16 +147,16 @@ model: 将 `YOUR_API_KEY` 替换为[阿里云百炼 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。可用模型请参考[Anthropic 兼容 API](https://help.aliyun.com/zh/model-studio/anthropic-api-messages#07833dedefft7)。 -`base_url` 按地域设置，API Key 需与所选地域对应： +`base_url` 按地域设置，API Key 需与所选地域对应，并将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)： -- 华北2（北京）：`https://dashscope.aliyuncs.com/apps/anthropic` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/apps/anthropic`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/apps/anthropic` ``` hermes config set model.provider alibaba -hermes config set model.base_url https://dashscope.aliyuncs.com/apps/anthropic +hermes config set model.base_url https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic hermes config set model.api_mode anthropic_messages hermes config set model.api_key YOUR_API_KEY hermes config set model.default qwen3.7-max @@ -121,7 +170,7 @@ config.yaml 配置示例 model: default: qwen3.7-max provider: alibaba - base_url: https://dashscope.aliyuncs.com/apps/anthropic + base_url: https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic api_mode: anthropic_messages api_key: YOUR_API_KEY ``` @@ -150,4 +199,6 @@ hermes chat -m qwen3.7-max - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 个人版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/kilo-cli.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/kilo-cli.md index aec56898..4f710ed8 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/kilo-cli.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/kilo-cli.md @@ -1,6 +1,6 @@ # Kilo CLI -Kilo CLI 是 Kilo Code 的命令行客户端，可以通过按量计费、Coding Plan 或 Token Plan 团队版接入阿里云百炼。 +Kilo CLI 是 Kilo Code 的命令行客户端，可以通过按量计费、Coding Plan、Token Plan 个人版或 Token Plan 团队版接入阿里云百炼。 ## **安装 Kilo CLI** @@ -23,6 +23,8 @@ Kilo CLI 是 Kilo Code 的命令行客户端，可以通过按量计费、Coding 使用文本编辑器打开配置文件 `~/.config/kilo/config.json`，根据所选方案写入对应配置： +- **Token Plan 个人版**：个人订阅，按 token 消耗抵扣 Credits。 + - **Token Plan 团队版**：按坐席订阅，按 token 消耗抵扣 Credits。 - **Coding Plan**：固定月费订阅，按模型调用次数计量。 @@ -30,6 +32,89 @@ Kilo CLI 是 Kilo Code 的命令行客户端，可以通过按量计费、Coding - **按量计费**：按实际调用量后付费。 +### Token Plan 个人版 + +需先购买 Token Plan 个人版套餐且套餐处于有效期内。可在[Token Plan 个人版页面](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)购买套餐。 + +将 `YOUR_API_KEY` 替换为 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)。可用模型：qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-flash、glm-5.2、deepseek-v4-pro，完整列表请参考 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)。 + +``` +{ + "$schema": "https://kilo.ai/config.json", + "provider": { + "bailian-token-plan-personal": { + "npm": "@ai-sdk/openai-compatible", + "name": "Alibaba Cloud Model Studio", + "options": { + "baseURL": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "apiKey": "YOUR_API_KEY" + }, + "models": { + "qwen3.8-max-preview": { + "name": "Qwen3.8 Max Preview", + "contextWindow": 983616, + "maxOutputTokens": 131072, + "reasoning": true, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 262144 + } + } + }, + "qwen3.7-max": { + "name": "Qwen3.7 Max", + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 8192 + } + } + }, + "qwen3.7-plus": { + "name": "Qwen3.7 Plus", + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 8192 + } + } + }, + "qwen3.6-plus": { + "name": "Qwen3.6 Plus", + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 8192 + } + } + }, + "qwen3.6-flash": { + "name": "Qwen3.6 Flash", + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 8192 + } + } + } + } + } + } +} +``` + +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Token Plan 团队版 需先购买 Token Plan 团队版套餐且套餐处于有效期内。可在[Token Plan 团队版页面](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/overview)购买套餐。 @@ -48,6 +133,18 @@ Kilo CLI 是 Kilo Code 的命令行客户端，可以通过按量计费、Coding "apiKey": "YOUR_API_KEY" }, "models": { + "qwen3.8-max-preview": { + "name": "Qwen3.8 Max Preview", + "contextWindow": 983616, + "maxOutputTokens": 131072, + "reasoning": true, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 262144 + } + } + }, "qwen3.7-max": { "name": "Qwen3.7 Max", "options": { @@ -156,6 +253,17 @@ Kilo CLI 是 Kilo Code 的命令行客户端，可以通过按量计费、Coding } ``` +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Coding Plan 将 `YOUR_API_KEY` 替换为 Coding Plan 专属 [API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan)。可用模型请参考 Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-plan-overview#b01f82a4218kx)。 @@ -242,11 +350,11 @@ Kilo CLI 是 Kilo Code 的命令行客户端，可以通过按量计费、Coding 将 `YOUR_API_KEY` 替换为[阿里云百炼 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。可用模型请参考[OpenAI 兼容 - 支持的模型](https://help.aliyun.com/zh/model-studio/anthropic-api-messages#07833dedefft7)。 -`baseURL` 按地域设置，API Key 需与所选地域对应： +`baseURL` 按地域设置（请将 URL 中的 `{WorkspaceId}` 替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)），API Key 需与所选地域对应： -- 华北2（北京）：`https://dashscope.aliyuncs.com/compatible-mode/v1` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` - 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1` @@ -259,7 +367,7 @@ Kilo CLI 是 Kilo Code 的命令行客户端，可以通过按量计费、Coding "npm": "@ai-sdk/openai-compatible", "name": "Alibaba Cloud Model Studio", "options": { - "baseURL": "https://dashscope.aliyuncs.com/compatible-mode/v1", + "baseURL": "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", "apiKey": "YOUR_API_KEY" }, "models": { @@ -305,4 +413,6 @@ Kilo CLI 是 Kilo Code 的命令行客户端，可以通过按量计费、Coding - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 个人版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/lingma-agent.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/lingma-agent.md index addc7866..73ccab14 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/lingma-agent.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/lingma-agent.md @@ -1,6 +1,6 @@ # Qoder CN（原 Lingma） -Qoder CN（原 Lingma）是阿里云智能编码助手，提供独立 IDE，可以通过 Token Plan、Coding Plan 或按量付费接入阿里云百炼。 +Qoder CN（原 Lingma）是阿里云智能编码助手，提供独立 IDE，可以通过 Token Plan 个人版、Token Plan 团队版、Coding Plan 或按量付费接入阿里云百炼。 **说明** @@ -31,7 +31,7 @@ Qoder CN 个人社区版和个人专业版均支持接入百炼，企业版不 类型 - 根据计费方案选择 **Token Plan**、**Coding Plan** 或 **按量付费** + 根据计费方案选择 **Token Plan**（个人版或团队版）、**Coding Plan** 或 **按量付费** 模型 @@ -41,7 +41,9 @@ Qoder CN 个人社区版和个人专业版均支持接入百炼，企业版不 填写对应方案的专属 API Key： - - Token Plan 团队版：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list) + - Token Plan 个人版：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/personal) + + - Token Plan 团队版：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/enterprise) - Coding Plan：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan) @@ -55,7 +57,7 @@ Qoder CN 个人社区版和个人专业版均支持接入百炼，企业版不 ## 了解更多 -如需进一步了解 Qoder CN 的智能体、MCP、Skills 等扩展能力，请参考 [Qoder CN 官方文档](https://help.aliyun.com/zh/lingma/product-overview/introduction-of-lingma)。 +如需进一步了解 Qoder CN 的智能体、MCP、Skills 等扩展能力，请参考 [Qoder CN 官方文档](https://help.aliyun.com/zh/lingma/introduction-of-lingma)。 ## 常见问题 @@ -65,6 +67,8 @@ Qoder CN 个人社区版和个人专业版均支持接入百炼，企业版不 - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) +- Token Plan 个人版：[Token Plan 常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) + - Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) - 按量计费：[错误码](https://help.aliyun.com/zh/model-studio/error-code) @@ -85,7 +89,7 @@ Qoder CN 个人社区版和个人专业版均支持接入百炼，企业版不 - **提供商或类型与实际套餐不一致**：在 Qoder CN 模型配置中，**提供商**与**类型**需与所购套餐保持一致。例如使用 Token Plan 团队版的 API Key，但**类型**选成了 Coding Plan。 -- **选用了套餐不支持的模型**：仅支持当前套餐覆盖的文本生成模型。例如，Token Plan 团队版的支持模型列表可在[Token Plan 团队版页面](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/overview)查看。 +- **选用了套餐不支持的模型**：仅支持当前套餐覆盖的文本生成模型。例如，Token Plan 团队版的支持模型列表可在[Token Plan 团队版页面](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/token-plan/enterprise)查看。 - **临时网络或服务波动**：稍后重试。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/more-tools.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/more-tools.md index d3f45eae..ff17120c 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/more-tools.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/more-tools.md @@ -1,9 +1,31 @@ # 更多工具 -除已列出的工具外，阿里云百炼还支持接入兼容 OpenAI / Anthropic API 协议且支持自定义服务端点的第三方编程工具。可通过按量计费、Coding Plan 或 Token Plan 团队版接入。 +除已列出的工具外，阿里云百炼还支持接入兼容 OpenAI / Anthropic API 协议且支持自定义服务端点的第三方编程工具。可通过按量计费、Coding Plan、Token Plan 个人版或 Token Plan 团队版接入。 ## **配置接入凭证** +### Token Plan 个人版 + +**API 协议** + +**Base URL** + +**API Key** + +**支持模型** + +OpenAI + +`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` + +Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview) + +[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)（仅文本生成类） + +Anthropic + +`https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` + ### Token Plan 团队版 **API 协议** @@ -60,9 +82,11 @@ Anthropic OpenAI -华北2（北京）：`https://dashscope.aliyuncs.com/compatible-mode/v1` +根据地域，填入对应 URL（请将 URL 中的 `{WorkspaceId}` 替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)）： + +华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -新加坡：`https://[workspace-id].ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` +新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1` @@ -72,9 +96,9 @@ OpenAI Anthropic -华北2（北京）：`https://dashscope.aliyuncs.com/apps/anthropic` +华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic` -新加坡：`https://[workspace-id].ap-southeast-1.maas.aliyuncs.com/apps/anthropic` +新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/apps/anthropic` [支持的模型](https://help.aliyun.com/zh/model-studio/anthropic-api-messages#ae1b2c3d4e5f6) @@ -97,7 +121,7 @@ Trae 支持接入自定义模型，无需安装插件即可直接配置上述任 ## **不支持的工具类型** -Token Plan 团队版和 Coding Plan 仅限在 AI 编程工具和 OpenClaw 类型 Agent 中使用，以下类型的工具**不支持**接入： +Token Plan 个人版、Token Plan 团队版和 Coding Plan 仅限在 AI 编程工具和 OpenClaw 类型 Agent 中使用，以下类型的工具**不支持**接入： - **工作流/自动化平台**：如 Dify、n8n、Coze 等。 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/openclaw.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/openclaw.md index 6f73f11e..a1252907 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/openclaw.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/openclaw.md @@ -1,6 +1,6 @@ # OpenClaw -OpenClaw 是一个开源的个人 AI 助手平台，支持通过多种消息渠道与 AI 交互。通过配置可接入阿里云百炼平台上的 AI 模型，支持按量付费、Coding Plan、Token Plan 团队版三种接入方式。 +OpenClaw 是一个开源的个人 AI 助手平台，支持通过多种消息渠道与 AI 交互。通过配置可接入阿里云百炼平台上的 AI 模型，支持按量付费、Coding Plan、Token Plan 个人版、Token Plan 团队版四种接入方式。 ## **安装 OpenClaw** @@ -84,6 +84,142 @@ How do you want to hatch your bot? ## **配置接入凭证** +### **Token Plan 个人版** + +将 `YOUR_API_KEY` 替换为 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)。可用模型包括 qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-flash、glm-5.2、deepseek-v4-pro，完整列表请参考 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)。 + +**API Key** + +Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview) + +**Base URL** + +`https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` + +**可用模型** + +Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview) + +配置文件位于 `~/.openclaw/openclaw.json`，OpenClaw 启动时会自动读取。 + +**说明** + +示例禁用了网关鉴权（`auth.mode: none`），仅适合单机本地使用。如需共享或远程访问，请运行 `openclaw doctor --fix` 启用 token 鉴权。 + +**首次配置**：复制以下内容到配置文件，将 `YOUR_API_KEY` 替换为 Token Plan 个人版 API Key。 + +**已有配置**：若需保留已有配置，请勿直接全量替换，详见[已有配置如何安全修改](#cp-openclaw-faq-safe-modify)。 + +``` +{ + "meta": { + "lastTouchedVersion": "2026.2.1", + "lastTouchedAt": "2026-02-03T08:20:00.000Z" + }, + "models": { + "mode": "merge", + "providers": { + "bailian-token-plan": { + "baseUrl": "https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic", + "apiKey": "YOUR_API_KEY", + "api": "anthropic-messages", + "models": [ + { + "id": "qwen3.8-max-preview", + "name": "qwen3.8-max-preview", + "reasoning": true, + "input": ["text", "image"], + "contextWindow": 983616, + "maxTokens": 131072, + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "compat": { "thinkingFormat": "openai" } + }, + { + "id": "qwen3.7-max", + "name": "qwen3.7-max", + "reasoning": false, + "input": ["text"], + "contextWindow": 1000000, + "maxTokens": 65536, + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "compat": { "thinkingFormat": "openai" } + }, + { + "id": "qwen3.7-plus", + "name": "qwen3.7-plus", + "reasoning": false, + "input": ["text", "image"], + "contextWindow": 1000000, + "maxTokens": 65536, + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "compat": { "thinkingFormat": "openai" } + }, + { + "id": "qwen3.6-flash", + "name": "qwen3.6-flash", + "reasoning": false, + "input": ["text", "image"], + "contextWindow": 1000000, + "maxTokens": 32768, + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "compat": { "thinkingFormat": "openai" } + }, + { + "id": "glm-5.2", + "name": "glm-5.2", + "reasoning": false, + "input": ["text"], + "contextWindow": 1000000, + "maxTokens": 16384, + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "compat": { "thinkingFormat": "openai" } + }, + { + "id": "deepseek-v4-pro", + "name": "deepseek-v4-pro", + "reasoning": false, + "input": ["text"], + "contextWindow": 163840, + "maxTokens": 32768, + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 } + } + ] + } + } + }, + "agents": { + "defaults": { + "model": { + "primary": "bailian-token-plan/qwen3.8-max-preview" + }, + "models": { + "bailian-token-plan/qwen3.8-max-preview": {}, + "bailian-token-plan/qwen3.7-max": {}, + "bailian-token-plan/qwen3.7-plus": {}, + "bailian-token-plan/qwen3.6-flash": {}, + "bailian-token-plan/glm-5.2": {}, + "bailian-token-plan/deepseek-v4-pro": {} + } + } + }, + "gateway": { + "mode": "local", + "auth": { "mode": "none" } + } +} +``` + +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### **Token Plan 团队版** 将 `YOUR_API_KEY` 替换为 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)。可用模型请参考 Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-overview)。 @@ -134,6 +270,16 @@ Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/to "apiKey": "YOUR_API_KEY", "api": "anthropic-messages", "models": [ + { + "id": "qwen3.8-max-preview", + "name": "qwen3.8-max-preview", + "reasoning": true, + "input": ["text", "image"], + "contextWindow": 983616, + "maxTokens": 131072, + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "compat": { "thinkingFormat": "openai" } + }, { "id": "qwen3.7-max", "name": "qwen3.7-max", @@ -278,9 +424,10 @@ Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/to "agents": { "defaults": { "model": { - "primary": "bailian-token-plan/qwen3.7-plus" + "primary": "bailian-token-plan/qwen3.8-max-preview" }, "models": { + "bailian-token-plan/qwen3.8-max-preview": {}, "bailian-token-plan/qwen3.7-max": {}, "bailian-token-plan/qwen3.7-plus": {}, "bailian-token-plan/qwen3.6-plus": {}, @@ -350,6 +497,16 @@ Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/to "apiKey": "YOUR_API_KEY", "api": "anthropic-messages", "models": [ + { + "id": "qwen3.8-max-preview", + "name": "qwen3.8-max-preview", + "reasoning": true, + "input": ["text", "image"], + "contextWindow": 983616, + "maxTokens": 131072, + "cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 }, + "compat": { "thinkingFormat": "openai" } + }, { "id": "qwen3.7-max", "name": "qwen3.7-max", @@ -494,9 +651,10 @@ Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/to "agents": { "defaults": { "model": { - "primary": "bailian-token-plan/qwen3.7-plus" + "primary": "bailian-token-plan/qwen3.8-max-preview" }, "models": { + "bailian-token-plan/qwen3.8-max-preview": {}, "bailian-token-plan/qwen3.7-max": {}, "bailian-token-plan/qwen3.7-plus": {}, "bailian-token-plan/qwen3.6-plus": {}, @@ -526,6 +684,17 @@ Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/to 先单击 **Save** 按钮将配置写入磁盘，再单击 **Apply** 按钮重启网关使配置生效。 +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### **Coding Plan** 将 `YOUR_API_KEY` 替换为 Coding Plan 专属 [API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan)（格式为 `sk-sp-xxxxx`）。可用模型请参考 Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-plan)。 @@ -888,11 +1057,11 @@ Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-pla **Base URL** -请确保 Base URL、API Key 和模型归属同一地域： +请确保 Base URL、API Key 和模型归属同一地域，并将 `WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) -- 华北2（北京）：`https://dashscope.aliyuncs.com/apps/anthropic` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/apps/anthropic`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/apps/anthropic` **可用模型** @@ -931,7 +1100,7 @@ Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-pla "mode": "merge", "providers": { "bailian": { - "baseUrl": "https://dashscope.aliyuncs.com/apps/anthropic", + "baseUrl": "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic", "apiKey": "YOUR_API_KEY", "api": "anthropic-messages", "models": [ @@ -1039,7 +1208,7 @@ Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-pla "mode": "merge", "providers": { "bailian": { - "baseUrl": "https://dashscope.aliyuncs.com/apps/anthropic", + "baseUrl": "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic", "apiKey": "YOUR_API_KEY", "api": "anthropic-messages", "models": [ @@ -1215,6 +1384,14 @@ Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-pla 以上配置中 `dmPolicy` 和 `groupPolicy` 均设为 `open`，适用于测试或个人使用场景。生产环境中建议设为 `allowlist`，通过白名单限制可访问的用户和群组，降低安全风险。 +切换到 `allowlist` 模式时，将 `dmPolicy` 和 `groupPolicy` 改为 `"allowlist"`，并添加 `allowFrom` 字段，填入允许访问的工号和群 ID。示例： + +``` +"dmPolicy": "allowlist", +"groupPolicy": "allowlist", +"allowFrom": ["你的工号", "群ID"] +``` + #### 步骤四：测试 1. 执行以下命令重启网关。 @@ -1790,7 +1967,7 @@ Skill 是可扩展的能力模块，Agent 会根据请求自动匹配并加载 ### 接入 MCP 服务 -OpenClaw 支持通过 MCP（Model Context Protocol）插件扩展 Agent 的工具调用能力，例如联网搜索、网页抓取等。具体案例可以参考[添加联网搜索MCP](https://help.aliyun.com/zh/model-studio/web-search-for-coding-plan)。 +OpenClaw 支持通过 MCP（Model Context Protocol）插件扩展 Agent 的工具调用能力，例如联网搜索、网页抓取等。具体案例可以参考[添加联网搜索MCP](https://help.aliyun.com/zh/model-studio/web-search-mcp)。 ## 常见问题 diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/opencode.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/opencode.md index 2f2b6d2a..e8a3a751 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/opencode.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/opencode.md @@ -1,6 +1,6 @@ # OpenCode -OpenCode 是一款终端 AI 编程工具，可以通过按量计费、Coding Plan 或 Token Plan 团队版接入阿里云百炼。 +OpenCode 是一款终端 AI 编程工具，可以通过按量计费、Coding Plan、Token Plan 个人版或 Token Plan 团队版接入阿里云百炼。 ## **安装 OpenCode** @@ -30,6 +30,8 @@ OpenCode 是一款终端 AI 编程工具，可以通过按量计费、Coding Pla 根据所选方案写入对应配置： +- **Token Plan 个人版**：个人订阅，按 token 消耗抵扣 Credits。 + - **Token Plan 团队版**：按坐席订阅，按 token 消耗抵扣 Credits。 - **Coding Plan**：固定月费订阅，按模型调用次数计量。 @@ -37,6 +39,102 @@ OpenCode 是一款终端 AI 编程工具，可以通过按量计费、Coding Pla - **按量计费**：按实际调用量后付费。 +### Token Plan 个人版 + +需先购买 Token Plan 个人版套餐且套餐处于有效期内。可在[Token Plan 个人版页面](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)购买套餐。 + +将 `YOUR_API_KEY` 替换为 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview)。可用模型包括 qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-flash、glm-5.2、deepseek-v4-pro，详细列表请参考 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)。 + +``` +{ + "$schema": "https://opencode.ai/config.json", + "provider": { + "bailian-token-plan-personal": { + "npm": "@ai-sdk/anthropic", + "name": "Alibaba Cloud Model Studio", + "options": { + "baseURL": "https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic/v1", + "apiKey": "YOUR_API_KEY" + }, + "models": { + "qwen3.8-max-preview": { + "name": "Qwen3.8 Max Preview", + "contextWindow": 983616, + "maxOutputTokens": 131072, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 99072 + }, + "temperature": 0.6, + "reasoning": true + } + }, + "qwen3.7-max": { + "name": "Qwen3.7 Max", + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 8192 + } + } + }, + "qwen3.7-plus": { + "name": "Qwen3.7 Plus", + "modalities": { + "input": ["text", "image"], + "output": ["text"] + }, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 8192 + } + } + }, + "qwen3.6-plus": { + "name": "Qwen3.6 Plus", + "modalities": { + "input": ["text", "image"], + "output": ["text"] + }, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 8192 + } + } + }, + "qwen3.6-flash": { + "name": "Qwen3.6 Flash", + "modalities": { + "input": ["text", "image"], + "output": ["text"] + }, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 8192 + } + } + } + } + } + } +} +``` + +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Token Plan 团队版 需先购买 Token Plan 团队版套餐且套餐处于有效期内。可在[Token Plan 团队版页面](https://bailian.console.aliyun.com/?tab=plan#/efm/subscription/overview)购买套餐。 @@ -55,6 +153,19 @@ OpenCode 是一款终端 AI 编程工具，可以通过按量计费、Coding Pla "apiKey": "YOUR_API_KEY" }, "models": { + "qwen3.8-max-preview": { + "name": "Qwen3.8 Max Preview", + "contextWindow": 983616, + "maxOutputTokens": 131072, + "options": { + "thinking": { + "type": "enabled", + "budgetTokens": 99072 + }, + "temperature": 0.6, + "reasoning": true + } + }, "qwen3.7-max": { "name": "Qwen3.7 Max", "options": { @@ -187,6 +298,17 @@ OpenCode 是一款终端 AI 编程工具，可以通过按量计费、Coding Pla } ``` +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Coding Plan 将 `YOUR_API_KEY` 替换为 Coding Plan 专属 [API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan)。可用模型请参考 Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-plan)。 @@ -289,11 +411,11 @@ OpenCode 是一款终端 AI 编程工具，可以通过按量计费、Coding Pla 将 `YOUR_API_KEY` 替换为[阿里云百炼 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。可用模型请参考[Anthropic 兼容 API](https://help.aliyun.com/zh/model-studio/anthropic-api-messages#07833dedefft7)。 -`baseURL` 按地域设置，API Key 需与所选地域对应： +`baseURL` 按地域设置（将 `{WorkspaceId}` 替换为真实的 [Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)），API Key 需与所选地域对应： -- 华北2（北京）：`https://dashscope.aliyuncs.com/apps/anthropic/v1` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/apps/anthropic/v1`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` ``` @@ -301,10 +423,10 @@ OpenCode 是一款终端 AI 编程工具，可以通过按量计费、Coding Pla "$schema": "https://opencode.ai/config.json", "provider": { "bailian-payg": { - "npm": "@ai-sdk/anthropic", + "npm": "@ai-sdk/openai-compatible", "name": "Alibaba Cloud Model Studio", "options": { - "baseURL": "https://dashscope.aliyuncs.com/apps/anthropic/v1", + "baseURL": "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", "apiKey": "YOUR_API_KEY" }, "models": { @@ -368,4 +490,6 @@ OpenCode 是一款终端 AI 编程工具，可以通过按量计费、Coding Pla - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 个人版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qoder-agent.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qoder-agent.md index 1de85109..04e9150e 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qoder-agent.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qoder-agent.md @@ -1,6 +1,6 @@ # Qoder -Qoder 是面向软件开发的 Agentic 编码平台，支持桌面 IDE、CLI 和 JetBrains 插件，可以通过按量付费、Coding Plan 或 Token Plan 团队版接入阿里云百炼。 +Qoder 是面向软件开发的 Agentic 编码平台，支持桌面 IDE、CLI 和 JetBrains 插件，可以通过按量付费、Coding Plan、Token Plan 个人版或 Token Plan 团队版接入阿里云百炼。 ## Qoder IDE @@ -27,7 +27,7 @@ Qoder 是面向软件开发的 Agentic 编码平台，支持桌面 IDE、CLI 和 类型 - 根据计费方案选择 **Token Plan**、**Coding Plan** 或 **按量付费** + 根据计费方案选择 **Token Plan**（个人版或团队版）、**Coding Plan** 或 **按量付费** 模型 @@ -37,7 +37,9 @@ Qoder 是面向软件开发的 Agentic 编码平台，支持桌面 IDE、CLI 和 填写对应方案的专属 API Key： - - Token Plan 团队版：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list) + - Token Plan 个人版：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/personal) + + - Token Plan 团队版：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/enterprise) - Coding Plan：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan) @@ -99,7 +101,7 @@ Qoder 是面向软件开发的 Agentic 编码平台，支持桌面 IDE、CLI 和 1. 在对话框中输入 `/model`，通过 Tab 键切换至 `Custom`。 -2. 回车选择 Add custom model，提供商选择 Alibaba Cloud Model Studio - China，类型根据计费方案选择 **Token Plan**、**Coding Plan** 或 **按量付费**。 +2. 回车选择 Add custom model，提供商选择 Alibaba Cloud Model Studio - China，类型根据计费方案选择 **Token Plan**（个人版或团队版）、**Coding Plan** 或 **按量付费**。 3. 选择模型后输入对应方案的专属 API Key，确认后等待配置生效。 @@ -135,7 +137,7 @@ Qoder 是面向软件开发的 Agentic 编码平台，支持桌面 IDE、CLI 和 类型 - 根据计费方案选择 **Token Plan**、**Coding Plan** 或 **按量付费** + 根据计费方案选择 **Token Plan**（个人版或团队版）、**Coding Plan** 或 **按量付费** 模型 @@ -145,7 +147,9 @@ Qoder 是面向软件开发的 Agentic 编码平台，支持桌面 IDE、CLI 和 填写对应方案的专属 API Key： - - Token Plan 团队版：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list) + - Token Plan 个人版：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/personal) + + - Token Plan 团队版：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/enterprise) - Coding Plan：[获取 API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan) @@ -195,8 +199,12 @@ Qoder 是面向软件开发的 Agentic 编码平台，支持桌面 IDE、CLI 和 - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) +- Token Plan 个人版：[Token Plan 个人版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) + - Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- 按量计费：[错误码](https://help.aliyun.com/zh/model-studio/error-code) + ### 为什么在 Qoder 设置中找不到模型选项？ diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qwen-code.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qwen-code.md index 7ca6ebb6..ec3143da 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qwen-code.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qwen-code.md @@ -1,6 +1,6 @@ # Qwen Code -Qwen Code 是一款终端 AI 编程工具，可以通过按量计费、Coding Plan 或 Token Plan 团队版接入阿里云百炼。 +Qwen Code 是一款终端 AI 编程工具，可以通过按量计费、Coding Plan、Token Plan 个人版或 Token Plan 团队版接入阿里云百炼。 ## **安装 Qwen Code** @@ -33,8 +33,10 @@ Qwen Code 是一款终端 AI 编程工具，可以通过按量计费、Coding Pl ## **配置接入凭证** -启动 Qwen Code 后输入 `/auth` 命令进行可视化配置。阿里云百炼提供三种计费方案，根据需要选择： +启动 Qwen Code 后输入 `/auth` 命令进行可视化配置。阿里云百炼提供四种计费方案，根据需要选择： +- **Token Plan 个人版**：按 token 消耗抵扣个人 Credits。 + - **Token Plan 团队版**：按坐席订阅，按 token 消耗抵扣 Credits。 - **Coding Plan**：固定月费订阅，按模型调用次数计量。 @@ -42,6 +44,98 @@ Qwen Code 是一款终端 AI 编程工具，可以通过按量计费、Coding Pl - **按量计费**：按实际调用量后付费。 +### Token Plan 个人版 + +启动 Qwen Code 后输入 `/auth`，依次选择 **订阅计划** > **阿里云百炼 Token Plan**，输入 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/overview) 即可完成配置。可用模型请参考 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)。 + +高级配置：通过 settings.json 配置文件 + +编辑或新建 `settings.json` 文件，将 `YOUR_API_KEY` 替换为 Token Plan 个人版专属 API Key。文件路径如下： + +- macOS/Linux：`~/.qwen/settings.json` + +- Windows：`C:\Users\\.qwen\settings.json` + + +``` +{ + "env": { + "BAILIAN_TOKEN_PLAN_API_KEY": "YOUR_API_KEY" + }, + "modelProviders": { + "openai": [ + { + "id": "qwen3.8-max-preview", + "name": "[Token Plan 个人版] qwen3.8-max-preview", + "baseUrl": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "envKey": "BAILIAN_TOKEN_PLAN_API_KEY", + "generationConfig": { + "extra_body": { + "enable_thinking": true + } + } + }, + { + "id": "qwen3.7-max", + "name": "[Token Plan 个人版] qwen3.7-max", + "baseUrl": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "envKey": "BAILIAN_TOKEN_PLAN_API_KEY", + "generationConfig": { + "extra_body": { + "enable_thinking": true + } + } + }, + { + "id": "qwen3.7-plus", + "name": "[Token Plan 个人版] qwen3.7-plus", + "baseUrl": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "envKey": "BAILIAN_TOKEN_PLAN_API_KEY", + "generationConfig": { + "extra_body": { + "enable_thinking": true + } + } + }, + { + "id": "qwen3.6-flash", + "name": "[Token Plan 个人版] qwen3.6-flash", + "baseUrl": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "envKey": "BAILIAN_TOKEN_PLAN_API_KEY", + "generationConfig": { + "extra_body": { + "enable_thinking": true + } + } + }, + { + "id": "glm-5.2", + "name": "[Token Plan 个人版] glm-5.2", + "baseUrl": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "envKey": "BAILIAN_TOKEN_PLAN_API_KEY" + }, + { + "id": "deepseek-v4-pro", + "name": "[Token Plan 个人版] deepseek-v4-pro", + "baseUrl": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "envKey": "BAILIAN_TOKEN_PLAN_API_KEY" + } + ] + } +} +``` + +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Token Plan 团队版 启动 Qwen Code 后输入 `/auth`，依次选择 **订阅计划** > **阿里云百炼 Token Plan**，输入 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list) 即可完成配置。可用模型请参考 Token Plan 团队版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-overview)。 @@ -62,6 +156,17 @@ Qwen Code 是一款终端 AI 编程工具，可以通过按量计费、Coding Pl }, "modelProviders": { "openai": [ + { + "id": "qwen3.8-max-preview", + "name": "[Token Plan 团队版] qwen3.8-max-preview", + "baseUrl": "https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", + "envKey": "BAILIAN_TOKEN_PLAN_API_KEY", + "generationConfig": { + "extra_body": { + "enable_thinking": true + } + } + }, { "id": "qwen3.7-max", "name": "[Token Plan 团队版] qwen3.7-max", @@ -207,12 +312,23 @@ Qwen Code 是一款终端 AI 编程工具，可以通过按量计费、Coding Pl "region": "china" }, "model": { - "name": "qwen3.7-plus" + "name": "qwen3.8-max-preview" }, "$version": 3 } ``` +**重要** + +**qwen3.8-max-preview 思考模式说明**： + +- thinking：始终开启，不支持关闭。 + +- temperature：思考模式下默认值为 0.6；传入值小于 0.6 时自动调整为 0.6。 + +- reasoning\_effort：控制推理深度，可选 xhigh、medium、low，默认 xhigh。 + + ### Coding Plan 启动 Qwen Code 后输入 `/auth`，依次选择 **订阅计划** > **阿里云百炼 Coding Plan**，选择 Coding Plan 区域（china），输入 Coding Plan 专属 [API Key](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/efm/coding_plan) 即可完成配置。可用模型请参考 Coding Plan [支持的模型](https://help.aliyun.com/zh/model-studio/coding-plan)。 @@ -348,11 +464,11 @@ Qwen Code 是一款终端 AI 编程工具，可以通过按量计费、Coding Pl - Windows：`C:\Users\\.qwen\settings.json` -`baseUrl` 按地域设置，API Key 需与所选地域对应： +`baseUrl` 按地域设置（URL 中的 `{WorkspaceId}` 需替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)），API Key 需与所选地域对应： -- 华北2（北京）：`https://dashscope.aliyuncs.com/compatible-mode/v1` +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`，请将`WorkspaceId`替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu) +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` - 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1` @@ -367,7 +483,7 @@ Qwen Code 是一款终端 AI 编程工具，可以通过按量计费、Coding Pl { "id": "qwen3.6-plus", "name": "[Bailian] qwen3.6-plus", - "baseUrl": "https://dashscope.aliyuncs.com/compatible-mode/v1", + "baseUrl": "https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", "envKey": "BAILIAN_API_KEY", "generationConfig": { "extra_body": { @@ -414,6 +530,27 @@ Qwen Code 支持在 VS Code 中以插件方式使用，在 IDE 中提供 AI 编 3. 点击右上角图标启动 Qwen Code，通过输入或点击`/`，选择`Switch model`切换模型。 +## **Qwen Code 桌面版** + +Qwen Code Desktop 是 Qwen Code 的图形界面应用，内置 CLI 运行时，无需额外安装命令行工具。桌面版与 CLI 共用同一份 `settings.json` 配置文件，如果已通过 CLI 完成配置，桌面版打开即可直接使用，无需重复配置。 + +如果尚未配置，可通过桌面版图形界面完成： + +1. 打开 Qwen Code Desktop，点击左侧**设置**，选择 **AI** 页签。 + +2. 在**模型 Provider**区域点击**\+ 连接**。 + +3. 在弹窗中选择 **Alibaba ModelStudio** 页签，根据计费方案选择对应项： + + - **Coding Plan**：个人开发者，按次计量。 + + - **Token Plan**：团队使用，按 token 消耗抵扣。 + + - **Standard API Key**：使用已有的百炼 API Key 按量计费。 + +4. 按照页面提示输入 API Key，完成配置。 + + ## **常见命令** **说明** @@ -547,7 +684,7 @@ Qwen Code 支持在 VS Code 中以插件方式使用，在 IDE 中提供 AI 编 4. 输入以下内容安装 skill。 ``` - 查看我是否有find skills，没有就直接帮我安装：npx skills add https://github.com/vercel-labs/skills --skill find-skills -y -a qwen-code，然后帮我安装 web-component-design 到当前目录qwen code skills中。 + 查看我是否有find skills，没有就直接帮我安装：npx skills add https://github.com/vercel-labs/skills --skill find-skills -y -a qwen-code，然后从 wshobson/agents 帮我安装 web-component-design 到当前目录：npx skills add https://github.com/wshobson/agents --skill web-component-design -y ``` 5. 下载[website.png](https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20260318/ymehla/website.png)到项目目录，输入以下内容，将自动识别截图的布局、样式，生成网页代码。 @@ -593,7 +730,9 @@ Qwen Code 支持在 VS Code 中以插件方式使用，在 IDE 中提供 AI 编 - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) ### **如何切换模型？** diff --git a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qwenpaw.md b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qwenpaw.md index 00dddafd..fd715d26 100644 --- a/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qwenpaw.md +++ b/skills/bailian-docs-llm-wiki/raw/model-user-guide/use-chat-client-or-development-tool/qwenpaw.md @@ -1,6 +1,6 @@ # QwenPaw -QwenPaw（原 CoPaw）是 AgentScope 团队开源的个人 AI 助手，支持本地或云端部署，可通过 Token Plan 团队版、Coding Plan 或按量计费接入阿里云百炼。 +QwenPaw（原 CoPaw）是 AgentScope 团队开源的个人 AI 助手，支持本地或云端部署，可通过 Token Plan 个人版、Token Plan 团队版、Coding Plan 或按量计费接入阿里云百炼。 ## **安装 QwenPaw** @@ -52,6 +52,22 @@ qwenpaw app 在 Console 点击 **设置** > **模型**，根据计费方案配置对应的提供商。 +### Token Plan 个人版 + +进入内置的 **Aliyun Token Plan** 提供商**设置**页面，填入 API Key。 + +**配置项** + +**说明** + +**API 密钥** + +填入 Token Plan 个人版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/personal)。 + +**模型** + +已预设常用模型。新增模型点击**添加模型**，**模型 ID**填入 Token Plan 个人版[支持的模型](https://help.aliyun.com/zh/model-studio/token-plan-personal-overview)。 + ### Token Plan 团队版 进入内置的 **Aliyun Token Plan** 提供商**设置**页面，填入 API Key。 @@ -62,7 +78,7 @@ qwenpaw app **API 密钥** -填入 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/uac-admin/organization/members/list)。 +填入 Token Plan 团队版专属 [API Key](https://bailian.console.aliyun.com/cn-beijing?tab=plan#/efm/subscription/token-plan/enterprise)。 **模型** @@ -98,9 +114,9 @@ qwenpaw app **基础 URL** -根据模型部署地域，在下拉菜单选择对应 URL： +根据模型部署地域，在下拉菜单选择对应 URL（请将 URL 中的 `{WorkspaceId}` 替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)）： -- China (Beijing)：`https://dashscope.aliyuncs.com/compatible-mode/v1` +- China (Beijing)：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` - International (Singapore)：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`，将 `WorkspaceId` 替换为真实的[获取Workspace ID](https://help.aliyun.com/zh/model-studio/obtain-the-app-id-and-workspace-id#d3eb3cd37b7fu)。 @@ -125,7 +141,9 @@ qwenpaw app - Coding Plan：[Coding Plan 常见问题](https://help.aliyun.com/zh/model-studio/coding-plan-faq) -- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-faq) +- Token Plan 个人版：[Token Plan 个人版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) + +- Token Plan 团队版：[Token Plan 团队版常见问题](https://help.aliyun.com/zh/model-studio/token-plan-team-faq) ### 报错 401 Incorrect API key provided diff --git a/skills/bailian-docs-llm-wiki/wiki/api/3d-generation.md b/skills/bailian-docs-llm-wiki/wiki/api/3d-generation.md index 74caf7e8..c4156a5a 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/3d-generation.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/3d-generation.md @@ -1,66 +1,138 @@ # 3d generation -百炼平台基于 Tripo 模型提供 3D 资产生成能力，支持文生 3D、单图生 3D 与多图生 3D 三种输入方式，产出带贴图的 PBR 材质 GLB 模型或无贴图基础模型。由于生成耗时较长，API 采用[异步调用](../concepts/async-invocation.md)，整体流程为「创建任务 → 轮询获取结果」。详细接口与参数见 [Tripo-3D模型生成](../../raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md)。 +百炼平台通过集成 Tripo 系列模型，提供 3D 模型生成能力，支持**文生3D**、**单图生3D**和**多图生3D**三种输入方式，输出 GLB 格式的带贴图或无贴图三维模型。API 采用[异步调用](../concepts/async-invocation.md)模式，整体流程分为"创建任务 → 轮询查询"两步。详细接口规范参见 [Tripo-3D模型生成](../../raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md)。 -## 适用范围 +> **注意**：本接口**仅适用于华北2（北京）地域**，必须使用北京地域的 [API Key](../concepts/api-key.md)，其他地域的 Key 将无法调用成功。 -- 仅适用于**华北2（北京）**地域，且必须使用该地域的 [API Key](../concepts/api-key.md)。 -- 需先在百炼控制台模型市场搜索「Tripo」并开通服务、完成授权，再配置好 [API Key](../concepts/api-key.md) 环境变量。具体开通与配置步骤见 [Tripo-3D模型生成](../../raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md)。 - -## 调用流程 - -API 仅支持[异步调用](../concepts/async-invocation.md)，包含两个步骤： +## 支持的模型 -1. **创建任务**：`POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation` -2. **轮询查询结果**：`GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}` +| 模型名称 | 说明 | 最高面数 | +| --- | --- | --- | +| `Tripo/Tripo-H3.1` | 高精度3D模型，对应 Tripo 官方 API `v3.1-20260211` | 200 万面（ultra 模式） | +| `Tripo/Tripo-P1.0` | 专业3D模型，速度更快，对应 Tripo 官方 API `P1-20260311` | 2 万面 | -创建任务时必须携带 `X-DashScope-Async: enable` 请求头，否则会报错 `current user api does not support synchronous calls`。成功创建后返回 `task_id`，有效期 24 小时，**请勿重复创建任务**，轮询获取即可。 +## 前置配置 -轮询建议间隔约 15 秒，任务状态流转为 `PENDING`（排队中）→ `RUNNING`（处理中）→ `SUCCEEDED` / `FAILED`。查询接口默认 RPS 为 20，如需更高频查询或事件通知建议配置异步任务回调。 +1. 在[阿里云百炼控制台](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/all)搜索"Tripo"并开通服务。 +2. 获取**北京地域** [API Key](../concepts/api-key.md) 并配置到环境变量 `DASHSCOPE_API_KEY`。 -## 支持的模型 +## 调用方式 -| 模型名 | 定位 | 输出面数 | 对应官方 API 版本 | -| --- | --- | --- | --- | -| `Tripo/Tripo-H3.1` | 高精度生成 | 最高 200 万面 | `v3.1-20260211` | -| `Tripo/Tripo-P1.0` | 专业生成，速度更快 | 最高 2 万面 | `P1-20260311` | +所有请求均为异步 HTTP，需在请求头加 `X-DashScope-Async: enable`，否则报错 `"current user api does not support synchronous calls"`。 -## 输入方式 +### 步骤一：创建任务 -`input` 中 `prompt`、`image`、`images` 三者**互斥**，只能选其一，同时传多个将报错。 +``` +POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation +``` -- **文生 3D**：`prompt` 必填，支持中英文等多语言，每个字符计 1 个字符，最大 1024 字符。 -- **单图生 3D**：`image` 必填，传入单张图像公网 URL。图像格式限 JPEG/PNG，宽高范围 [20, 6000] 像素（建议边长大于 256），文件不超过 20MB，支持 HTTP/HTTPS。 -- **多图生 3D**：`images` 必填，数组长度固定为 4，对应视角顺序为**前、左、后、右**；不需要的视角传空对象 `{}`。实际有效图片数为 2~4 张，多张图像的分辨率和宽高比不要求一致。每个对象含 `type`（`jpeg` 或 `png`）与 `file_token`（公网 URL）字段。 +请将 `{WorkspaceId}` 替换为真实的[业务空间](../concepts/workspace.md) ID。 -## 关键参数（parameters） +#### input 参数（三选一） -| 参数 | 适用模型 | 默认值 | 说明 | +| 字段 | 类型 | 使用场景 | 限制 | | --- | --- | --- | --- | -| `texture_quality` | 全部 | `standard` | 贴图质量，可选 `standard`（标清）/`detailed`（高清） | -| `geometry_quality` | `Tripo/Tripo-H3.1` | `standard` | 几何精度，`standard` 最高 150 万面，`ultra` 最高 200 万面 | -| `pbr` | 全部 | `true` | 是否生成 PBR 材质模型。设为 `true` 时强制启用贴图，返回 `pbr_model_url` | -| `texture` | 全部 | `true` | 是否生成贴图。生成无贴图模型需**同时**将 `texture` 和 `pbr` 设为 `false`，返回 `base_model_url` | - -## 响应与产物 +| `prompt` | string | 文生3D | 最大 1024 字符，支持中英文 | +| `image` | string（URL） | 单图生3D | JPEG/PNG，宽高 [20, 6000]px，≤ 20MB | +| `images` | array[object] | 多图生3D | 固定长度 4，对应前/左/后/右，有效图片 2～4 张 | -成功响应的 `output.results` 仅在 `task_status` 为 `SUCCEEDED` 时返回，包含以下字段： +`prompt`、`image`、`images` 三者互斥，同时传入多个将报错。 -- `pbr_model_url`：PBR 材质模型（GLB）下载 URL，当 `pbr` 为 `true`（默认）时返回。 -- `base_model_url`：无贴图基础模型（GLB）下载 URL，当 `texture` 与 `pbr` 均为 `false` 时返回。 -- `rendered_image_url`：3D 模型预览渲染图（1 张）URL。 +多图模式下，`images` 数组长度固定为 4，不需要的视角传入空对象 `{}` 即可，顺序为**前、左、后、右**。 -> **注意**：以上下载链接有效期均为 **2 小时**，请及时下载。 +#### parameters 参数 -`usage` 字段记录任务类型（`text-to-3d` / `image-to-3d` / `multi-image-to-3d`）、生成数量 `count`、贴图质量与几何精度，仅对成功结果计数。任务状态 `task_status` 的完整枚举为 `PENDING` / `RUNNING` / `SUCCEEDED` / `FAILED` / `CANCELED` / `UNKNOWN`，其中 `UNKNOWN` 表示任务不存在或超过 24 小时有效期。更多响应字段说明见 [Tripo-3D模型生成](../../raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md)。 +| 字段 | 默认值 | 说明 | 适用模型 | +| --- | --- | --- | --- | +| `texture_quality` | `standard` | 贴图质量：`standard`（标清）/ `detailed`（高清） | 全部 | +| `geometry_quality` | `standard` | 几何精度：`standard`（最高150万面）/ `ultra`（最高200万面） | 仅 `Tripo/Tripo-H3.1` | +| `pbr` | `true` | 是否生成 PBR 材质，设为 `true` 时强制启用贴图 | 全部 | +| `texture` | `true` | 是否生成贴图；无贴图时需同时设 `texture=false` 和 `pbr=false` | 全部 | + +#### 创建成功响应示例 + +```json +{ + "output": { + "task_status": "PENDING", + "task_id": "0385dc79-5ff8-4d82-bcb6-xxxxxx" + }, + "request_id": "4909100c-7b5a-9f92-bfe5-xxxxxx" +} +``` + +保存 `task_id`，有效期 **24 小时**，切勿重复创建任务。 + +### 步骤二：查询任务结果 + +``` +GET https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/tasks/{task_id} +``` + +建议每隔 **15 秒**轮询一次，查询接口默认 RPS 为 20。任务状态流转： + +``` +PENDING → RUNNING → SUCCEEDED / FAILED +``` + +| 状态 | 说明 | +| --- | --- | +| `PENDING` | 排队中 | +| `RUNNING` | 处理中 | +| `SUCCEEDED` | 成功，返回模型 URL | +| `FAILED` | 失败，返回错误码和信息 | +| `CANCELED` | 已取消 | +| `UNKNOWN` | 不存在或超过 24 小时有效期 | + +#### 成功响应中的 results 字段 + +| 字段 | 说明 | 返回条件 | +| --- | --- | --- | +| `pbr_model_url` | PBR 材质模型（GLB），有效期 2 小时 | `pbr=true`（默认） | +| `base_model_url` | 无贴图基础模型（GLB），有效期 2 小时 | `texture=false` 且 `pbr=false` | +| `rendered_image_url` | 模型预览渲染图（1张），有效期 2 小时 | 始终返回 | +| `orig_prompt` | 原始 [prompt](../guides/prompt.md) 输入 | 文生3D 模式 | + +> **注意**：模型文件和渲染图 URL **有效期仅 2 小时**，请及时下载，过期后需重新生成任务。 + +## 示例代码 + +以下示例均来自 [Tripo-3D模型生成](../../raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md)，仅展示关键差异。 + +**文生3D（有贴图）** + +```bash +curl --location 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/video-generation/3d-generation' \ + -H 'X-DashScope-Async: enable' \ + -H "Authorization: Bearer $DASHSCOPE_API_KEY" \ + -H 'Content-Type: application/json' \ + -d '{ + "model": "Tripo/Tripo-P1.0", + "input": { "prompt": "一只可爱的猫" }, + "parameters": { "texture_quality": "standard" } +}' +``` + +**无贴图生成**（需同时设 `texture` 和 `pbr` 为 `false`） + +```bash +# 参数中同时设置 texture 和 pbr 为 false +-d '{ + "model": "Tripo/Tripo-P1.0", + "input": { "prompt": "一只可爱的猫" }, + "parameters": { "texture": false, "pbr": false } +}' +``` ## 限制与注意事项 -- 仅限北京地域 [API Key](../concepts/api-key.md) 调用，地域不匹配将无法使用。 -- `task_id` 查询有效期 24 小时，超时返回 `UNKNOWN` 且无法再查询。 -- 查询接口默认 RPS 限制为 20，建议通过异步任务回调获取更高频通知。 -- 产物下载链接有效期仅 2 小时。 -- 调用失败时响应中会返回 `code` 与 `message`，可参照百炼错误码文档排查。 +- **地域限制**：仅支持华北2（北京）地域，[API Key](../concepts/api-key.md) 也必须是北京地域的，详见 [Tripo-3D模型生成](../../raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md)。 +- **异步强制**：HTTP 调用必须携带 `X-DashScope-Async: enable`，不支持同步调用。 +- **task_id 有效期**：24 小时，超时后状态返回 `UNKNOWN`，无法补查结果。 +- **input 互斥**：`prompt`、`image`、`images` 三者只能选其一，混用会报错。 +- **多图视角顺序**：`images` 数组固定长度为 4，顺序为前、左、后、右，缺省视角传 `{}`。 +- **模型 URL 时效**：生成结果 URL（模型文件、渲染图）有效期仅 2 小时，请及时下载。 +- **错误排查**：调用失败可参考[错误码文档](https://help.aliyun.com/zh/model-studio/error-code)。 ## 来源文档 @@ -71,18 +143,4 @@ API 仅支持[异步调用](../concepts/async-invocation.md)，包含两个步 - - - - - - - - - - - - - - diff --git a/skills/bailian-docs-llm-wiki/wiki/api/application-call.md b/skills/bailian-docs-llm-wiki/wiki/api/application-call.md index eb343a80..305fb682 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/application-call.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/application-call.md @@ -1,6 +1,6 @@ # application call -阿里云百炼平台提供两套 API 来调用智能体和工作流应用：**OpenAI 兼容的 Responses API** 和 **DashScope API**。两者均支持同步/[异步调用](../concepts/async-invocation.md)、多轮对话、[流式输出](../concepts/streaming.md)等核心能力，开发者可根据生态兼容性和功能需求选择合适的接入方式。调用前需先获取 APP ID（以及子[业务空间](../concepts/workspace.md)场景下的 Workspace ID）和 [API Key](../concepts/api-key.md)。 +阿里云百炼平台提供两套 API 来调用智能体和[工作流](../concepts/workflow.md)应用：**OpenAI 兼容的 Responses API** 和 **DashScope API**。两者均支持同步/[异步调用](../concepts/async-invocation.md)、多轮对话、[流式输出](../concepts/streaming.md)等核心能力，开发者可根据生态兼容性和功能需求选择合适的接入方式。调用前需先获取 APP ID（以及子[业务空间](../concepts/workspace.md)场景下的 Workspace ID）和 [API Key](../concepts/api-key.md)。 ## 前置准备 @@ -16,7 +16,7 @@ ### 其他前提 - 已获取 [API Key](../concepts/api-key.md) 并配置到环境变量 `DASHSCOPE_API_KEY`。 -- 已创建并发布百炼应用（智能体或工作流）。 +- 已创建并发布百炼应用（智能体或[工作流](../concepts/workflow.md)）。 - 如使用 SDK 调用，需安装对应的 SDK（OpenAI SDK 或 [DashScope SDK](../concepts/dashscope-sdk.md)）。 ## 两套 API 对比 @@ -81,7 +81,7 @@ Responses API 支持在 `content` 数组中混合多种输入类型： ### [流式输出](../concepts/streaming.md) -设置 `stream=true` 可边生成边输出，适用于需要实时展示生成内容的场景。若应用类型为工作流，需在结束节点或流程输出节点中启用「[流式输出](../concepts/streaming.md)」开关并重新发布。 +设置 `stream=true` 可边生成边输出，适用于需要实时展示生成内容的场景。若应用类型为[工作流](../concepts/workflow.md)，需在结束节点或流程输出节点中启用「[流式输出](../concepts/streaming.md)」开关并重新发布。 ## DashScope API @@ -145,6 +145,19 @@ response = await client.responses.create( + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/application-component-api-reference.md b/skills/bailian-docs-llm-wiki/wiki/api/application-component-api-reference.md index 65de9baf..1a24b8a9 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/application-component-api-reference.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/application-component-api-reference.md @@ -1,165 +1,57 @@ # application component api reference -百炼平台应用组件 API（`bailian/2023-12-29`）提供了数据连接、知识库、Prompt 模板、长期记忆等核心能力的 OpenAPI 接口，采用 ROA 签名风格。开发者可通过阿里云百炼 SDK 直接调用，也可使用自签名方式对接。所有接口均需传入 `WorkspaceId`（[业务空间](../concepts/workspace.md) ID），RAM 子账号需要先获取对应权限策略并加入[业务空间](../concepts/workspace.md)后才能调用。 +阿里云百炼平台组件 OpenAPI（`bailian/2023-12-29`）是一组管控类接口，覆盖数据连接（原应用数据）、Prompt 模板、知识库（Index）三大能力域，以及切片管理、长期记忆等扩展接口。该 API 采用 ROA 签名风格，官方强烈建议通过多语言预置 SDK 调用而非自签名对接，详见 [API概览](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-overview.md)。 -## 服务接入点与鉴权 +## 接入与鉴权 -当前支持两个地域的接入点： +- **服务接入点**：目前支持华北2（北京，`bailian.cn-beijing.aliyuncs.com`）与新加坡（`bailian.ap-southeast-1.aliyuncs.com`）两个地域，均提供公网与 VPC 接入地址，完整列表见 [服务接入点](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-endpoint.md)。 +- **凭证**：使用阿里云 AccessKey（非百炼 API-Key）签名调用。建议创建仅具备 API 访问权限的 RAM 用户，并按最小权限原则（PoLP）授权。 +- **RAM 授权**：产品 RamCode 为 `sfm`，授权粒度为操作级。RAM 用户（子账号）通常需要 `AliyunBailianDataFullAccess` 策略（只读类接口也可用 `AliyunBailianDataReadOnlyAccess`），并**加入一个[业务空间](../concepts/workspace.md)**后才能调用；阿里云账号（主账号）可直接调用。各接口对应的 Action（如 `sfm:CreateIndex`、`sfm:Retrieve`）见 [授权信息](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-ram.md)。 +- **公共路径参数**：几乎所有接口都以 `WorkspaceId`（[业务空间](../concepts/workspace.md) ID，形如 `llm-xxxx`）作为路径参数，调用前需先确认目标资源所在的[业务空间](../concepts/workspace.md)。 -| 地域 | 地域 ID | 公网接入地址 | VPC 接入地址 | -|------|---------|-------------|-------------| -| 华北2（北京） | cn-beijing | bailian.cn-beijing.aliyuncs.com | bailian-vpc.cn-beijing.aliyuncs.com | -| 新加坡 | ap-southeast-1 | bailian.ap-southeast-1.aliyuncs.com | bailian-vpc.ap-southeast-1.aliyuncs.com | +## 能力分组与核心接口 -调用前需准备 AccessKey，建议使用 RAM 用户而非主账号以降低安全风险。RAM 权限策略的 RamCode 为 `sfm`，授权粒度为操作级。大多数写操作需要 `AliyunBailianDataFullAccess` 策略，部分只读接口（如 DescribeFile、GetIndexJobStatus）也支持 `AliyunBailianDataReadOnlyAccess`。详见[授权信息](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-ram.md)。 +### 数据连接（原应用数据） -## 数据连接（原应用数据） +管理类目、文件、解析设置与连接器： -数据连接相关 API 用于管理类目、文件、解析设置、表格和连接器，是构建知识库的数据基础。 +- **类目**：`AddCategory`（每个业务空间最多 500 个类目）、`ListCategory`、`DeleteCategory`。 +- **文件导入**：标准流程为 `ApplyFileUploadLease`（申请上传租约）→ 上传文件 → `AddFile`（凭 `LeaseId` 导入，并指定 `Parser` 解析器，如 `DOCMIND`、`DOCMIND_LLM_VERSION`、`DASH_QWEN_VL_PARSER`、`AUTO_SELECT` 等）；也可用 `AddFilesFromAuthorizedOss` 直接从同账号已授权的 OSS Bucket 导入。 +- **文件管理**：`DescribeFile`（查解析状态）、`ListFile`、`UpdateFileTag` / `BatchUpdateFileTag`（标签最多 100 个、总长不超过 700 字符）、`DeleteFile` / `DeleteFiles`（单次最多 20 个，仅能删除 `PARSE_FAILED` 或 `PARSE_SUCCESS` 状态的文件）。 +- **解析设置**：`GetParseSettings`、`GetAvailableParserTypes`（按扩展名查询可用解析器）、`ChangeParseSetting`（为特定文件类型指定解析方式）。 +- **连接器与表格**：`AddConnector`（当前仅支持文件类型连接器）、`GetConnector`、`UpdateConnector`、`AddTable`、`UpdateTableFromAuthorizedOss`。 -### 类目管理 +> **注意**：[AddConnector](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-addconnector.md) 的接口说明中写的是需要 `sfm:AddCategory` 权限点，而其授权信息表为 `sfm:AddConnector`，两处不一致，疑为文档笔误，实际授权建议以 `AliyunBailianDataFullAccess` 策略整体覆盖。 -| API | 说明 | 限流 | 幂等性 | -|-----|------|------|--------| -| AddCategory | 在[业务空间](../concepts/workspace.md)中新建类目，每空间最多 500 个 | 5 次/秒 | 否 | -| ListCategory | 查询类目列表，支持分页 | 5 次/秒 | 是 | -| DeleteCategory | 永久删除指定类目 | 5 次/秒 | 是 | +> **注意**：数据表（结构化表格）的新增、上传数据与删除均**不支持通过 API 完成**，需在控制台的应用数据页面操作；`ListCategory` 也暂不支持查询数据表。 -> **注意**：当前不支持通过 API 查询或新增数据表，数据表操作请通过控制台完成。 +### Prompt 工程 -### 文件管理 +`CreatePromptTemplate`、`GetPromptTemplate`、`UpdatePromptTemplate`（PATCH，增量更新）、`DeletePromptTemplate`、`ListPromptTemplates`（支持按名称关键字、`System`/`Custom` 类型过滤及 `nextToken` 分页）。模板内容用 `${变量名}` 声明变量，返回参数中会解析出 `variables` 列表。暂不支持文生图 Prompt 模板。 -文件上传采用两步流程：先调用 ApplyFileUploadLease 获取上传租约，使用返回的 URL 上传文件后，再调用 AddFile 将文件导入百炼。也可通过 AddFilesFromAuthorizedOss 直接从已授权的 OSS Bucket 导入。详见[ApplyFileUploadLease - 申请文件上传租约](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-data-connection-original-application-data/api-bailian-2023-12-29-applyfileuploadlease.md)。 +### 知识库（Index） -| API | 说明 | 限流 | -|-----|------|------| -| ApplyFileUploadLease | 申请上传租约（知识库文件或会话交互文件） | 10 次/秒 | -| AddFile | 将临时存储文件导入数据连接 | 10 次/秒 | -| AddFilesFromAuthorizedOss | 从已授权 OSS Bucket 批量导入文件 | 5 次/秒 | -| DescribeFile | 查询文件基本信息（名称、类型、状态等） | 10 次/秒 | -| ListFile | 分页查询指定类目下的文件列表 | 5 次/秒 | -| UpdateFileTag | 更新单个文件的标签 | 5 次/秒 | -| BatchUpdateFileTag | 批量更新文件标签 | 5 次/秒 | -| DeleteFile | 删除单个文件 | 5 次/秒 | -| DeleteFiles | 批量删除文件 | 5 次/秒 | +- **创建**：`CreateIndex` 仅初始化知识库作业，**必须**随后调用 `SubmitIndexJob` 才能真正完成构建，否则得到空知识库，完整流程见 [CreateIndex - 创建知识库](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-createindex.md)。`CreateIndex` 不具幂等性，重复调用会产生多个同名知识库，建议“先查询、后创建”。 +- **追加与状态**：`SubmitIndexAddDocumentsJob`（追加前需先用 `AddFile` 完成导入；不支持数据查询/图片问答类知识库）、`GetIndexJobStatus`（凭 `JobId` + `IndexId` 轮询任务状态，高峰期任务可能耗时数小时）。 +- **检索**：`Retrieve` 在指定知识库中检索，可直接使用而不依赖百炼应用；响应包含复杂检索匹配，延迟可能较长，需合理设置超时与重试。也可通过应用调用的 `rag_options` 传入 `IndexId` 关联使用。 +- **管理**：`ListIndices`、`UpdateIndex`（可调整名称、`RerankMinScore`、`DenseSimilarityTopK` 与 `SparseSimilarityTopK`——二者之和须 ≤ 200，以及标准版/旗舰版规格）、`DeleteIndex`（若知识库被应用引用需先在控制台解除关联；删除不可逆，但不影响应用数据中的文件）、`ListIndexDocuments` / `ListIndexFileDetails`（按 `DocumentStatus` 如 `FINISH`、`INSERT_ERROR` 过滤）、`DeleteIndexDocument`（仅能删除 `FINISH` 或 `INSERT_ERROR` 状态的文件）。 +- **切片**：`ListChunks`（查询文本切片）、`UpdateChunk`（修改切片 content/title、控制是否参与检索，仅支持文档搜索类知识库）、`DeleteChunk`、`AddChunk`。 -AddFile 接口的 `Parser` 参数支持以下解析器类型： -- `DOCMIND`（智能文档解析） -- `DOCMIND_DIGITAL`（电子文档解析） -- `DOCMIND_LLM_VERSION`（大模型文档解析） -- `DASH_QWEN_VL_PARSER`（Qwen VL 解析） -- `DOCMIND_LLM_VERSION_MEDIA`（音视频解析） -- `AUTO_SELECT`（自动选择解析器） +> **注意**：`ListIndexDocuments` 与 `ListIndexFileDetails` 的接口说明均写需要 `sfm:ListIndexFiles` 权限点，但 `ListIndexFileDetails` 授权表中的操作名为 `sfm:ListIndexFileDetails`，命名不一致；同样 `ListChunks` 对应的权限点是 `sfm:ChunkList`、`ListIndices` 对应 `sfm:ListIndex`，与接口名并非一一同名，编写细粒度 RAM 策略时需逐一核对。 -### 解析设置 +### 其他 -| API | 说明 | -|-----|------| -| GetParseSettings | 获取类目的解析设置 | -| GetAvailableParserTypes | 获取指定文件支持的解析器类型列表 | -| ChangeParseSetting | 修改类目的解析设置 | +临时存储租约（`ApplyTempStorageLease`）、支付宝打赏（`GetAlipayUrl` / `GetAlipayTransferStatus`）以及长期记忆体与记忆片段的完整 CRUD（`CreateMemory` / `GetMemory` / `UpdateMemory` / `DeleteMemory` / `ListMemories`、`CreateMemoryNode` 等）。 -### 表格与连接器 +## 使用要点与限制 -| API | 说明 | -|-----|------| -| AddTable | 添加表格 | -| UpdateTableFromAuthorizedOss | 从已授权 OSS Bucket 更新表格 | -| AddConnector | 新增连接器 | -| GetConnector | 获取连接器信息（当前仅支持文件连接器） | -| UpdateConnector | 编辑连接器名称和描述 | +- **[限流](../concepts/rate-limit.md)**：各接口独立[限流](../concepts/rate-limit.md)，常见为 5 次/秒（类目、连接器、列表类）或 10 次/秒（文件、知识库操作类），`ListIndexDocuments` 为 15 次/秒。触发[限流](../concepts/rate-limit.md)后应退避重试。 +- **幂等性**：查询与删除类接口普遍幂等；创建类接口（`AddCategory`、`AddFile`、`CreateIndex`、`SubmitIndexJob` 等）不幂等，需要调用方自行防重。 +- **分页**：数据连接类接口使用 `MaxResults` + `NextToken` 令牌分页；知识库列表类接口多用 `PageNumber` + `PageSize`。 +- **删除语义**：`DeleteFile` 删除应用数据中的文件不影响已构建知识库；删除知识库中的文件需用 `DeleteIndexDocument`，且不会反向删除应用数据，两者相互独立。 +- **版本变更**：该 API 仍在持续演进（如 `CreateIndex` 入参多次变更、新增 `UpdateIndex` / `GetIndexMonitor`），升级 SDK 前建议核对 [版本说明](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-changeset.md)。 -连接器的 `StorageType` 支持 `OSS_CUSTOM`（自有 OSS 存储）和 `OSS_PLATFORM`（平台 OSS 存储）。 - -## Prompt 工程 - -Prompt 模板 API 支持对 Prompt 模板的完整 CRUD 操作。模板内容支持变量占位符（如 `${theme}`），系统会自动提取变量列表。详见[CreatePromptTemplate - 创建Prompt模板](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-prompt-engineering/api-bailian-2023-12-29-createprompttemplate.md)。 - -| API | 方法 | 说明 | -|-----|------|------| -| CreatePromptTemplate | POST | 创建模板（暂不支持文生图模板） | -| GetPromptTemplate | GET | 按模板 ID 获取详情 | -| UpdatePromptTemplate | PATCH | 增量更新模板名称或内容 | -| DeletePromptTemplate | DELETE | 按模板 ID 删除 | -| ListPromptTemplates | GET | 分页查询模板列表，支持按名称和类型（System/Custom）过滤 | - -## 知识库 - -知识库 API 是百炼 RAG 能力的核心，覆盖知识库的创建、数据导入、检索、文件与切片管理全流程。 - -### 知识库生命周期 - -创建知识库的典型流程为：CreateIndex -> SubmitIndexJob -> 轮询 GetIndexJobStatus 直到完成。详见[CreateIndex - 创建知识库](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-createindex.md)。 - -| API | 说明 | 限流 | -|-----|------|------| -| CreateIndex | 创建知识库（非结构化或结构化），不具幂等性 | 10 次/秒 | -| SubmitIndexJob | 提交知识库创建任务，必须在 CreateIndex 后调用 | 10 次/秒 | -| SubmitIndexAddDocumentsJob | 向已有知识库追加文件（不支持数据查询/图片问答类） | 10 次/秒 | -| GetIndexJobStatus | 查询任务状态，调用间隔建议 5 秒以上 | - | -| UpdateIndex | 更新知识库配置（名称、描述、检索参数等） | - | -| ListIndices | 分页查询[业务空间](../concepts/workspace.md)下的知识库列表 | 10 次/秒 | -| DeleteIndex | 永久删除知识库（不可逆，不删除源文件） | 10 次/秒 | -| GetIndexMonitor | 获取知识库监控数据 | - | - -> **注意**：CreateIndex 仅初始化知识库，必须后续调用 SubmitIndexJob 才能完成创建，否则将得到空知识库。CreateIndex 不具幂等性，重复调用会创建多个同名知识库。 - -UpdateIndex 支持调整检索参数： -- `DenseSimilarityTopK`：向量检索 Top K，范围 [0-100]，默认 100 -- `SparseSimilarityTopK`：关键词检索 Top K，范围 [0-100]，默认 100 -- 两者之和不超过 200 -- `RerankMinScore`：排序最低分数，范围 [0-1] -- `PipelineCommercialType`：知识库规格（standard / enterprise） - -### 知识库检索 - -Retrieve 接口用于在指定知识库中检索信息，支持通过百炼 SDK（AccessKey 鉴权）或 Spring AI Alibaba（API-Key 鉴权）调用。接口具有幂等性，但因包含复杂检索逻辑，响应时间可能较长，建议合理设置超时和重试策略。详见[Retrieve - 检索知识库](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-knowledge-base/api-bailian-2023-12-29-retrieve.md)。 - -### 文件与切片管理 - -| API | 说明 | -|-----|------| -| ListIndexFileDetails | 查询知识库中文件的详细信息，支持按状态和名称过滤 | -| ListIndexDocuments | 查询知识库中文件的概要信息 | -| DeleteIndexDocument | 从知识库中删除指定文件 | -| ListChunks | 查询文件的切片列表（文档搜索类查指定文件，数据查询类查全部） | -| UpdateChunk | 修改切片内容和标题（仅支持文档搜索类知识库） | -| DeleteChunk | 删除指定切片 | - -文件导入状态包括：`RUNNING`（构建中）、`FINISH`（成功）、`INSERT_ERROR`（导入失败）、`PARSE_FAILED`（解析失败）、`DOC_PARSING`（解析中）、`DELETED`（已删除）。 - -## 长期记忆 - -长期记忆 API 用于管理智能体的记忆能力，包括记忆体（Memory）和记忆片段（MemoryNode）两个层级。 - -| API | 说明 | -|-----|------| -| CreateMemory | 创建长期记忆体 | -| GetMemory | 获取记忆体详情 | -| UpdateMemory | 更新记忆体 | -| DeleteMemory | 删除记忆体 | -| ListMemories | 查询记忆体列表 | -| CreateMemoryNode | 创建记忆片段 | -| GetMemoryNode | 获取记忆片段详情 | -| UpdateMemoryNode | 更新记忆片段 | -| DeleteMemoryNode | 删除记忆片段 | -| ListMemoryNodes | 查询记忆片段列表 | - -## 其他 - -| API | 说明 | -|-----|------| -| ApplyTempStorageLease | 申请临时文件上传许可 | -| GetAlipayTransferStatus | 查询支付宝打赏状态 | -| GetAlipayUrl | 获取支付宝打赏 URL | - -## 通用注意事项 - -- 所有接口均需 `WorkspaceId` 路径参数，获取方式参见[业务空间](../concepts/workspace.md)文档 -- 建议使用官方 SDK 调用而非自签名，自签名对接复杂度高（约需 5 个工作日） -- 分页查询使用 `NextToken` / `MaxResults` 模式（部分接口使用 `PageNumber` / `PageSize`） -- 各接口限流频率为 5-15 次/秒不等，触发限流后需等待后重试 -- 版本变更历史可查看[版本说明](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-changeset.md)，近期变更包括 CreateIndex 入参调整、UpdateIndex 新增、GetIndexMonitor 新增等 +> **注意**：`GetIndexJobStatus` 文档自身建议“调用间隔 5 秒以上”，而 `SubmitIndexAddDocumentsJob` 文档中提示其限流为“不高于 20 次/分钟”（约 3 秒一次），两处口径略有出入，保守起见按每 5 秒以上轮询一次处理。 ## 来源文档 @@ -220,13 +112,6 @@ Retrieve 接口用于在指定知识库中检索信息，支持通过百炼 SDK - [UpdateMemoryNode - 更新记忆片段](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-updatememorynode.md) - [DeleteMemoryNode - 删除记忆片段](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-deletememorynode.md) - [ListMemoryNodes - 获取记忆片段列表](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-dir-long-term-memory/api-bailian-2023-12-29-listmemorynodes.md) - - - - - - - - +- [AddChunk - 新增切片](../../raw/application-api-reference/application-component-api-reference/api-bailian-2023-12-29-dir/api-bailian-2023-12-29-dir-others/api-bailian-2023-12-29-addchunk.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/api/file-management-api.md b/skills/bailian-docs-llm-wiki/wiki/api/file-management-api.md index 912b6483..ee87e244 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/file-management-api.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/file-management-api.md @@ -1,6 +1,6 @@ # file management api -文件管理 API 用于管理上传至百炼平台的文件，覆盖上传、查询、列举和删除等基础操作。它是使用需要文件输入的模型能力（如文档解析、多模态理解、批量任务等）的前置步骤，开发者需先将文件上传到平台并获取文件标识，再在后续调用中引用。详见 [文件管理](../../raw/model-api-reference/file-management-api.md)。 +文件管理 API 用于管理上传至百炼平台的文件，覆盖上传、查询、列举和删除等基础操作。它是使用需要文件输入的模型能力（如文档解析、[多模态](../concepts/multimodal.md)理解、批量任务等）的前置步骤，开发者需先将文件上传到平台并获取文件标识，再在后续调用中引用。详见 [文件管理](../../raw/model-api-reference/file-management-api.md)。 ## 核心功能 @@ -36,3 +36,16 @@ + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/frameworks.md b/skills/bailian-docs-llm-wiki/wiki/api/frameworks.md index c4de7ce6..27fe13ce 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/frameworks.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/frameworks.md @@ -7,9 +7,9 @@ | 框架 | 语言 | 主要能力 | | --- | --- | --- | | LlamaIndex | Python 3.9+ | 读取本地文件上传到百炼应用数据、构建云端[知识库](../concepts/knowledge-base.md)、构建检索引擎与 RAG 应用 | -| Spring AI Alibaba | Java（Spring Boot 3.x，JDK 17+） | 调用百炼[智能体应用](../concepts/agent-application.md)/[工作流](../concepts/workflow.md)应用（流式与非流式）、检索百炼知识库 | +| Spring AI Alibaba | Java（Spring Boot 3.x，JDK 17+） | 调用百炼[智能体应用](../concepts/agent-application.md)/[工作流](../concepts/workflow.md)应用（流式与非流式）、检索百炼[知识库](../concepts/[knowledge](knowledge.md)-base.md) | -- LlamaIndex 路线将知识库部署在云端，使用默认的智能文档切分与官方向量模型，**不支持**自定义文档切分方式或自定义嵌入模型。如需本地知识库或灵活切分，应改用本地知识库方案，详见[通过LlamaIndex API构建RAG应用](../../raw/application-api-reference/frameworks/llamaindex.md)。 +- LlamaIndex 路线将[知识库](../concepts/[knowledge](knowledge.md)-base.md)部署在云端，使用默认的智能文档切分与官方向量模型，**不支持**自定义文档切分方式或自定义嵌入模型。如需本地[知识库](../concepts/[knowledge](knowledge.md)-base.md)或灵活切分，应改用本地知识库方案，详见[通过LlamaIndex API构建RAG应用](../../raw/application-api-reference/frameworks/llamaindex.md)。 - Spring AI Alibaba 的应用集成**仅支持**[智能体应用](../concepts/agent-application.md)与[工作流](../concepts/workflow.md)应用两类，需提前在百炼控制台创建并获取应用 ID。 ## 前提条件 @@ -121,7 +121,7 @@ this.chatClient = builder ## 限制与注意事项 - LlamaIndex 云端方案不支持自定义文档切分与嵌入模型；本地需可访问公网，文件上传与生成回答均需等待。 -- Spring AI Alibaba 应用集成仅支持[智能体应用](../concepts/agent-application.md)与工作流应用，其他应用类型不在支持范围。 +- Spring AI Alibaba 应用集成仅支持[智能体应用](../concepts/agent-application.md)与[工作流](../concepts/workflow.md)应用，其他应用类型不在支持范围。 - 知识库检索需提前创建好知识库并获取其名称；检索默认[业务空间](../concepts/workspace.md)知识库无需配置 `workspace-id`。 - 子[业务空间](../concepts/workspace.md)场景必须配置对应的[业务空间](../concepts/workspace.md) ID 环境变量，否则会鉴权或定位失败。 - [API Key](../concepts/api-key.md) 一律通过环境变量注入，切勿硬编码到源码或配置文件中。 @@ -145,6 +145,19 @@ this.chatClient = builder + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/image-generation.md b/skills/bailian-docs-llm-wiki/wiki/api/image-generation.md index 18e498c4..d5db1c47 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/image-generation.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/image-generation.md @@ -1,55 +1,112 @@ # image generation -阿里云百炼平台提供了一整套图像生成与编辑 API，覆盖文生图、图像编辑、图像翻译以及大量垂直创意工具（虚拟模特、鞋靴模特、扩图、擦除补全、海报生成等）。这些接口以 DashScope 网关为基础，模型来自千问（Qwen-Image）、通义万相（Wan/WanX）、Z-Image、可灵（Kling）、Vidu 等多个系列。本文面向开发者，梳理各类模型能力、调用方式、关键参数及常见限制。 +百炼平台提供丰富的图像生成与编辑 API，涵盖文生图、图像编辑、图像翻译、涂鸦作画、局部重绘、虚拟模特、试衣、海报生成、背景生成、擦除补全、画面扩展、人物分割、写真生成、创意文字等多种能力。底层模型包括千问图像（Qwen-Image）、万相（Wan）、Z-Image、可灵（Kling）、Vidu 等多个系列，支持同步与异步两种调用模式。 -## 支持的模型与功能 +## 模型概览 -按能力可将图像模型大致分为四类： +### 文生图模型 -- **通用文生图**：千问文生图（qwen-image 系列，擅长复杂文本渲染）、万相文生图 V2（wan2.6-t2i / wan2.5-t2i-preview / wan2.2-t2i-* / wanx2.1-t2i-*）、万相文生图 V1（wanx-v1，仅存量）、轻量快速的 z-image-turbo，以及可灵、Vidu 系列。详见 [千问-文生图API参考](../../raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-api.md) 与 [万相-文生图V2版API参考](../../raw/model-api-reference/image-generation/wan-image-api-reference/text-to-image-v2-api-reference.md)。 -- **图像编辑 / 多图融合**：千问图像编辑（qwen-image-edit 系列，支持多图输入输出、改文字/增删物体/风格迁移）、万相通用图像编辑 2.5/2.6/2.7、万相通用图像编辑（wanx2.1-imageedit，支持风格化、指令编辑、局部重绘、去水印、扩图、超分、上色、线稿生图）、图像局部重绘（wanx-x-painting）。参见 [千问-图像编辑API参考](../../raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-edit-api.md) 与 [万相-通用图像编辑API参考](../../raw/model-api-reference/image-generation/wan-image-api-reference/wanx-image-edit-api-reference.md)。 -- **图像翻译**：千问图像翻译（qwen-mt-image），精准翻译图中文字并保留排版。见 [千问-图像翻译API参考](../../raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-mt-image-api.md)。 -- **垂直创意工具**：人像风格重绘（wanx-style-repaint-v1）、虚拟模特（wanx-virtualmodel / virtualmodel-v2）、鞋靴模特（shoemodel-v1）、图像画面扩展（image-out-painting）、创意海报生成（wanx-poster-generation-v1）、人物实例分割（image-instance-segmentation）、AI 试衣 OutfitAnyone（aitryon 系列）、图像背景生成（wanx-background-generation-v2）、图像擦除补全（image-erase-completion）、人物写真 FaceChain、创意文字 WordArt 锦书。 +| 模型系列 | 推荐模型名称 | 输出图像规格 | 调用方式 | +| --- | --- | --- | --- | +| 千问图像 | qwen-image-2.0-pro、qwen-image-2.0、qwen-image-max、qwen-image-plus | 分辨率 512×512~2048×2048，PNG，1-6张 | 同步 | +| 万相文生图 | wan2.7-image-pro、wan2.7-image、wan2.6-t2i、wan2.5-t2i-preview、wan2.2-t2i-flash/plus、wanx2.1-t2i-turbo/plus | PNG，尺寸依模型不同 | 同步（wan2.6+）/异步（2.5及以下） | +| Z-Image | z-image-turbo | 512×512~2048×2048，PNG，1张 | 同步 | +| 可灵图像 | kling/kling-v3-image-generation、kling/kling-v3-omni-image-generation | 1K/2K/4K，1-9张，PNG | 异步 | +| Vidu图像 | vidu/vidu-image_reference2image 等 | 1K/2K/4K，1张，PNG | 异步 | + +千问图像模型擅长复杂文字渲染和真实质感，万相文生图支持多种艺术风格与写实摄影，Z-Image 为轻量快速生图模型，可灵支持文生图和参考图生图（含组图分镜模式），Vidu 擅长中英文字精准渲染和 UI/图表像素级还原。详见[千问-文生图API参考](../../raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-api.md)和[万相-文生图V2版API参考](../../raw/model-api-reference/image-generation/wan-image-api-reference/text-to-image-v2-api-reference.md)。 + +### 图像编辑模型 + +| 模型系列 | 推荐模型名称 | 核心能力 | +| --- | --- | --- | +| 千问图像编辑 | qwen-image-2.0-pro、qwen-image-2.0、qwen-image-edit-max、qwen-image-edit-plus | 多图输入/输出、精确修改文字、增删移动物体、风格迁移、细节增强 | +| 万相图像编辑 | wan2.7-image-pro、wan2.6-image、wan2.5-i2i-preview、wanx2.1-imageedit | 单图编辑、多图融合、风格化、去水印、扩图、超分、上色、线稿生图 | +| 万相局部重绘 | wanx-x-painting | 基于涂抹区域和提示词的局部重绘 | +| 万相涂鸦作画 | wanx-sketch-to-image-lite | 手绘图案+文字描述生成涂鸦绘画 | + +千问图像编辑模型支持多图融合和精确文字修改，万相通用编辑覆盖风格化、局部重绘、去水印、扩图、超分等场景。详见[千问-图像编辑API参考](../../raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-edit-api.md)。 + +### 图像翻译 + +千问-图像翻译模型（qwen-mt-image）可精准翻译图像中的文字并保留原始排版，支持中/英文与日/韩/西/法等语种互译（不支持非中英语种间直接翻译），还支持领域提示、敏感词过滤、术语干预。详见[千问-图像翻译API参考](../../raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-mt-image-api.md)。 + +### 创意工具类模型 + +| 模型名称 | 功能 | 状态 | +| --- | --- | --- | +| wanx-style-repaint-v1 | 人像风格重绘 | 付费 | +| wanx-virtualmodel / virtualmodel-v2 | 虚拟模特生成 | 免费体验 | +| shoemodel-v1 | 鞋靴模特试穿 | 免费体验 | +| image-out-painting | 图像画面扩展 | 付费 | +| wanx-poster-generation-v1 | 创意海报生成 | 免费体验 | +| image-instance-segmentation | 人物实例分割 | 免费体验 | +| aitryon / aitryon-plus / aitryon-refiner / aitryon-parsing-v1 | AI试衣（基础/Plus/精修/分割） | 付费 | +| wanx-background-generation-v2 | 图像背景生成 | 付费 | +| image-erase-completion | 图像擦除补全 | 免费体验 | +| FaceChain | 人物写真生成 | — | +| WordArt锦书 | 创意文字变形/纹理生成 | — | + +> **注意**：wanx-x-painting、wanx-virtualmodel、virtualmodel-v2、image-erase-completion、image-instance-segmentation、shoemodel-v1、wanx-poster-generation-v1 等模型当前仅提供免费体验，额度用完后不可调用且不支持付费。官方推荐参考千问图像编辑或万相2.1图像编辑获取替代方案。 ## 调用方式 -图像 API 主要有两种调用协议，选择取决于模型版本： +### 同步调用（推荐） -- **异步调用（传统主流）**：由于生成耗时较长（通常 1-2 分钟），多数模型仅支持异步。流程为「创建任务 → 轮询获取结果」两步：先 POST 创建任务拿到 `task_id`，再用 `task_id` 查询状态直至 `SUCCEEDED` 并取回图像 URL。任务创建请求必须携带请求头 `X-DashScope-Async: enable`，否则会报错 `current user api does not support synchronous calls`。返回的图像 URL 有效期为 24 小时，`task_id` 有效期也为 24 小时，请勿重复创建任务。 -- **HTTP 同步调用（新版协议）**：仅新版模型支持，一次请求即可拿到结果，流程更简单，推荐大多数场景使用。目前支持同步的有 **wan2.6 / wan2.7 图像模型**、**z-image-turbo** 等，走 `multimodal-generation/generation` 端点。 +适用于千问图像、万相2.6+、Z-Image、千问3.0等新版模型。一次请求即可获得结果： -> **注意**：同步调用仅限新版模型。以万相文生图为例，wan2.6 支持 HTTP 同步/异步与 SDK 调用，而 **wan2.5 及以下版本不支持 HTTP 同步调用**，只能异步 + SDK。请勿把同步协议用在旧模型上。 +``` +POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation +``` -任务状态取值：`PENDING`（排队）、`RUNNING`（处理中）、`SUSPENDED`（挂起）、`SUCCEEDED`（成功）、`FAILED`（失败）。 +请求体采用 `model` + `input.messages`（含 `text`/`image` 内容块）+ `parameters` 结构，类似[多模态](../concepts/multimodal.md)对话格式。 -不同模型使用的服务端点也不同，常见的有： +### [异步调用](../concepts/async-invocation.md) -- `.../aigc/text2image/image-synthesis`（万相 V1、创意海报等文生图） -- `.../aigc/image2image/image-synthesis`（图像编辑、涂鸦、局部重绘、图像翻译、擦除补全等） -- `.../aigc/multimodal-generation/generation`（wan2.6/2.7、z-image 等新版） -- `.../aigc/image-generation/generation`（可灵、Vidu、人像风格重绘） -- `.../aigc/virtualmodel/generation`（虚拟模特、鞋靴模特） -- `.../aigc/image2image/out-painting`（图像画面扩展） -- `.../aigc/background-generation/generation`（图像背景生成） +适用于万相2.5及以下版本、可灵、Vidu、以及大部分创意工具类模型。分两步： -## 关键参数 +1. **创建任务**：发送请求，获取 `task_id`（有效期24小时）。 +2. **轮询结果**：使用 `task_id` 查询任务状态，成功后获取图像URL（有效期24小时）。 -- **鉴权与请求头**：`Authorization: Bearer $DASHSCOPE_API_KEY`（必选）、`Content-Type: application/json`（必选）、异步接口需 `X-DashScope-Async: enable`。子账号调用可通过 `X-DashScope-WorkSpace` 指定业务空间 ID。 -- **input**：文生图通常传 `prompt`（可选 `negative_prompt` 反向提示词）；图像编辑/参考图任务传 `image_url` / `images` / `base_image_url` / `mask_image_url` 等；新版多模态模型使用 `messages`（含 `text` 与 `image` 的 content 数组）结构。 -- **parameters**：`size`（分辨率，格式 `宽*高` 或档位如 `1K`/`2K`/`4K`）、`n`（生成张数）、`style`、`watermark`、`prompt_extend`（智能改写/思考，如 z-image、wan2.6）、`thinking_mode`、`aspect_ratio`/`resolution`（可灵）等，随模型而异。 +创建任务时必须携带 `X-DashScope-Async: enable` 请求头，否则会报错。接口路径因功能不同而异： +- 文生图：`/api/v1/services/aigc/text2image/image-synthesis` +- 图生图/编辑：`/api/v1/services/aigc/image2image/image-synthesis` +- 可灵/Vidu图像生成：`/api/v1/services/aigc/image-generation/generation` +- 虚拟模特：`/api/v1/services/aigc/virtualmodel/generation` +- 背景生成：`/api/v1/services/aigc/background-generation/generation` +- 画面扩展：`/api/v1/services/aigc/image2image/out-painting` -输出图像规格差异较大：例如千问 Pro/Plus 系列总像素需在 512\*512 至 2048\*2048 之间、可 1-6 张；万相 2.6 总像素在 [1280\*1280, 1440\*1440]、宽高比 [1:4, 4:1]；可灵支持 1k/2k/4k 及组图；z-image 固定 1 张。具体以各模型文档为准。 +## 关键参数 -## 限制与注意事项 +| 参数 | 适用场景 | 说明 | +| --- | --- | --- | +| `prompt` | 文生图/编辑 | 正向提示词，描述期望生成的图像内容 | +| `negative_prompt` | 文生图 | 反向提示词，描述不希望出现的元素（V1版本支持） | +| `size` | 通用 | 输出图像分辨率，如 `1024*1024`、`1K`、`2K`、`4K` | +| `n` | 通用 | 输出图像张数，范围1-9（因模型而异） | +| `style` | 文生图 | 风格预设，如 ``、`` | +| `watermark` | 通用 | 是否添加水印，默认 true | +| `prompt_extend` | 万相/Z-Image/千问3.0 | 开启智能提示词扩展，优化生成效果 | +| `thinking_mode` | 万相2.7 | 开启智能思考模式 | +| `enable_interleave` | 万相2.6 | 开启图文混排输出（仅流式） | +| `aspect_ratio` / `resolution` | 可灵 | 宽高比（16:9/9:16/1:1）和分辨率（1k/2k/4k） | + +## [计费](../concepts/billing.md)与[限流](../concepts/rate-limit.md) + +- **免费额度**：开通百炼服务后自动发放，有效期90天，主账号与RAM子账号共享。仅成功生成的输出图片占用免费额度。 +- **限时免费**：公测阶段模型，免费额度用尽后不可使用。 +- **商业化模型**：如 wanx-v1（0.16元/张）、wanx2.1-imageedit（0.14元/张）、image-out-painting（0.18元/张）等，按成功输出图片[计费](../concepts/billing.md)。 +- **[限流](../concepts/rate-limit.md)**：任务下发接口QPS限制通常为2，同时处理中任务数量1-5（因模型而异），主账号与RAM子账号共享。详见[常见问题](../../raw/model-api-reference/image-generation/image-faq.md)。 -- **地域隔离**：华北2（北京）、新加坡、美国（弗吉尼亚）等地域拥有**独立的 API Key 与请求地址，不可混用**，跨地域调用会导致鉴权失败或报错。相当一部分创意工具（如虚拟模特、鞋靴模特、人像风格重绘、图像翻译、可灵、Vidu 等）**仅在华北2（北京）地域可用**。 -- **专属域名迁移**：百炼为北京/新加坡地域推出业务空间专属域名（`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` / `...ap-southeast-1.maas.aliyuncs.com`），性能与稳定性更佳，建议从 `https://dashscope.aliyuncs.com` 迁移。旧域名仍可用。 -- **图片 URL 必须公网可访问**：使用自有图片时若报 `BadRequest.InputDownloadFailed`（下载图片失败），需确认 URL 完整、支持公网访问，可上传至 OSS 等云存储；URL 中不能包含中文字符。相关排查见 [常见问题](../../raw/model-api-reference/image-generation/image-faq.md)。 -- **计费与限流**：只对成功生成的输出图片计费，输入图片和失败任务不计费；免费额度（通常 500 张）有效期 90 天，主账号与 RAM 子账号共享额度与限流。部分模型标注「限时免费」（公测阶段，额度用尽即不可用）。 +## 限制与注意事项 -> **注意**：多个模型（如 wanx-x-painting 局部重绘、wanx-virtualmodel/virtualmodel-v2 虚拟模特、wanx-poster-generation-v1 海报生成、image-erase-completion 擦除补全等）当前**仅供免费体验，额度用完后不可调用且不支持付费**，官方推荐迁移到千问图像编辑或万相 2.1 等替代方案。新项目集成前请确认目标模型的商业化状态。 +- **地域隔离**：华北2（北京）、新加坡、美国（弗吉尼亚）地域拥有独立的 [API Key](../concepts/api-key.md) 与请求地址，不可混用，跨地域调用将导致鉴权失败。建议使用[业务空间](../concepts/workspace.md)专属域名 `{WorkspaceId}.cn-beijing.maas.aliyuncs.com`。 +- **图片URL要求**：输入图片URL需支持公网访问。若报错"下载图片失败"，请确保URL完整且可公网访问，可将图片上传至OSS等云存储。 +- **图片格式与大小**：支持 JPEG、PNG、BMP、WEBP 等格式，图片大小建议小于5-10MB，分辨率不低于512×512。 +- **task_id与结果有效期**：task_id 有效期24小时，生成结果图像URL有效期24小时，请及时下载。 +- **SDK支持**：千问、万相等模型支持 [DashScope SDK](../concepts/dashscope-sdk.md)（Python/Java），部分创意工具类模型（如人像风格重绘）仅提供HTTP API。 -> **注意**：万相文生图 V1（wanx-v1）已被 V2 版全面替代，官方推荐使用 V2；旧版仅适用于北京地域。选择模型时优先考虑最新版本。 +> **注意**：万相文生图V1版（wanx-v1）为早期模型，官方推荐使用升级后的V2版模型。万相2.5及以下版本不支持HTTP同步调用，仅支持[异步调用](../concepts/async-invocation.md)。 ## 来源文档 @@ -79,5 +136,12 @@ - [人物写真生成FaceChain](../../raw/model-api-reference/image-generation/image-creative-tools-api-reference/facechain-portrait-generation.md) - [创意文字WordArt锦书](../../raw/model-api-reference/image-generation/image-creative-tools-api-reference/wordart-quick-start.md) - [Vidu-图像生成API参考](../../raw/model-api-reference/image-generation/vidu-image-models/vidu-image-generation-api-reference.md) +- [千问-图像生成与编辑3.0 API参考](../../raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-generation-and-editing-api-reference.md) + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/knowledge.md b/skills/bailian-docs-llm-wiki/wiki/api/knowledge.md index 6af4d028..712467b6 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/knowledge.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/knowledge.md @@ -1,12 +1,12 @@ # knowledge -百炼平台「知识检索与问答」相关的 HTTP REST API 概览，提供跨知识库语义检索与基于知识库的智能问答两个接口。这两个接口属于 DashScope 应用网关体系，通过 [API Key](../concepts/api-key.md) Bearer 鉴权调用，与 `CreateIndex`、`Retrieve` 等 OpenAPI RPC 接口不同。详见 [知识检索与问答](../../raw/application-api-reference/knowledge.md)。 +百炼平台「知识检索与问答」相关的 HTTP REST API 概览，提供跨[知识库](../concepts/knowledge-base.md)语义检索与基于[知识库](../concepts/knowledge-base.md)的智能问答两个接口。这两个接口属于 DashScope 应用网关体系，通过 [API Key](../concepts/api-key.md) Bearer 鉴权调用，与 `CreateIndex`、`Retrieve` 等 OpenAPI RPC 接口不同。详见 [知识检索与问答](../../raw/application-api-reference/knowledge.md)。 ## 接口列表 | 接口 | 描述 | 路径 | | --- | --- | --- | -| 知识检索 | 跨多个知识库执行联合语义检索，返回按相关性排序的切片 | `POST /api/v1/indices/knowledge/search` | +| 知识检索 | 跨多个[知识库](../concepts/knowledge-base.md)执行联合语义检索，返回按相关性排序的切片 | `POST /api/v1/indices/knowledge/search` | | 知识问答 | 基于知识库的智能问答，通过 SSE [流式输出](../concepts/streaming-output.md)，依次返回规划、工具调用、生成三个阶段 | `POST /api/v2/apps/knowledge/chat` | ## 鉴权与 Base URL @@ -19,9 +19,9 @@ https://{workspaceId}.cn-beijing.maas.aliyuncs.com 其中 `{workspaceId}` 为[业务空间](../concepts/workspace.md) ID。[API Key](../concepts/api-key.md) 在控制台 [API Key 页面](https://rag.console.aliyun.com/settings/apikey) 获取，[业务空间](../concepts/workspace.md) ID 在控制台 [业务空间管理](https://bailian.console.aliyun.com/cn-beijing?tab=globalset#/efm/business_management) 获取。 -## 限流 +## [限流](../concepts/rate-limit.md) -默认用户维度 25 QPS。如遇限流，请稍后重试。更多信息参见 [知识检索与问答](../../raw/application-api-reference/knowledge.md)。 +默认用户维度 25 QPS。如遇[限流](../concepts/rate-limit.md)，请稍后重试。更多信息参见 [知识检索与问答](../../raw/application-api-reference/knowledge.md)。 ## 与 OpenAPI 的区别 @@ -35,7 +35,7 @@ https://{workspaceId}.cn-beijing.maas.aliyuncs.com - 知识检索接口适合需要自定义生成流程的场景：拿到排序后的切片后，自行拼接 [prompt](../guides/prompt.md) 调用大模型。 - 知识问答接口适合开箱即用的问答场景：服务端自动完成规划、检索、生成，通过 SSE 流式返回三个阶段的结果。 -- 调用前确认 API Key 与[业务空间](../concepts/workspace.md) ID 已正确配置，详见 [知识检索与问答](../../raw/application-api-reference/knowledge.md)。 +- 调用前确认 [API Key](../concepts/api-key.md) 与[业务空间](../concepts/workspace.md) ID 已正确配置，详见 [知识检索与问答](../../raw/application-api-reference/knowledge.md)。 ## 来源文档 @@ -49,6 +49,19 @@ https://{workspaceId}.cn-beijing.maas.aliyuncs.com + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/long-term-memory-new.md b/skills/bailian-docs-llm-wiki/wiki/api/long-term-memory-new.md index 75c168a8..cb8085c1 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/long-term-memory-new.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/long-term-memory-new.md @@ -1,6 +1,6 @@ # long term memory new -百炼平台的「长期记忆（新）」提供一组 RESTful API，用于存储、检索、更新和删除用户记忆片段，并支持通过画像模板（profile schema）维护用户画像。记忆片段会从对话中自动提取关键信息，可在后续对话中通过语义检索召回，从而实现跨会话的个性化上下文。完整接口参考见 [长期记忆（新）API 参考](../../raw/application-api-reference/long-term-memory-new/long-term-memory-api-reference.md)。 +百炼平台的「[长期记忆](../concepts/long-term-memory.md)（新）」提供一组 RESTful API，用于存储、检索、更新和删除用户记忆片段，并支持通过画像模板（profile schema）维护用户画像。记忆片段会从对话中自动提取关键信息，可在后续对话中通过语义检索召回，从而实现跨会话的个性化上下文。完整接口参考见 [长期记忆（新）API 参考](../../raw/application-api-reference/long-term-memory-new/long-term-memory-api-reference.md)。 ## 公共请求信息 @@ -12,7 +12,7 @@ ## 接口概览 -长期记忆（新）提供以下 API 接口： +[长期记忆](../concepts/long-term-memory.md)（新）提供以下 API 接口： | 接口名称 | HTTP 方法 | 路径 | 说明 | | --- | --- | --- | --- | @@ -30,7 +30,7 @@ ## 使用限制 -| API 接口 | 限流（阿里云账号级别） | +| API 接口 | [限流](../concepts/rate-limit.md)（阿里云账号级别） | | --- | --- | | 全部接口 | 总计不超过 3000 QPM | | 记忆片段 add 接口 | 120 QPM | @@ -199,7 +199,7 @@ asyncio.run(add_memory_example()) ## 限制和注意事项 -- **限流**：全部接口合计 3000 QPM；`add` 单独 120 QPM，`search` 单独 300 QPM。 +- **[限流](../concepts/rate-limit.md)**：全部接口合计 3000 QPM；`add` 单独 120 QPM，`search` 单独 300 QPM。 - **消息上限**：AddMemory 的 `messages` 最多 50 条对话记录（一问一答算 2 条）。 - **内容长度**：`custom_content` 与 UpdateMemory 的 `custom_content` 均限制 512 个字符。 - **互斥参数**：AddMemory 中 `messages` 与 `custom_content` 互斥，传 `custom_content` 会忽略 `messages`。 @@ -223,6 +223,19 @@ asyncio.run(add_memory_example()) + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/managed-agents-api.md b/skills/bailian-docs-llm-wiki/wiki/api/managed-agents-api.md index 021b21a6..0b776e88 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/managed-agents-api.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/managed-agents-api.md @@ -1,6 +1,6 @@ # [managed agents](../guides/managed-agents.md) api -Managed Agents API 是百炼平台提供的智能体托管运行时，由平台负责会话管理、沙箱执行、工具调用与事件流推送。开发者通过 REST API 或 SDK 完成 Agent 定义、Environment 配置、Session 创建与事件交互，五分钟即可跑通端到端流程。详细的认证方式与 SDK 版本要求见 [API 总览与认证](../../raw/application-api-reference/managed-agents-api/managed-agents-api-overview.md)。 +Managed Agents API 是百炼平台提供的智能体托管运行时，由平台负责会话管理、沙箱执行、工具调用与[事件流](../concepts/event-stream.md)推送。开发者通过 REST API 或 SDK 完成 Agent 定义、Environment 配置、Session 创建与事件交互，五分钟即可跑通端到端流程。详细的认证方式与 SDK 版本要求见 [API 总览与认证](../../raw/application-api-reference/managed-agents-api/managed-agents-api-overview.md)。 ## 核心概念与资源模型 @@ -138,6 +138,19 @@ Managed Agents 模块通过 [DashScope SDK](../concepts/dashscope-sdk.md) 接入 + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/model-production.md b/skills/bailian-docs-llm-wiki/wiki/api/model-production.md index f2c1d6e0..ecff7caf 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/model-production.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/model-production.md @@ -1,33 +1,33 @@ # model production -百炼平台提供模型生产相关的 API，覆盖从模型微调训练到部署上线的完整流程。开发者可以通过[模型调优](../../raw/model-api-reference/model-production/fine-tuning-jobs-api.md)接口定制专属模型，再通过[模型部署](../../raw/model-api-reference/model-production/deployments-api.md)接口将其发布为在线推理服务。 +模型生产（model production）涵盖百炼平台上从定制模型到上线服务的完整链路：先通过微调训练得到专属模型，再将其部署为在线推理服务。本页汇总 [模型调优](../../raw/model-api-reference/model-production/fine-tuning-jobs-api.md) 与 [模型部署](../../raw/model-api-reference/model-production/deployments-api.md) 两个 API 的核心信息，帮助开发者快速打通"调优 → 部署 → 调用"的流程。 -## 模型调优 +## 能力概览 -模型调优（Fine-tuning）允许开发者通过微调训练定制专属模型，以适配特定业务场景。调优流程通常包括： +| 环节 | 能力 | 对应文档 | +| --- | --- | --- | +| 模型调优 | 通过微调（fine-tuning）训练定制专属模型 | [模型调优](../../raw/model-api-reference/model-production/fine-tuning-jobs-api.md) | +| 模型部署 | 将微调产出的模型或外部导入的模型部署为在线推理服务 | [模型部署](../../raw/model-api-reference/model-production/deployments-api.md) | -- **创建调优任务**：指定基础模型、训练数据集和超参数，提交微调训练任务 -- **查询任务状态**：轮询或监听训练任务进度，获取训练指标 -- **管理调优产物**：训练完成后获取调优模型，用于后续部署或评估 +## 典型使用流程 -详细的接口定义和参数说明请参考[模型调优](../../raw/model-api-reference/model-production/fine-tuning-jobs-api.md)文档。 +1. **准备并提交微调任务**：按照 [模型调优](../../raw/model-api-reference/model-production/fine-tuning-jobs-api.md) 中的接口创建微调训练任务，产出定制模型。 +2. **部署为在线服务**：微调完成后（或已有导入的模型），按 [模型部署](../../raw/model-api-reference/model-production/deployments-api.md) 中的接口将模型部署为在线推理服务。 +3. **调用推理服务**：部署成功后，即可通过在线推理接口以部署得到的模型/服务标识发起调用。 -## 模型部署 +## 支持的模型来源 -模型部署将微调或导入的模型发布为在线推理服务，使其可通过 API 调用进行推理。部署流程通常包括： +部署环节支持两类模型来源： -- **创建部署**：选择调优完成的模型或外部导入的模型，配置推理资源和服务参数 -- **管理部署实例**：查看部署状态、调整资源配置、启停服务 -- **调用推理服务**：部署成功后，通过标准 API 端点发送推理请求 +- **微调产出的模型**：由模型调优（fine-tuning）任务训练得到的专属模型。 +- **导入的模型**：从外部导入到平台的模型。 -详细的接口定义和参数说明请参考[模型部署](../../raw/model-api-reference/model-production/deployments-api.md)文档。 +## 限制和注意事项 -## 典型工作流 +- 微调与部署是两个独立步骤：微调完成不代表模型可直接调用，必须先完成部署才能提供在线推理服务。 +- 部署会占用推理资源，实际的计费方式、算力规格与配额请以对应 API 文档和控制台展示为准。 -1. 准备训练数据集 -2. 通过调优 API 提交微调训务，等待训练完成 -3. 通过部署 API 将调优产物部署为在线服务 -4. 调用部署后的模型端点进行推理 +> **注意**：当前两篇原始文档内容较为简略，仅给出能力定位，未包含具体的请求参数、返回结构与错误码。集成前请以最新版 API 参考为准，避免依赖过时字段。 ## 来源文档 @@ -35,12 +35,3 @@ - [模型部署](../../raw/model-api-reference/model-production/deployments-api.md) - - - - - - - - - diff --git a/skills/bailian-docs-llm-wiki/wiki/api/more-about-models.md b/skills/bailian-docs-llm-wiki/wiki/api/more-about-models.md index b2130305..0287b479 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/more-about-models.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/more-about-models.md @@ -14,9 +14,9 @@ POST https://dashscope.aliyuncs.com/api/v1/tokens?expire_in_seconds= - `expire_in_seconds`：有效期，范围 [1, 1800] 秒，默认 60 秒。 - 返回的 `token` 字段即为临时 [API Key](../concepts/api-key.md)，`expires_at` 为 UNIX 过期时间戳。 -- 临时 API Key 继承生成它的永久 API Key 的全部权限，到期后自动失效，无法手动删除。 +- 临时 [API Key](../concepts/api-key.md) 继承生成它的永久 [API Key](../concepts/api-key.md) 的全部权限，到期后自动失效，无法手动删除。 -> **注意**：各地域的 API Key 不同，新加坡地域需将 Endpoint 中的 WorkspaceId 替换为实际值。 +> **注意**：各地域的 [API Key](../concepts/api-key.md) 不同，新加坡地域需将 Endpoint 中的 WorkspaceId 替换为实际值。 ## 异步任务管理 @@ -50,7 +50,7 @@ POST https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}/cancel ## 异步任务完成通知 -频繁轮询任务结果接口会浪费资源且可能触发限流。百炼已接入阿里云事件总线 EventBridge，支持在任务完成后主动推送通知。详见[通过HTTP回调URL或MQ接收异步任务完成通知](../../raw/model-api-reference/more-about-models/async-task-api.md)。 +频繁轮询任务结果接口会浪费资源且可能触发[限流](../concepts/rate-limit.md)。百炼已接入阿里云事件总线 EventBridge，支持在任务完成后主动推送通知。详见[通过HTTP回调URL或MQ接收异步任务完成通知](../../raw/model-api-reference/more-about-models/async-task-api.md)。 两种接入方案： @@ -63,13 +63,13 @@ POST https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}/cancel ## 子[业务空间](../concepts/workspace.md)的模型调用 -默认[业务空间](../concepts/workspace.md)的 API Key 拥有所有模型的调用权限。如需按业务线隔离权限或分账，可使用[子业务空间的模型调用](../../raw/model-api-reference/more-about-models/model-calling-in-sub-workspace.md)。 +默认[业务空间](../concepts/workspace.md)的 [API Key](../concepts/api-key.md) 拥有所有模型的调用权限。如需按业务线隔离权限或分账，可使用[子业务空间的模型调用](../../raw/model-api-reference/more-about-models/model-calling-in-sub-workspace.md)。 **使用要点**： -- 必须使用子[业务空间](../concepts/workspace.md)自身的 API Key 进行调用。 +- 必须使用子[业务空间](../concepts/workspace.md)自身的 [API Key](../concepts/api-key.md) 进行调用。 - 调用标准模型（如 `qwen-plus`）前，需为该空间设置模型调用权限。 -- 调用在百炼上调优并部署的模型无需额外授权，但仅能由其所在空间的 API Key 调用。 +- 调用在百炼上调优并部署的模型无需额外授权，但仅能由其所在空间的 [API Key](../concepts/api-key.md) 调用。 - 支持 OpenAI 兼容方式和 DashScope 方式调用，但调优后模型仅支持 DashScope 方式。 ## 上传本地文件获取临时 URL @@ -80,8 +80,8 @@ POST https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}/cancel - 文件有效期 48 小时，过期自动清理。 - 上传时必须指定模型名称，且与后续调用的模型一致，不同模型无法共享文件。 -- 上传与调用的 API Key 必须属于同一阿里云主账号。 -- 单文件不超过 1GB，上传凭证接口限流 100 QPS。 +- 上传与调用的 [API Key](../concepts/api-key.md) 必须属于同一阿里云主账号。 +- 单文件不超过 1GB，上传凭证接口[限流](../concepts/rate-limit.md) 100 QPS。 - 使用 `oss://` 形式的 URL 调用模型时，HTTP 请求头中必须添加 `X-DashScope-OssResourceResolve: enable`。 > **注意**：临时 URL 不适用于生产环境。生产环境建议使用阿里云 OSS 等稳定存储方案。 @@ -122,6 +122,19 @@ POST https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}/cancel + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/more-models.md b/skills/bailian-docs-llm-wiki/wiki/api/more-models.md index 65fa010b..0efaf781 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/more-models.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/more-models.md @@ -105,7 +105,7 @@ completion = client.chat.completions.create( ## 限制和注意事项 -- **限流**：各模型有独立的限流条件，`farui-plus` 等模型限流详见[限流](https://help.aliyun.com/zh/model-studio/rate-limit)。 +- **[限流](../concepts/rate-limit.md)**：各模型有独立的[限流](../concepts/rate-limit.md)条件，`farui-plus` 等模型[限流](../concepts/rate-limit.md)详见[限流](https://help.aliyun.com/zh/model-studio/rate-limit)。 - **上下文与[计费](../concepts/billing.md)**：`farui-plus` 上下文 12k、最大输入 12k、最大输出 2k，输入成本 20 元/百万 [Token](../concepts/token.md)；`tongyi-intent-detect-v3` 上下文 8,192、最大输入 8,192、最大输出 1,024，输入 0.4 元、输出 1 元/百万 [Token](../concepts/token.md)，开通后 90 天内赠送 100 万 [Token](../concepts/token.md) 免费额度。 - **协议限制**：`qwen-deep-research` 仅支持 Python [DashScope SDK](../concepts/dashscope-sdk.md)；`gui-plus-2026-02-26` 仅在华北2（北京）地域提供。 - **SDK 线程安全**：DashScope Java SDK 的 `Generation` 等对象非线程安全，需自行管理同步机制或及时关闭进程。 @@ -134,6 +134,19 @@ completion = client.chat.completions.create( + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/more.md b/skills/bailian-docs-llm-wiki/wiki/api/more.md index 43d4533d..7eb1826a 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/more.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/more.md @@ -59,9 +59,9 @@ curl -X POST "https://dashscope.aliyuncs.com/api/v1/tokens?expire_in_seconds=180 **删除前注意事项**：删除 SLR 会导致对应功能不可用，须先清理依赖资源。例如删除 `AliyunServiceRoleForSFMAccessFC` 前须先删除所有已发布[工作流](../concepts/workflow.md)应用和流程中的函数计算节点并重新发布；删除 `AliyunServiceRoleForAccessOSS` 前须在安全存储空间中断开所有 OSS 连接；删除 `AliyunServiceRoleForSFMDataHubOSSImport` 前须确保没有进行中的 OSS 数据导入任务。具体删除步骤参见 [服务关联角色](https://help.aliyun.com/zh/ram/user-guide/service-linked-roles)。 -## 知识库 SearchFilters +## [知识库](../concepts/[knowledge](knowledge.md)-base.md) SearchFilters -在调用知识库 [Retrieve](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-retrieve) 接口时，若返回结果包含较多与 Query 无关的干扰信息（尤其适合结构化数据场景），可在请求体中传入 `searchFilters` 对语义检索结果做进一步过滤。 +在调用[知识库](../concepts/[knowledge](knowledge.md)-base.md) [Retrieve](https://help.aliyun.com/zh/model-studio/api-bailian-2023-12-29-retrieve) 接口时，若返回结果包含较多与 Query 无关的干扰信息（尤其适合结构化数据场景），可在请求体中传入 `searchFilters` 对语义检索结果做进一步过滤。 **效果对比**：未传入 `searchFilters` 时，Retrieve 可能返回多条低相关切片（如查询「张三」却返回李四、王五）；传入后可仅保留命中过滤条件的切片。 @@ -85,7 +85,7 @@ curl -X POST "https://dashscope.aliyuncs.com/api/v1/tokens?expire_in_seconds=180 | 范围查询-等值 | 数值、字符串 | 支持 `eq`（等于）、`neq`（不等于）；一个字段不可配多个值 | | 范围查询-区间 | 数值（long/double） | 支持 `gt`/`gte`/`lt`/`lte` | | 模糊查询 | 字符串 | 支持 `like` 属性，`%` 匹配任意字符（含零个） | -| 标签（Tag）查询 | 仅文档搜索、音视频搜索类知识库 | `tags` 字段，多个标签之间为 OR 关系 | +| 标签（Tag）查询 | 仅文档搜索、音视频搜索类[知识库](../concepts/[knowledge](knowledge.md)-base.md) | `tags` 字段，多个标签之间为 OR 关系 | **前置条件**：子账号需获取 `AliyunBailianDataFullAccess` 策略并加入[业务空间](../concepts/workspace.md)（主账号可操作所有[业务空间](../concepts/workspace.md)），获取[业务空间](../concepts/workspace.md) ID，安装百炼 SDK（2023-12-29 版本）并配置 `ALIBABA_CLOUD_ACCESS_KEY_ID` / `ALIBABA_CLOUD_ACCESS_KEY_SECRET` 环境变量。 @@ -127,6 +127,19 @@ resp = client.retrieve('请传入实际的业务空间ID', retrieve_request) + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/omni-realtime-api.md b/skills/bailian-docs-llm-wiki/wiki/api/omni-realtime-api.md index 9a40d452..b2b6205d 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/omni-realtime-api.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/omni-realtime-api.md @@ -1,164 +1,89 @@ # omni realtime api -Qwen-Omni-Realtime API 是阿里云百炼平台提供的实时[多模态](../concepts/multimodal.md)交互接口，基于 WebSocket 协议实现低延迟的音视频对话。该 API 支持语音输入/输出、图像输入、语音活动检测（VAD）、工具调用（Function Calling）、联网搜索及声音复刻等功能，适用于智能客服、语音助手等实时对话场景。 +Qwen-Omni-Realtime API 是百炼平台基于 WebSocket 的实时[多模态](../concepts/multimodal.md)对话接口，支持音频、图像（视频帧）与文本的流式输入输出，适用于语音助手、音视频客服、实时同传等场景。本页汇总其支持的模型、交互模式、事件协议、SDK 用法与声音复刻能力。 -## 支持的模型 +## 支持的模型与功能 -| 模型系列 | 模型名称 | 特性 | -| --- | --- | --- | -| Qwen3.5-Omni-Realtime | qwen3.5-omni-plus-realtime、qwen3.5-omni-flash-realtime | 支持 semantic_vad、联网搜索、工具调用、idle_timeout_ms | -| Qwen3-Omni-Flash-Realtime | qwen3-omni-flash-realtime | 支持 smooth_output 参数 | -| Qwen-Omni-Turbo-Realtime | qwen-omni-turbo-realtime | 大部分生成参数不支持修改 | +- **模型系列**：`qwen3.5-omni-realtime` 系列（含 `qwen3.5-omni-plus-realtime`、`qwen3.5-omni-flash-realtime`）、`qwen3-omni-flash-realtime`、`qwen-omni-turbo-realtime`。 +- **输出模态**：`["text"]` 仅文本，或 `["text","audio"]`（默认）文本+音频。 +- **默认音色**：Qwen3.5 系列为 `Tina`，Qwen3-Omni-Flash 为 `Cherry`，Qwen-Omni-Turbo 为 `Chelsie`；音色列表见实时文档。 +- **音频格式**：输入仅支持 16 kHz PCM，输出仅支持 24 kHz PCM，不支持自定义输出采样率。 +- **图像输入**：JPG/JPEG，建议 480p/720p（最高 1080p），Base64 编码后 ≤256KB，建议 1 张/秒，且须在发送过音频之后再发图像。 +- **能力差异**： + - `semantic_vad`、联网搜索（`enable_search`）、工具调用（`tools`）仅 Qwen3.5-Omni-Realtime 系列支持； + - `idle_timeout_ms`（静默超时后模型主动引导对话，范围 [5000, 30000]）仅 `qwen3.5-omni-plus/flash-realtime` + `server_vad` 生效； + - `smooth_output`（口语化/书面化风格）仅 Qwen3-Omni-Flash-Realtime 支持； + - `qwen-omni-turbo` 系列**不支持修改** temperature / top_p / top_k / max_tokens / repetition_penalty / presence_penalty / seed 等采样参数。 -各模型的默认音色不同：Qwen3.5-Omni-Realtime 系列为 `Tina`，Qwen3-Omni-Flash-Realtime 为 `Cherry`，Qwen-Omni-Turbo-Realtime 为 `Chelsie`。 +> **注意**：`tools` 与 `enable_search` 不兼容，不可同时开启。 -## 交互模式 +## 交互流程 -根据[实时多模态交互流程](../../raw/model-api-reference/omni-realtime-api/omni-realtime-interaction-process.md)，API 支持两种交互模式： +详见[实时多模态交互流程](../../raw/model-api-reference/omni-realtime-api/omni-realtime-interaction-process.md)。 -### VAD 模式（默认） +- **VAD 模式**（默认，`turn_detection.type = "server_vad"`）：服务端自动检测语音起止、自动提交缓冲区并生成响应，支持语音打断。事件序列：`speech_started` → `speech_stopped` → `input_audio_buffer.committed` → `conversation.item.created` → 响应事件。 +- **Manual 模式**（`turn_detection = null`）：客户端显式发送 `input_audio_buffer.commit` 提交音频，再发 `response.create` 触发响应，适用于"按下即说"场景。 +- **工具调用**：服务端通过 `response.function_call_arguments.delta/done` 下发调用参数（含 `call_id`）→ 客户端本地执行 → 通过 `conversation.item.create`（type 为 `function_call_output`）回传结果 → VAD 模式下服务端自动生成最终响应，Manual 模式需客户端再发 `response.create`。命中工具调用时模型不生成音频，仅返回参数。 -将 `session.turn_detection` 设为 `server_vad` 或 `semantic_vad`。服务端自动检测语音起止并触发模型响应，适用于持续音频流场景。支持语音打断。 +## 事件协议 -### Manual 模式 +### 客户端事件 -将 `session.turn_detection` 设为 `null`。客户端通过 `input_audio_buffer.commit` + `response.create` 手动控制对话节奏，适用于按下即说场景。 +完整字段定义见[客户端事件](../../raw/model-api-reference/omni-realtime-api/client-events.md)： -## 连接地址 - -``` -wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/realtime # 北京地域 -wss://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api-ws/v1/realtime # 新加坡地域 -``` - -将 `{WorkspaceId}` 替换为[业务空间](../concepts/workspace.md) ID。建议使用[业务空间](../concepts/workspace.md)专属域名以获得更好的性能和稳定性。 - -## 客户端事件 - -详细参数说明参见[客户端事件](../../raw/model-api-reference/omni-realtime-api/client-events.md)。 - -| 事件 | 用途 | +| 事件 | 说明 | | --- | --- | -| `session.update` | 更新会话配置（模态、音色、VAD、工具等） | -| `input_audio_buffer.append` | 追加音频数据（Base64 编码） | -| `input_audio_buffer.commit` | 提交音频缓冲区（Manual 模式必需） | +| `session.update` | 更新会话配置（modalities、voice、instructions、turn_detection、tools、enable_search、采样参数等） | +| `response.create` | 触发模型生成响应（VAD 模式下通常无需发送） | +| `response.cancel` | 取消进行中的响应 | +| `input_audio_buffer.append` | 追加 Base64 音频到缓冲区 | +| `input_audio_buffer.commit` | 提交缓冲区创建用户消息项（Manual 模式必需；同时提交图像缓冲区） | | `input_audio_buffer.clear` | 清空音频缓冲区 | -| `input_image_buffer.append` | 追加图像数据（JPG/JPEG，Base64 编码） | -| `response.create` | 触发模型生成响应 | -| `response.cancel` | 取消正在进行的响应 | -| `conversation.item.create` | 回传工具调用结果 | +| `input_image_buffer.append` | 追加 Base64 图像到缓冲区 | +| `conversation.item.create` | 回传工具执行结果（仅支持 `function_call_output` 类型） | -## 服务端事件 +### 服务端事件 -详细参数说明参见[服务端事件](../../raw/model-api-reference/omni-realtime-api/server-events.md)。 +完整定义见[服务端事件](../../raw/model-api-reference/omni-realtime-api/server-events.md)，主要包括： -| 事件 | 含义 | -| --- | --- | -| `session.created` | 连接建立，返回默认配置 | -| `session.updated` | 会话配置更新成功 | -| `error` | 错误信息 | -| `input_audio_buffer.speech_started` | VAD 检测到语音开始 | -| `input_audio_buffer.speech_stopped` | VAD 检测到语音结束 | -| `input_audio_buffer.committed` | 音频缓冲区已提交 | -| `response.audio.delta` | 增量音频输出 | -| `response.audio_transcript.delta` | 增量文本转录 | -| `response.done` | 响应完成 | -| `response.function_call_arguments.done` | 工具调用参数完成 | -| `conversation.item.input_audio_transcription.delta` | 实时语音识别中间结果 | +- **会话类**：`session.created`（连接后首个事件，含默认配置）、`session.updated`、`error`。 +- **缓冲区类**：`input_audio_buffer.speech_started/speech_stopped/committed/cleared`。 +- **对话项类**：`conversation.item.created`（type 为 `message` 或 `function_call`）。 +- **输入转录类**：`conversation.item.input_audio_transcription.delta`（实时预览 = `text` + `stash`，附带 `language`、`emotion`）与 `.completed`；转录模型固定为 `qwen3-asr-flash-realtime`，转录结果仅供参考，可能与模型理解存在差异。 +- **响应类**：`response.created`、`response.output_item.added`、`response.content_part.added`、`response.audio.delta`、`response.audio_transcript.delta/done`、`response.audio.done`、`response.content_part.done`、`response.output_item.done`、`response.done`，以及工具调用的 `response.function_call_arguments.delta/done`。 ## 关键会话参数 -通过 `session.update` 事件配置： - -| 参数 | 说明 | 默认值 | -| --- | --- | --- | -| `modalities` | 输出模态：`["text"]` 或 `["text","audio"]` | `["text","audio"]` | -| `voice` | 音色名称 | 因模型而异 | -| `input_audio_format` | 输入音频格式，仅支持 `pcm`（16kHz） | `pcm` | -| `output_audio_format` | 输出音频格式，仅支持 `pcm`（24kHz） | `pcm` | -| `instructions` | 系统消息 | - | -| `turn_detection.type` | VAD 类型：`server_vad` / `semantic_vad` | `server_vad` | -| `turn_detection.threshold` | VAD 灵敏度，范围 [-1.0, 1.0] | 0.5 | -| `turn_detection.silence_duration_ms` | 静音触发时间（ms），范围 [200, 6000] | 800 | -| `turn_detection.idle_timeout_ms` | 静默超时（ms），范围 [5000, 30000]，仅 qwen3.5 系列 | - | -| `enable_search` | 联网搜索，仅 Qwen3.5-Omni-Realtime | `false` | -| `tools` | 工具定义列表，仅 Qwen3.5-Omni-Realtime | `[]` | -| `smooth_output` | 口语化风格，仅 Qwen3-Omni-Flash-Realtime | `true` | - -> **注意**：`tools` 和 `enable_search` 不兼容，不可同时开启。 - -### 生成参数 - -| 参数 | Qwen3.5-Omni-Realtime | Qwen3-Omni-Flash-Realtime | Qwen-Omni-Turbo-Realtime | -| --- | --- | --- | --- | -| `temperature` | 0.7 | 0.9 | 1.0（不可改） | -| `top_p` | 0.8 | 1.0 | 0.01（不可改） | -| `top_k` | 20 | 50 | 20（不可改） | -| `repetition_penalty` | 1.0 | 1.05 | 1.05（不可改） | -| `presence_penalty` | 1.5 | 0.0 | 0.0（不可改） | - -> **注意**：`qwen-omni-turbo` 系列模型的生成参数不支持修改。 - -## SDK 使用 - -### Python SDK - -需要 [DashScope SDK](../concepts/dashscope-sdk.md) >= 1.25.17。核心类为 `OmniRealtimeConversation`，通过 `from dashscope.audio.qwen_omni import OmniRealtimeConversation` 引入。详见 [Python SDK](../../raw/model-api-reference/omni-realtime-api/omni-realtime-python-sdk.md)。 - -```python -from dashscope.audio.qwen_omni import MultiModality, OmniRealtimeCallback, OmniRealtimeConversation - -conv = OmniRealtimeConversation(model="qwen3.5-omni-plus-realtime", callback=callback, url=url) -conv.connect() -conv.update_session( - output_modalities=[MultiModality.AUDIO, MultiModality.TEXT], - voice="Tina", - enable_turn_detection=True -) -conv.append_audio(audio_base64) -conv.close() -``` - -### Java SDK - -需要 DashScope Java SDK >= 2.22.15。核心类为 `OmniRealtimeConversation`，通过 `OmniRealtimeParam` 和 `OmniRealtimeConfig` 配置参数。详见 [Java SDK](../../raw/model-api-reference/omni-realtime-api/omni-realtime-java-sdk.md)。 - -```java -OmniRealtimeParam param = OmniRealtimeParam.builder() - .model("qwen3.5-omni-plus-realtime") - .url(url) - .build(); -OmniRealtimeConversation conversation = new OmniRealtimeConversation(param, callback); -conversation.connect(); -conversation.updateSession(OmniRealtimeConfig.builder() - .modalities(Arrays.asList(OmniRealtimeModality.AUDIO, OmniRealtimeModality.TEXT)) - .voice("Tina") - .enableTurnDetection(true) - .build()); -``` - -> **注意**：Java SDK 中 `instructions`、`smooth_output`、`enable_search`、`search_options`、`tools` 及生成参数（temperature/top_p/top_k 等）需通过 `OmniRealtimeConfig` 的 `parameters` 方法设置。 - -## 工具调用（Function Calling） - -仅 Qwen3.5-Omni-Realtime 模型支持。流程如下： - -1. 通过 `session.update` 配置 `tools` 列表 -2. 服务端识别到需要调用工具时，通过 `response.function_call_arguments.done` 返回函数名和参数 -3. 客户端执行工具函数，通过 `conversation.item.create` 回传结果（`type: "function_call_output"`） -4. VAD 模式下服务端自动生成响应；Manual 模式下需额外发送 `response.create` +- **turn_detection**：`server_vad`（默认）或 `semantic_vad`；`threshold` ∈ [-1.0, 1.0]（默认 0.5，越低越灵敏），`silence_duration_ms` ∈ [200, 6000]（默认 800）。设为 `null` 即 Manual 模式。 +- **采样参数默认值**（temperature / top_p / top_k）：qwen3.5 系列 0.7 / 0.8 / 20；qwen3-omni-flash 0.9 / 1.0 / 50；qwen-omni-turbo 1.0 / 0.01 / 20。temperature 与 top_p 建议只设其一。 +- **presence_penalty**：qwen3.5 系列默认 1.5，其他默认 0.0；范围 [-2.0, 2.0]。 +- **max_tokens**：仅截断输出，不影响生成过程。 +- **seed**：0 ~ 2³¹−1，默认 -1，用于结果复现。 + +## SDK 使用方式 + +- **Python SDK**（dashscope ≥ 1.25.17）：核心类 `OmniRealtimeConversation`，方法包括 `connect()`、`update_session()`、`append_audio()`、`append_video()`、`commit()`、`create_response()`、`cancel_response()`、`create_item()`、`close()`，服务端事件通过 `OmniRealtimeCallback` 回调下发。详见 [Python SDK](../../raw/model-api-reference/omni-realtime-api/omni-realtime-python-sdk.md)。 +- **Java SDK**（dashscope ≥ v2.22.15）：对应类 `OmniRealtimeConversation` + `OmniRealtimeParam` / `OmniRealtimeConfig`；注意 `instructions`、`smooth_output`、`enable_search`、`search_options`、`tools` 及采样参数需通过 `OmniRealtimeConfig.parameters` 传入。详见 [Java SDK](../../raw/model-api-reference/omni-realtime-api/omni-realtime-java-sdk.md)。 +- **接入地址**（WebSocket）： + - 北京：`wss://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api-ws/v1/realtime` + - 新加坡：`wss://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api-ws/v1/realtime` + + 官方建议从旧域名 `wss://dashscope.aliyuncs.com` / `wss://dashscope-intl.aliyuncs.com` 迁移至[业务空间](../concepts/workspace.md)专属域名（旧域名仍可用）。 ## 声音复刻 -通过 `qwen-voice-enrollment` 模型创建自定义音色，然后在实时对话中使用。音频要求：WAV/MP3/M4A 格式，10-20 秒，采样率 >= 24kHz，单声道，文件 < 10MB。创建音色时指定的 `target_model` 必须与后续对话使用的模型一致。 +通过 `qwen-voice-enrollment` 模型上传 10~20 秒音频（WAV 16bit / MP3 / M4A，≥24 kHz，单声道，<10 MB）即可免训练创建定制音色，随后在 Realtime 会话的 `voice` 参数中使用。详见[声音复刻API参考](../../raw/model-api-reference/omni-realtime-api/qwen-omni-voice-cloning.md)。 -支持的驱动模型：qwen3.5-omni-plus-realtime、qwen3.5-omni-flash-realtime。 +> **注意**：创建音色时的 `target_model` 必须与后续对话调用的模型一致（支持 qwen3.5-omni-plus/flash-realtime 及非实时 qwen3.5-omni-plus/flash），否则合成失败。 -## 输入限制 +## 限制与注意事项 -- 音频输入：16kHz 采样率 PCM，音频缓冲区最大 15MiB -- 图像输入：JPG/JPEG 格式，建议 480p-720p（最高 1080p），Base64 编码后不超过 256KB，建议 1 帧/秒 -- 图像需在至少一次 `input_audio_buffer.append` 之后发送，通过 `input_audio_buffer.commit` 与音频一起提交 +- 输入/输出音频格式均仅支持 PCM，采样率固定（输入 16 kHz、输出 24 kHz）。 +- VAD 模式下推荐使用耳机播放，避免回声触发语音打断。 +- 提交音频缓冲区（commit）本身不会触发模型响应。 +- Manual 模式下 `append_audio` 单次最多 15 MiB。 +- 联网搜索与工具调用互斥；`semantic_vad` 仅 qwen3.5 系列可用。 +- qwen-omni-turbo 系列采样参数不可修改。 ## 来源文档 @@ -173,7 +98,3 @@ conversation.updateSession(OmniRealtimeConfig.builder() - - - - diff --git a/skills/bailian-docs-llm-wiki/wiki/api/preparations.md b/skills/bailian-docs-llm-wiki/wiki/api/preparations.md index b4840377..4e4ee180 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/preparations.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/preparations.md @@ -1,73 +1,73 @@ # preparations -本页汇总在阿里云百炼平台调用模型 API 前的准备工作，涵盖获取鉴权凭证（API Key）、安装官方或兼容 SDK、使用百炼 CLI 快速集成，以及常见错误码的排查思路。面向开发者，帮助你从零完成环境搭建并稳定发起第一次调用。 +本页汇总调用阿里云百炼模型 API 前的准备工作：获取并安全配置 API Key、安装 SDK（DashScope 或 OpenAI 兼容）、安装百炼 CLI，以及调用出错时的错误码排查。完成这些步骤后，即可通过代码、第三方工具或命令行调用百炼平台的模型与应用。 -## 获取并配置 API Key +## 1. 获取 API Key -调用模型或应用前，需先获取 API Key 作为鉴权凭证。需使用主账号，或具备 `管理员` / `API-Key` 页面权限的子账号，在[阿里云百炼控制台](https://bailian.console.aliyun.com/)对应地域的 **API Key** 页面创建。详见 [获取API Key](../../raw/model-api-reference/preparations/get-api-key.md)。 +API Key 是调用百炼模型和应用的鉴权凭证，详见[获取API Key](../../raw/model-api-reference/preparations/get-api-key.md)。 -创建时的关键选项： +- **操作入口**：百炼控制台右上角选择地域（华北2（北京）、新加坡、日本（东京）、德国（法兰克福），或美国（弗吉尼亚）走独立控制台），进入 API Key 页面创建。 +- **权限要求**：需主账号，或具备 `管理员` / `API-Key` 页面权限的子账号。 +- **关键配置**： + - **归属[业务空间](../concepts/workspace.md)**：API Key 的调用权限完全由归属[业务空间](../concepts/workspace.md)决定，同一空间内权限相同，无需按模型类型创建多个 Key；子空间 Key 只能调用已授权模型和该空间内的应用；调优后的模型仅能用其所在空间的 Key 调用。 + - **权限**：`全部` 或 `自定义`（IP 白名单最多 20 个 IPv4/IPv6 地址或网段 + 可访问模型范围；美国（弗吉尼亚）地域不支持自定义配置）。 +- **安全升级**：新创建的 Key 统一以 `sk-ws` 开头，仅在创建时展示一次明文，关闭弹窗后无法再次查看，丢失需重置或重建；旧的 `sk-` 开头 Key 可继续使用。 +- **管理操作**：控制台支持编辑、删除、禁用、重置（禁用/重置在美国（弗吉尼亚）地域不可用）；也可通过 OpenAPI（CreateApiKey / ListApiKeys / ResetApiKey 等，需阿里云 AccessKey 签名而非 API Key）编程管理。 -- **归属业务空间**：决定该 Key 的调用权限。同一空间内的 Key 权限相同，无需为不同模态（文生文、文生图、语音等）分别创建。默认业务空间的 Key 可调用所有标准模型及默认空间内的应用；子业务空间的 Key 只能调用已授权的模型及本空间应用。 -- **权限**：可选 **全部**（调用所有模型与应用），或 **自定义**（配置 IP 白名单最多 20 个 IPv4/IPv6 地址或网段，以及可访问的模型/应用范围）。 +> **注意**：Token Plan / Coding Plan 使用以 `sk-sp-` 开头的专属 API Key，与本文按量付费 API Key 获取方式不同。 -> **注意**：百炼已对按量付费 API Key 做安全升级（美国（弗吉尼亚）地域除外）。升级后新建的 Key 以 `sk-ws` 开头，且**仅在创建时展示一次明文**，关闭弹窗后无法再次查看，务必立即复制保存；升级前 `sk-` 开头的旧 Key 仍可正常使用。此外，Token Plan / Coding Plan 使用以 `sk-sp-` 开头的专属 Key，不同于本文的按量付费 Key。 +### 配置到环境变量 -推荐将 API Key 配置到环境变量 `DASHSCOPE_API_KEY`，避免硬编码泄漏。各系统配置方式（`~/.bashrc`、`~/.zshrc`、`~/.bash_profile`、Windows 系统属性 / `setx` / PowerShell）参见原文。调用时除 API Key 外，还需指定**服务端点** `base_url`（即创建弹窗中的 API Host），且 OpenAI 兼容协议与 Anthropic 兼容协议的 `base_url` 不同、随地域变化，请以对应接口文档为准。 +建议将 API Key 配置为环境变量 `DASHSCOPE_API_KEY`，避免硬编码泄露： -除控制台外，百炼还提供 OpenAPI（`CreateApiKey` / `GetApiKey` / `ListApiKeys` / `UpdateApiKey` / `DeleteApiKey` / `EnableApiKey` / `DisableApiKey` / `ResetApiKey`）以编程方式管理 Key，调用需使用阿里云账号 AccessKey 签名认证并具备相应 RAM 权限。 +- Linux：`echo "export DASHSCOPE_API_KEY='YOUR_KEY'" >> ~/.bashrc && source ~/.bashrc` +- macOS：按 `echo $SHELL` 判断写入 `~/.zshrc` 或 `~/.bash_profile` +- Windows：系统属性图形界面，或 CMD `setx DASHSCOPE_API_KEY "YOUR_KEY"`，或 PowerShell `[Environment]::SetEnvironmentVariable(...)` -## 安装 SDK +调用时除 API Key 外还需指定服务端点（`base_url`）。百炼同时提供 **OpenAI 兼容**与 **Anthropic 兼容**两种协议，`base_url` 不同且随地域变化，以对应接口文档为准。 -百炼同时支持官方 **DashScope SDK**（Python、Java）与通过 **[OpenAI 兼容接口](../concepts/openai-compatible-interface.md)**调用的多语言 SDK。详见 [安装SDK](../../raw/model-api-reference/preparations/install-sdk.md)。 +## 2. 安装 SDK -- **Python**（需 `python >= 3.8`）：`pip install -U openai` 或 `pip install -U dashscope` -- **Java**：DashScope 用 `com.alibaba:dashscope-sdk-java`；OpenAI 用 `com.openai:openai-java`（需 Java 8+，推荐 `3.5.0`），均通过 Maven / Gradle 引入。 -- **Node.js**：`npm install --save openai`（或 `yarn add openai`）；安装失败可配置镜像源 `npm config set registry https://registry.npmmirror.com/`。 -- **Go**（需 `Go 1.22+`）：`go get 'github.com/openai/openai-go/v3'`；超时可设 `go env -w GOPROXY=https://mirrors.aliyun.com/goproxy/,direct`。 +支持官方 DashScope SDK 与 OpenAI 官方 SDK 两条路线，详见[安装SDK](../../raw/model-api-reference/preparations/install-sdk.md)： -安装后即可调用文本生成、图像生成、视频生成、语音合成/识别、向量、排序等模型。 +| 语言 | SDK | 安装方式 | 版本要求 | +| --- | --- | --- | --- | +| Python | OpenAI / DashScope | `pip install -U openai` / `pip install -U dashscope` | Python >= 3.8 | +| Java | DashScope / OpenAI | Maven/Gradle 添加 `dashscope-sdk-java` 或 `com.openai:openai-java`（推荐 3.5.0） | Java 8+ | +| Node.js | OpenAI | `npm install --save openai` | — | +| Go | OpenAI | `go get github.com/openai/openai-go/v3` | Go 1.22+ | -## 使用百炼 CLI +网络不畅时可配置国内镜像：npm 用 `registry.npmmirror.com`，Go 用 `GOPROXY=https://mirrors.aliyun.com/goproxy/,direct`。 -百炼 CLI（npm 包 `bailian-cli`，命令 `bl` / `bailian`）是面向 AI Agent 的命令行工具，可将平台能力集成到各类 AI 工具中。安装前置要求 **Node.js ≥ 22.12.0**，且**仅支持 npm 安装**（勿用 pnpm / yarn 安装该包）。详见 [使用百炼 CLI](../../raw/model-api-reference/preparations/use-model-studio-cli.md)。 +## 3. 使用百炼 CLI（可选） -```bash -# 1. 安装 CLI -npm install -g bailian-cli -# 2. 安装 Skills（注册能力描述文件到各 Agent） -npx skills add modelstudioai/cli --all -g -# 3. 验证 -bl --version -``` +百炼 CLI（npm 包 `bailian-cli`，命令 `bl` / `bailian`）是面向 AI Agent 的命令行工具，详见[使用百炼 CLI](../../raw/model-api-reference/preparations/use-model-studio-cli.md)： -**认证方式**（可组合使用，互不覆盖）： +- **安装**：要求 Node.js >= 22.12.0，仅支持 npm：`npm install -g bailian-cli`，可配套 `npx skills add modelstudioai/cli --all -g` 安装 Skills。 +- **认证**：控制台浏览器登录 `bl auth login --console`（推荐，同时打通应用管理能力）、API Key 登录 `bl auth login --api-key sk-xxx`、环境变量、`bl config set --key api_key`（不校验 Key 有效性）、单次 `--api-key` 临时传入。 +- **能力**：文本对话（`bl text chat`）、全模态（`bl omni`）、图像生成/编辑（`bl image generate/edit`）、视频生成/编辑/参考生成（`bl video generate/edit/ref`）、视觉理解（`bl vision describe`）、语音合成（`bl speech synthesize`）等。 +- **常用全局参数**：`--region cn|us|intl`（默认 cn）、`--output text|json`、`--non-interactive`（Agent/CI 场景）、`--dry-run`、`--concurrent `。 +- **验证**：`bl auth status --output json`、`bl text chat --message "ping" --non-interactive --output json`。 -| 方式 | 命令 | 适用场景 | -| --- | --- | --- | -| 控制台登录（推荐） | `bl auth login --console` | 模型调用 + 应用管理（浏览器 OAuth） | -| API Key | `bl auth login --api-key sk-xxx` | 模型调用；会先校验 Key 有效性 | -| 环境变量 | 配置 API Key 环境变量 | CI/CD、无界面环境 | -| 配置文件 | `bl config set --key api_key --value sk-xxx` | 持久化，**不校验** Key 有效性 | -| 临时传入 | `bl text chat --api-key sk-xxx ...` | 单次调用，不落盘 | +> **注意**：切勿把真实 API Key 写入仓库、日志或公开渠道；CI 环境请用密钥管理或环境变量注入。 -常用全局参数：`--region `（默认 cn）、`--base-url`、`--output `、`--non-interactive`（Agent/CI）、`--dry-run`、`--concurrent ` 等。子命令覆盖文本对话（`bl text chat`）、全模态（`bl omni`）、图像（`bl image generate/edit`）、视频（`bl video generate/edit/ref`）、视觉理解（`bl vision describe`）、语音合成（`bl speech synthesize`）等。 +## 4. 错误码排查 -> **注意**：CLI 文档中示例默认模型（如 `qwen3.7-max`、`qwen3.5-omni-plus`、`qwen-image-2.0`、`happyhorse-1.0-t2v` 等）为工具内置默认值，可能随版本变化；实际可用模型请以模型列表 / 控制台为准。安全约束上，禁止将真实 API Key 写入仓库、日志、Skill 或聊天记录的可公开部分。 +调用失败时可对照[错误码](../../raw/model-api-reference/preparations/error-code.md)排查，也可将报错信息直接粘贴给阿里云 AI 助理分析。常见 400-InvalidParameter 类问题： -## 常见错误码与排查 +- **思考模式相关**：思考模式模型必须流式调用（`enable_thinking=true` 时需 `stream` + `incremental_output=true`，DashScope 协议下 `result_format="message"`）；部分模型（如 `qwen3-235b-a22b-thinking-2507`）不允许 `enable_thinking=false`；结构化输出（`json_object`）不能与思考模式同时开启。 +- **参数范围**：`temperature` 在 [0.0, 2.0)、`top_p` 在 (0.0, 1.0]、`top_k >= 0`、`presence_penalty` 在 [-2.0, 2.0]、`n` 在 [1, 4]、`max_tokens` 不超过模型最大输出 Token 数、输入长度不超过模型上限（连续对话超限时开启新对话）。 +- **模型与开通**：`Model not exist` 多为模型名大小写/格式错误（使用百炼模型 ID 而非开源社区名称）；`The product is not activated` 需先在模型市场开通目标模型或开通百炼服务。 +- **messages 构造**：纯文本模型 `content` 只能是字符串，混入 `image_url` 等[多模态](../concepts/multimodal.md)元素会报错（需改用 Qwen-VL 等[多模态](../concepts/multimodal.md)模型）；Tool Message 前必须先追加含 `tool_calls` 的 Assistant Message；检测到连续重复的相同工具调用会被拦截，需在应用侧增加终止逻辑。 +- **文件类**（Qwen-Long 等）：file-id 需有效且属于当前账号、数量小于 100、单文件小于 150 MB、仅支持纯文本格式（TXT/DOCX/PDF/EPUB/MOBI/MD）。 +- **欠费**：`Arrearage` 表示账号欠费，需前往费用与成本页面充值。 -调用过程中的报错多为 **400-InvalidParameter** 类的参数问题，可对照错误信息定位。完整清单见 [错误码](../../raw/model-api-reference/preparations/error-code.md)，以下为高频场景： +## 推荐流程 -- **思考模式相关**：思考模式模型需 `enable_thinking=true` 时配合[流式输出](../concepts/streaming.md)，并设 `incremental_output=true`、`result_format="message"`；部分模型（如 `qwen3-235b-a22b-thinking-2507`）不允许将 `enable_thinking` 设为 `false`。 -- **参数取值范围**：`temperature` ∈ [0.0, 2.0)、`top_p` ∈ (0.0, 1.0]、`top_k` ≥ 0、`presence_penalty` ∈ [-2.0, 2.0]、`n` ∈ [1, 4]；`max_tokens` 与输入长度上限以模型列表为准。 -- **模型不存在（Model not exist）**：核对 `model` 名称大小写与空格，勿混用开源社区名与百炼模型 ID（用 `qwen3-235b-a22b-instruct-2507` 而非 `Qwen/Qwen3-235B-A22B-Instruct-2507`）。 -- **content 类型错误**：纯文本模型的 `content` 必须为字符串，不能传数组或图片等多模态元素；需要图片输入请改用 Qwen-VL / Qwen3-VL 等多模态模型。 -- **结构化输出**：使用 `response_format` 的 `json_object` 时，提示词须包含 `json` 关键词，且不能同时开启思考模式。 -- **文件类（Qwen-Long）**：仅支持纯文本格式（TXT/DOCX/PDF/EPUB/MOBI/MD），单文件 < 150 MB、< 15000 页，file-id 数量 < 100。 -- **账号状态（Arrearage）**：账号欠费会导致访问被拒绝，需在费用与成本页面充值后等待系统更新。 - -排障时可借助[阿里云 AI 助理](https://www.aliyun.com/ai-assistant/)，直接粘贴报错信息即可获得原因与解决方案。 +1. 创建 API Key 并立即保存（新 Key 明文只展示一次）。 +2. 将 Key 配置到 `DASHSCOPE_API_KEY` 环境变量。 +3. 按语言安装 SDK（或安装百炼 CLI），确认对应协议的 `base_url`。 +4. 发起首次调用验证，报错时对照错误码文档定位原因。 ## 来源文档 @@ -77,4 +77,3 @@ bl --version - [使用百炼 CLI](../../raw/model-api-reference/preparations/use-model-studio-cli.md) - diff --git a/skills/bailian-docs-llm-wiki/wiki/api/qwen-api-reference.md b/skills/bailian-docs-llm-wiki/wiki/api/qwen-api-reference.md index f4dc1072..0aba030b 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/qwen-api-reference.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/qwen-api-reference.md @@ -22,7 +22,7 @@ ## 使用方式与注意事项 -- [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)可直接复用官方 OpenAI 客户端库，仅需替换 base URL 和 API Key，改动量小。 +- [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)可直接复用官方 OpenAI 客户端库，仅需替换 base URL 和 [API Key](../concepts/api-key.md)，改动量小。 - 若依赖联网搜索、代码解释器等内置工具，需使用 Responses 接口，而非普通的 Chat Completions。 - 不同接口在参数集合和功能覆盖上存在差异：DashScope 参数最全，OpenAI/Anthropic 兼容接口以对应生态的字段约定为准，跨接口迁移时需核对参数映射。 @@ -35,3 +35,16 @@ + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/toolkits-and-frameworks.md b/skills/bailian-docs-llm-wiki/wiki/api/toolkits-and-frameworks.md index fad0f724..3eea86fb 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/toolkits-and-frameworks.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/toolkits-and-frameworks.md @@ -1,98 +1,108 @@ # toolkits and [frameworks](frameworks.md) -阿里云百炼的通义千问等模型提供了一套与 OpenAI 高度兼容的接口体系，覆盖 Chat Completions、Responses、Completions、Embedding、文件、Batch、Conversations 等能力，并可直接接入 LangChain/LangChain4j 等主流框架。对于已有 OpenAI 应用，通常只需替换 `api_key`、`base_url` 与 `model` 三项即可完成迁移，无需改动业务逻辑。 +阿里云百炼提供了一整套 [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)（Chat、Responses、Vision、Embedding、Completions、Files、Batch、Conversations），开发者只需替换 `base_url`、`api_key` 和 `model` 三个参数，即可将原有 OpenAI 代码或 LangChain 等框架应用迁移到百炼服务。本文汇总各兼容接口的支持范围、接入方式与关键限制。 -## 迁移三要素与服务地址 +## 服务地址与鉴权 -将 OpenAI 应用迁移到百炼的核心是配置以下三项（详见 [OpenAI Chat接口兼容](../../raw/model-api-reference/toolkits-and-frameworks/compatibility-of-openai-with-dashscope.md)）： +所有 [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)统一使用 `https:///compatible-mode/v1` 作为 base_url，按地域区分： -- **`api_key`**：替换为[百炼 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)。**各地域的 API Key 不同**，切换地域时需同步更换。建议配置到环境变量 `DASHSCOPE_API_KEY` 以降低泄露风险。 -- **`base_url`**：OpenAI SDK 调用统一使用 `/compatible-mode/v1` 路径；HTTP 调用在其后追加具体资源路径（如 `/chat/completions`、`/responses`、`/embeddings`、`/files`）。 -- **`model`**：替换为百炼支持的模型名称。 - -各地域 SDK `base_url`： - -| 地域 | base_url | +| 地域 | SDK base_url | | --- | --- | | 华北2（北京） | `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` | | 新加坡 | `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` | | 日本（东京） | `https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1` | -| 德国（法兰克福） | `https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/compatible-mode/v1` | | 美国（弗吉尼亚） | `https://dashscope-us.aliyuncs.com/compatible-mode/v1` | +| 德国（法兰克福，仅 Responses API） | `https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/compatible-mode/v1` | -其中 `{WorkspaceId}` 为业务空间 ID，可在百炼控制台**业务空间详情**页面查看。 +其中 `{WorkspaceId}` 为[业务空间](../concepts/workspace.md) ID，可在百炼控制台「[业务空间](../concepts/workspace.md)详情」页查看。各地域的 [API Key](../concepts/api-key.md) 不通用，切换地域需同步更换 Key。详见 [OpenAI Chat接口兼容](../../raw/model-api-reference/toolkits-and-frameworks/compatibility-of-openai-with-dashscope.md)。 -> **注意**：百炼为北京、新加坡地域推出了业务空间专属域名，性能与稳定性更佳，建议从旧域名迁移：北京 `https://dashscope.aliyuncs.com` → `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com`；新加坡 `https://dashscope-intl.aliyuncs.com` → `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`。现有域名仍可正常使用。 +> **注意**：北京和新加坡地域已推出[业务空间](../concepts/workspace.md)专属域名（`{WorkspaceId}.cn-beijing.maas.aliyuncs.com` / `{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com`），旧域名 `dashscope.aliyuncs.com` / `dashscope-intl.aliyuncs.com` 仍可用但建议迁移。此外 Responses API 旧路径 `/api/v2/apps/protocols/compatible-mode/v1/responses` 与 Conversations 旧路径即将停止维护，应迁移到 `/compatible-mode/v1/...` 新路径，参见 [OpenAI Responses接口兼容](../../raw/model-api-reference/toolkits-and-frameworks/compatibility-with-openai-responses-api.md)。 -> **注意**：Responses 与 Conversations 接口的旧版路径 `/api/v2/apps/protocols/compatible-mode/v1/...` 即将停止维护，请尽快迁移至新版 `/compatible-mode/v1/...` 路径。 +## 各接口能力概览 -## 各兼容接口一览 +### Chat Completions（对话） -### Chat Completions（对话补全） +- **支持模型**：Qwen 商业版/开源版、Qwen-VL、Qwen-Coder、Qwen-Omni、Qwen-Math，以及 DeepSeek、Kimi、GLM、MiniMax 等第三方直供模型（三方直供仅限中国站中国内地，需先在控制台开通）。 +- **不支持**：Qwen-Audio 仅支持 DashScope 协议，不走 OpenAI 兼容。 +- 支持非流式/流式（`stream=True` + `stream_options={"include_usage": True}` 可在末尾返回 token 用量）、function call（`tools` 参数，支持多轮工具调用）。 -最常用的兼容接口，支持非流式、流式（`stream=True`，配合 `stream_options={"include_usage": True}` 返回 Token 统计）与 function call（工具调用）。支持模型广泛：Qwen 大语言模型（商业版/开源版）、Qwen-VL、Qwen-Coder、Qwen-Omni、Qwen-Math，以及 DeepSeek、Kimi、GLM、MiniMax 等三方模型。 +### Responses API（Chat 的演进版本） -> **注意**：三方直供模型仅在中国站的中国内地地域可用，调用前需先在百炼控制台开通对应服务。Qwen-Audio 不支持 OpenAI 兼容协议，仅支持 DashScope 协议。 +- 支持 `qwen3-max`、`qwen3.7-max`、`qwen3.7-plus`、`qwen3.6-plus`、`qwen3.5-plus`、`qwen3.7-flash`、`qwen3.6-flash`、`qwen3.5-flash`、`qwen3-coder-plus/flash/next`、`qwen-plus`、`qwen-flash` 等模型。 +- 输入更灵活：可直接传字符串，也兼容 Chat 消息数组；通过 `previous_response_id` 自动关联上下文（响应 id 有效期 7 天），无需手动维护消息历史。 +- 内置联网搜索、网页抓取、代码解释器、文搜图、图搜图等工具。 -### Responses（智能体原生接口） - -作为 Chat Completions 的演进版本，Responses API 内置联网搜索、网页抓取、代码解释器、文搜图/图搜图等工具，输入更灵活（可直接传字符串），并通过 `previous_response_id` 自动管理多轮上下文，无需手动拼接消息历史。详见 [OpenAI Responses接口兼容](../../raw/model-api-reference/toolkits-and-frameworks/compatibility-with-openai-responses-api.md)。 +### Vision（视觉理解） -- 支持模型示例：`qwen3-max`、`qwen3.7-plus`、`qwen-plus`、`qwen-flash`、`qwen3-coder-plus` 等。 -- `previous_response_id` 需传入上一轮响应的顶层 `id`（`resp_xxx`），而非 `output` 数组内消息的 `id`；当前响应 `id` 有效期为 **7 天**。 +- 支持 Qwen-VL、QVQ、Qwen-OCR 系列，各地域可用模型有差异，以控制台模型广场为准。 +- 通过 `messages[].content` 中的 `image_url` 类型传入图片 URL 或 Base64；QVQ 仅支持[流式输出](../concepts/streaming.md)。示例见 [OpenAI Vision接口兼容](../../raw/model-api-reference/toolkits-and-frameworks/qwen-vl-compatible-with-openai.md)。 -### Conversations（会话管理） +### Embedding（文本向量） -提供会话的创建、查询、更新、删除及消息项管理。配合 Responses API 可自动注入历史上下文，实现跨设备、跨会话的对话延续。初始消息项 `items` 最多 20 条，`metadata` 最多 16 对键值对（key ≤ 64 字符、value ≤ 512 字符）。删除会话时其消息项不会被删除。 +- 支持 `text-embedding-v1` ~ `text-embedding-v4`。v4/v3 支持 `dimensions` 参数指定向量维度（v4 可选 64~2048，默认 1024；v2 固定 1536 维）。 +- v4 单行最大 8,192 [Token](../concepts/token.md)、最多 10 行；v2/v1 单行最大 2,048 [Token](../concepts/token.md)、最多 25 行。 +- [多模态](../concepts/multimodal.md) Embedding 模型（如 qwen3-vl-embedding）**不支持** [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)，需走 DashScope。详见 [OpenAI Embedding接口兼容](../../raw/model-api-reference/toolkits-and-frameworks/embedding-interfaces-compatible-with-openai.md)。 ### Completions（文本补全） -专为代码补全、内容续写设计，当前仅支持 `qwen-coder-turbo`，且**仅适用于中国内地（北京地域）**。通过 `<|fim_prefix|>...<|fim_suffix|>...<|fim_middle|>` 模板可实现「前缀生成后续」或「前缀+后缀生成中间」两种补全（暂不支持仅凭后缀生成前缀）。关键参数包括 `max_tokens`、`temperature`、`top_p`、`stop`、`seed`、`presence_penalty` 等，详见 [completions 接口](../../raw/model-api-reference/toolkits-and-frameworks/completions.md)。 +- 仅支持 `qwen-coder-turbo`，且仅适用于华北2（北京）地域。 +- 使用 FIM 模板：前缀补全 `<|fim_prefix|>{prefix}<|fim_suffix|>`；前后缀补中间 `<|fim_prefix|>{prefix}<|fim_suffix|>{suffix}<|fim_middle|>`。不支持仅给后缀生成前缀。参数说明见 [completions 接口](../../raw/model-api-reference/toolkits-and-frameworks/completions.md)。 -### Embedding（文本向量） +### Files（文件管理） -兼容 OpenAI Embedding 规范，支持 `text-embedding-v1/v2/v3/v4`。其中 v3、v4 支持通过 `dimensions` 参数指定向量维度（v4 可选 64~2048 多档，默认 1024）。 +通过 `client.files.create(file=..., purpose=...)` 上传，按 `purpose` 区分用途： -> **注意**：多模态 Embedding 模型（如 qwen3-vl-embedding、tongyi-embedding-vision 系列）不支持 [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)，需改用[多模态向量接口](https://help.aliyun.com/zh/model-studio/multimodal-embedding-api-reference)。 +| purpose | 用途 | 单文件上限 | +| --- | --- | --- | +| `file-extract` | Qwen-Long / Qwen-Doc-Turbo 文档问答与数据提取（支持 TXT/DOCX/PDF/XLSX/EPUB/MD/CSV/JSON 及图片） | 150 MB | +| `batch` | 批量推理输入（JSONL） | 500 MB | +| `fine-tune` | 模型调优数据集（JSONL） | 300 MB | -### Vision（视觉理解） +存储空间上限：10,000 个文件、总计 100 GB，无有效期限制；达到上限需删除旧文件后才能继续上传。文件 ID 可复用，无需重复上传。 -Qwen-VL、QVQ、Qwen-OCR 兼容 OpenAI Chat 接口，通过 `content` 数组中的 `image_url` 传入图片。各地域支持的模型有差异。QVQ 模型仅支持[流式输出](../concepts/streaming.md)。 +### Batch（批量推理，费用为实时调用的 50%） -### 文件接口与 Batch +两种形态： -文件上传接口（`client.files.create`）通过 `purpose` 区分用途，详见 [OpenAI文件接口兼容](../../raw/model-api-reference/toolkits-and-frameworks/openai-file-interface.md)： +1. **Batch File API（文件输入）**：上传 JSONL → 创建任务 → 轮询状态 → 下载结果/错误文件。JSONL 每行格式为 `{"custom_id", "method": "POST", "url": "/v1/chat/completions", "body": {...}}`。北京地域支持千问 Max/Plus/Flash/Long、DeepSeek、[多模态](../concepts/multimodal.md)（qwen3-vl 等）、OCR、qwen3.5-omni-plus 及 text-embedding-v1~v4；新加坡仅支持 qwen-max/plus/turbo。可先用 `batch-test-model` 做全链路测试（不产生推理费用）。详见 [OpenAI兼容-Batch（文件输入）](../../raw/model-api-reference/toolkits-and-frameworks/batch-interfaces-compatible-with-openai.md)。 +2. **Batch Chat（同步调用）**：保持与实时 API 一致的同步调用方式，仅需将 base_url 改为 `https://batch.dashscope.aliyuncs.com/compatible-mode/v1`。默认等待超时 3600 秒，可自定义 60~3600 秒；超时则断开连接返回错误。目前仅华北2（北京）地域。参见 [OpenAI兼容-Batch Chat](../../raw/model-api-reference/toolkits-and-frameworks/openai-compatible-batch-chat.md)。 -| purpose | 用途 | 单文件大小上限 | -| --- | --- | --- | -| `file-extract` | Qwen-Long / Qwen-Doc-Turbo 文档问答与数据提取 | 150 MB | -| `batch` | 批量推理输入（jsonl） | 500 MB | -| `fine-tune` | 模型调优数据集（jsonl） | 300 MB | +> **注意**：Batch 场景下 qwen3.7/3.6/3.5 系列单次请求上下文最大 256K，且这些模型默认开启思考模式（会产生额外 reasoning tokens），建议显式设置 `enable_thinking`；在 JSONL 中该参数须与 `model` 同级放在 `body` 顶层，不能放入 `extra_body`。 -百炼存储空间上限为 10000 个文件、总计 100 GB，达到任一上限后新上传会失败，需删除文件释放配额。上传返回的文件 ID（如 `file-batch-xxx`）可重复使用。 +### Conversations（会话管理） -百炼提供两种批量推理方式，费用均约为实时调用的 **50%**： +配合 Responses API 实现服务端上下文托管，适用于跨设备/长时间中断的对话延续。提供 Create / Retrieve / Update / Delete conversation 以及 Items 增删查接口。创建会话时 `items` 最多 20 条初始消息，`metadata` 最多 16 对键值对（key ≤ 64 字符，value ≤ 512 字符）。目前支持北京和新加坡地域。详见 [OpenAI Conversations接口兼容](../../raw/model-api-reference/toolkits-and-frameworks/openai-compatible-conversations.md)。 -- **Batch（文件输入）**：上传 jsonl 文件异步批处理，适合大批量、时效性要求不高的场景（数据分析、模型评测）。可先用测试模型 `batch-test-model` 做全链路验证（文件 ≤ 1 MB、≤ 100 行、最大并行 2 个任务，不产生推理费用）。 -- **Batch Chat**：保持与实时 API 一致的同步调用方式，仅需将 `base_url` 改为 `https://batch.dashscope.aliyuncs.com/compatible-mode/v1`，单次仅支持一个请求；默认等待超时 3600 秒（可设 60~3600 秒）。 +## 在 LangChain 中使用百炼 -> **注意**：Batch 场景下 `enable_thinking` 须作为请求 body 的顶层参数（与 `model` 同级）传入，不能放在 `extra_body` 中；`qwen3.7`/`qwen3.6`/`qwen3.5` 系列默认开启思考模式，会产生额外思考 Token 成本，建议显式设置。 +两种接入路径，详见 [在LangChain中使用阿里云百炼](../../raw/model-api-reference/toolkits-and-frameworks/use-bailian-in-langchain.md)： -## 框架集成（LangChain） +| 方式 | 依赖 | 模型覆盖 | +| --- | --- | --- | +| OpenAI 兼容（`ChatOpenAI` / `@langchain/openai` / LangChain4j） | `langchain_openai` 等 | 仅 OpenAI 兼容模式支持的部分模型 | +| DashScope 原生（`ChatTongyi` / `ChatAlibabaTongyi`） | `langchain-community` + `dashscope` | 百炼所有文本生成模型（含部署后的模型） | + +Python 示例（OpenAI 兼容路径）： -百炼可通过两条路径接入 LangChain（Python / JavaScript / Java），详见 [在LangChain中使用阿里云百炼](../../raw/model-api-reference/toolkits-and-frameworks/use-bailian-in-langchain.md)： +```python +from langchain_openai import ChatOpenAI +import os -- **OpenAI 兼容路径**：使用 `langchain_openai.ChatOpenAI`（JS 为 `@langchain/openai`，Java 为 `langchain4j-open-ai`），配置 `base_url` 指向 `compatible-mode/v1`。**仅支持 OpenAI 兼容模式覆盖的部分模型**。 -- **DashScope 原生路径**：使用 `ChatTongyi`（`langchain-community` + `dashscope`）或 JS 的 `ChatAlibabaTongyi`，**支持百炼所有文本生成模型（含部署后的模型）**。 +llm = ChatOpenAI( + api_key=os.getenv("DASHSCOPE_API_KEY"), + base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", + model="qwen-plus", +) +``` -> **注意**：LangChain4j 1.0.0-beta3 需要 Java 17 及以上版本，使用 Java 11 编译会报 `Unsupported class file major version 61` 错误。 +> **注意**：LangChain4j 1.0.0-beta3 需要 Java 17+，低版本编译会报 `Unsupported class file major version 61`。 -## 限制与注意事项 +## 通用注意事项 -- **地域隔离**：API Key 与 `base_url` 均按地域区分，跨地域调用必须成对更换；不同接口/模型在各地域的可用性存在差异，以[百炼控制台](https://bailian.console.aliyun.com/)为准。 -- **协议差异**：并非所有模型都支持 OpenAI 兼容协议（如 Qwen-Audio、多模态 Embedding），此类模型需使用 DashScope 原生协议。 -- **端点区别**：普通请求走各地域 `compatible-mode/v1`，而 Batch Chat 使用独立的 `batch.dashscope.aliyuncs.com` 域名。 -- 调用失败时请参考[错误码](https://help.aliyun.com/zh/model-studio/error-code)排查。 +- [API Key](../concepts/api-key.md) 建议配置到环境变量 `DASHSCOPE_API_KEY`，避免硬编码泄露。 +- 流式调用建议设置 `stream_options={"include_usage": True}` 以获取 token 统计。 +- 错误排查统一参考百炼错误码文档；异常响应通过 `error.code` / `error.message` 指明原因。 +- 各地域支持的模型列表有差异，以百炼控制台模型广场为准。 ## 来源文档 @@ -109,3 +119,4 @@ Qwen-VL、QVQ、Qwen-OCR 兼容 OpenAI Chat 接口，通过 `content` 数组中 + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/vector-and-sort.md b/skills/bailian-docs-llm-wiki/wiki/api/vector-and-sort.md index 2a8da565..79723fbd 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/vector-and-sort.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/vector-and-sort.md @@ -1,90 +1,77 @@ # vector and sort -百炼平台围绕"向量化"与"排序"提供了一整套模型 API，覆盖通用文本向量、多模态向量与文本/多模态重排序三大能力。它们共同服务于语义搜索、推荐、聚类、分类与 RAG 检索：向量模型负责把文本、图片、视频编码为同一语义空间中的数值向量，排序（rerank）模型则在召回阶段之后对候选结果做二次精排，提升最终相关性。 +本页汇总百炼平台"向量化与排序"类模型的 API 参考，覆盖通用文本向量（同步/批处理）、[多模态](../concepts/multimodal.md)向量以及文本排序（rerank）四组接口。所有接口均通过 [API Key](../concepts/api-key.md)（环境变量 `DASHSCOPE_API_KEY`）鉴权，HTTP 端点中的 `{WorkspaceId}` 需替换为真实的[业务空间](../concepts/workspace.md) ID。 -## 能力与模型总览 +## 通用文本向量（Text Embedding） -按用途可分为三类接口，分别对应不同的 endpoint 与调用方式： +### 模型与规格 -- **通用文本向量（同步）**：将字符串 / 字符串列表 / 文件转为向量，实时返回。支持 `qwen3.7-text-embedding`、`text-embedding-v4/v3/v2/v1`。详见 [同步接口API详情](../../raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-synchronous-api.md)。 -- **通用文本向量（批处理）**：面向大规模离线向量化，仅支持异步模式，通过文件 URL 输入。支持 `text-embedding-async-v2/v1`。详见 [批处理接口API详情](../../raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-batch-api.md)。 -- **多模态向量**：将文本、图片、视频编码到同一语义空间，支持跨模态检索与融合表征。支持 `qwen3-vl-embedding`、`qwen2.5-vl-embedding`、`tongyi-embedding-vision-plus/flash`（含 `2026-03-06` 快照版）、`multimodal-embedding-v1`。详见 [Multimodal-Embedding API详情](../../raw/model-api-reference/vector-and-sort/multimodal-vector/multimodal-embedding-api-reference.md)。 -- **文本 / 多模态排序**：对召回文档做精排，支持 `qwen3-rerank`、`qwen3-vl-rerank`（多模态）、`gte-rerank-v2`。详见 [文本排序](../../raw/model-api-reference/vector-and-sort/rerank-model/text-rerank-api.md)。 +同步接口支持以下模型（详见[同步接口API详情](../../raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-synchronous-api.md)）： -> **注意**：`gte-rerank` 模型将于 2026-05-30 下线，官方推荐迁移到 `qwen3-rerank`。新项目请直接选用 `qwen3-rerank` / `qwen3-vl-rerank`。 +| 模型 | 向量维度 | 单行最大 [Token](../concepts/token.md) | 最大行数 | 单价（千 [Token](../concepts/token.md)） | +| --- | --- | --- | --- | --- | +| qwen3.7-text-embedding | 2560/2048/1536/1024（默认）/768/512/256 | 128,000 | 20 | 0.0005 元 | +| text-embedding-v4 | 2048/1536/1024（默认）/768/512/256/128/64 | 8,192 | 10 | 0.0005 元 | +| text-embedding-v3 | 1024（默认）/768/512/256/128/64 | 8,192 | 10 | 0.0005 元 | +| text-embedding-v2 | 1,536 固定 | 2,048 | 25 | 0.0007 元 | +| text-embedding-v1 | 1,536 固定 | 2,048 | 25 | 0.0007 元 | -## 通用文本向量 +### 调用方式 -### 同步接口 - -- **兼容方式**：提供 [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)，可用 OpenAI SDK 直连。 - - base_url：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` - - endpoint：`POST .../compatible-mode/v1/embeddings` - - 调用前需将 `{WorkspaceId}` 替换为真实业务空间 ID。 -- **关键参数**： - - `model`（必选）：模型名称。 - - `input`（必选）：`string` / `array` / `file` 三种形态。 - - `dimensions`（可选）：仅 `text-embedding-v3/v4`（及 `qwen3.7-text-embedding` 的 2560 维）支持自定义维度，取值 2560/2048/1536/1024/768/512/256/128/64，默认 1024。 - - `encoding_format`（可选）：当前仅支持 `float`。 -- **输入上限（按模型区分）**： - - `qwen3.7-text-embedding`：单条字符串最长 128,000 Token；列表/文件最多 20 条。 - - `text-embedding-v3/v4`：单条 8,192 Token；列表/文件最多 10 条。 - - `text-embedding-v1/v2`：单条 2,048 Token；列表/文件最多 25 条。 - -> **注意**：`dimensions` 只对部分模型生效——`text-embedding-v1/v2` 为固定维度（分别 1536 / 1536），传入该参数无意义；`v4` 才支持 2048/1536 等高维度。选维度前请对照模型概览表。 +- **OpenAI 兼容**：`POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/embeddings`，请求体包含 `model`、`input`（字符串/字符串列表/文件）、`dimensions`（仅 v3/v4 及 qwen3.7 支持）、`encoding_format`（仅 `float`）。 +- 输入既支持单条字符串，也支持字符串列表；Python SDK 还支持直接传入文件对象。 ### 批处理接口 -批处理专用于大批量离线场景，特点是**仅支持异步**： +大规模离线向量化使用异步批处理（详见[批处理接口API详情](../../raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-batch-api.md)）： + +- 模型：`text-embedding-async-v1` / `text-embedding-async-v2`，单次请求最多 100,000 行、单行最长 2,048 [Token](../concepts/token.md)、文件不超过 200 MB。 +- HTTP 仅支持异步：创建任务时必须携带请求头 `X-DashScope-Async: enable`（缺失会报 "current user api does not [support](../guides/support.md) synchronous calls"），随后用 `GET .../api/v1/tasks/{task_id}` 轮询结果。 +- 任务数据仅保留 24 小时，需及时下载结果 URL；[限流](../concepts/rate-limit.md)：任务下发 RPS 为 1，并发运行作业最多 3 个，排队+运行总数不超过 50。 +- 参数 `text_type` 区分 `query` / `document`（默认 `document`），非对称检索场景建议显式区分。 -- endpoint：`POST https://dashscope.aliyuncs.com/api/v1/services/embeddings/text-embedding/text-embedding`。 -- HTTP 请求**必须**带请求头 `X-DashScope-Async: enable`，否则报错 `current user api does not support synchronous calls`。 -- 输入通过 `input.url` 传入文件 HTTP URL（一行一条），单行最长 2,048 Token、最多 100,000 行、文件不超过 200MB。 -- `parameters.text_type` 可选 `document`（默认）或 `query`；检索类非对称任务建议区分 query / document。 -- 调用两步走：创建任务拿到 `task_id` → `GET .../api/v1/tasks/{task_id}` 轮询结果。任务状态含 PENDING / RUNNING / SUCCEEDED / FAILED / CANCELED / UNKNOWN。 -- **数据时效**：任务结果 URL 仅保留 24 小时，务必及时下载，详见 [批处理接口API详情](../../raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-batch-api.md)。 -- **限流**：`text-embedding-async-v2` 任务下发 RPS 为 1，排队+运行作业不超过 50 个，同时并发运行不超过 3 个。 +> **注意**：批处理接口文档中"查询结果"部分将 `url` 字段描述为"模型生成图片的URL地址"，系从图像类文档复制而来的笔误，实际为向量结果文件的下载地址。 -## 多模态向量 +## [多模态](../concepts/multimodal.md)向量（Multimodal Embedding） -多模态向量把 text / image / video 编码进**同一语义空间**，可直接用余弦相似度做跨模态匹配（以文搜图、以图搜视频等）。 +将文本、图片、视频映射到同一语义空间，支持跨模态检索、相似度计算与聚类（详见[Multimodal-Embedding API详情](../../raw/model-api-reference/vector-and-sort/multimodal-vector/multimodal-embedding-api-reference.md)）。 -- endpoint：`POST https://dashscope.aliyuncs.com/api/v1/services/embeddings/multimodal-embedding/multimodal-embedding`。 -- 输入通过 `input.contents` 数组传入，每个元素为 `{"模态类型": "值"}`，支持 `text` / `image` / `video` / `multi_images` 四种类型。图片可用 URL 或 Base64 Data URI，视频仅支持公开 URL。 -- **独立向量 vs 融合向量**： - - 独立向量：为每个输入分别生成一个向量，适合逐项对比（以图搜图）。 - - 融合向量：将所有输入融合为 1 个向量，适合整体语义表征（如商品图+描述文本）。 - - `qwen3-vl-embedding` 通过 `enable_fusion=true` 开启融合；`tongyi-embedding-vision-*-2026-03-06` 则通过把 text/image/video 放进同一个 content 对象来生成融合向量（不使用 `enable_fusion`）。 -- **关键参数（在 `parameters` 内）**：`dimension`（不同模型取值不同）、`output_type`（仅 `dense`）、`fps`（视频帧率比例 [0,1]）、`instruct`（任务说明，建议英文）、`res_level`（分辨率档位 0/1/2/3，仅 2026-03-06 版）、`max_video_frames`（最大采样帧，≤64，仅 2026-03-06 版）。 +- 端点：`POST https://dashscope.aliyuncs.com/api/v1/services/embeddings/multimodal-embedding/multimodal-embedding`。 +- 输入：`input.contents` 数组，元素支持 `text`、`image`（URL 或 Base64 Data URI）、`video`（仅 URL）、`multi_images`（仅 tongyi 系列支持）。 +- 向量类型： + - **独立向量**：每个输入各生成一个向量； + - **融合向量**：`qwen3-vl-embedding` 通过 `parameters.enable_fusion=true` 开启；`tongyi-embedding-vision-*-2026-03-06` 通过将 text/image/video 放入**同一个 content 对象**实现，不使用 `enable_fusion`；`qwen2.5-vl-embedding` 仅支持融合向量。 +- 关键参数：`dimension`（各模型可选值不同，`multimodal-embedding-v1` 与旧版 tongyi 系列固定维度）、`fps`（视频抽帧比例）、`instruct`（英文任务说明，约 1%–5% 效果提升）、`res_level` 与 `max_video_frames`（仅 2026-03-06 快照版本）。 +- 请求条数限制因模型而异，例如 `qwen3-vl-embedding` 单次内容元素不超过 20、图片不超过 10 张、视频不超过 1 个。 -> **注意**：各模型的向量类型能力差异明显——`qwen2.5-vl-embedding` **仅**支持融合向量、不支持独立向量与多图；`tongyi-embedding-vision-plus/flash`（非快照版）**仅**支持独立向量；`multimodal-embedding-v1` 与 `tongyi-embedding-vision-plus/flash` 不支持 `dimension` 参数（维度固定）。选型前务必核对 [Multimodal-Embedding API详情](../../raw/model-api-reference/vector-and-sort/multimodal-vector/multimodal-embedding-api-reference.md) 中的"模型能力对照"表。 +## 文本排序（Rerank） -## 文本 / 多模态排序（Rerank） +对召回结果做二次精排，提升 RAG 与搜索的相关性（详见[文本排序](../../raw/model-api-reference/vector-and-sort/rerank-model/text-rerank-api.md)）。 -排序模型对召回文档二次精排，返回相关性分数。不同模型使用不同接口： +### 模型与接口 -- `qwen3-rerank`：`POST .../compatible-api/v1/reranks`，且 `query` / `documents` / `top_n` / `instruct` 与 `model` **同层级**（不使用 `input` / `parameters` 包装）。 -- `qwen3-vl-rerank`（多模态）/ `gte-rerank-v2`：`POST .../api/v1/services/rerank/text-rerank/text-rerank`，参数需包装进 `input` 与 `parameters` 对象。 +| 模型 | 端点 | 特点 | +| --- | --- | --- | +| qwen3-rerank | `POST .../compatible-api/v1/reranks` | 最多 500 文档，单条 4,000 [Token](../concepts/token.md)，请求上限 120,000 [Token](../concepts/token.md)，100+ 语种 | +| qwen3-vl-rerank | `POST .../api/v1/services/rerank/text-rerank/text-rerank` | [多模态](../concepts/multimodal.md)（文本 100 / 图片 40 / 视频 4），query 支持 text 或 image | +| gte-rerank-v2 | 同上 | 50+ 语种 | -关键参数与返回： +> **注意**：gte-rerank 模型将于 **2026 年 5 月 30 日下线**，请迁移到 qwen3-rerank。 -- `query`（必选）：最大 4,000 Token；`qwen3-vl-rerank` 支持 `{"text": ...}` 或 `{"image": ...}` 对象形式。 -- `documents`（必选）：候选文档数组；`qwen3-vl-rerank` 每项可为 `text` / `image` / `video`。 -- `top_n`（可选）：返回前 N 条，默认全部。 -- `return_documents`（可选，默认 `false`）：是否回带原文，仅 `gte-rerank-v2` / `qwen3-vl-rerank` 支持。 -- `instruct`（可选）：仅 `qwen3-rerank` / `qwen3-vl-rerank` 生效，用于切换问答检索 / 语义相似度等排序策略，建议英文。 -- `fps`（可选）：仅 `qwen3-vl-rerank` 支持，控制视频抽帧比例。 -- 返回 `results` 按 `relevance_score`（0.0–1.0）降序排列，`index` 对应输入原始位置。 +### 请求结构差异 -> **注意**：`relevance_score` 是**单次请求内的相对分数**，仅用于本次请求内排序，不可作为跨请求比较的绝对阈值。此外两类接口响应结构不同——`qwen3-rerank` 的 `results` 位于响应顶层且无 `output` 对象，其余模型结果在 `output.results` 内，详见 [文本排序](../../raw/model-api-reference/vector-and-sort/rerank-model/text-rerank-api.md)。 +- **qwen3-rerank** 使用扁平结构：`query`、`documents`、`top_n`、`instruct` 与 `model` 同级，响应中 `results` 位于顶层。 +- **qwen3-vl-rerank / gte-rerank-v2** 使用嵌套结构：`input.query` / `input.documents` + `parameters.top_n` / `return_documents` / `fps`，响应位于 `output.results`。 +- `instruct` 用于声明排序任务类型（如问答检索或语义相似度），建议英文撰写，仅 qwen3 系列生效。 +- `relevance_score` 取值 0–1，是**当前请求内的相对分数**，不可跨请求比较。 +- [DashScope SDK](../concepts/dashscope-sdk.md) 将参数扁平化封装（如 `dashscope.TextReRank.call(model=..., query=..., documents=..., top_n=...)`），与 HTTP 嵌套结构不同，开发时注意区分。 -## 通用限制与注意事项 +## 通用注意事项 -- **前提条件**：所有接口都需先[获取 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)并配置到环境变量 `DASHSCOPE_API_KEY`；SDK 调用还需安装 DashScope SDK。 -- **地域**：同步向量与 rerank 走 `maas.aliyuncs.com`（需替换 `{WorkspaceId}`），多模态与批处理走 `dashscope.aliyuncs.com`；新加坡地域需将 base_url 换为 `dashscope-intl.aliyuncs.com`。 -- **SDK 与 HTTP 差异**：HTTP 使用嵌套的 `input` / `parameters` 结构，DashScope SDK 多为扁平参数，开发时注意区分。 -- **超长截断**：rerank 中单条超过"单条最大输入 Token"会被截断，API 仅基于截断后内容计算，可能影响排序准确性。 -- **限流与错误码**：触发条件参见平台[限流](https://help.aliyun.com/zh/model-studio/rate-limit)文档，失败响应通过 `code` / `message` 指明原因，对照[错误码](https://help.aliyun.com/zh/model-studio/error-code)排查。 +- 调用前需获取并配置 [API Key](../concepts/api-key.md)；SDK 调用需安装 [DashScope SDK](../concepts/dashscope-sdk.md)。 +- 不同地域 URL 不同（北京为 `cn-beijing`，新加坡为 `ap-southeast-1`），调用时替换 `{WorkspaceId}`。 +- 输入超长会被截断，可能影响向量/排序质量，请按模型的 [Token](../concepts/token.md) 上限切分文本。 +- 各模型[限流](../concepts/rate-limit.md)规则参见百炼[限流](../concepts/rate-limit.md)文档；错误码对照参见百炼错误码文档。 ## 来源文档 @@ -94,3 +81,6 @@ - [批处理接口API详情](../../raw/model-api-reference/vector-and-sort/general-text-vector/text-embedding-batch-api.md) + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/api/video-generation-api.md b/skills/bailian-docs-llm-wiki/wiki/api/video-generation-api.md index f4a55920..d0a11466 100644 --- a/skills/bailian-docs-llm-wiki/wiki/api/video-generation-api.md +++ b/skills/bailian-docs-llm-wiki/wiki/api/video-generation-api.md @@ -1,66 +1,120 @@ # video generation api -阿里云百炼平台提供覆盖多家厂商（万相 Wan、爱诗 PixVerse、Vidu、可灵 Kling、HappyHorse 等）的视频生成 API，支持文生视频、图生视频（首帧/首尾帧）、参考生视频、视频编辑、数字人、人像驱动、视频超清与对口型等能力。所有视频生成任务均通过统一的异步调用模式完成，开发者先提交任务拿到 `task_id`，再轮询查询结果。 +百炼平台视频生成 API 覆盖文生视频、图生视频（首帧/首尾帧）、参考生视频、视频编辑、视频风格重绘、视频超清、对口型、动作模仿、数字人/人像动画等任务，统一采用异步任务模式（创建任务 → 轮询结果）。本页汇总各模型家族的能力差异、接入地址、关键参数与注意事项，供开发者选型与集成参考。 -## 统一调用模式：异步「创建任务 → 轮询获取」 +## 支持的模型与功能 -由于视频生成耗时较长（通常 1-5 分钟，个别统一编辑模型约 5-10 分钟），API 全部采用异步方式，流程分两步： +### 万相（Wan）系列 -1. **创建任务**：向 `video-synthesis` 端点发起 `POST` 请求，请求头必须带 `X-DashScope-Async: enable`（缺少会报错 `current user api does not support synchronous calls`），返回一个 `task_id`。 -2. **轮询获取**：用 `task_id` 发起 `GET https:///api/v1/tasks/{task_id}` 查询任务状态，直到完成并拿到视频 URL。 +- **wan2.7（新版协议，推荐）**： + - 图生视频：支持[多模态](../concepts/multimodal.md)输入（文本/图像/音频/视频），可完成首帧生视频、首尾帧生视频、视频续写，详见[万相2.7-图生视频API参考](../../raw/model-api-reference/video-generation-api/wan-api-reference/image-to-video-general-api-reference.md)。 + - 文生视频：支持自然语言控制单/多镜头叙事（通过 [prompt](../guides/prompt.md) 描述分镜，无需 `shot_type`），详见[万相2.7-文生视频API参考](../../raw/model-api-reference/video-generation-api/wan-api-reference/text-to-video-api-reference.md)。 + - 参考生视频（wan2.7-r2v）：[多模态](../concepts/multimodal.md)输入（图片/视频/音频），保持角色形象与音色一致性，支持多主体参考，详见[万相2.7-参考生视频API参考](../../raw/model-api-reference/video-generation-api/wan-api-reference/wan-video-to-video-api-reference.md)。 + - 视频编辑（wan2.7-videoedit）：支持纯指令编辑（如风格转换）与视频迁移，详见[万相2.7-视频编辑API参考](../../raw/model-api-reference/video-generation-api/wan-api-reference/wan-video-editing-api-reference.md)。 +- **旧版模型（2.1–2.6，旧版协议）**：wan2.6/2.5/2.2/wanx2.1 的文生视频、图生视频（首帧）、首尾帧生视频（wan2.2-kf2v）、参考生视频（wan2.6）、视频编辑统一模型（wanx2.1-vace-plus，通过 `function` 字段区分任务，如 `image_reference`）。 +- **人像/动作类**： + - wan2.2-animate-move（图生动作）：将参考视频的动作与表情迁移到人物图片，提供标准模式 `wan-std` 与专业模式 `wan-pro`，详见[万相-图生动作API参考](../../raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-move-api.md)。 + - wan2.2-animate-mix（视频换人）：保留原视频场景/光照/动作，替换主角，详见[万相-视频换人API参考](../../raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-mix-api.md)。 + - wan2.2-s2v（数字人）：单张图片 + 音频生成说话/唱歌/表演视频；须先调用 `wan2.2-s2v-detect` 检测图像合规，详见[万相-数字人](../../raw/model-api-reference/video-generation-api/wan-api-reference/wan-s2v-overview.md)。 -其余通用约定： +### HappyHorse 系列 -- `task_id` 有效期为 **24 小时**，过期无法查询（返回状态 `UNKNOWN`）；请勿重复创建任务，轮询即可。 -- 请求头 `Content-Type: application/json`、`Authorization: Bearer $DASHSCOPE_API_KEY` 为必填。 -- 新手可参考 [Postman 首次调用指引](../../raw/model-api-reference/video-generation-api/wan-api-reference/text-to-video-api-reference.md)。 +支持文生视频（happyhorse-1.1-t2v）、图生视频（首帧）、参考生视频（多张参考图像 + 提示词）、视频编辑（视频 + 参考图 + 文本指令，完成风格变换/局部替换），详见[HappyHorse-文生视频API参考](../../raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-text-to-video-api-reference.md)与[HappyHorse-视频编辑API参考](../../raw/model-api-reference/video-generation-api/happyhorse-api-reference/happyhorse-video-edit-api-reference.md)。 -> **注意**：绝大多数视频生成模型使用端点路径 `/api/v1/services/aigc/video-generation/video-synthesis`，但部分数字人/换人/图生动作类模型（[万相-数字人](../../raw/model-api-reference/video-generation-api/wan-api-reference/wan-s2v-overview.md)、[万相-图生动作](../../raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-move-api.md)、[万相-视频换人](../../raw/model-api-reference/video-generation-api/wan-api-reference/wan-animate-mix-api.md)、[万相2.2-首尾帧](../../raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-by-first-and-last-frame-api-reference.md)）使用的是 `/api/v1/services/aigc/image2video/video-synthesis`。接入时请以对应文档端点为准。 +### 爱诗（PixVerse）系列 -## 支持的模型与功能 +- 文生视频：`pixverse/pixverse-c1-t2v`、`pixverse/pixverse-v6-t2v`、`pixverse/pixverse-v5.6-t2v`，c1 支持 [prompt](../guides/prompt.md) 描述多镜头。 +- 图生视频（首帧）：`pixverse-c1-it2v`、`pixverse-v6-it2v`、`pixverse-v5.6-it2v`。 +- 首尾帧生视频：`pixverse-c1-kf2v`、`pixverse-v6-kf2v`、`pixverse-v5.6-kf2v`，通过 `media` 中 `first_frame` / `last_frame` 传入。 +- 参考生视频：`pixverse-c1-r2v`。 +- 视频增强类：`pixverse-upscale`（固定输出 4K 超清）、`pixverse-lipsync`（音频或 TTS 文本驱动对口型）、`pixverse-motioncontrol`（动作模仿）。 +- 调用前需在控制台搜索 PixVerse 并开通，详见[爱诗-文生视频API参考](../../raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-text-to-video-api-reference.md)。 -按厂商与任务类型划分，主要能力如下： - -- **万相 Wan（2.7 新版协议）**： - - 文生视频（`wan2.7-t2v-*`），支持通过 `prompt` 自然语言控制单/多镜头。 - - 图生视频（`wan2.7-i2v-*`），支持多模态输入（文本/图像/音频/视频），可完成首帧生视频、首尾帧生视频、视频续写三大任务。 - - 参考生视频（`wan2.7-r2v-*`），多主体参考（图像+视频+音色）。 - - 视频编辑（`wan2.7-videoedit`），指令编辑与视频迁移。 -- **万相 Wan（旧版协议，2.1-2.6）**：文生视频、图生视频-基于首帧、参考生视频（`wan2.6-r2v-flash`）、首尾帧生视频（`wan2.2-kf2v-flash`）、视频编辑统一模型（`wanx2.1-vace-plus`，支持多图参考、视频重绘等 `function`）。 -- **万相人物/数字人系列**：数字人 `wan2.2-s2v`（图片+音频，需先用 `wan2.2-s2v-detect` 检测图片）、图生动作 `wan2.2-animate-move`、视频换人 `wan2.2-animate-mix`（均含 `wan-std`/`wan-pro` 两种模式）。 -- **爱诗 PixVerse**：文生视频、图生视频、首尾帧生视频（`pixverse/pixverse-c1-*`、`-v6-*`、`-v5.6-*`）、参考生视频（`-r2v`）、视频超清（`pixverse/pixverse-upscale`，固定输出 4K）、视频对口型（`pixverse/pixverse-lipsync`，支持音频驱动或 TTS 文本）、视频动作模仿（`pixverse/pixverse-motioncontrol`）。 -- **Vidu**：文生视频、图生视频、首尾帧生视频、参考生视频（`vidu/viduq3-*`、`viduq2_*`）。 -- **可灵 Kling**：一个模型（`kling/kling-v3-video-generation`、`kling/kling-v3-omni-video-generation`）统一支持文生视频、图生视频（首帧/首尾帧）、参考生视频、视频编辑，并支持智能分镜/多镜头（`multi_shot`、`shot_type`、`multi_prompt`）。 -- **HappyHorse**：文生视频、图生视频-基于首帧、参考生视频（多图像）、视频编辑。 -- **人像驱动系列（两步调用：先检测后生成）**：舞动人像 AnimateAnyone（图生舞蹈）、悦动人像 EMO（图生唱演，`style_level` 控制风格强度）、灵动人像 LivePortrait（图生播报）、表情包 Emoji（预设模板 `driven_id`）、声动人像 VideoRetalk（口型替换）、视频风格重绘 `video-style-transform`（8 种预设风格）。 +### Vidu 系列 -## 关键参数 +- 文生视频：`vidu/viduq3-turbo_text2video` 等。 +- 图生视频（首帧）：`vidu/viduq3-pro-fast_img2video` 等。 +- 首尾帧生视频：`vidu/viduq3-turbo_start-end2video`。 +- 参考生视频：`vidu/viduq3-ad_reference2video`（广告场景）等。 +- 调用前需在控制台搜索 Vidu 并开通，详见[Vidu-文生视频API参考](../../raw/model-api-reference/video-generation-api/vidu-api-reference/vidu-text-to-video-api-reference.md)。 + +### 可灵（Kling） + +支持文生视频、图生视频（首帧/首尾帧）、参考生视频、视频编辑，模型如 `kling/kling-v3-video-generation`、`kling/kling-v3-omni-video-generation`（支持智能分镜），参数含 `mode`（std）、`aspect_ratio`、`duration`、`audio`，详见[可灵-视频生成API文档](../../raw/model-api-reference/video-generation-api/kling-api-reference/kling-video-generation-api-reference.md)。 + +### 人像动画与风格类 + +- 舞动人像 AnimateAnyone：detect → template → 生成三步流程，支持后付费调用与独立部署，详见[图生舞蹈视频-舞动人像AnimateAnyone](../../raw/model-api-reference/video-generation-api/portrait-animation-api-reference/animateanyone-quick-start.md)。 +- 悦动人像 EMO：肖像 + 人声音频生成唱演视频，`style_level` 可选 active/normal/calm。 +- 灵动人像 LivePortrait：肖像 + 音频快速生成播报视频。 +- 声动人像 VideoRetalk：视频 + 音频替换口型。 +- 表情包 Emoji：肖像 + 预设模板（driven_id）生成表情视频。 +- 视频风格重绘（video-style-transform）：8 种预设风格（日式漫画、3D 卡通、国风水墨等），参数 `style`（0–7）与 `video_fps`，详见[视频风格重绘API参考](../../raw/model-api-reference/video-generation-api/portrait-animation-api-reference/video-style-transform-api-reference.md)。 + +## 调用方式 + +所有视频生成/编辑任务耗时较长（通常 1–5 分钟，wanx2.1-vace 约 5–10 分钟），均采用异步两步流程： + +1. **创建任务**：`POST` 提交请求，必须携带请求头 `X-DashScope-Async: enable`（缺失会报 "current user api does not [support](../guides/support.md) synchronous calls"）与 `Authorization: Bearer $DASHSCOPE_API_KEY`，返回 `task_id`（有效期 24 小时）。请勿重复创建任务。 +2. **轮询结果**：`GET /api/v1/tasks/{task_id}` 查询状态，成功后获取视频 URL。 -请求体主要由 `model`、`input`、`parameters` 三部分构成： +**Endpoint 路径**（按模型而异）： -- `model`：模型名称，决定能力与协议版本。 -- `input`：任务输入。文生类用 `prompt`；图生/参考/编辑类多用 `media` 数组（`type` 可为 `image_url`/`first_frame`/`last_frame`/`video_url`/`audio_url`/`reference_image` 等），部分旧版模型用 `image_url`/`video_url`/`audio_url`/`first_frame_url`/`last_frame_url`/`ref_images_url` 等独立字段。 -- `parameters`：常见有 `resolution`（如 `480P`/`540P`/`720P`/`1080P`）、`size`（如 `1280*720`）、`duration`（秒）、`watermark`、`prompt_extend`（智能改写）、`audio`、`shot_type`/`multi_shot`（分镜）、`seed`、`style`/`style_level` 等，具体取值随模型不同。 +- 通用路径 `/api/v1/services/aigc/video-generation/video-synthesis`：wan2.7、HappyHorse、PixVerse、Vidu、可灵、视频风格重绘等。 +- 旧路径 `/api/v1/services/aigc/image2video/video-synthesis`：wan2.2-animate-move/mix、wan2.2-s2v、wan2.2 首尾帧（wan2.2-kf2v）等，详见[万相-首尾帧生视频API参考（2.2）](../../raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-image-to-video-by-first-and-last-frame-api-reference.md)。 -> **注意**：多镜头控制方式在不同模型间不一致。万相 2.7 与 PixVerse-c1 通过 `prompt` 自然语言描述控制，设置 `shot_type` 不生效；而旧版万相 2.6（见 [万相-文生视频（2.1-2.6）](../../raw/model-api-reference/video-generation-api/wan-api-reference/legacy-video-models/legacy-wan-text-to-video-api-reference.md)）需显式设置 `shot_type: "multi"` 且 `prompt_extend: true` 才能启用多镜头。接入前务必确认所用模型的具体协议。 +**地域与域名**：模型、Endpoint URL、[API Key](../concepts/api-key.md) 必须属于同一地域，跨地域调用会失败。 -## 地域与域名 +- 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` +- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` +- 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com`（HappyHorse 与旧版万相支持） +- 德国（法兰克福）：`https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com` +- 日本（东京）：`https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com`（HappyHorse 支持） -- **必须保证模型、Endpoint URL 与 API Key 属于同一地域**，跨地域调用会失败（鉴权失败或服务报错）。 -- 多数第三方模型（PixVerse、Vidu、Kling、数字人等）**仅支持华北2（北京）地域**；万相与 HappyHorse 部分能力还支持新加坡、美国（弗吉尼亚）、德国（法兰克福）等地域。 -- 百炼为华北2（北京）、新加坡地域推出了业务空间专属域名，提供更高性能与稳定性，建议迁移： - - 华北2（北京）：`https://dashscope.aliyuncs.com` → `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` - - 新加坡：`https://dashscope-intl.aliyuncs.com` → `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` - - `{WorkspaceId}` 为业务空间 ID，可在控制台「业务空间详情」查看；现有域名仍可正常使用。 +其中 `{WorkspaceId}` 为[业务空间](../concepts/workspace.md) ID。旧域名 `dashscope.aliyuncs.com` / `dashscope-intl.aliyuncs.com` 仍可用，但官方建议迁移至[业务空间](../concepts/workspace.md)专属域名以获得更好的性能与稳定性。 + +**请求体结构**（新版协议示例）： + +```json +{ + "model": "wan2.7-t2v-2026-06-12", + "input": { + "prompt": "...", + "media": [{ "type": "image_url", "url": "..." }] + }, + "parameters": { + "resolution": "720P", + "duration": 5, + "watermark": true + } +} +``` + +## 关键参数 + +| 参数 | 说明 | +| --- | --- | +| `model` | 模型名称，决定协议版本与能力（如 `wan2.7-*` 为新版协议，`wanx2.1-*` 为旧版） | +| `input.prompt` | 文本提示词；wan2.7 与 pixverse-c1 可在 [prompt](../guides/prompt.md) 中用自然语言/时间戳描述多镜头 | +| `input.media` | [多模态](../concepts/multimodal.md)素材数组，`type` 取值如 `image_url`、`image`、`first_frame`、`last_frame`、`video`、`video_url`、`audio_url`、`reference_image` | +| `parameters.resolution` / `size` | 分辨率（如 `720P`、`1080P`）或尺寸（如 `1280*720`），不同模型支持的档位不同 | +| `parameters.duration` | 视频时长（秒），常见 5 | +| `parameters.watermark` | 是否添加水印 | +| `parameters.audio` | 是否生成音频（wan2.7、可灵等支持） | +| `parameters.seed` | 随机种子，用于结果复现 | +| `parameters.shot_type` | 仅旧版 wan2.6 系列支持（`multi` 多镜头）；wan2.7 与 pixverse-c1 通过 [prompt](../guides/prompt.md) 控制，设置该参数不生效 | ## 限制与注意事项 -- **版本选型**：万相已推出 2.7 新版协议，[万相2.7-图生视频](../../raw/model-api-reference/video-generation-api/wan-api-reference/image-to-video-general-api-reference.md) 支持首帧/首尾帧/续写三大任务，官方推荐优先选用；旧版 wan2.6 及更早模型仅支持首帧生视频。新旧协议接口不通用，`wan2.7-*` 只走新版协议。 -- **服务开通**：PixVerse、Vidu、Kling 等第三方模型需先在百炼控制台模型广场搜索并「立即开通」授权后方可调用。 -- **两步式模型**：数字人、AnimateAnyone、EMO、LivePortrait、Emoji 等需先调用对应的 `-detect` 检测模型确认图片合规（如清晰度、单人、正面），再调用生成模型。检测模型多为同步调用（如 `wan2.2-s2v-detect` 0.004 元/张）。 -- **限流**：视频生成模型通常「同时处理中任务数量」限制较低（多为 1，即同一时刻仅 1 个作业运行，其余排队），任务下发接口 RPS/QPS 约为 5，接入时需做好排队与重试。 -- **计费**：多按生成视频时长计费（如 LivePortrait 0.02 元/秒、EMO/VideoRetalk/AnimateAnyone 0.08 元/秒、数字人 720P 0.9 元/秒），`wan-pro` 等专业模式价格高于标准模式。 -- VideoRetalk 目前仅支持 API 调用，不支持控制台在线体验。 +> **注意**：wan2.7 系列使用**新版协议**（`media` 数组传参），旧版 wan2.6/2.5/2.2/wanx2.1 使用**旧版协议**（如 `img_url`、`reference_urls`、`function` 等字段），两套协议不兼容；选型时请优先使用 wan2.7 新版接口（参见[万相2.7-图生视频API参考](../../raw/model-api-reference/video-generation-api/wan-api-reference/image-to-video-general-api-reference.md)中的迁移说明）。 + +- **地域限制**：PixVerse、Vidu、可灵、人像动画系列（EMO/LivePortrait/VideoRetalk/Emoji/AnimateAnyone）、视频风格重绘、wan2.2-s2v 等**仅支持华北2（北京）地域**；万相 2.7 支持北京与新加坡；HappyHorse 与旧版万相支持更多海外地域。 +- **服务开通**：PixVerse、Vidu、可灵为第三方模型，需先在模型广场搜索并单击"立即开通"。 +- **task_id 有效期 24 小时**，过期查询返回 `UNKNOWN`。 +- **[限流](../concepts/rate-limit.md)**：多数视频模型任务下发 RPS/QPS 限制为 5，同时处理中任务数为 1（排队机制）；检测类接口（detect）为同步接口、无并发限制。 +- **人像类模型需先检测**：AnimateAnyone、EMO、LivePortrait、Emoji、wan2.2-s2v 均须先调用对应 `-detect` 模型确认图像合规，再提交生成任务。 +- **计费**：按生成视频时长计费（如 0.02–0.16 元/秒不等）或按张计费（detect 0.004 元/张）；部分模型提供免费额度，详见各文档的资费表。 +- **输出说明**：AnimateAnyone 生成的视频不含音频；视频超清固定输出 4K（3840×2160）。 ## 来源文档 @@ -100,3 +154,6 @@ - [爱诗-视频动作模仿API参考](../../raw/model-api-reference/video-generation-api/pixverse-api-reference/pixverse-motioncontrol-api-reference.md) + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/agent-extensibility.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/agent-extensibility.md new file mode 100644 index 00000000..6c011a5c --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/agent-extensibility.md @@ -0,0 +1,70 @@ +# 插件、MCP 与工具框架对比 + +阿里云百炼为大模型应用提供了三条主流"扩能"路径：**插件（Plug-in）**、**模型上下文协议（MCP）** 与 **工具框架（Toolkits & Frameworks，[OpenAI 兼容接口](../concepts/openai-compatible-interface.md)体系）**。三者定位不同——插件偏向平台预置能力的即插即用，MCP 面向跨系统标准化工具接入，工具框架则聚焦于让既有 OpenAI/LangChain 生态无缝迁移到百炼。本页从形态、集成方式、模型支持、调用协议、计费与典型场景等维度做一次横向对比，帮助开发者在技术选型阶段快速定位合适方案。 + +## 对比背景 + +- **插件**：百炼原生的工具集合概念，一个插件下挂多个工具（API），可通过智能体应用、工作流应用或 Assistant API 触发。分为官方插件、三方插件、自定义插件三类。 +- **MCP**：Anthropic 主导的开源协议，百炼提供官方托管与自定义部署两种服务形态，重点解决"一次接入、多处复用"的工具标准化问题。 +- **工具框架 / [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)**：百炼针对通义千问等模型提供的一整套与 OpenAI 高度兼容的接口（Chat Completions、Responses、Completions、Embedding、Files、Batch、Conversations 等），并可直接对接 LangChain / LangChain4j 等社区框架。 + +## 关键维度对比 + +| 维度 | 插件（Plug-in） | 模型上下文协议（MCP） | 工具框架（Toolkits & Frameworks） | +| --- | --- | --- | --- | +| 本质 | 平台内工具集合，扩展模型能力边界 | 大模型与外部工具间的开源标准协议 | OpenAI 兼容 API 与主流框架适配层 | +| 提供形态 | 官方插件 / 三方插件 / 自定义插件 | 官方 MCP 服务 / 自定义 MCP 服务（脚本部署 / AI 网关 / 阿里云 OpenAPI） | Chat Completions、Responses、Completions、Embedding、Files、Batch、Conversations 等接口 | +| 集成入口 | 智能体应用、工作流应用、Assistant API | 智能体应用、工作流应用；也支持外部第三方客户端（Cherry Studio、Cursor 等）与个人项目 SDK | OpenAI SDK / HTTP，直接调用模型；也可被 LangChain 等框架封装 | +| 与模型的交互协议 | 平台内部触发，由模型基于名称与描述判断是否调用工具 | 遵循 MCP 协议（Streamable HTTP，旧版 SSE 已升级） | OpenAI 兼容协议（REST / SSE 流式） | +| 支持模型 | qwen-turbo / qwen-plus / qwen-max / qwen-vl-max / qwen-vl-plus（以控制台实测为准） | 智能体或工作流中使用的模型（推荐千问 3 系列以获得更稳定的工具调用） | Qwen 商业版与开源版、Qwen-VL、Qwen-Coder、Qwen-Omni、Qwen-Math，以及 DeepSeek、Kimi、GLM、MiniMax 等三方模型 | +| 每个应用可挂载数量 | 智能体应用最多 **10 个工具** | 单个智能体最多 **5 个 MCP 服务**；工作流单节点仅支持 1 个工具 | 无平台侧数量限制（受模型上下文与开发者代码控制） | +| 典型 API 端点 | 通过 Assistant API 传 `tool_id`（如 `calculator`） | 云端：`https://dashscope.aliyuncs.com/api/v1/mcps//mcp`；`type` 必须与端点匹配（`sse` → GET `/sse`，`streamableHttp` → POST `/mcp`） | 各地域 `/compatible-mode/v1` 前缀 + 资源路径（`/chat/completions`、`/responses`、`/embeddings`、`/files` 等） | +| 鉴权方式 | 平台内部授权 + 服务关联角色（`AliyunServiceRoleForSFMAccessCloudAPI`）；自定义插件支持 `basic` / `bearer` / `appcode` | `Authorization: Bearer `；敏感信息可用 KMS 凭据加密 | `api_key` 使用[百炼 API Key](https://help.aliyun.com/zh/model-studio/get-api-key)（`DASHSCOPE_API_KEY`），各地域独立 | +| 是否可直连千问 API | 通过 Assistant API 可直接调用 | **不支持**直连千问 API，必须挂在智能体或工作流内 | 直接调用模型，无需应用层封装 | +| 本地资源访问 | 由自定义插件的后端实现决定 | 托管在函数计算 FC，**不能访问用户本地资源**；如需访问建议在本地部署 MCP Server | 由开发者的应用代码自行处理 | +| 上下文与多轮管理 | 由智能体/工作流内部拼接 | 工具返回内容作为上下文注入模型，会增加 Token 消耗 | Responses + Conversations 可通过 `previous_response_id`（有效期 7 天）自动管理多轮 | +| 计费方式 | 官方插件：多数免费，部分限时免费需申请；三方/自定义按第三方或业务实际计费 | 云部署：限时免部署费，联网搜索超 2000 次后 29 元/千次；自定义部署：基础模式 0.000156 元/秒；极速模式 0.000036 元/秒（部署）+ 0.000156 元/秒（调用） | 按所调用模型的推理 Token 计费；Batch、文件、Embedding 各接口按对应资源计费 | +| 生态兼容性 | 百炼平台专有 | 遵循 MCP 开源标准，可跨支持 MCP 的客户端/项目复用 | 与 OpenAI SDK、LangChain 等生态高度兼容，迁移成本低（改 `api_key`/`base_url`/`model` 三项） | +| 常见错误码 | RAM 授权 140052 等 | `11200044`~`11200060` 系列（连接、超时、鉴权、协议错配 405/404） | 参考 [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)的错误规范 | + +## 适用场景建议 + +- **优先选择"插件"的场景** + - 需要"开箱即用"的通用能力：Python 代码解释器、计算器、图片生成、夸克搜索、二维码、GitHub 搜索等。 + - 希望在百炼智能体或工作流中以最低成本让模型具备工具调用能力，且工具形态是较为传统的 REST API。 + - 团队的能力扩展只服务于百炼平台，不需要跨系统标准化。 + +- **优先选择"MCP"的场景** + - 已有多个 Agent 客户端（Cursor、Cherry Studio、自研项目）需要**复用同一套工具**，希望走标准协议避免重复接入。 + - 存在需要通过阿里云 OpenAPI 操作 OSS/ECS 等云资源、或将存量 RESTful API 通过 AI 网关升级为标准工具的诉求。 + - 需要在同一个智能体中组合多个高质量官方服务（如 Amap Maps + QuickChart + WebSearch）完成路径规划、数据可视化、联网检索等复合任务。 + - 对调用频次与冷启动敏感时，选择极速模式；对偶发调用敏感时，选择基础模式。 + +- **优先选择"工具框架 / OpenAI 兼容接口"的场景** + - 已有基于 OpenAI SDK 或 LangChain / LangChain4j 的成熟应用，希望以最小代价迁移到百炼。 + - 需要直接对模型进行编排、控制多轮上下文、批量推理、文档抽取、代码补全（如 `qwen-coder-turbo`）、文本或多模态 Embedding 等能力。 + - 需要在 Chat Completions 之上使用 **Responses API** 的内置工具（联网搜索、网页抓取、代码解释器、文搜图/图搜图）与 `previous_response_id` 免拼接多轮机制。 + - 关注多地域部署（北京、新加坡、东京、法兰克福、弗吉尼亚），需要业务空间专属域名（`{WorkspaceId}..maas.aliyuncs.com`）。 + +## 组合使用建议 + +三者并非互斥关系，实际落地时经常组合出现： + +1. **对话入口 = 智能体应用（挂载插件 + MCP）**：让模型可以按对话上下文自动路由至代码解释、地图、联网搜索等能力。 +2. **业务后端 = OpenAI 兼容接口**：由业务侧直接调用 `chat/completions`、`responses` 或 `embeddings` 完成结构化生成、向量检索、代码补全等确定性任务。 +3. **跨平台复用 = MCP**：将高价值内部工具沉淀为 MCP 服务，同时被百炼智能体、Cursor 等第三方客户端与自研 Agent 共享。 + +## 技术选型速查 + +- 只想"给通义千问加几个能力"、又不想写后端 → **插件**（尤其是官方插件）。 +- 已经/即将在多个 Agent 客户端之间复用工具，或需要操作阿里云资源 → **MCP**。 +- 已有 OpenAI/LangChain 代码或需要 Batch、Embedding、Responses 等原生接口 → **工具框架 / OpenAI 兼容接口**。 +- 复杂业务：三者组合，用工具框架承担确定性接口调用，用 MCP 承担跨端标准工具，用插件补齐平台内即用能力。 + +## 被对比主题页 + +- [plug in](../guides/plug-in.md) +- [model context protocol](../guides/model-context-protocol.md) +- [toolkits and frameworks](../api/toolkits-and-frameworks.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/agent-extension-options.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/agent-extension-options.md new file mode 100644 index 00000000..3c6ac682 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/agent-extension-options.md @@ -0,0 +1,76 @@ +# 智能体、插件与技能扩展方式对比 + +百炼平台提供多种扩展大模型能力的方式：Managed Agents 侧重托管运行时与长时任务执行，插件侧重把外部 API 或平台预置工具接入应用，Skill 侧重把特定任务能力封装成可被智能体自动触发的能力包。三者都能增强智能体能力，但在运行环境、开发复杂度、适用对象和交付形态上差异明显。本文面向开发者，对三种方案进行技术选型对比，帮助在构建百炼应用时选择合适的扩展路径。 + +## 关键维度对比 + +| 维度 | Managed Agents | 插件 | Skill | +| --- | --- | --- | --- | +| 核心定位 | 服务端托管的智能体运行时，支持多步工具调用、代码执行、文件处理与长时任务 | 工具集合，一个插件可包含多个工具或 API，用于补足模型外部能力 | 智能体应用的能力包，用于自动识别并处理特定类型任务 | +| 扩展粒度 | 智能体、运行环境、会话、事件组成完整运行链路 | 插件下的单个工具或多个 API | 一个可复用的 Skill 包，官方 Skill 或自定义 ZIP 包 | +| 输入格式 | 用户消息、会话事件、挂载资源、上传文件、系统提示词等 | 用户输入由模型识别，或通过业务透传参数传入工具；自定义插件可配置请求参数 | 用户对话中的任务意图与文件、数据等输入；自定义 Skill 通过 SKILL.md 描述触发条件 | +| 输出格式 | 会话级 SSE 事件流，包含用户消息、工具调用、工具输出、状态变更和最终结果 | 工具调用结果回传给应用，再由模型整合生成最终回复；工作流中可作为节点输出 | 智能体调用 Skill 后生成任务结果，例如处理后的文件、表格或分析结果 | +| 执行环境 | 百炼托管的独立云端容器沙箱，可执行命令、读写文件、安装依赖 | 由插件对应工具或外部 API 执行；官方插件由平台提供能力 | 由平台按 Skill 能力执行；自定义 Skill 以 ZIP 技能包形式上传并审查 | +| 状态管理 | 服务端维护会话状态，支持事件历史持久化、中断与续接 | 通常随应用调用链路执行，更多依赖应用或工作流上下文 | 随智能体应用触发，面向具体任务处理，不强调独立会话状态机 | +| 支持工具能力 | bash、read、write、edit、glob、grep、download_file、MCP 服务和 Skill | 官方插件、三方插件、自定义插件；官方工具包括计算器、Python 代码解释器、搜索、图片生成、二维码等 | 官方 Skill 和自定义 Skill；适合文件处理、数据分析、格式转换等特定任务 | +| 支持模型 | 创建智能体时选择模型，示例包括 qwen3-max；实际以控制台可选模型为准 | 文档列出 qwen-turbo、qwen-plus、qwen-max、qwen-vl-max、qwen-vl-plus 等 | 添加到智能体应用后由应用中的模型根据 Skill 描述判断是否触发 | +| API 或配置入口 | `POST /api/v1/agentstudio/agents`、`POST /api/v1/agentstudio/environments`、`POST /api/v1/agentstudio/sessions`、`POST /sessions/{id}/events`、`GET /sessions/{id}/events/stream` | 控制台插件页面、应用管理页面、Assistant API；自定义插件需配置插件 URL、工具路径、请求方法、参数等 | 控制台“组件 > Skill 管理”；自定义 Skill 上传 ZIP，根目录必须包含 SKILL.md | +| 开发复杂度 | 较高，需要理解 Agent、Environment、Session、Event、沙箱和事件流 | 中等到较高，官方插件接入简单，自定义插件需要设计 API、参数、鉴权和工具描述 | 官方 Skill 接入低，自定义 Skill 主要关注 SKILL.md 描述和 ZIP 包规范 | +| 计费方式 | 文档切片未给出独立计费规则，通常需结合模型调用、平台服务和具体能力确认 | 官方插件存在免费、限时免费或需申请开通等差异；具体以插件详情和计费说明为准 | 文档切片未列出独立计费规则，需以控制台和产品计费说明为准 | +| 典型场景 | 长时多步骤任务、代码执行、文件批处理、需要沙箱和可续接事件历史的自动化任务 | 实时搜索、精确计算、调用企业 API、图像处理、二维码生成、工作流节点集成 | 表格处理、文件解析、数据清洗、格式转换、行业专用任务封装 | +| 主要限制 | 上传文件不超过 10 MB；挂载路径遵循 `/mnt/session/uploads`；需关注沙箱隔离和会话状态 | 每个智能体应用最多添加 10 个工具；部分插件不支持访问详情页或外网；自定义工具参数有格式限制 | ZIP 包不超过 10 MB；自定义 Skill 名称需唯一；上传后需审查；description 质量影响触发准确率 | + +## 方案解读 + +### Managed Agents：适合“让智能体持续执行任务” + +Managed Agents 更像一个完整的托管执行平台，而不仅是给模型增加一个工具。开发者需要先配置智能体，再配置运行环境，随后创建会话并通过事件流观察执行过程。它的优势在于服务端托管状态、独立沙箱、事件历史持久化，以及对文件操作、命令执行、MCP 服务和 Skill 的组合支持。 + +当任务需要多轮规划、执行命令、读写项目文件、处理上传资源，或可能运行较长时间并需要中断续接时，应优先考虑 Managed Agents。它适合构建“云端自动化执行者”，例如代码分析、批量文件处理、复杂数据加工、带工具审批的长任务执行等。 + +### 插件：适合“把明确工具或 API 接给模型” + +插件的核心是工具调用。官方插件和三方插件可以快速补足模型能力，例如计算、搜索、生成二维码、图片生成等；自定义插件则适合把企业已有 API 封装成模型可调用的工具。插件既可以挂载到智能体应用，也可以在工作流中作为节点执行，还可通过 Assistant API 调用。 + +插件选型的关键是工具边界是否清晰。如果某项能力可以抽象为一次或少数几次 API 调用，且输入输出参数明确，插件通常比 Managed Agents 更轻量。对于自定义插件，工具名称、工具描述、参数配置和调用示例会直接影响模型是否能正确选择工具和填充参数。 + +### Skill：适合“把一类任务封装成可自动触发的能力包” + +Skill 面向智能体应用中的特定任务能力扩展。与插件强调 API 和工具参数不同，Skill 更强调通过描述让智能体判断“什么时候该调用这项能力”。官方 Skill 适合直接覆盖通用场景，自定义 Skill 则通过上传包含 SKILL.md 的 ZIP 包实现扩展。 + +如果开发者希望让智能体在用户提到某类文件、某类数据处理需求或某类固定任务时自动启用专门能力，Skill 是更自然的选择。Skill 的 description 是核心配置，应明确适用输入类型、支持操作、触发关键词和不适用场景，以降低误触发和漏触发。 + +## 技术选型建议 + +| 选型问题 | 推荐方案 | 原因 | +| --- | --- | --- | +| 任务是否需要独立沙箱、文件系统和命令执行？ | Managed Agents | 提供云端容器、文件读写、命令执行和资源挂载能力 | +| 是否需要服务端保存会话状态、事件历史和支持续接？ | Managed Agents | 会话级事件流和状态机更适合长时任务 | +| 能力是否可以表达为一个明确 API 或工具调用？ | 插件 | 插件对 API 封装、参数配置和模型工具选择最直接 | +| 是否已有企业内部 API，希望让大模型调用？ | 自定义插件 | 可配置插件 URL、鉴权、工具路径、请求方法和参数 | +| 是否只是想快速增加计算、搜索、二维码等通用能力？ | 官方插件 | 平台预置，接入成本低 | +| 是否需要在工作流中按固定编排执行某个能力？ | 插件 | 插件可作为工作流节点按流程执行 | +| 是否围绕文件处理、表格处理、数据清洗等一类任务自动触发？ | Skill | Skill 通过描述匹配任务类型，适合能力包式封装 | +| 是否希望平台维护并自动更新能力？ | 官方 Skill 或官方插件 | 官方能力由平台维护，接入简单 | +| 是否需要组合多种工具完成端到端任务？ | Managed Agents，可结合插件和 Skill | Managed Agents 可以承载更复杂的多步执行链路 | + +## 面向开发者的落地路径 + +1. **先判断任务形态**：如果是一次 API 调用，优先插件；如果是一类文件或数据任务，优先 Skill；如果是多步执行和长时运行，优先 Managed Agents。 +2. **再判断运行环境要求**：需要 shell、文件系统、依赖安装或沙箱隔离时，不应只用插件或 Skill，应考虑 Managed Agents。 +3. **明确触发与参数边界**：插件要写好工具名称、工具描述、参数和调用示例；Skill 要写好 SKILL.md 的 description；Managed Agents 要写好系统提示词和工具授权范围。 +4. **控制能力数量和误调用风险**：插件有工具数量限制，Skill 依赖描述触发，能力越多越需要清晰边界和不适用场景。 +5. **从低成本方案开始验证**：能用官方插件或官方 Skill 覆盖的场景先不要自定义；确认需要专属 API 或专属任务逻辑后，再建设自定义插件或自定义 Skill。 +6. **复杂场景可组合使用**：Managed Agents 可以作为上层执行运行时，在智能体配置中组合工具、MCP 服务和 Skill；插件和 Skill 则作为其中的能力模块。 + +## 总结 + +Managed Agents、插件和 Skill 并不是互斥关系，而是位于不同抽象层的扩展方式。插件适合封装明确工具，Skill 适合封装可自动触发的任务能力，Managed Agents 适合承载需要状态、沙箱和多步执行的复杂智能体任务。开发者选型时应优先从任务复杂度、输入输出边界、运行环境要求、状态管理需求和维护成本五个方面判断，再决定采用单一方案还是组合方案。 + +## 被对比主题页 + +- [managed agents](../guides/managed-agents.md) +- [plug in](../guides/plug-in.md) +- [skill](../guides/skill.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/app-eval-vs-monitoring.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/app-eval-vs-monitoring.md new file mode 100644 index 00000000..528de2b0 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/app-eval-vs-monitoring.md @@ -0,0 +1,79 @@ +# 应用评测与应用监控对比 + +百炼平台为应用质量保障提供了两条互补路径：**应用评测**（Evaluation）和**应用观测**（Monitoring/Tracing）。前者以"评测集 + 评估器"为核心，对应用输出进行系统化质量评分与归因分析；后者以"调用链路 + 性能指标"为核心，对线上运行状态进行实时追踪与监控。两者共享标签体系，且观测数据可直接导入评测集，形成"监控发现问题 → 评测验证改进"的闭环。本文从开发者技术选型角度对比两者的关键维度。 + +## 关键维度对比 + +| 维度 | 应用评测 | 应用观测（监控） | +|------|----------|------------------| +| 核心目标 | 评估应用输出质量，发现 BadCase 并给出调优建议 | 追踪调用链路，监控延时、[Token](../concepts/token.md) 消耗等运行指标 | +| 支持的应用类型 | 智能体应用、工作流应用（需已发布） | 智能体应用、工作流应用、高代码应用 | +| 数据来源 | 评测集（自动生成 / 手动上传 / 从观测导入） | 线上真实调用数据（自动采集，分钟级同步） | +| 评估方式 | LLM 评估器（语义评分）、Code 评估器（规则判断）、人工标注 | 无评分机制；提供指标统计与链路详情查看 | +| 核心产出 | 评测报告（正确率、BadCase 分析、调优建议） | 调用记录（Prompt/输出/延时/[Token](../concepts/token.md)）、监控统计图表、Trace 链路 | +| 数据保留时长 | 评测任务数据长期保留 | 调用记录最长 30 天 | +| 指标更新频率 | 评测任务完成后一次性生成 | 分钟级持续更新 | +| 操作方式 | 控制台操作 | 仅控制台操作，无 API | +| 计费 | LLM 评估器产生 [Token](../concepts/token.md) 费用（评估模型限 qwen-max / qwen-plus）；Code 评估器免费 | 功能本身免费；观测数据存储由 OpenTelemetry 服务收费 | +| 前提条件 | 应用已发布且关联知识库；须先开通应用观测 | 授权 OpenTelemetry 服务角色 → 开通服务 → 初始化 LogStore | +| 权限要求 | 管理员或「应用评测-操作」权限 | 管理员或「应用观测-操作」权限；子账号另需 `ram:CreateServiceLinkedRole` 策略 | +| 数据导出 | 评测报告在线查看 | 支持导出为 JSONL / EXCEL | +| 标签体系 | 共享标签管理（布尔值/分类/数字/文本四种类型） | 共享标签管理（与评测相同） | + +## 功能侧重对比 + +| 能力 | 应用评测 | 应用观测 | +|------|----------|----------| +| 多应用横向对比 | 支持（最多 8 个应用同时评测） | 不支持 | +| 调用链路追踪 | 不支持 | 支持（CHAIN/AGENT/RETRIEVER/LLM/TOOL 等节点类型） | +| 性能指标（延时/[Token](../concepts/token.md)） | 不提供 | 提供（调用次数、失败率、[Token](../concepts/token.md) 总量、首 [Token](../concepts/token.md) 耗时、平均调用时长） | +| 自动化质量评分 | 支持（预置模板 + 自定义 LLM/Code 评估器，每任务最多 10 个） | 不支持 | +| 人工标注 | 支持（普通模式 + 快速标注） | 支持（Span 级标签标注） | +| 版本管理 | 评测集支持版本管理 | 无版本概念 | +| 归因分析与调优建议 | 支持（Prompt/检索配置/知识库切片层面建议） | 不支持（需人工分析链路） | +| 线上数据回流 | 可从观测导入 Span 数据作为评测样本 | 支持将 Span 添加到评测集（最多 50 个字段映射） | + +## 适用场景建议 + +### 优先使用应用评测的场景 + +- **版本迭代验证**：Prompt 或知识库调整后，需要量化对比前后版本质量差异。 +- **应用选型决策**：多个智能体应用候选方案需要在统一基准下横向比较。 +- **上线前质量把关**：基于评测集系统性检测幻觉、相关性、格式合规等质量维度。 +- **标注经验固化**：将领域专家的人工标注经验抽象为可复用的 LLM 评估器。 + +### 优先使用应用观测的场景 + +- **线上问题排查**：用户反馈回答异常时，通过 Trace ID 定位具体节点（检索/重排/模型调用）的耗时与输出。 +- **性能与成本监控**：持续跟踪 Token 消耗趋势、平均延时、失败率，识别成本异常。 +- **链路瓶颈分析**：观察 RETRIEVER、RERANKER、LLM 等各节点耗时分布，定位性能瓶颈。 +- **真实数据积累**：将线上调用数据（尤其是错误状态的 Span）导入评测集，持续丰富评测样本。 + +### 两者协同的典型工作流 + +1. 通过应用观测发现线上失败率上升或延时异常； +2. 筛选问题 Span，添加标签标注错误类型； +3. 将标注后的 Span 数据导入评测集； +4. 创建评测任务，配置评估器进行系统化评分与归因； +5. 根据调优建议修改应用配置，重新发布后继续观测验证。 + +## 选型要点 + +| 决策因素 | 建议 | +|----------|------| +| 需要量化质量分数 | 选应用评测 | +| 需要实时运行状态感知 | 选应用观测 | +| 应用尚未关联知识库 | 只能用应用观测（自动评测要求关联知识库） | +| 高代码应用 | 只能用应用观测（评测暂不支持） | +| 预算敏感 | 观测免费（仅存储收费）；评测中 Code 评估器免费，LLM 评估器按 Token 计费 | +| 需要 API 集成 | 两者均仅支持控制台操作，暂无 API | + +> **提示**：应用评测的自动评测功能以开通应用观测为前提，两者并非互斥关系。建议生产环境同时启用：观测负责持续监控，评测负责定期质量审计。 + +## 被对比主题页 + +- [application evaluation](../guides/application-evaluation.md) +- [application monitoring](../guides/application-monitoring.md) + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/application-calling-approaches.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-calling-approaches.md new file mode 100644 index 00000000..8369ebd6 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-calling-approaches.md @@ -0,0 +1,91 @@ +# 应用调用方式对比 + +百炼平台围绕"应用"提供了多套 API 与接入路径，开发者在技术选型时常面临选择困难：是用 OpenAI 兼容的 Responses API，还是用 DashScope API？是走 SDK 还是裸 HTTP？管理类操作（知识库、数据连接、Prompt 模板）又该用哪套接口？本页从定位、端点、认证、功能覆盖、SDK 支持等维度对三种方式做横向对比，帮助开发者快速决策。 + +## 对比对象概览 + +| 对比项 | Responses API（OpenAI 兼容） | DashScope API（应用调用） | 应用组件 OpenAPI（bailian/2023-12-29） | +|--------|------------------------------|--------------------------|---------------------------------------| +| 定位 | 运行时调用智能体/工作流应用 | 运行时调用智能体/工作流应用 | 应用资源管理（数据、知识库、Prompt 模板、记忆等） | +| 协议风格 | OpenAI Responses 兼容 | 百炼自有 REST | 阿里云 ROA 签名风格 | +| 典型用途 | 对话交互、[多模态](../concepts/multimodal.md)输入、异步任务 | 对话交互、多轮会话、自定义参数透传 | 文件上传、知识库构建、Prompt 模板 CRUD、数据连接管理 | +| 是否触发模型推理 | 是 | 是 | 否（管理类操作） | + +## 关键维度对比 + +| 维度 | Responses API | DashScope API | 应用组件 OpenAPI | +|------|--------------|--------------|-----------------| +| Endpoint | `POST /api/v2/apps/agent/{APP_ID}/compatible-mode/v1/responses` | `POST /api/v1/apps/{APP_ID}/completion` | `https://bailian..aliyuncs.com` 下多路径 | +| 认证方式 | [API Key](../concepts/api-key.md)（`Authorization: Bearer`） | [API Key](../concepts/api-key.md)（SDK 自动读取 `DASHSCOPE_API_KEY`） | RAM 主账号 / RAM 子账号（需百炼权限 + [业务空间](../concepts/workspace.md)） | +| SDK | OpenAI Python/Java SDK | DashScope Python/Java SDK；同时支持 HTTP（Node.js/PHP/C#/Go/curl） | 阿里云 SDK（多语言 ROA 风格） | +| 多轮对话 | 通过 `input` 数组传完整历史 | `session_id`（云端托管，1 小时有效）或自行管理 `messages` | 不涉及 | +| [流式输出](../concepts/streaming.md) | `stream=true`（工作流需启用开关并重新发布） | 支持 | 不涉及 | +| [异步调用](../concepts/async-invocation.md) | `background=true`（创建→轮询→取结果） | 暂不支持 | 部分 Job 类接口异步执行（如 `SubmitIndexJob`） | +| [多模态](../concepts/multimodal.md)输入 | 支持 `input_image` / `input_file` | 支持（取决于应用配置） | 文件上传走 `ApplyFileUploadLease` → `AddFile` | +| 自定义参数透传 | 不直接支持 | `biz_params.user_defined_params` | 不涉及 | +| 适用地域 | 华北2（北京） | 华北2（北京） | 华北2（北京）、新加坡等 | +| 限流策略 | 按应用/模型维度 | 按应用/模型维度 | 文件类 5–10 次/秒，需退避重试 | + +## 各方案适用场景 + +### Responses API（OpenAI 兼容模式） + +**推荐场景：** +- 团队已有 OpenAI SDK 技术栈，希望最低成本迁移到百炼。 +- 需要异步执行长耗时任务（生成报告、多步骤工具调用），利用 `background` 模式避免请求超时。 +- 需要[多模态](../concepts/multimodal.md)输入（图像、文件），且使用 OpenAI SDK 的 `input_image` / `input_file` 类型。 + +**局限：** +- 暂不支持 `biz_params` 自定义参数透传。 +- 异步任务不支持[流式输出](../concepts/streaming.md)。 +- 地域仅华北2（北京）。 + +### DashScope API（应用调用） + +**推荐场景：** +- 需要多语言 SDK 支持（Python/Java 之外还有 Node.js/PHP/C#/Go 的 HTTP 调用）。 +- 需要通过 `session_id` 让百炼云端托管多轮对话上下文，简化客户端逻辑。 +- 智能体应用使用了自定义插件，需要通过 `biz_params.user_defined_params` 透传业务参数。 +- 工作流应用需通过 `messages` 自行管理对话历史，获得更灵活的上下文控制。 + +**局限：** +- 不支持[异步调用](../concepts/async-invocation.md)模式。 +- 多轮对话 `session_id` 有效期 1 小时，最多 50 轮。 + +### 应用组件 OpenAPI（bailian/2023-12-29） + +**推荐场景：** +- 需要通过 API 管理数据连接（文件上传、类目、解析器、连接器）。 +- 需要编程式创建和管理知识库（`CreateIndex` → `SubmitIndexJob` → `SubmitIndexAddDocumentsJob`）。 +- 需要通过 API 维护 Prompt 模板（`CreatePromptTemplate` / `ListPromptTemplates` 等）。 +- 需要管理记忆（Memory）、临时存储与支付流转等应用侧资源。 +- RAM 子账号需要操作级权限控制（`AliyunBailianDataFullAccess` / `AliyunBailianDataReadOnlyAccess`）。 + +**局限：** +- 这是管理类 API，不触发模型推理；运行时调用仍需走 Responses API 或 DashScope API。 +- 授权粒度为操作级，不支持资源级授权；细粒度隔离需通过[业务空间](../concepts/workspace.md)实现。 +- 文件类接口限流较严（5–10 次/秒），批量操作需注意退避重试。 + +## 选型决策流程 + +1. **需要管理应用资源（数据/知识库/Prompt 模板）吗？** → 是 → 使用应用组件 OpenAPI；运行时调用再从下面两选一。 +2. **已有 OpenAI SDK 技术栈或需要[异步调用](../concepts/async-invocation.md)吗？** → 是 → 使用 Responses API。 +3. **需要 `session_id` 托管多轮对话、多语言 HTTP 调用或自定义参数透传吗？** → 是 → 使用 DashScope API。 +4. **两者均可用时** → 优先选择与现有技术栈兼容性更好的一方；若需要异步能力则必须选 Responses API。 + +## 小结 + +三种方式并非互斥，而是面向不同层面的互补关系： + +- **Responses API 和 DashScope API** 都属于运行时调用层，负责触发应用执行并返回推理结果，开发者按生态兼容性和功能需求二选一即可。 +- **应用组件 OpenAPI** 属于管理层，负责应用所依赖的数据、知识库和模板的创建与维护，与运行时调用层配合使用才能完成端到端的应用集成。 + +实际项目中，典型组合是：先用应用组件 OpenAPI 完成知识库构建和文件上传，再通过 Responses API 或 DashScope API 调用已发布的应用进行对话交互。 + +## 被对比主题页 + +- [application call](../api/application-call.md) +- [bailian application calling](../guides/bailian-application-calling.md) +- [application component api reference](../api/application-component-api-reference.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/application-calling-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-calling-comparison.md index c3c897e5..d7b2b62a 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/application-calling-comparison.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-calling-comparison.md @@ -1,51 +1,64 @@ -# 应用调用方式对比：API 直调与百炼应用调用 +# 应用调用方式对比 -阿里云百炼平台为已编排好的应用（智能体、工作流、新版智能体 Agent 2.0）提供两套对外调用路径：一是面向 OpenAI 生态的 **Responses API（OpenAI 兼容模式）**，二是面向百炼原生的 **DashScope `Application.call` / `/completion` API**。两者底层均指向同一个 `APP_ID`，但在端点形态、SDK 选型、输入结构、多轮与多模态能力、扩展参数等方面存在差异。本文从技术选型视角对比两种方式，帮助开发者根据现有技术栈与功能需求做出取舍。 +本页对比阿里云百炼平台调用智能体应用和工作流应用的两种 API 方式——**OpenAI 兼容的 Responses API** 与 **DashScope API**，帮助开发者根据生态兼容性、异步能力、多模态支持等需求做出技术选型。两种方式均可调用智能体应用和工作流应用，核心凭证一致（APP ID + API Key），但在端点、SDK、多轮对话机制、异步调用等维度存在差异。 + +## 前提条件 + +无论选择哪种方式，均需完成以下准备： + +- 获取 **API Key**，配置到环境变量 `DASHSCOPE_API_KEY`。 +- 获取 **APP ID**（从控制台「应用管理」页面复制）。若应用位于子业务空间，还需提供 **Workspace ID**。 +- 已创建并发布百炼应用（智能体或工作流应用）。 +- 安装对应 SDK：Responses API 用 OpenAI SDK，DashScope API 用 DashScope SDK（或直接 HTTP 调用）。 + +> 目前 APP ID 和 Workspace ID 只能通过控制台手动获取，不支持 API 或 CLI 查询。 ## 关键维度对比 -| 维度 | OpenAI 兼容 Responses API（API 直调） | DashScope 原生 API（百炼应用调用） | -| --- | --- | --- | -| 调用端点 | `POST https://dashscope.aliyuncs.com/api/v2/apps/agent/{APP_ID}/compatible-mode/v1/responses` | `POST https://dashscope.aliyuncs.com/api/v1/apps/{APP_ID}/completion` | -| SDK 选型 | OpenAI SDK（多语言） | DashScope SDK（Python / Java），或直接 HTTP | -| base_url 配置 | `https://dashscope.aliyuncs.com/api/v2/apps/agent/{APP_ID}/compatible-mode/v1` | 无需 base_url，SDK 内置或直接 POST | -| 输入格式 | `input` 为字符串或消息数组，`role` 取 `system`/`user`/`assistant`，多模态 `content` 为数组（`input_text`/`input_image`/`input_file`） | `input.prompt` 字符串，或 `messages` 数组（自行管理多轮历史） | -| 输出格式 | OpenAI Responses 结构，`response.output` 等 | `{"output": {"finish_reason","session_id","text"}, "usage":{...}, "request_id":"..."}`，业务侧消费 `output.text` | -| 支持模型 | 智能体、工作流、新版智能体 Agent 2.0 | 智能体应用、工作流应用（[智能体编排](../concepts/agent-orchestration.md)应用已被工作流应用替代） | -| 同步/异步 | 支持 `background` 异步执行，同步默认；流式 `stream=true` | 主要为同步调用，`session_id` 由云端管理历史 | -| [流式输出](../concepts/streaming-output.md) | 支持（`stream=true`），异步暂不支持流式 | 通过 SDK / HTTP 支持（详见调用文档） | -| 多轮对话 | 传递完整 `input` 消息数组；基于 `pre_response_id`/`conversation_id` 的上下文能力后续支持 | 两种方式：`session_id`（云端托管，1 小时有效、最多 50 轮）或自行维护 `messages`（推荐，更灵活） | -| 多模态 | 原生支持文本、图像、文件（`input_file` 仅智能体应用支持） | 通过 `messages` 与应用内编排支持 | -| 自定义参数透传 | 通过 `input`/应用编排间接实现 | `biz_params.user_defined_params` 透传至自定义插件与工作流插件节点 | -| 业务空间 | 默认空间仅需 APP ID；子空间或海外地域需在请求中包含 Workspace ID | 同样需要 APP_ID；子空间按地域 Base URL 处理 | -| 典型场景 | 复用现有 OpenAI 代码库与工具链、多模态交互、统一 OpenAI 协议接入 | 全面功能与更高性能、自定义插件参数透传、Java/Node.js 直接 HTTP 集成 | - -## 适用场景建议 - -**OpenAI 兼容 Responses API 适合：** - -- 已有 OpenAI SDK 代码资产、希望以最小改动接入百炼应用的团队。 -- 需要多模态输入（文本 + 图像 + 文件）的智能体交互场景。 -- 希望统一在 OpenAI 协议生态下做模型/应用切换、保持代码中立。 -- 需要异步执行（`background`）与[流式输出](../concepts/streaming-output.md)能力的实时或长任务交互。 - -**DashScope 原生 `/completion` API 适合：** - -- 追求更全面功能与更高性能，使用百炼原生能力（如自定义插件参数透传 `biz_params`）。 -- Java/Node.js 项目希望直接以 HTTP 方式集成，不引入 OpenAI SDK 依赖。 -- 工作流应用需要通过 `session_id` 让云端托管对话历史，简化多轮实现。 -- 需要在工作流大模型节点中配合 `historyList` 变量精细控制提示词与上下文。 - -## 技术选型建议 - -1. **优先看协议生态**：若团队代码栈已围绕 OpenAI SDK 构建（含观测、重试、流式解析），选 Responses API 可降低迁移与维护成本；若以阿里云/DashScope 体系为主，选原生 API 更顺。 -2. **看扩展能力**：自定义插件参数透传（`biz_params.user_defined_params`）目前是原生 API 的明确能力，需要此能力的场景应选原生 API。 -3. **看多轮管理偏好**：希望云端托管历史、降低客户端状态复杂度，用原生 API 的 `session_id`；希望完全自控历史与上下文，两套 API 都支持 `messages` 数组方式。 -4. **看多模态需求**：图像、文件等多模态输入在 Responses API 中有标准化的 `content` 数组结构，接入更直接；原生 API 需结合应用编排实现。 -5. **看地域与业务空间**：两套 API 均支持默认空间仅凭 APP ID 调用；子业务空间或海外地域需携带 Workspace ID，选型不影响该约束，但需在请求中正确拼装。 -6. **凭证一致**：两套方式都使用同一份 `DASHSCOPE_API_KEY`，无需为不同调用方式分别管理密钥，切换成本主要在 SDK 与请求结构层面。 - -综上，两种方式并非互斥：同一 `APP_ID` 可同时被两套 API 调用，团队可按业务模块分别选型——面向外部生态集成用 Responses API，面向内部能力扩展用原生 API。 +| 维度 | Responses API（OpenAI 兼容） | DashScope API | +|------|------------------------------|---------------| +| API 端点 | `POST .../api/v2/apps/agent/{APP_ID}/compatible-mode/v1/responses` | `POST .../api/v1/apps/{APP_ID}/completion` | +| SDK | OpenAI Python / Java SDK | DashScope Python / Java SDK | +| HTTP 调用语言 | 任意支持 HTTP 的语言 | Python、Java、PHP、Node.js、C#、Go 等 | +| 多轮对话 | 通过 `input` 数组传递完整消息历史（role + content） | `session_id`（系统托管，有效期 1 小时，最多 50 轮）或自行维护 `messages` 数组 | +| 异步调用 | 支持，设置 `background=true`，返回任务 ID 后轮询 | 暂不支持 | +| 流式输出 | 支持，设置 `stream=true` | 支持（工作流应用需在结束节点启用流式开关并重新发布） | +| 多模态输入 | 支持图像（`input_image`）和文件（`input_file`） | 通过应用编排配置 | +| 自定义参数透传 | 未提供 | 支持，通过 `biz_params.user_defined_params` 传递业务参数 | +| 可用地域 | 仅华北2（北京） | 仅华北2（北京） | +| 响应结构 | OpenAI 标准响应格式 | `{"output": {"text", "session_id", "finish_reason"}, "usage": {...}, "request_id": "..."}` | +| 典型场景 | OpenAI 生态兼容、异步长任务、多模态交互 | 深度集成百炼能力、多语言调用、自定义参数透传 | + +## Responses API 适用场景 + +Responses API 提供 OpenAI 兼容接口，适合以下场景： + +- **OpenAI 生态复用**：已有基于 OpenAI SDK 的代码或服务，可平滑迁移到百炼，仅需替换 `base_url` 和 `api_key`。 +- **异步长任务**：如生成报告、多步骤工具调用等耗时任务，通过 `background=true` 异步执行，避免请求超时。流程为创建任务 → 轮询状态 → 获取结果。 +- **多模态输入**：需同时传入图像或文件的场景，通过 `input_image` / `input_file` 类型在 `content` 数组中混合多种输入。 + +> 异步任务暂不支持流式输出（`stream=true`）。 + +## DashScope API 适用场景 + +DashScope API 提供更全面的功能支持，适合以下场景： + +- **多语言集成**：除 Python/Java SDK 外，还支持 PHP、Node.js、C#、Go 等语言的 HTTP 直接调用，适合异构技术栈。 +- **自定义参数透传**：通过 `biz_params.user_defined_params` 向自定义插件或自定义节点传递业务参数，实现动态配置。 +- **Session 托管**：通过 `session_id` 让系统自动加载历史对话，减少客户端维护成本（有效期 1 小时，最多 50 轮）。如需更灵活的控制，也可自行管理 `messages` 数组。 +- **工作流应用深度集成**：工作流应用支持在调用时传递自定义参数、配置提示词变量 `historyList` 等能力。 + +## 选型建议 + +| 需求 | 推荐方案 | +|------|----------| +| 已有 OpenAI SDK 代码，希望快速接入 | Responses API | +| 需要异步执行长耗时任务 | Responses API（`background=true`） | +| 需要多模态（图像/文件）输入 | Responses API | +| 需要多语言 HTTP 调用（非 Python/Java） | DashScope API | +| 需要向自定义插件/节点透传业务参数 | DashScope API | +| 希望由系统托管对话上下文 | DashScope API（`session_id`） | +| 需要完全控制对话历史 | 两者皆可（Responses 用 `input` 数组，DashScope 用 `messages` 数组） | ## 被对比主题页 @@ -53,5 +66,3 @@ - [bailian application calling](../guides/bailian-application-calling.md) - - diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/application-eval-vs-model-eval.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-eval-vs-model-eval.md new file mode 100644 index 00000000..5ec2e77a --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-eval-vs-model-eval.md @@ -0,0 +1,74 @@ +# 应用评测与模型评测对比 + +百炼平台提供两套相互独立的评测体系：**应用评测**面向已发布的智能体/工作流应用，评估其端到端输出质量；**模型评测**面向基础模型本身，衡量模型的推理能力和回答质量。两者的评测对象、数据格式、评分机制和适用场景均有显著差异，开发者在做技术选型时需要根据实际需求加以区分。 + +## 核心维度对比 + +| 维度 | 应用评测 | 模型评测 | +|------|----------|----------| +| **评测对象** | 智能体应用、工作流应用（端到端） | 文本生成类基础模型 | +| **评测目的** | 评估 RAG 流程、知识库配置、提示词等整体输出质量 | 评估模型本身的推理、理解、生成能力 | +| **数据输入格式** | 旧版：`.xls`/`.xlsx`（对话分析）或 `.jsonl`（知识问答）；新版：根据应用出入参定义的评测集 | 含 Prompt + Completion 列的评测数据集，或已有推理结果集 | +| **评分机制** | 自动评测（qwen-max/qwen-plus 评分）、手动人工打分、新版评估器（LLM/Code） | 大模型评估（裁判模型打分）、规则评估（ROUGE/BLEU/Cosine/字符串匹配）、人工评估 | +| **评分器/维度** | 新版评估器：预置模板 + LLM 评估器 + Code 评估器，每任务最多 10 个 | 评测维度：大模型评估（数值型/分类型）、规则评估（相似度/字符串匹配）、人工评估（分类型） | +| **基线评测** | 不支持 | 支持（C-Eval、MMLU、GSM8K、BBH 等公开数据集，仅北京地域） | +| **多对象横向对比** | 支持（最多 8 个应用或版本同台对比） | 支持（通过排行榜绑定进行模型间排名对比） | +| **归因分析** | 有（将 BadCase 定位到 RAG 流程具体环节：检索/重排/切片/模型理解） | 无内置归因；通过逐条评分数据自行分析 | +| **计费构成** | 评估模型 Token 费用（LLM 评估器产生）；Code 评估器无额外费用 | 被评测模型推理费用 + 裁判模型评分费用；使用推理结果集可免去推理费用 | +| **操作方式** | 控制台 + 新旧两套系统 | 仅控制台，无公开 API/SDK | +| **前提条件** | 应用须已发布并配置知识库；须开通应用观测 | 无需应用，仅需准备数据集和评测维度 | +| **数据规模建议** | 视应用复杂度而定，评测集基于知识库自动生成或手动上传 | 小规模验证 50-100 条，正式评测 200-500 条，全面评估 500+ 条 | + +## 评估器 / 评测维度对比 + +| 类型 | 应用评测（评估器） | 模型评测（评测维度） | +|------|------------------|------------------| +| 大模型语义打分 | LLM 评估器（裁判模型） | 大模型评估-数值型 / 分类型 | +| 规则/代码判断 | Code 评估器（Python 脚本） | 规则评估-文本相似度 / 字符串匹配 | +| 人工标注 | 标签（分类/布尔/数字/文本） | 人工评估-分类型（Pass/Fail） | +| 预置模板 | 有（通用质量、智能体、格式校验等） | 无专用模板，自行配置评分器 Prompt | +| 需要参考答案 | 否（LLM 评估器）；是（部分 Code 评估器） | 规则评估需要；大模型评估和人工评估不需要 | + +## 各方案适用场景建议 + +### 选择应用评测的场景 + +- 已有上线的智能体或工作流应用，希望评估知识库配置、RAG 流程或提示词的整体效果。 +- 需要对同一应用的多个版本进行 A/B 对比，或在多个候选应用中做横向选型。 +- 需要将 BadCase 归因到 RAG 具体环节（检索策略、切片配置等），从而指导优化。 +- 知识库更新后，需要持续回归评测确认质量没有下降。 + +### 选择模型评测的场景 + +- 在选型阶段，需要横向对比不同基础模型（如 Qwen 各规格）的推理能力和任务表现。 +- 完成模型微调/调优后，需要验证调优效果是否达到预期。 +- 需要在标准公开基准（C-Eval、MMLU 等）上快速获得模型基础能力数据。 +- 评测对象是裸模型而非应用，无需关注 RAG 流程的影响。 + +### 两者结合使用 + +在完整的 LLM 应用开发生命周期中，两者可以形成互补：先用**模型评测**选出基础能力最优的模型，再用**应用评测**验证该模型在具体应用场景下的端到端表现，并持续迭代优化 RAG 配置。 + +## 技术选型参考 + +``` +需要评测的对象是： +├─ 基础模型本身 → 模型评测 +│ ├─ 有公开基准需求 → 基线评测（仅北京地域） +│ └─ 自有数据集 → 自定义评测 +│ ├─ 有标准答案且格式固定 → 规则评估（ROUGE/字符串匹配） +│ ├─ 有标准答案但表述多样 → 规则评估（文本相似度） +│ ├─ 无标准答案需语义理解 → 大模型评估 +│ └─ 需主观判断 → 人工评估 +└─ 智能体/工作流应用 → 应用评测 + ├─ 需要 RAG 归因分析 → 自动评测 + ├─ 多应用/版本对比 → 自动评测（横向评测模式） + └─ 领域专家主观评审 → 手动评测 / 新版标签标注 +``` + +## 被对比主题页 + +- [application evaluation](../guides/application-evaluation.md) +- [model evaluation introduction](../guides/model-evaluation-introduction.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/application-evaluation-vs-monitoring.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-evaluation-vs-monitoring.md index 044ec3fc..4c4678a3 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/application-evaluation-vs-monitoring.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-evaluation-vs-monitoring.md @@ -1,41 +1,70 @@ -# 应用评测与应用监控对比 +# 应用[评测](../concepts/evaluation.md)与应用监控对比 -阿里云百炼平台同时提供"应用评测"与"应用监控（应用观测）"两套围绕应用质量的能力：前者面向**离线**场景，用评测集 + 评估器系统化衡量应用输出质量；后者面向**在线**场景，端到端追踪已发布应用的真实调用链路并采集延时、Token 量等运行时指标。两者通过共享的"标签"组件打通——线上 Span 可标注后直接沉淀为评测样本，实现"线上观测 → 离线评测 → [模型调优](../concepts/fine-tuning.md)"的闭环。本文从目的、数据来源、输入输出、计费、典型场景等维度对比两者，帮助开发者做技术选型。 +应用[评测](../concepts/evaluation.md)与应用监控都是百炼平台面向应用质量改进的关键能力，但关注点不同：应用[评测](../concepts/evaluation.md)更偏向“离线质量验证”和“版本效果对比”，用于判断应用回答是否正确、是否符合预期；应用监控更偏向“线上运行观测”和“调用链路追踪”，用于发现真实调用中的性能、成本、错误与数据分布问题。开发者通常不应二选一，而应将应用监控产生的真实样本沉淀为评测集，再通过应用评测验证调优效果，形成持续改进闭环。 -## 关键维度对比 +## 核心差异 -| 对比维度 | 应用评测 | 应用监控（应用观测） | +| 维度 | 应用评测 | 应用监控 / 应用观测 | | --- | --- | --- | -| 核心目的 | 离线评估应用输出质量，产出评分、BadCase、归因与调优建议 | 在线追踪应用真实调用链路，采集延时、Token 量等运行时指标 | -| 数据来源 | 评测集（手动上传或大模型自动生成） | 已发布应用在线接收的真实 Prompt 与调用 | -| 触发方式 | 手动发起评测任务，运行评测集 | 应用添加到观测列表后自动追踪，分钟级同步 | -| 应用范围 | 智能体应用（自动评测需配置知识库）、[工作流](../concepts/workflow.md)应用；新版支持自定义类型 | 智能体应用、[工作流](../concepts/workflow.md)应用、高代码应用；暂不支持 Assistant API 创建的智能体应用 | -| 输入格式 | 旧版：`.xls`/`.xlsx`（对话分析）、`.jsonl`（知识问答）；新版：`.xls`/`.xlsx`，按应用出入参生成模板，单文件 ≤ 20MB，单次最多 10 个文件 | 真实线上请求，无需上传文件；导出支持 JSONL 与 EXCEL | -| 输出形式 | 评分（1-5 / 0-100 / 0-1 等）、评估器结果、BadCase 分析、归因报告、调优建议 | 调用链路树（CHAIN/AGENT/RETRIEVER/LLM 等节点）、延时、Token 量、监控统计图表 | -| 评估器/评分机制 | 预置模板 + LLM 评估器 + Code 评估器；每任务最多 10 个评估器 | 不做评分，仅记录指标；可对 Span 人工打标签 | -| 标签能力 | 分类/布尔值/数字/文本四种类型，与监控共用 | 同一套标签体系，可对每个 Span 标注，自动保存 | -| API 端点 | 通过控制台操作（评测集、评测任务、评估器、标签管理） | 无 API，仅控制台操作；底层依赖 OpenTelemetry 服务 | -| 数据时效 | 任务执行后产出，按版本留存 | 指标分钟级更新，调用记录最长可查 30 天 | -| 计费方式 | 评测任务调用大模型产生 Token 费用，正常计费 | 应用观测功能本身不收费；观测数据存储费用由 OpenTelemetry 服务收取 | -| 关键限制 | 自动评测仅面向已发布且配置知识库的智能体应用，单次最多 8 个应用横向评测；评测任务发起后配置不可修改 | 高代码应用仅能观测到入口 CHAIN 节点，不支持内部链路追踪；暂不支持长期记忆检索过程观测 | -| 版本差异 | 区分"旧版"与"新版"两套界面，评测集类型、关联应用范围、评估器机制不同 | 无新旧版本区分 | - -## 两者协同关系 - -应用监控与评测并非孤立能力，平台在数据层做了打通： - -- **Span → 评测集**：应用观测支持将 Span 数据直接加入评测集（选择目标评测集、导入方式、字段映射，每评测集最多 50 个字段映射），把真实线上调用沉淀为评测样本。 -- **共享标签**：标签管理为评测与监控共用，线上 Span 标注的标签可在评测任务中复用，反之亦然，便于跨阶段质量分析。 +| 核心目标 | 评估应用输出质量，发现 BadCase，验证 Prompt、[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)、检索配置或版本迭代效果 | 观测应用真实调用过程，追踪链路、延时、Token、错误状态和节点明细 | +| 典型使用阶段 | 发布前验证、版本上线前回归、[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)更新后质量检查、多个应用或版本横向对比 | 发布后运行期观测、线上问题排查、性能与成本分析、真实样本采集 | +| 输入数据来源 | 评测集；可由[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)自动生成、手动上传，也可从应用观测导入 | 已发布应用的真实调用数据，包括 Prompt、输出、Trace、Span、Token、延时等 | +| 输出结果 | 评测报告、评分、正确率、BadCase 分析、归因分析、调优建议、人工标注结果 | Trace 列表、调用链路、节点详情、监控统计图表、导出数据、Span 标签 | +| 支持应用类型 | 新版评测支持智能体应用、工作流应用，也支持不关联应用的纯人工标注；旧版能力存在自动评测和手动评测差异 | 支持智能体应用、工作流应用、高代码应用；高代码应用目前只能观测入口 CHAIN 节点 | +| 前提条件 | 自动评测通常要求应用已发布、配置知识库、开通应用观测并加入观测列表；多应用横向评测要求共享至少一个知识库 | 首次使用需授权 OpenTelemetry 服务角色、开通 OpenTelemetry 服务并初始化 LogStore；被观测应用需已发布并加入观测列表 | +| 数据格式 | 旧版对话分析支持 .xls / .xlsx；旧版知识问答支持 .jsonl；新版评测集可按智能体、工作流或自定义结构定义字段 | 控制台查看 Trace / Span；可按筛选条件导出 JSONL 或 EXCEL | +| 自动化能力 | 自动评测可使用评估器评分；新版支持 LLM 评估器、Code 评估器、预置模板、自定义评估器与标签 | 自动采集线上调用与指标，分钟级更新；支持过滤、标注、导出和加入评测集 | +| 模型与评估器 | 自动评测集生成和评估模型当前涉及 qwen-max、qwen-plus；LLM 评估器会产生 Token 费用，Code 评估器无额外费用 | 不直接选择评估模型；记录应用实际调用中的模型节点、Embedding、Reranker、Retriever、Tool 等链路信息 | +| API 端点 | 文档强调控制台评测任务、评测集、评估器和标签配置；未体现面向应用评测的通用公开 API 端点 | 应用观测本身没有 API，只能通过控制台操作 | +| 计费方式 | LLM 评估器会产生 Token 费用；Code 评估器无额外费用；应用自身调用仍按相关模型或服务计费 | 应用观测功能本身不收费，但观测数据存储在 OpenTelemetry 服务中，相关存储费用由 OpenTelemetry 服务收取 | +| 数据保留与查询 | 以评测集版本、评测任务和报告为核心，可长期用于回归和对比 | 详情页可查看最长 30 天内调用记录，支持按 Request ID、Trace ID、Span ID、时间范围和过滤条件查询 | +| 标签体系 | 标签用于评测任务人工标注、数据筛选和指标统计 | Span 数据可添加标签，且与评测标签管理共享，可作为后续评测样本筛选依据 | +| 主要使用者 | 负责质量验收、评估体系建设、Prompt / RAG 调优、版本发布决策的开发者或算法工程师 | 负责线上排障、性能优化、成本分析、运营监控和数据闭环的开发者或运维人员 | ## 适用场景建议 -- **选应用评测** when：需要系统化、可重复地评估应用输出质量；要对响应做相关性、有害性、幻觉等语义判断；有领域专家介入做端到端标注；发布前做回归验证或横向对比多个应用（最多 8 个）。 -- **选应用监控** when：应用已上线，需要排查真实调用的延时、失败率、Token 成本；想查看智能体/[工作流](../concepts/workflow.md)内部检索、重写、重排、LLM 等子节点的执行链路；需要分钟级性能监控与最长 30 天的调用记录检索；想从线上真实数据中沉淀评测样本。 -- **组合使用** when：希望构建"线上观测 → 标注/BadCase 沉淀 → 离线评测 → 调优迭代"的闭环质量运营体系。此时建议先开通应用观测（自动评测的前置条件之一），再用观测到的 Span 数据补充评测集。 +### 优先使用应用评测的场景 -## 技术选型小结 +- 需要在上线前判断应用回答质量是否达标。 +- 知识库、Prompt、检索策略、重排配置或模型发生变化后，需要做回归验证。 +- 需要比较多个智能体应用、多个版本或不同配置在同一评测基准下的效果。 +- 需要将人工专家经验沉淀为标签或评估器，并逐步自动化质量判断。 +- 需要定位 RAG 类应用的质量问题，例如模型理解有误、重排不佳、检索无效、切片不完整或知识缺失。 -若问题偏"质量好不好"（输出是否准确、完整、合规）→ 用应用评测；若问题偏"跑得稳不稳"（延时、失败、成本、链路追踪）→ 用应用监控；若希望用线上真实数据驱动持续调优，则两者配合使用，通过共享标签与 Span 导入评测集的能力打通闭环。开通自动评测前需先开通应用观测，这也是两者协同的一个隐含前置条件。 +### 优先使用应用监控的场景 + +- 应用已经发布，需要观察真实用户调用中的延时、Token 消耗、失败率和调用次数。 +- 需要按 Request ID、Trace ID 或 Span ID 排查单次调用异常。 +- 需要理解智能体或工作流内部节点的执行过程，例如检索、改写、向量化、重排、LLM 调用、工具调用或条件分支。 +- 需要收集真实线上样本，并导出或加入评测集，作为后续评测和调优依据。 +- 需要按标签、状态、输入、输出、Token、延时等条件筛选问题调用。 + +## 技术选型参考 + +如果目标是“证明某个应用版本效果更好”，应以应用评测为主。它提供评测集、评估器、标签、评分和报告，适合做可复现的质量基准。对于 RAG 智能体，自动评测还能给出 BadCase 归因和调优建议，帮助判断问题来自模型理解、检索、重排、切片还是知识库内容。 + +如果目标是“定位线上一次调用为什么慢、为什么错、Token 为什么高”，应以应用监控为主。它提供 Trace / Span 级别的链路信息和分钟级统计，能够看到应用内部节点、状态、延时和 Token 分布。需要注意，应用观测本身没有 API，且高代码应用目前只能看到入口 CHAIN 节点，不能完整追踪内部调用链路。 + +如果目标是“建立持续质量闭环”，建议组合使用两者： + +1. 先开启应用观测，把已发布应用加入观测列表。 +2. 通过监控统计和 Trace 明细发现高频问题、错误调用、高延时调用或高 Token 调用。 +3. 对 Span 数据进行标签标注，并将典型样本加入评测集。 +4. 使用新版评测任务配置评估器和标签，对应用版本进行离线评测。 +5. 根据评测报告和归因分析调整 Prompt、知识库、检索、重排或模型配置。 +6. 发布新版本后继续通过应用监控观察真实效果，并定期回灌新的评测样本。 + +## 开发者落地建议 + +- 对发布流程：将应用评测作为上线门禁，尤其是在知识库更新、Prompt 修改、模型切换和工作流调整后执行回归评测。 +- 对线上运维：将应用监控作为日常观测入口，重点关注失败率、平均调用时长、首 Token 耗时、Token 总量和异常 Trace。 +- 对数据闭环：优先从应用监控中筛选真实问题样本，沉淀到评测集，而不是只依赖人工构造的理想问题。 +- 对成本控制：使用应用监控定位高 Token 链路，使用应用评测验证优化方案是否在质量不下降的前提下降低成本。 +- 对评估器选择：语义类、相关性、幻觉检测等场景适合 LLM 评估器；格式校验、精确匹配、数值计算等场景优先使用 Code 评估器以降低额外成本。 + +## 总结 + +应用评测回答的是“应用质量是否符合预期、哪个版本更好、应该如何调优”；应用监控回答的是“应用在线上实际如何运行、问题发生在哪个链路、性能和成本表现如何”。前者适合建立质量基准和评估闭环，后者适合线上观测和样本采集。对于生产级智能体或工作流应用，推荐同时配置应用监控与应用评测，让真实调用数据驱动评测集迭代，再用评测结果指导应用优化。 ## 被对比主题页 diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/application-lifecycle-management.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-lifecycle-management.md new file mode 100644 index 00000000..4ef8a0c2 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/application-lifecycle-management.md @@ -0,0 +1,74 @@ +# 应用评估、监控与权限管理对比 + +在阿里云百炼平台上，应用从开发到上线运营涉及三个互补的能力：**应用评估**（Application Evaluation）用于回答质量的系统化打分与调优，**应用观测**（Application Monitoring）用于运行期链路追踪与指标监控，**权限管理**（Application Permission Management）用于组织内资源与功能的访问控制。三者共同构成应用生命周期管理的闭环，但在使用主体、数据形态、开通方式和计费模式上差异显著。本文横向对比三项能力，为开发者做技术选型和落地路径决策提供参考。 + +## 定位速览 + +- **应用评估**：面向应用回答质量，"生成 → 评分 → 归因 → 迭代"的离线质量闭环。 +- **应用监控**：面向应用运行时行为，端到端追踪调用链路、延时和 Token 消耗，指标分钟级更新。 +- **权限管理**：面向组织与资源，围绕业务空间与角色划分模型、页面、API-Key 的访问范围。 + +## 关键维度对比 + +| 维度 | 应用评估 | 应用监控 | 权限管理 | +| --- | --- | --- | --- | +| 目的 | 评估回答质量、辅助调优 | 追踪运行链路、观测指标 | 控制资源与功能的访问 | +| 生效阶段 | 离线 / 版本迭代期 | 应用发布后运行时 | 全生命周期，先决条件 | +| 使用主体 | 应用开发者、算法工程师 | 应用运营者、SRE | 主账号 / 业务空间管理员 | +| 主要对象 | 智能体应用、工作流应用 | 智能体、工作流、高代码应用 | 业务空间、模型、页面、API-Key | +| 输入形态 | 评测集（JSONL / XLSX / 自定义） | 应用运行期 Prompt 与调用日志 | RAM 用户、角色策略、模型清单 | +| 输出形态 | 评测报告（正确率、BadCase、归因、调优建议） | Trace 列表、监控图表、导出 JSONL / EXCEL | 权限矩阵、限流规则、API-Key | +| 关键组件 | 评测集、评估器（LLM / Code）、标签、评测任务 | CHAIN / AGENT / LLM / TOOL / RETRIEVER 等节点 | 超级管理员、业务空间管理员、普通用户 | +| 触发方式 | 手动创建评测任务（自动 / 人工） | 添加"被观测应用"后自动追踪 | 全局管理 / 业务空间「权限管理」页面 | +| API / OpenAPI | 无独立 OpenAPI，走控制台 | 无 API，仅控制台操作 | OpenAPI 权限须主账号在 RAM 单独授予 | +| 数据时效 | 任务级（分钟到小时） | 分钟级同步，保留最长 30 天 | 授权分钟级生效，高峰期略有延迟 | +| 前置条件 | 已发布应用；须开通应用观测并加入观测列表；子账号需评测操作权限 | 首次需授权 OpenTelemetry 服务角色、开通服务并初始化 LogStore | 授予对应 RAM 策略（如 AliyunBailianFullAccess） | +| 支持模型 / 范围 | 评测集生成与评估仅支持 `qwen-max`、`qwen-plus` | 智能体 / 工作流 / 高代码（高代码不追踪内部链路） | 覆盖所有可管控模型与页面 | +| 计费方式 | 评测任务调用大模型按 Token 计费 | 功能免费；观测数据存储由 OpenTelemetry 服务收取 | 无计费 | +| 与其他能力的关系 | 依赖应用观测的观测列表；共享标签体系 | 数据可导入评测集；标签体系与评估共享 | 是评估、观测能力的准入门槛 | +| 典型场景 | 版本迭代对比、Prompt / 检索调优、RAG BadCase 归因 | 上线后性能监控、线上问题排查、成本分析 | 多地域组织授权、模型调用限流、API-Key 治理 | + +## 数据流示意 + +- **权限管理** → 授予业务空间与页面访问权 → 用户才能创建、发布应用并开通后续能力。 +- **应用监控** → 追踪线上 Trace / Span → 导出或"加入评测集" → 为**应用评估**提供真实样本。 +- **应用评估** → 生成评测报告与调优建议 → 迭代应用 → 新版本再次进入监控与评测循环。 + +## 适用场景建议 + +### 应用评估 + +- 应用完成一轮开发或调整（Prompt / 知识库 / 检索 / 重排 / 模型）后需要客观验证效果。 +- 需要横向对比多个候选应用或同一应用不同版本，用于选型或灰度决策。 +- 需要将人工标注经验固化为可复用的自动评估器，构建长期质量基线。 +- 需要针对 RAG BadCase 做归因（模型理解 / 检索 / 重排 / 切片 / 未获取知识）。 + +### 应用监控 + +- 生产环境的应用需要感知延时、Token 消耗与失败率等运营指标。 +- 排查线上问题：某次调用为什么慢、卡在哪个节点、Guardrail 是否被触发。 +- 希望把真实线上样本沉淀到评测集，用于持续回归。 +- 需要为应用做成本核算（输入 / 输出 Token、按分钟 / 小时 / 天粒度聚合）。 + +### 权限管理 + +- 组织内存在多个 RAM 用户、多个业务空间、多个地域，需要精细化拆分授权。 +- 需要限制特定模型的调用范围、请求数与 Token 限流，或控制模型调优 / 部署权限。 +- 需要为不同团队分配独立的 API-Key，并保证归属业务空间的策略生效。 +- 需要通过 OpenAPI 调用百炼数据 / 知识库 / Prompt 工程 / 长期记忆等能力（须单独授予 `AliyunBailianDataFullAccess` / `AliyunBailianDataReadOnlyAccess`）。 + +## 技术选型参考 + +- **上线前 → 上线后 → 治理**：先用权限管理搭好组织与资源边界，再用应用监控保障运行时可见性，最后用应用评估驱动质量迭代。三者顺序不可颠倒——没有权限即无法访问功能；没有观测，自动评测的前置条件（观测列表、真实样本）就无从谈起。 +- **构建质量闭环**：把应用监控里的高价值 Span 通过"加入评测集"沉淀为回归样本，配合 3-5 个评估器（相关性 LLM + 格式 Code + 幻觉检测等）形成"线上采样 → 离线评测 → 迭代发布"的持续闭环。 +- **控制成本**：应用监控功能本身不收费，但 OpenTelemetry 存储与评测调用 Token 均产生费用；对大规模应用建议在权限管理中先做模型限流与业务空间隔离，避免评测任务或线上流量突增引发失控成本。 +- **子账号协作**：三项能力都对子账号有额外授权要求（评测的"应用评测-操作"、观测的 `ram:CreateServiceLinkedRole`、OpenAPI 的独立系统策略），落地时建议先梳理 RAM 策略清单再逐项开通。 +- **新旧版本兼容**：应用评估存在新旧两套系统，新版围绕"评测集 + 评估器 + 标签 + 评测任务"更灵活；如无历史包袱建议直接采用新版体系，并复用其与应用监控共享的标签管理。 + +## 被对比主题页 + +- [application evaluation](../guides/application-evaluation.md) +- [application monitoring](../guides/application-monitoring.md) +- [application permission management](../guides/application-permission-management.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/content-generation-apis.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/content-generation-apis.md new file mode 100644 index 00000000..cc445223 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/content-generation-apis.md @@ -0,0 +1,58 @@ +# [多模态](../concepts/multimodal.md)生成 API 对比 + +百炼平台提供图像生成、3D 资产生成和视频生成三大类[多模态](../concepts/multimodal.md)生成 API，分别覆盖静态视觉内容、三维空间资产和动态视频内容的生产需求。三者底层模型、调用模式、输入输出格式各有差异，本文从关键维度进行横向对比，帮助开发者在技术选型时快速定位合适的 API 方案。 + +## 关键维度对比 + +| 对比维度 | 图像生成 API | 3D 生成 API | 视频生成 API | +| --- | --- | --- | --- | +| **输入格式** | 文本（[prompt](../guides/prompt.md)）、图像 URL（图生图/编辑）、涂抹区域、手绘涂鸦 | 文本（[prompt](../guides/prompt.md)，最大1024字符）、单张图像 URL、多图（固定4视角，前/左/后/右） | 文本（[prompt](../guides/prompt.md)）、图像 URL（首帧/首尾帧/参考图）、视频 URL、音频 URL | +| **输出格式** | PNG 图像（1-9张，分辨率512×512~4K） | GLB 格式 3D 模型（PBR材质/无贴图基础模型）+ 预览渲染图 | 视频（分辨率480P~4K，时长5/8秒等） | +| **支持模型** | 千问图像（Qwen-Image）、万相（Wan）、Z-Image、可灵（Kling）、Vidu，及多个创意工具模型 | Tripo/Tripo-H3.1（高精度）、Tripo/Tripo-P1.0（专业，速度更快） | 万相（wan2.7/wan2.6/早期）、HappyHorse、PixVerse、Vidu、可灵（Kling）、人像动画系列、万相动作/数字人 | +| **调用模式** | 同步（千问/万相2.6+/Z-Image）+ 异步（万相2.5及以下/可灵/Vidu/创意工具） | 仅异步 | 仅异步 | +| **API 端点** | 同步：`/api/v1/services/aigc/multimodal-generation/generation`；异步：`/api/v1/services/aigc/text2image/image-synthesis` 等多路径 | 创建：`/api/v1/services/aigc/video-generation/3d-generation`；查询：`/api/v1/tasks/{task_id}` | 创建：`/api/v1/services/aigc/video-generation/video-synthesis`；查询：`/api/v1/tasks/{task_id}` | +| **地域要求** | 支持多地域（北京、新加坡、弗吉尼亚、法兰克福） | 仅限华北2（北京），须使用北京地域 API Key | 支持多地域，但模型、Endpoint、API Key 必须属同一地域 | +| **异步任务有效期** | task_id 有效期24小时，图像 URL 有效期24小时 | task_id 有效期24小时，产物下载链接有效期仅2小时 | task_id 有效期24小时，过期返回 UNKNOWN | +| **轮询间隔建议** | — | 约15秒，查询 RPS 限制为20 | 约15秒，状态流转 PENDING→RUNNING→SUCCEEDED/FAILED | +| **典型场景** | 文生图、图像编辑、虚拟模特、海报生成、试衣、背景生成、创意文字等 | 文生3D、单图生3D、多图生3D，输出可编辑 GLB 模型 | 文生视频、图生视频、参考生视频、视频编辑、视频超分、对口型、数字人 | + +## 调用模式差异 + +三者最显著的技术差异在于调用模式： + +- **图像生成**：部分模型支持同步调用（千问图像、万相2.6+、Z-Image），一次请求即可获得结果，开发体验最佳；旧版模型及创意工具仍需异步轮询。 +- **3D 生成**和**视频生成**：由于生成耗时较长，均采用纯异步模式。创建任务时必须携带 `X-DashScope-Async: enable` 请求头，否则报错 `current user api does not support synchronous calls`。 + +异步流程统一为「创建任务获取 task_id → 轮询查询结果」，但产物有效期差异较大：3D 生成的下载链接仅2小时，远短于图像和视频的24小时，需特别注意及时下载。 + +## 模型丰富度与功能覆盖 + +| 能力类别 | 图像生成 | 3D 生成 | 视频生成 | +| --- | --- | --- | --- | +| 文生内容 | 支持 | 支持 | 支持 | +| 图生内容 | 支持（图生图/编辑/局部重绘） | 支持（单图生3D/多图生3D） | 支持（首帧/首尾帧/参考生视频） | +| 内容编辑/增强 | 支持（风格迁移、超分、去水印、扩图） | 不支持 | 支持（视频编辑、超清、对口型、动作模仿） | +| 文字渲染 | 支持（千问擅长复杂文字） | 不适用 | 部分支持（Vidu 中英文字精准渲染） | +| 模型数量 | 最多（5+系列，十余个创意工具） | 最少（2个模型） | 较多（7+家族） | + +## 选型建议 + +- **静态视觉内容生产**（海报、模特图、产品图、创意设计）→ 优先选择图像生成 API。千问图像擅长文字渲染和真实质感，万相支持多种艺术风格，Z-Image 适合轻量快速生图。 +- **三维空间资产生产**（游戏道具、电商3D展示、AR/VR内容）→ 选择 3D 生成 API。需要北京地域 API Key，且注意产物下载链接仅2小时有效期。 +- **动态视频内容生产**（短视频、广告、数字人播报、视频编辑增强）→ 选择视频生成 API。万相2.7为新版协议推荐使用，可灵支持智能分镜，PixVerse 支持4K超清和动作模仿，人像动画系列适合舞蹈/唱演等场景。 +- **混合需求**：若需先生成静态素材再制作动态内容，可组合使用。例如先用图像生成 API 产出首帧图，再用视频生成 API 做图生视频；或用3D生成 API 产出模型后截图作为视频参考素材。 + +## 来源文档 + +- [千问-文生图API参考](../../raw/model-api-reference/image-generation/qwen-image-api-reference/qwen-image-api.md) +- [万相-文生图V2版API参考](../../raw/model-api-reference/image-generation/wan-image-api-reference/text-to-image-v2-api-reference.md) +- [Tripo-3D模型生成](../../raw/model-api-reference/3d-generation/tripo-3d-generation-api-reference.md) +- [万相2.7-图生视频API参考](../../raw/model-api-reference/video-generation-api/wan-api-reference/image-to-video-general-api-reference.md) + +## 被对比主题页 + +- [image generation](../api/image-generation.md) +- [3d generation](../api/3d-generation.md) +- [video generation api](../api/video-generation-api.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/evaluation-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/evaluation-comparison.md index deadf313..6373370c 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/evaluation-comparison.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/evaluation-comparison.md @@ -1,67 +1,71 @@ # 应用评测与模型评测对比 -百炼平台提供两套独立的评测体系：**应用评测**面向已构建的智能体应用和工作流应用，评估端到端的输出质量与 RAG 链路效果；**模型评测**面向底层大模型本身，评估模型的推理能力和指令遵循表现。两者的评测对象、数据流、评分机制和适用场景均有显著差异，开发者需要根据当前所处的开发阶段选择合适的评测方式。 +百炼平台提供两套独立的评测能力：**应用评测**面向已发布的智能体应用和工作流应用，评估端到端的输出质量；**模型评测**面向文本生成类模型本身，评估模型的基础推理能力。两者在评测对象、数据流转、评分机制和适用场景上存在显著差异，开发者应根据实际选型需求选择对应能力。 ## 关键维度对比 | 维度 | 应用评测 | 模型评测 | |------|----------|----------| -| **评测对象** | 智能体应用、工作流应用（已发布的完整应用） | 文本生成类大模型（基础模型或调优后模型） | -| **核心目标** | 验证应用端到端输出质量，定位 RAG 链路问题 | 评估模型推理能力，辅助模型选型或调优验证 | -| **评测方式** | 自动评测（单应用 / 多应用横向）、手动评测 | 自定义评测（AI / 规则 / 人工）、基线评测（公开数据集） | -| **评测集来源** | 基于应用关联知识库自动生成，或手动上传 | 手动上传评测数据集，或使用公开标准数据集（C-Eval、MMLU 等） | -| **评估器 / 评分机制** | 新版评估器（LLM 评估器 + Code 评估器 + 预置模板）；旧版由平台内置评分 | 评测维度模板（大模型评估数值型/分类型、规则评估相似度/匹配、人工评估） | -| **归因分析** | 支持 RAG 链路归因（模型理解有误、重排不佳、检索无效、切片不完整、未获取知识） | 不提供链路归因，仅输出维度得分和通过率 | -| **横向对比能力** | 最多 8 个应用同基准横向对比 | 支持多模型评测结果排行榜对比 | -| **人工标注** | 新版通过标签体系支持四种类型标注（分类 / 布尔值 / 数字 / 文本） | 人工评估维度（Pass/Fail 标注） | -| **前提条件** | 应用已发布、已配置知识库、已开通应用观测 | 无特殊前提，上传数据集即可评测 | -| **地域限制** | 无特殊地域限制 | 基线评测仅北京地域可用 | -| **API 支持** | 通过控制台操作 | 仅控制台操作，不提供公开 API/SDK（可参考 PAI Judge Model API） | -| **计费构成** | 评测集生成 + 应用调用 + 评估器模型的 Token 费用 | 被评测模型推理费用 + 裁判模型评分费用 | - -## 评分体系差异 - -| 对比项 | 应用评测 | 模型评测 | -|--------|----------|----------| -| **评分范围** | 1-5 分（正确率 = 得分 >= 4 的占比） | 可自定义整数区间（默认 0-5，建议不超过 10） | -| **自动评分方式** | LLM 评估器（语义）+ Code 评估器（规则） | 大模型评估（裁判模型）+ 规则评估（ROUGE/BLEU/Cosine/字符串匹配） | -| **评估器数量** | 每任务最多 10 个，建议组合 3-5 个 | 按评测维度配置，无上限说明 | -| **评分模型** | 评测集生成和评估仅支持 qwen-max 和 qwen-plus | 裁判模型推荐千问-Max，被评测模型不限 | +| 评测对象 | 已发布的智能体应用、工作流应用 | 文本生成类模型（含已部署的调优模型） | +| 评测方式 | 自动评测（LLM 生成评测集 + 自动评分）、手动评测（人工标注） | 自定义评测（自有数据集 + 自定义维度）、基线评测（公开标准数据集） | +| 前置条件 | 应用须已发布并配置知识库；须开通应用观测 | 拥有被评测模型的调用权限；基线评测仅北京地域可用 | +| 输入数据格式 | 对话分析 `.xls`/`.xlsx`；知识问答 `.jsonl`；新版支持智能体、工作流、自定义类型评测集 | 评测数据集（含 Prompt + Completion 列）；或推理结果集（已含 Output，免去推理费用） | +| 评测集创建方式 | 旧版：自动生成（基于知识库）或手动上传；新版：手动上传、从应用观测导入 | 在数据管理模块上传评测集类型数据 | +| 评分机制 | 评估器（LLM 评估器 + Code 评估器）+ 标签（分类、布尔值、数字、文本） | 评测维度（大模型评估、规则评估、人工评估） | +| 自动评分支持模型 | `qwen-max`、`qwen-plus` | 推荐千问-Max 作为裁判模型 | +| 人工标注 | 支持，新版通过评测任务 + 标签组合实现 | 支持人工评估维度（Pass/Fail 逐条标注） | +| 横向对比 | 支持最多 8 个应用（或同应用不同版本）横向评测 | 支持排行榜功能，多个任务结果参与排名 | +| 报告内容 | 总正确率、BadCase 分析、调优建议、RAG 归因分析 | 综合得分、通过率、逐条评分明细 | +| [计费](../concepts/billing.md)构成 | LLM 评估器产生 Token 费用；Code 评估器无额外费用 | 被评测模型推理费用 + 裁判模型评分费用；规则评估和人工评估无裁判模型费用 | +| API/SDK 支持 | 新版支持控制台操作 | 仅支持控制台操作，不提供公开 API/SDK | +| 地域限制 | 无特殊限制 | 基线评测仅北京地域可用 | +| 维度/评估器上限 | 每个评测任务最多 10 个评估器，建议组合 3-5 个 | 每个评测任务可关联多个评测维度模板 | + +## 评测流程对比 + +| 步骤 | 应用评测（自动模式） | 模型评测 | +|------|----------------------|----------| +| 1 | 创建评测任务 | 准备数据集（上传含 Prompt + Completion 的数据） | +| 2 | 设置评测集（自动生成或上传） | 创建评测维度（选择评分器类型并配置参数） | +| 3 | 配置评测规则（评估器 + 标签） | 创建评测任务（选模型、关联数据集和维度） | +| 4 | 执行评测，生成报告 | 提交评测，查看指标统计和数据明细 | + +## 评分方式对比 + +| 评分方式 | 应用评测 | 模型评测 | +|----------|----------|----------| +| LLM 语义评分 | LLM 评估器（相关性、有害性、幻觉检测） | 大模型评估-数值型（0-5 打分）、大模型评估-分类型（Pass/Fail 标签） | +| 规则/代码评分 | Code 评估器（Python 代码规则，格式校验、数值计算） | 规则评估-文本相似度（ROUGE/BLEU/Cosine）、规则评估-字符串匹配 | +| 人工评分 | 标签标注（分类、布尔值、数字、文本） | 人工评估-分类型（Pass/Fail） | ## 适用场景建议 -### 优先选择应用评测的场景 +| 场景 | 推荐能力 | 理由 | +|------|----------|------| +| 评估智能体应用的端到端回答质量 | 应用评测 | 支持基于知识库自动生成评测集，提供 RAG 归因分析定位检索/重排/切片问题 | +| 对比多个应用或同一应用不同版本的表现 | 应用评测 | 支持最多 8 个应用横向评测，适合选型决策和迭代验证 | +| 验证模型在公开基准上的基础能力 | 模型评测（基线评测） | 无需自备数据集，使用 C-Eval、MMLU、GSM8K 等标准数据集快速评估 | +| 选择最优模型或验证调优效果 | 模型评测（自定义评测） | 支持自定义数据集和评分维度，可对比不同模型的推理质量 | +| 需要格式校验、数值计算等确定性评分 | 两者均可 | 应用评测用 Code 评估器，模型评测用规则评估-字符串匹配/文本相似度 | +| 需要人工主观判断（创意写作、专业领域） | 两者均可 | 应用评测用标签人工标注，模型评测用人工评估-分类型 | +| 优化 RAG 流程（检索、重排、切片） | 应用评测 | 归因分析可定位到具体环节并提供调优建议 | +| 编程化、批量化的自动化评测 | 模型评测（参考 PAI Judge Model API） | 模型评测当前不提供公开 API，但可参考 PAI Judge Model API 实现编程化评测 | -- 智能体应用已发布上线，需要持续监控输出质量 -- 需要定位 RAG 链路中的具体瓶颈（检索、重排、切片、模型理解） -- 知识库更新或 Prompt 调整后需要回归验证 -- 多个应用版本之间需要横向对比，选出最优配置 -- 需要将人工标注经验固化为自动评估规则（通过评估器模板化） +## 数据量与成本建议 -### 优先选择模型评测的场景 - -- 项目初期的模型选型，需要在多个候选模型间对比基础能力 -- 模型微调（SFT）后需要验证调优效果是否达标 -- 使用公开基准（C-Eval、MMLU、GSM8K、BBH）快速了解模型通用能力 -- 需要用规则评估（ROUGE/BLEU）做确定性指标验证（如翻译、摘要场景） -- 关注模型推理能力本身，而非上层应用的端到端效果 - -### 组合使用建议 - -典型的开发流程中，两种评测可以分阶段配合使用：先通过**模型评测**完成基础模型选型和调优验证，确定最优模型后构建应用，再通过**应用评测**验证端到端效果并持续迭代优化。 - -## 成本优化对比 - -| 策略 | 应用评测 | 模型评测 | +| 维度 | 应用评测 | 模型评测 | |------|----------|----------| -| **减少推理费用** | 缩小评测集规模 | 使用推理结果集（复用已有推理输出） | -| **减少评分费用** | 使用 Code 评估器替代 LLM 评估器 | 使用规则评估或人工评估替代大模型评估 | -| **渐进式评测** | 先小规模自动评测，再针对 BadCase 人工复核 | 先 50-100 条验证，再扩大到 200-500 条正式评测 | +| 数据量建议 | 自动评测基于知识库生成，无明确条数限制；手动评测单文件不超过 20MB | 小规模验证 50-100 条，正式评测 200-500 条，全面评估 500 条以上 | +| 成本优化策略 | 组合 Code 评估器减少 LLM 调用；利用归因分析精准定位优化点 | 先小规模验证 → 保存推理结果集复用 → 确定性场景优先用规则评估 | +| 评分一致性注意 | 无特殊说明 | LLM 评分器存在位置偏差和自我偏好偏差，1-3% 分数差异属噪声，建议定期人工抽查校准 | -## 来源文档 +## 选型决策路径 -- [application evaluation](../guides/application-evaluation.md) (guides/application-evaluation.md) -- [model evaluation introduction](../guides/model-evaluation-introduction.md) (guides/model-evaluation-introduction.md) +1. **要评估的对象是应用还是模型？** 应用选应用评测，模型选模型评测。 +2. **是否有标准答案？** 有标准答案且格式固定优先规则评估（模型评测的字符串匹配 / 应用评测的 Code 评估器）；有标准答案但表述多样用文本相似度；无标准答案需语义理解用 LLM 评估器/大模型评估。 +3. **是否需要 RAG 归因分析？** 需要则选应用评测，其报告会定位 BadCase 到检索、重排、切片等具体环节。 +4. **是否需要横向对比？** 对比多个应用选应用评测（最多 8 个），对比多个模型选模型评测排行榜。 +5. **是否需要基线评测？** 需要公开基准数据集评估选模型评测的基线评测（仅北京地域）。 ## 被对比主题页 diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/evaluation-vs.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/evaluation-vs.md new file mode 100644 index 00000000..0d3cedc6 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/evaluation-vs.md @@ -0,0 +1,67 @@ +# 应用评测与模型评测对比 + +百炼平台提供两套独立的评测体系：**应用评测**针对智能体应用和工作流应用的端到端输出质量进行评估，**模型评测**针对底层大模型本身的推理能力进行评估。两者在评测对象、数据组织、评分机制和典型场景上有显著差异。本页帮助开发者根据实际需求选择合适的评测路径。 + +## 关键维度对比 + +| 维度 | 应用评测 | 模型评测 | +|------|----------|----------| +| 评测对象 | 已发布的智能体应用、工作流应用 | 文本生成类大模型（含已部署的调优模型） | +| 评测目标 | 应用端到端输出质量（含 RAG 检索、Prompt、知识库等环节） | 模型推理能力本身（问答质量、翻译、摘要等） | +| 评测模式 | 自动评测（LLM 生成评测集并打分）、手动评测（人工标注） | 自定义评测（自有数据集）、基线评测（公开标准数据集） | +| 评测集类型 | 旧版：对话分析(.xls/.xlsx)、知识问答(.jsonl)；新版：智能体、工作流、自定义 | 评测数据集（含 Prompt + Completion）或推理结果集（已含 Output） | +| 评测集来源 | 基于知识库自动生成、手动上传、从应用观测导入 | 数据管理模块上传、公开标准数据集（C-Eval、MMLU、GSM8K、BBH 等） | +| 评分方式 | LLM 评估器、Code 评估器、人工标注 | 大模型评估（数值型/分类型）、规则评估（文本相似度/字符串匹配）、人工评估 | +| 评分器配置 | 评估器（预置模板 / 自定义 LLM/Code / 基于历史任务创建），单任务最多 10 个 | 评测维度（五种类型，创建为模板可复用），维度类型创建后不可修改 | +| 参考答案要求 | 视评估器类型而定，LLM 评估器可不需参考答案 | 规则评估需参考答案，大模型评估和人工评估不需要 | +| 裁判模型 | 评测集生成和评估仅支持 qwen-max、qwen-plus | 推荐千问-Max，可配置评分器 Prompt | +| 标签体系 | 支持分类、布尔值、数字、文本四种标签类型，用于人工标注和数据筛选 | 支持 Pass/Fail 标签（分类型/人工评估） | +| 报告与归因 | 总正确率、BadCase 分析、RAG 归因分析、调优建议 | 综合得分、通过率、逐条评分明细、可选排行榜 | +| 基线评测 | 不适用 | 支持（C-Eval、MMLU、GSM8K、BBH 等，仅北京地域） | +| 多应用对比 | 支持最多 8 个应用横向评测 | 支持排行榜参与，多模型排名对比 | +| 计费方式 | LLM 评估器产生 [Token](../concepts/token.md) 费用，Code 评估器无额外费用 | 被评测模型推理费用 + 裁判模型评分费用；推理结果集可免推理费 | +| API/SDK | 控制台操作 | 控制台操作，无公开 API/SDK（可参考 PAI Judge Model API） | +| 地域限制 | 无特殊限制 | 基线评测仅北京地域可用 | + +## 评测流程对比 + +**应用评测**流程：创建评测任务 → 设置评测集 → 配置评测规则（评估器 + 标签）→ 执行评测 → 查看报告与归因分析。自动评测模式下，平台基于应用关联的知识库自动生成评测集，并由 LLM 完成打分；手动评测模式下，需人工下载模板填充数据并逐条标注。 + +**模型评测**流程：准备数据集 → 创建评测维度 → 创建评测任务（选模型 + 关联数据集和维度）→ 查看结果。自定义评测需开发者自行上传数据并配置评分规则；基线评测使用公开标准数据集，无需准备数据或配置维度，适合快速验证模型基础能力。 + +## 评分机制对比 + +应用评测的评估器体系更侧重于**应用层面的质量维度**，如回答相关性、格式规范性、幻觉检测等，并支持将人工标注经验固化为自动化 LLM 评估器。评估器分 LLM 和 Code 两类，建议组合 3-5 个从不同维度评估。 + +模型评测的评测维度体系更侧重于**模型推理能力的标准化评测**，提供五种明确的维度类型，且有清晰的选型决策路径：有标准答案且格式固定用字符串匹配；有标准答案但表述多样用文本相似度；无标准答案需语义理解用大模型评估；需主观判断用人工评估。 + +## 适用场景建议 + +| 场景 | 推荐评测方式 | 原因 | +|------|-------------|------| +| 验证智能体应用回答质量 | 应用评测（自动评测） | 可基于知识库自动生成评测集，含 RAG 归因分析定位检索/Prompt 问题 | +| 多版本智能体应用对比选型 | 应用评测（多应用横向评测） | 同一基准下对比最多 8 个应用，支持版本迭代验证 | +| 需要人工主观判断的评测 | 应用评测（手动评测 / 新版标签标注） | 支持自定义评测维度和标签体系，灵活适配领域专家判断 | +| 评估模型基础能力（MMLU/GSM8K 等） | 模型评测（基线评测） | 使用公开标准数据集，无需准备数据，快速获得模型能力基线 | +| 对比不同模型的问答/翻译/摘要能力 | 模型评测（自定义评测） | 支持规则评估和大模型评估，可绑定排行榜做多模型排名 | +| 有标准答案的确定性任务评测 | 模型评测（规则评估） | ROUGE/BLEU/Cosine 等算法直接计算，无裁判模型费用 | +| 无标准答案的语义质量评测 | 两者均可 | 应用评测用 LLM 评估器，模型评测用大模型评估维度 | +| 验证已部署调优模型效果 | 模型评测（自定义评测） | 已部署调优模型不额外计费，推理费用含在部署算力费用中 | + +## 成本优化建议 + +- **应用评测**：优先组合 Code 评估器（无额外费用）与少量 LLM 评估器，减少 [Token](../concepts/token.md) 消耗；利用应用观测数据导入评测集，避免重复生成。 +- **模型评测**：先用 50-100 条小规模验证，保存推理结果集复用以免去推理费用；有确定性标准的场景优先用规则评估（无裁判模型费用）；已部署调优模型评测不额外计费。 + +## 注意事项 + +- 应用评测存在新旧两套系统，新版在评测任务管理、评估器和标签体系上做了较大升级，手动评测属于旧版功能。 +- 模型评测当前仅支持文本生成类模型，基线评测仅北京地域可用，任务提交后不可更换目标模型。 +- 模型评测的 LLM 评分器存在位置偏差和自我偏好偏差，建议定期人工抽查校准；1-3% 的分数差异通常为评测噪声。 + +## 被对比主题页 + +- [application evaluation](../guides/application-evaluation.md) +- [model evaluation introduction](../guides/model-evaluation-introduction.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/fine-tuning-vs-compression-vs-deployment.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/fine-tuning-vs-compression-vs-deployment.md index 58583e15..a69f1adb 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/fine-tuning-vs-compression-vs-deployment.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/fine-tuning-vs-compression-vs-deployment.md @@ -1,68 +1,86 @@ -# 模型微调、压缩与部署对比 - -在百炼平台上，把一个模型从「原始能力」推向「专属生产服务」通常经过三个环节：**模型微调 → 模型压缩（可选）→ 模型部署**。三者构成一条完整的自定义模型生产链路，但目标、输入输出、计费方式和适用场景各不相同。本页横向对比这三个环节的关键差异，帮助开发者理清「先做什么、要不要做、怎么上线」的技术选型问题。 - -需要先明确三者的关系： - -- **模型微调**解决「模型能不能做这件事」——把领域知识、指令遵循、人类偏好或特定音色写入参数。 -- **模型压缩**解决「部署贵不贵」——把全精度微调模型量化为低精度版本，降低部署所需的 MU 规格与推理成本，是链路中的**可选**环节。 -- **模型部署**解决「怎么对外提供服务」——为预置模型或自定义模型分配资源专享的推理服务，满足高并发、低延迟需求。 - -> 三者均**仅在华北2（北京）地域可用**，且需使用该地域的 API Key；子账号（RAM 用户）需预先获得相应授权。 - -## 关键维度对比 - -| 维度 | 模型微调（Fine-tuning） | 模型压缩（量化） | 模型部署（Deployment） | -| --- | --- | --- | --- | -| 核心目标 | 定制模型能力（知识/指令/偏好/音色） | 降低部署规格与推理成本 | 提供资源专享的推理服务 | -| 链路位置 | 第一步（必选） | 中间步骤（可选） | 最后一步（上线必选） | -| 输入 | 训练/验证数据集（JSONL、ZIP、OSS 挂载） | 上游全精度微调模型 + 可选校准数据 | 预置模型或自定义（含压缩后）模型 | -| 输出 | 微调后模型（`finetuned_output`） | 低精度量化模型 | 专属推理服务（`deployed_model`） | -| 支持模型范围 | 千问文本/VL、万相图像/视频、CosyVoice 等多模态 | 仅平台微调产出的自定义模型（如 qwen3.5-flash） | 部分预置模型 + 所有调优后模型 | -| 主要方式 | CPT / SFT（全参、LoRA）/ DPO | 量化（不含剪枝、蒸馏） | PTU / 模型单元(MU) / 按 Token | -| API 端点 | `POST /api/v1/fine-tunes`（配合 `/files`） | 无公开 API，仅控制台创建压缩任务 | `POST/GET/DELETE /api/v1/deployments` | -| 控制台路径 | 模型调优页面 | 模型 → 模型训练 → 模型压缩 | 我的模型 / 部署页面 | -| 计费方式 | API 仅按 Token；训练单元须走控制台 | 压缩任务限时免费，成本体现在部署阶段 | PTU（按 TPM）/ MU（按时长×单元）/ 按 Token | -| 可逆性 | 可继续微调、可组合多阶段 | **不可逆**，不支持继续微调或二次压缩 | 可下线重建；计费方式创建后不可改 | -| 典型场景 | 效果不达标、需深度领域/风格定制 | 微调模型上线成本高、需降本 | 高并发/低延迟生产、专属推理服务 | - -## 各环节适用场景建议 - -### 模型微调：什么时候做 - -- 当 Prompt 工程、插件调用等手段**仍无法满足效果**时，才引入微调。 -- 文本生成推荐按 `CPT（可选）→ SFT → DPO（可选）` 递进组合：补领域知识用 CPT（千万级 Token 无标签文本），学会遵循指令用 SFT（1000+ 条 ChatML 问答对），对齐人类偏好用 DPO（100+ 组 chosen/rejected）。 -- 训练模式在模型支持全参时**优先选全参**（费用相同、效果更好）；数据集小或对时间/成本敏感则用 LoRA 高效训练。 -- 多模态定制（视觉理解、图像/视频、语音合成）走各自独立的超参与数据格式，其中 CosyVoice 当前**只能通过 API 发起**。 -- 入门建议用控制台零代码流程；批量/自动化建议走 API 四步流程（上传数据 → 创建任务 → 轮询状态 → 部署调用），注意 **API 创建的任务仅支持按 Token 计费**。 - -### 模型压缩：要不要做 - -- 仅当已有**平台微调产出的自定义模型**且**部署成本偏高**时才考虑，属于可选优化环节。 -- 收益示例：qwen3.5-flash 微调模型从 MU1*2（108 元/小时）压缩到 MU8*1（47 元/小时），成本节省约 56%。 -- 权衡精度与成本：量化模板中 MU 编号越大，部署规格越小、成本越低，但精度损失可能越大；用与推理场景语义相近的校准数据可提升量化精度。 -- **注意不可逆**：压缩后不能继续微调、不能二次压缩，调整需从上游全精度模型重新压缩。 -- 压缩任务当前**限时免费**，建议在免费期内对同一模型尝试多个量化模板，用业务测试集验证后再选最优方案上线。 - -### 模型部署：怎么上线 - -按流量特征选择互斥的计费方式（创建后不可更改，切换须先下线再重部署）： - -- **预置吞吐（PTU）**：预留资源保障 TPM，额度内不限速，TPS 通常提升 1.5～2.0 倍，支持长输入（部分模型达 200K）与前缀缓存折扣。适合流量可预估的高负载生产（智能客服、实时内容审核）。超额或超长输入自动转按量计费，业务不中断。 -- **模型单元（MU）**：按时长×单元数计费，资源独占、性能可自定义，支持 PD 分离降低首 Token 延迟。适合需要独占资源、性能可控的生产场景，也是压缩后模型的落地方式。 -- **按 Token 使用量**：不使用不计费，仅支持基础模型 SFT 高效训练后的自定义模型，主要用于**调优效果验证**；扩缩容需控制台人工审核。 - -LoRA 模型导入需满足约束：仅支持 LoRA（不支持全参微调）、rank 为 8/16/32/64、不得修改 vocab 与 chat_template、VL 模型须冻结 VIT。API 部署通过 `plan` 字段区分计费方式（`ptu` / `mu` / `lora`），流程为创建部署 → 轮询至 `RUNNING` → 调用推理 → 用完 DELETE 下线停止计费。 - -## 技术选型参考 - -1. **只想验证微调效果**：微调（SFT 高效训练）→ 直接用「按 Token」部署验证，成本最低、无需长期占用资源。 -2. **要正式上线且流量可预估**：微调 →（评估成本后可选压缩）→ PTU 部署，享受不限速吞吐与缓存折扣。 -3. **要独占资源、性能可控**：微调 → 压缩降本 → MU 部署，用量化模板在成本与精度间取平衡。 -4. **本地已训练 LoRA**：跳过平台微调，直接经「我的模型」导入（满足 rank 与配置约束）→ 部署。 -5. **成本敏感的高负载场景**：优先评估「压缩 + PTU/MU」组合，先在压缩免费期内多模板试验，用业务测试集选出精度/成本最优版本再上线。 - -总体决策顺序：先用微调确保能力达标，再判断是否需要压缩降本，最后按流量与性能诉求选择部署计费方式。 +# 模型微调、压缩与部署方案对比 + +在百炼平台的模型生产链路中，微调、压缩与部署是三个相互衔接的核心环节。微调负责让基础模型习得特定领域的知识或能力；压缩是可选的优化步骤，用于降低已微调模型的推理成本；部署则将最终模型暴露为可调用的推理服务。理解三者的定位与差异，有助于开发者在成本、性能和效果之间做出合理的技术选型。 + +## 核心维度对比 + +| 维度 | 模型微调（Fine-tuning） | 模型压缩（Compression） | 模型部署（Deployment） | +|------|------------------------|------------------------|----------------------| +| **主要目的** | 让模型学习领域知识、指令遵循或偏好对齐 | 降低已微调模型的部署规格与推理成本 | 将模型发布为独立推理服务，接受线上调用 | +| **在链路中的位置** | 起点：基础模型 → 自定义模型 | 中间（可选）：微调模型 → 压缩模型 | 终点：（压缩）模型 → 在线服务 | +| **是否必须** | 可选（Prompt 工程等无法满足时才用） | 可选（仅用于降成本） | 必须（用模型前必须部署） | +| **支持的模型类型** | 预置基础模型（文本/图像/视频/语音） | 仅百炼平台微调产出的自定义模型 | 预置模型 + 微调/压缩后的自定义模型 | +| **操作入口** | 控制台 或 API（`POST /api/v1/fine-tunes`） | 仅控制台（控制台 → 模型训练 → 模型压缩） | 控制台 或 API（`POST /api/v1/deployments`） | +| **计费方式** | 按训练 Token 消耗计费（API 方式）；控制台支持训练单元 | 压缩任务本身限时免费；部署后按压缩后的 MU 规格计费 | PTU（预置吞吐）/ 模型单元（按时长）/ 按 Token 用量 | +| **地域限制** | 仅华北2（北京） | 仅华北2（北京） | 仅华北2（北京）（API 部署） | +| **可逆性** | 训练完成后可继续微调或压缩 | 不可逆：压缩后不能继续微调或二次压缩 | 可下线后重新部署 | +| **耗时** | 较长（依数据集规模，从数小时到数天） | 中等（排队 + 量化计算） | 较快（分钟级变为 RUNNING） | +| **典型输入** | 训练数据集（JSONL / ZIP）+ 超参配置 | 微调模型 ID + 量化模板 + 可选校准数据 | 模型 ID + 计费方式 + 资源规格 | +| **典型输出** | 自定义模型（`finetuned_output`） | 低精度压缩模型（MU 规格降低） | 推理服务端点（`deployed_model` ID） | + +## 调优方式对比（微调内部） + +| 调优方式 | 代码 | 典型数据量 | 适用场景 | +|----------|------|-----------|---------| +| 继续预训练 | `cpt` | ≥ 1000 万 Token 无标签文本 | 补充垂直领域背景知识 | +| 监督微调（全参） | `sft` | ≥ 1000 条问答对 | 指令遵循、对话风格塑造 | +| 高效微调（LoRA） | `efficient_sft` | ≥ 1000 条问答对 | 数据集较小或快速验证 | +| 直接偏好优化 | `dpo_full` / `dpo_lora` | ≥ 100 组 chosen/rejected 对 | 对齐人类偏好、减少有害输出 | + +## 部署计费方式对比 + +| 计费方式 | 适用场景 | 性能可控性 | 备注 | +|----------|---------|-----------|------| +| 预置吞吐（PTU） | 高负载、流量稳定的生产环境 | 吞吐/并发由平台预置，超出可自动溢出 | 支持前缀缓存、长输入阶梯系数 | +| 模型单元（MU） | 大规模推理、需自定义延迟吞吐 | 完全自定义，资源独占 | 微调/压缩模型必选此方式才能独占资源 | +| 按 Token 用量 | 效果验证、低并发场景 | 不可调 | 仅支持部分高效微调（LoRA）模型 | + +## 压缩对部署成本的影响示例 + +| 模型 | 压缩前部署规格 | 压缩前成本 | 压缩后部署规格 | 压缩后成本 | 成本节省 | +|------|--------------|----------|--------------|----------|---------| +| qwen3.5-flash-2026-02-23 微调模型 | MU1×2 | ¥108/小时 | MU8×1 | ¥47/小时 | ≈56% | + +## 各方案适用场景建议 + +### 只需要部署（无微调需求） +直接选择部署预置模型。对于流量稳定、吞吐要求高的生产业务，优先选择 PTU 计费；需要完全自定义性能参数时选择模型单元；效果验证阶段选择按 Token 用量（最灵活，按需付费）。 + +### 需要微调后部署 +1. 评估是否真的需要微调（Prompt 工程 / Function Calling 能解决的问题不要微调）。 +2. 选择调优方式：数据量大、效果优先 → 全参 SFT；数据量小、快速迭代 → LoRA；需要对齐偏好 → SFT 后接 DPO。 +3. 微调完成后直接部署为模型单元服务，若对推理成本敏感则先做压缩再部署。 + +### 需要控制推理成本 +在微调完成后，利用免费期对同一模型尝试多个量化模板，分别部署后用业务测试集评估效果，选择成本与精度最优的方案。注意：压缩不可逆，一旦压缩无法继续微调，务必在压缩前保留全精度微调模型。 + +### 本地训练的 LoRA 模型上云 +通过 OSS 导入功能将本地 LoRA 权重导入百炼平台，再以"按 Token 用量"或"模型单元"方式部署。需满足 rank 限制（8/16/32/64）、不可修改词汇表等约束。 + +## 技术选型决策树 + +``` +业务需求 +├── 基础模型能力已满足需求？ +│ └── 是 → 直接部署预置模型 +│ ├── 流量稳定、高吞吐 → PTU +│ ├── 需自定义性能 → 模型单元 +│ └── 效果验证 → 按 Token 用量 +└── 否 → 需要微调 + ├── 选调优方式（CPT / SFT / DPO） + ├── 微调完成 → 推理成本是否敏感？ + │ ├── 是 → 压缩（量化）→ 部署（模型单元） + │ └── 否 → 直接部署（模型单元） + └── 本地 LoRA 模型 → OSS 导入 → 部署 +``` + +## 关键注意事项 + +- **地域统一**：微调、压缩、部署均仅支持华北2（北京），API Key 须属于该地域。 +- **子账号授权**：RAM 用户需提前授予模型调用、训练、部署相关权限。 +- **计费方式锁定**：部署后计费方式不可更改，切换须先下线再重新部署，下线后即停止计费。 +- **压缩不可逆**：压缩后的模型无法继续微调或二次压缩，操作前须确认全精度模型已备份。 +- **按量微调限制**：通过 API 创建的文本生成训练任务仅支持按 Token 计费，如需训练单元须在控制台操作。 ## 被对比主题页 diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/fine-tuning-vs-compression.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/fine-tuning-vs-compression.md index 025bfd66..c1e1d670 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/fine-tuning-vs-compression.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/fine-tuning-vs-compression.md @@ -1,51 +1,55 @@ # 模型微调与模型压缩对比 -[模型微调（Fine-tuning）](../concepts/fine-tuning.md)与模型压缩（量化）是百炼平台模型生产链路中两个相邻但目标截然不同的环节。微调解决的是「模型能不能做好某项任务」的**能力问题**，通过训练修改模型参数来提升特定行业/业务表现；模型压缩解决的是「模型部署贵不贵」的**成本问题**，通过量化把全精度微调模型转为低精度版本，从而降低部署所需的 MU 规格。二者在完整链路中的位置为：**模型调优 →（可选）模型压缩 → 模型部署**——压缩的输入正是微调的产出。本文从技术选型角度对比两者的关键差异，帮助开发者判断在什么阶段该用哪个能力。 +模型微调（Fine-tuning）和模型压缩（Model Compression）是百炼平台模型生产链路中的两个关键环节：微调用于让模型学习新的知识、能力或风格，压缩用于降低微调模型的部署成本。两者并非互斥关系，而是串联协作——完整链路为"模型调优 → 模型压缩（可选）→ 模型部署"。本文从功能定位、输入输出、支持范围、计费和典型场景等维度进行对比，帮助开发者在技术选型时做出合理决策。 ## 关键维度对比 -| 维度 | [模型微调（Fine-tuning）](../concepts/fine-tuning.md) | 模型压缩（量化） | -| --- | --- | --- | -| 核心目的 | 提升模型在特定任务/领域的能力 | 降低部署规格与推理成本 | -| 处理对象 | 基础模型（千问系列、VL、图像/视频、语音等） | 仅限百炼平台微调产出的自定义模型 | -| 输入 | 训练数据集（ChatML / 纯文本 / ZIP 音视频等） | 上游全精度微调模型 + 可选校准数据集 | -| 输出 | 新的自定义微调模型 | 低精度（量化）版本的自定义模型 | -| 技术手段 | CPT / SFT / DPO（全参或 LoRA 高效训练） | 量化（不含结构剪枝、知识蒸馏） | -| 使用方式 | 控制台可视化 或 API/命令行（DashScope HTTP） | 控制台：模型 → 模型训练 → 模型压缩 → 创建压缩任务 | -| API 端点 | `POST /api/v1/files`、`/api/v1/fine-tunes`、`/api/v1/deployments` | 以控制台操作为主（未提供公开 API 流程） | -| 关键配置 | learning_rate、n_epochs、max_length、lora_rank、training_type 等 | 量化模板（MU 编号）、量化产出后缀、校准数据 | -| 计费方式 | 按训练 Token 计费（CosyVoice 0.2 元/千 Tokens；部署另计） | 压缩任务本身限时免费；压缩后模型按部署 MU 规格计费 | -| 地域限制 | 仅华北2（北京），须用该地域 API Key | 仅华北2（北京） | -| 可逆性/复用 | 微调模型可继续被压缩、部署 | 不可逆；压缩后不支持继续微调或二次压缩 | -| 典型收益 | 指令遵循、领域知识、偏好对齐、风格定制 | 部署成本下降（示例约节省 56%） | -| 典型场景 | 通用模型无法满足行业需求、需注入专业知识或对齐偏好 | 微调模型已达标、需在保持能力前提下压低上线成本 | +| 维度 | 模型微调 | 模型压缩 | +|------|---------|---------| +| 功能定位 | 改变模型能力：注入领域知识、学习任务格式、对齐偏好、复现风格/IP | 降低部署成本：将全精度微调模型量化为低精度版本，减少推理所需 MU 规格 | +| 技术手段 | CPT（继续预训练）、SFT（监督微调）、DPO（直接偏好优化）、LoRA 高效微调 | 量化（降低参数精度），不涉及结构剪枝或知识蒸馏 | +| 输入 | 训练数据集（ChatML JSONL、纯文本 JSONL、图像/视频/音频素材等） | 已微调完成的全精度自定义模型 + 量化模板 + 可选校准数据 | +| 输出 | 新的自定义模型 ID（需部署后调用） | 低精度版本的自定义模型（需部署后调用） | +| 支持的模型类型 | 文本生成模型、千问 VL 视觉理解模型、万相图像/视频生成模型、CosyVoice 语音合成模型 | 仅百炼平台微调产出的自定义模型（当前已知支持 qwen3.5-flash-2026-02-23 微调模型） | +| 可用地域 | 华北2（北京）为主（万相图像/视频微调、CosyVoice 仅北京）；OSS 数据支持北京和新加坡 | 仅华北2（北京） | +| 发起方式 | 控制台 + API（`/api/v1/fine-tunes`）；CosyVoice 仅 API | 仅控制台（模型 → 模型训练 → 模型压缩） | +| 计费方式 | 按训练资源消耗计费（训练费用因模型和训练方式而异） | 压缩任务限时免费；部署阶段按压缩后更低的 MU 规格计费 | +| 成本效果示例 | — | qwen3.5-flash 微调模型：MU1*2（108 元/小时）→ MU8*1（47 元/小时），节省约 56% | +| 可逆性 | 产出模型可继续微调（支持多轮迭代训练） | 不可逆：压缩后模型不能继续微调，也不能二次压缩 | +| 任务状态 | 创建 → 运行 → 成功/失败 | PENDING → QUEUING → RUNNING → SUCCEEDED / FAILED / CANCELING → CANCELED | +| 对模型能力的影响 | 增强或改变模型能力（知识、格式、风格、偏好） | 可能有精度损失，MU 编号越大成本越低但精度损失风险越大 | ## 适用场景建议 -### 优先选择模型微调 +### 选择模型微调的场景 -- Prompt 工程、插件调用等方法已用尽，仍无法满足业务对准确率、专业性或风格的要求。 -- 需要向模型注入领域词汇与事实（CPT）、教会模型遵循特定指令或执行任务（SFT）、或对齐人类偏好并抑制幻觉（DPO）。这三种方式可按 `CPT（可选）→ SFT → DPO（可选）` 递进组合。 -- 涉及多模态定制：视觉理解（VL，仅 SFT）、图像/视频生成（SFT-LoRA，需触发词）、语音合成（CosyVoice，仅 API 发起）。 -- 建议：若模型支持全参训练则优先全参（效果更好且与高效训练计费相同）；对成本/时间敏感或数据集较小时选 LoRA 高效训练。 +- **行业知识适配**：基础模型缺乏特定领域知识，需要通过 CPT 补充大规模领域语料。 +- **任务格式与流程定制**：需要模型严格遵循特定对话格式、业务流程或安全合规要求（SFT）。 +- **输出偏好对齐**：需要模型的输出风格、质量更符合人类偏好（DPO）。 +- **视觉风格复现**：固定 IP 形象、画风、特效或动作复现（万相图像/视频 LoRA 微调）。 +- **专属音色定制**：同一发音人的高还原度语音定制（CosyVoice 调优）。 -### 优先选择模型压缩 +### 选择模型压缩的场景 -- 已经拥有一个训练达标的全精度微调模型，主要痛点是**部署/推理成本偏高**。 -- 希望在尽量保持模型能力的前提下降低部署 MU 规格（MU 编号越大规格越小、成本越低，但精度损失可能越大）。 -- 建议：利用压缩任务限时免费的窗口，对同一微调模型尝试多个量化模板，分别部署后用业务测试集验证推理效果，选出成本与精度平衡最优的方案再正式上线；若面向特定场景（如客服问答），校准数据应选语义相近的数据集以提升量化精度。 +- **降低推理成本**：微调模型效果已满足需求，但部署规格过高、推理成本不可接受。 +- **规模化部署**：需要在有限预算内部署更多模型实例。 +- **多模板实验**：利用免费期对同一微调模型尝试多个量化模板，用业务测试集验证后选择最优方案。 -### 二者结合的典型链路 +### 组合使用的典型链路 -对绝大多数生产落地而言，两者不是「二选一」而是「先后使用」：先用**微调**把模型能力打磨到位并验证效果，再用**压缩**在上线前压低部署成本。需特别注意压缩不可逆——若后续还想继续微调或调整，必须回到上游全精度微调模型重新压缩，因此应先冻结微调版本、确认效果达标后再进入压缩环节。 +1. 准备数据集，创建微调任务（SFT/LoRA 等），获得效果达标的全精度微调模型。 +2. 评估部署成本：若 MU 规格超出预算，创建压缩任务选择合适的量化模板。 +3. 使用与目标场景语义相近的校准数据提升量化精度。 +4. 部署压缩后模型，用业务测试集验证推理效果是否仍在可接受范围内。 -## 技术选型速查 +## 技术选型注意事项 -- 目标是「让模型更会做事」→ 微调。 -- 目标是「让模型更省钱部署」→ 压缩。 -- 需要通过 API/命令行自动化全流程（上传数据 → 训练 → 部署）→ 微调具备完整 HTTP 接口；压缩目前以控制台操作为主。 -- 数据集较小或预算有限 → 微调选 LoRA 高效训练；压缩选较大 MU 编号模板并做好精度验证。 -- 两者均受**华北2（北京）地域**约束，须使用该地域 API Key 与工作空间。 +1. **先微调、后压缩**：压缩只能作用于微调产出模型，不能直接压缩基础模型或第三方模型。 +2. **压缩不可逆**：如需调整量化方案或继续迭代训练，必须从上游全精度微调模型重新操作。建议在确认模型效果稳定后再压缩。 +3. **精度与成本的权衡**：量化模板 MU 编号越大，部署成本越低，但精度损失可能越大。务必用真实业务数据验证压缩后模型的推理质量。 +4. **训练方式选择**：微调阶段如果成本敏感或需要快速验证，优先选择高效训练（LoRA）；生产环境对效果要求高且模型支持时，可评估全参训练。 +5. **地域限制**：两个功能当前均主要在华北2（北京）地域可用，创建任务前需确认使用对应地域的 API Key 和资源。 +6. **数据准备差异**：微调需要准备符合格式要求的训练数据集（不同模型类型格式差异大）；压缩的校准数据为可选项，但建议使用与推理场景相近的数据集以提升量化精度。 ## 被对比主题页 diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/frameworks-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/frameworks-comparison.md new file mode 100644 index 00000000..6fbbc367 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/frameworks-comparison.md @@ -0,0 +1,47 @@ +# 框架与工具包对比 + +百炼平台为开发者提供了两条主要的框架集成路径：一是通过官方适配的开源框架（LlamaIndex、Spring AI Alibaba）直接调用百炼云端能力，二是通过 [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)将已有应用迁移至百炼并接入 LangChain 等主流框架。两者在定位、语言生态、功能范围和迁移成本上存在显著差异。本页面帮助开发者根据技术栈和业务需求选择合适的集成方案。 + +## 关键维度对比 + +| 维度 | LlamaIndex + Spring AI Alibaba（官方框架适配） | [OpenAI 兼容接口](../concepts/openai-compatible-interface.md) + LangChain（兼容迁移路线） | +| --- | --- | --- | +| 定位 | 深度集成百炼云端知识库、智能体与工作流 | 以 OpenAI 兼容协议接入百炼模型推理能力 | +| 支持语言 | Python（LlamaIndex）、Java（Spring AI Alibaba） | 多语言（Python/Node.js/Go 等任意支持 OpenAI SDK 的语言） | +| 核心能力 | 云端知识库构建、RAG 应用、智能体/工作流调用、知识库检索 | Chat Completions、Responses、Completions、Embedding、Vision、文件、Batch、Conversations | +| 知识库支持 | 原生支持云端知识库（智能切分 + 官方向量模型） | 通过 Embedding 接口 + 文件接口可自建知识库管线 | +| 自定义切分/嵌入 | LlamaIndex 云端方案不支持；需改用本地知识库方案 | 完全可控，可自定义切分策略与嵌入模型 | +| 模型范围 | LlamaIndex：qwen-max 等对话模型；Spring AI Alibaba：智能体/工作流应用 | Qwen 全系列（含 VL/Coder/Omni/Math）、DeepSeek、Kimi、GLM、MiniMax 等三方模型 | +| [流式输出](../concepts/streaming.md) | Spring AI Alibaba 支持流式（`Flux`，`text/event-stream`） | Chat Completions 支持 `stream=True`；Responses 支持流式 | +| 迁移成本 | 需学习百炼专用 SDK 与配置方式 | 已有 OpenAI 应用仅需替换 `api_key`、`base_url`、`model` 三项 | +| 鉴权方式 | [API Key](../concepts/api-key.md)（环境变量 `DASHSCOPE_API_KEY` 等） | [API Key](../concepts/api-key.md)（环境变量 `DASHSCOPE_API_KEY`） | +| 地域支持 | LlamaIndex 按百炼通用约定；Spring AI Alibaba 需配置[业务空间](../concepts/workspace.md) ID | 北京、新加坡、东京、法兰克福、弗吉尼亚等多地域，各地域 [API Key](../concepts/api-key.md) 不同 | +| 典型场景 | 企业级 RAG 应用、Java/Spring 生态集成百炼智能体 | OpenAI 应用迁移、多模型统一调用、代码补全、批量推理、视觉理解 | + +## 各方案适用场景建议 + +### LlamaIndex（Python） + +适合使用 Python 技术栈、希望快速搭建 RAG 应用的开发者。核心优势是将本地文件上传至百炼云端自动构建知识库，无需自行管理切分与向量模型。但云端方案不支持自定义切分和嵌入模型，对切分粒度有精细控制需求时应改用本地知识库方案。 + +### Spring AI Alibaba（Java） + +适合 Java/Spring Boot 生态的企业应用，需要直接调用百炼控制台中已创建的智能体应用或工作流应用。支持流式与非流式调用，可获取文档引用与思考过程元数据。也支持检索百炼知识库。前提是 JDK 17+ 与 Spring Boot 3.x，并需提前在控制台创建应用获取 App ID。 + +### [OpenAI 兼容接口](../concepts/openai-compatible-interface.md) + LangChain + +适合已有 OpenAI 应用希望低成本迁移到百炼的场景，或需要使用多语言 SDK 的项目。通过替换三项配置即可完成迁移，同时支持 LangChain/LangChain4j 等主流框架。该路线覆盖能力最广（对话、补全、嵌入、视觉、文件、批量），但不提供百炼云端知识库的托管能力，知识库需自行构建。 + +## 选型建议 + +- **已有 OpenAI 应用或使用 LangChain 生态**：优先选择 OpenAI 兼容接口路线，迁移成本最低，模型覆盖最广。 +- **Python 新项目、需要云端托管知识库**：选择 LlamaIndex，快速构建 RAG 应用，无需运维基础设施。 +- **Java/Spring 企业应用、需要调用百炼智能体或工作流**：选择 Spring AI Alibaba，与 Spring 生态无缝集成。 +- **需要自定义文档切分或嵌入模型**：OpenAI 兼容接口 + 自建知识库管线，或 LlamaIndex 本地知识库方案。 + +## 被对比主题页 + +- [frameworks](../api/frameworks.md) +- [toolkits and frameworks](../api/toolkits-and-frameworks.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/frameworks-vs.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/frameworks-vs.md new file mode 100644 index 00000000..c8fa9876 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/frameworks-vs.md @@ -0,0 +1,46 @@ +# 框架与工具包对比 + +阿里云百炼提供两条不同的框架与工具包集成路径：一条是通过 **LlamaIndex / Spring AI Alibaba** 等原生框架深度集成百炼的云端知识库与智能体应用能力；另一条是通过 **OpenAI 兼容接口**体系，以最小改动迁移已有 OpenAI 应用或接入 LangChain 等主流框架。本文从集成方式、语言生态、核心能力、适用场景等维度对两条路径进行对比，帮助开发者根据自身技术栈和业务需求做出选型。 + +## 关键维度对比 + +| 维度 | [frameworks](../api/frameworks.md)（原生框架集成） | toolkits-and-[frameworks](../api/frameworks.md)（OpenAI 兼容接口） | +| --- | --- | --- | +| **集成方式** | 使用框架原生 SDK 深度对接百炼云端知识库、智能体应用、工作流 | 替换 `api_key`、`base_url`、`model` 三项即可迁移已有 OpenAI 应用 | +| **语言/生态** | Python（LlamaIndex）、Java（Spring AI Alibaba，JDK 17+） | 任意支持 OpenAI SDK 的语言；可直接接入 LangChain / LangChain4j | +| **核心能力** | 云端知识库构建、RAG 检索引擎、智能体应用调用（流式/非流式）、工作流应用调用 | Chat Completions、Responses、Completions、Embedding、文件上传、Batch 批量推理、Conversations 会话管理、Vision 视觉理解 | +| **知识库** | 支持将本地文件上传到百炼构建云端知识库，使用默认智能切分与官方向量模型；Spring AI Alibaba 可检索百炼知识库 | 不直接提供知识库构建能力，但可通过 Embedding 接口自行实现向量检索 | +| **文档切分/嵌入模型** | 云端方案不支持自定义切分与嵌入模型；需本地知识库方案方可灵活控制 | 可自由选择嵌入模型（text-embedding-v1~v4），v3/v4 支持自定义向量维度 | +| **智能体/工作流** | Spring AI Alibaba 直接调用百炼智能体应用与工作流应用，获取思考过程、文档引用等元数据 | Responses API 提供内置联网搜索、网页抓取、代码解释器等工具，通过 `previous_response_id` 自动管理多轮上下文 | +| **API 端点** | 通过框架 SDK 封装调用百炼原生接口 | 统一使用 `/compatible-mode/v1` 路径，HTTP 调用追加具体资源路径 | +| **鉴权** | API Key 配置到环境变量（`DASHSCOPE_API_KEY` 等） | API Key 配置到环境变量 `DASHSCOPE_API_KEY`；各地域 API Key 不同，切换地域需同步更换 | +| **多地域支持** | 主要面向百炼主站 | 支持华北2（北京）、新加坡、日本（东京）、德国（法兰克福）、美国（弗吉尼亚）等多地域，含[业务空间](../concepts/workspace.md)专属域名 | +| **迁移成本** | 需按框架文档搭建工程、配置依赖、编写集成代码 | 已有 OpenAI 应用通常零代码改动，仅改配置即可完成迁移 | +| **典型场景** | 从零构建 RAG 应用、集成百炼智能体/工作流到 Java/Python 工程 | 迁移已有 OpenAI 应用、多语言接入、使用 LangChain 生态、批量推理、代码补全 | + +## 适用场景建议 + +### 选择 [frameworks](../api/frameworks.md)（原生框架集成）的情况 + +- **需要使用百炼云端知识库**：希望将本地文档上传到百炼、由平台托管知识库并完成智能切分与向量索引，无需自建向量数据库。 +- **需要调用百炼智能体应用或工作流应用**：已在百炼控制台创建了智能体或工作流，希望在 Java 后端通过 Spring AI Alibaba 集成，获取[流式输出](../concepts/streaming.md)、思考过程等结构化结果。 +- **使用 Python 构建 RAG**：技术栈为 Python，希望用 LlamaIndex 快速搭建[检索增强生成](../concepts/rag.md)应用，复用百炼的向量模型与重排能力。 + +### 选择 toolkits-and-frameworks（OpenAI 兼容接口）的情况 + +- **已有 OpenAI 应用需要迁移**：应用已使用 OpenAI SDK，希望以最低成本切换到百炼通义千问等模型，不改动业务逻辑。 +- **多语言或 LangChain 生态**：使用的语言或框架不在原生框架覆盖范围内（如 Go、Node.js、LangChain/LangChain4j），通过 OpenAI 兼容接口可无缝接入。 +- **需要多地域部署**：业务需要在中国内地、新加坡、日本、德国、美国等多地域调用模型，通过切换 `base_url` 即可实现。 +- **需要批量推理或代码补全**：有大规模批量推理需求（Batch 接口）或代码补全场景（Completions 接口，支持 `qwen-coder-turbo`）。 +- **需要 Responses API 的高级能力**：希望使用内置联网搜索、网页抓取、代码解释器等工具，以及通过 `previous_response_id` 自动管理多轮上下文。 + +### 两者结合使用 + +在实际项目中，两条路径并非互斥。常见的组合方式是：使用 **toolkits-and-frameworks** 的 OpenAI 兼容接口完成模型调用（Chat Completions、Embedding 等），同时使用 **frameworks** 的 LlamaIndex 或 Spring AI Alibaba 对接百炼云端知识库与智能体应用，兼顾迁移便利性与平台原生能力。 + +## 被对比主题页 + +- [frameworks](../api/frameworks.md) +- [toolkits and frameworks](../api/toolkits-and-frameworks.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/generation-api-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/generation-api-comparison.md new file mode 100644 index 00000000..2c1106b3 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/generation-api-comparison.md @@ -0,0 +1,73 @@ +# 生成类 API 对比 + +百炼平台提供了三大类生成式 AI API——图像生成、3D 资产生成和视频生成。它们在输入方式、输出产物、调用模式、支持模型和适用场景上各有差异。本页对三类 API 做横向对比，帮助开发者根据业务需求快速选型。 + +## 关键维度对比 + +| 维度 | 图像生成 | 3D 生成 | 视频生成 | +| --- | --- | --- | --- | +| **输入格式** | 文本提示词（[prompt](../guides/prompt.md)）；参考图 URL（图生图/编辑）；多图参考 | 文本 [prompt](../guides/prompt.md)（≤1024 字符）；单张图像 URL；多图（固定 4 张：前/左/后/右视角） | 文本 [prompt](../guides/prompt.md)；media 多模态数组（首帧/尾帧/参考图/视频/音频）；直接 URL 字段（旧版协议） | +| **输出格式** | 图像文件（PNG/JPG），像素范围 512×512 ~ 2048×2048，1~6 张 | GLB 格式 3D 模型（PBR 材质或无贴图基础模型）+ 预览渲染图 | 视频文件，分辨率 480P~4K，时长 5~8 秒（视模型而定） | +| **调用模式** | 混合：wan2.6/2.7、z-image、qwen-image 支持同步调用；长耗时任务走异步 | 仅异步（必须 `X-DashScope-Async: enable`） | 仅异步（必须 `X-DashScope-Async: enable`） | +| **API 端点** | 多个端点（按模型系列不同）；同步与异步并存 | `POST .../aigc/video-generation/3d-generation` + `GET .../tasks/{task_id}` | `POST .../aigc/video-generation/video-synthesis` + `GET .../tasks/{task_id}` | +| **支持模型数** | 最多（Qwen-Image、万相全系列、z-image、Kling、Vidu、创意工具等 20+ 模型） | 最少（Tripo-H3.1、Tripo-P1.0 两个模型） | 较多（万相 wan2.7/2.6/2.2、HappyHorse、PixVerse、Vidu、Kling、人像动画系列等） | +| **地域限制** | 部分北京独占（创意工具、wanx-v1）；其余支持多地域 | 仅华北2（北京） | 北京、新加坡、美国（弗吉尼亚）、德国（法兰克福） | +| **task_id 有效期** | 视模型而定（异步任务一般为 24 小时） | 24 小时 | 24 小时 | +| **产物下载有效期** | 视模型而定 | 2 小时 | 视模型而定 | +| **轮询建议间隔** | 视模型而定 | 约 15 秒 | 视模型而定（查询 RPS 限制 20） | +| **计费方式** | 按张计费（如 wanx-v1 0.16 元/张）或按调用次数 | 按任务类型（text-to-3d / image-to-3d / multi-image-to-3d）计费 | 按任务类型与分辨率/时长计费 | +| **同步调用支持** | 是（wan2.6/2.7、z-image、qwen-image 系列新模型） | 否 | 否 | +| **任务状态枚举** | — | PENDING → RUNNING → SUCCEEDED / FAILED / CANCELED / UNKNOWN | PENDING → RUNNING → SUCCEEDED / FAILED / CANCELED / UNKNOWN | + +## 各方案适用场景建议 + +### 图像生成 + +- **适用场景**：营销素材批量生成、商品图编辑、内容配图、创意海报、虚拟试衣、图像翻译等。 +- **选型建议**： + - 需要复杂文本渲染与段落排版 → Qwen-Image 系列。 + - 追求高性价比、轻量快速 → z-image-turbo。 + - 需要多图参考、4K 高清输出 → 万相 wan2.7-image-pro。 + - 需要风格化、去水印、超分等专项编辑 → 万相通用图像编辑 `wanx2.1-imageedit`。 + - 第三方风格偏好 → Kling 或 Vidu 系列。 + +### 3D 生成 + +- **适用场景**：游戏资产生成、电商 3D 展示、建筑可视化、教育/文物数字化的快速 3D 建模。 +- **选型建议**： + - 需要高精度、高面数模型 → Tripo-H3.1（最高 200 万面，支持 `ultra` 几何精度）。 + - 追求生成速度、对面数要求不高 → Tripo-P1.0（最高 2 万面，速度更快）。 + - 需要带贴图的 PBR 材质模型 → `pbr` 设为 `true`（默认）。 + - 只需基础白模 → `texture` 和 `pbr` 同时设为 `false`。 + - 注意：仅限北京地域 API Key 调用，需提前在控制台开通 Tripo 服务。 + +### 视频生成 + +- **适用场景**：短视频内容创作、广告生成、视频编辑、数字人/口型替换、视频超清、动作模仿等。 +- **选型建议**： + - 需要最新协议、多能力集成 → 万相 wan2.7（文生/图生/参考/编辑，推荐）。 + - 需要首尾帧、多镜头分镜 → Kling kling-v3 系列。 + - 需要 4K 超清、对口型、动作模仿 → PixVerse 系列。 + - 数字人/人像动画 → animate-anyone、emo、liveportrait 等人像动画系列。 + - 注意：所有视频 API 均为异步调用，跨地域调用会失败，务必保证模型、域名、API Key 同地域。 + +## 技术选型参考 + +| 需求特征 | 推荐方案 | 理由 | +| --- | --- | --- | +| 低延迟、即时返回 | 图像生成（同步模型） | wan2.6/2.7、z-image 等支持 HTTP 同步调用，无需轮询 | +| 批量处理、可容忍异步 | 任意（图像异步 / 3D / 视频） | 异步模式支持任务队列，适合批量场景 | +| 多模态输入（图+文+音） | 视频生成 | media 数组支持首帧/尾帧/参考图/视频/音频等多模态组合 | +| 多视角输入 | 3D 生成 | 支持前/左/后/右四视角多图生 3D | +| 需要文本渲染能力 | 图像生成（Qwen-Image） | 擅长复杂文本渲染与多行段落排版 | +| 需要多地域部署 | 视频生成 | 支持北京/新加坡/美国/德国四地域 | +| 成本敏感 | 图像生成（z-image-turbo / wanx-v1） | 轻量模型或 V1 版价格更低 | +| 高保真输出 | 3D 生成（Tripo-H3.1 ultra）或图像生成（wan2.7-image-pro 4K） | 高精度模型支持超高分辨率/面数输出 | + +## 被对比主题页 + +- [image generation](../api/image-generation.md) +- [3d generation](../api/3d-generation.md) +- [video generation api](../api/video-generation-api.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/generation-api-vs.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/generation-api-vs.md new file mode 100644 index 00000000..fbd132d2 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/generation-api-vs.md @@ -0,0 +1,78 @@ +# 图像/3D/视频生成 API 对比 + +百炼平台提供三类[多模态](../concepts/multimodal.md)生成 API：图像生成、3D 资产生成和视频生成。它们共享 DashScope 网关和异步任务轮询的基础调用范式，但在输入模态、输出产物、模型矩阵、地域限制和计费粒度上存在显著差异。本页面向需要在这些生成能力之间做技术选型的开发者，从关键维度逐一对比，帮助快速判断哪种 API 最适合你的业务场景。 + +## 关键维度对比 + +| 维度 | 图像生成 | 3D 生成 | 视频生成 | +| --- | --- | --- | --- | +| **输入格式** | 文本提示词；参考图（图生图/编辑）；多图参考（部分模型） | 文本提示词；单张图像 URL；4 视角多图（前/左/后/右） | 文本提示词；首帧/尾帧图像；参考图；视频/音频输入（视模型而定） | +| **输出格式** | 图像（JPEG/PNG），1~6 张 | GLB 3D 模型（PBR 材质或无贴图基础模型）+ 预览渲染图 | 视频文件 | +| **支持模型** | Qwen-Image、万相 wan2.7/2.6/2.5/2.2/2.1/2.0、Z-Image、Kling、Vidu 等系列，另含图像翻译、涂鸦作画、虚拟模特等专项工具 | Tripo/Tripo-H3.1（高精度）、Tripo/Tripo-P1.0（专业快速） | 万相 wan2.7/2.6/2.2、HappyHorse、PixVerse、Vidu、Kling、人像动画、数字人等 | +| **API 端点** | 各模型独立端点（见对应 API 参考） | `POST /api/v1/services/aigc/video-generation/3d-generation` | `POST /api/v1/services/aigc/video-generation/video-synthesis` | +| **调用模式** | 大多数异步（`X-DashScope-Async: enable`）；wan2.6/2.7、z-image、qwen-image 系列支持 HTTP 同步调用 | 仅异步（必须携带 `X-DashScope-Async: enable`） | 仅异步（必须携带 `X-DashScope-Async: enable`） | +| **结果轮询** | 异步任务通过 `GET /api/v1/tasks/{task_id}` 轮询 | 同左；建议轮询间隔约 15 秒，RPS 默认 20 | 同左；`task_id` 有效期 24 小时 | +| **地域要求** | 多地域可用（北京为主，部分模型仅限北京） | 仅华北2（北京） | 北京、新加坡、弗吉尼亚、法兰克福；PixVerse/Vidu/Kling/HappyHorse/人像动画仅限北京 | +| **计费方式** | 按张计费（如 wanx-v1 为 0.16 元/张）；部分专项工具提供免费体验 | 按任务类型计费（text-to-3d / image-to-3d / multi-image-to-3d） | 按任务计费（分辨率、时长影响费用） | +| **产物下载有效期** | 视模型而定 | 2 小时 | 视模型而定 | +| **典型场景** | 文生图、图像编辑、图像翻译、创意海报、虚拟模特、试衣、涂鸦作画、局部重绘 | 文生 3D、单图生 3D、多视角生 3D、PBR 材质模型生成 | 文生视频、图生视频、首尾帧视频、参考生视频、视频编辑、视频超分、对口型、动作模仿、数字人 | + +## 调用流程差异 + +三类 API 都遵循"创建任务 → 轮询获取结果"的异步范式，但图像生成有同步与异步两种模式可选，而 3D 和视频生成均只支持异步： + +- **图像生成**：较新的 wan2.6/wan2.7、z-image、qwen-image 系列支持 HTTP 同步调用，可直接在单次请求中获取生成结果；其余模型需走异步流程。如果对延迟敏感且使用的是支持同步的模型，优先选同步模式。 +- **3D 生成**：强制异步，创建任务时若不携带 `X-DashScope-Async: enable` 会直接报错。轮询间隔建议 15 秒，任务状态流转为 PENDING → RUNNING → SUCCEEDED/FAILED。 +- **视频生成**：同样强制异步。需注意部分万相模型（图生动作 `wan2.2-animate-move`、视频换人 `wan2.2-animate-mix`、数字人 `wan2.2-s2v`）使用 `image2video/video-synthesis` 路径而非 `video-generation/video-synthesis`，调用时需区分端点。 + +## 输入模态对比 + +| 输入类型 | 图像生成 | 3D 生成 | 视频生成 | +| --- | --- | --- | --- | +| 纯文本提示词 | 支持（文生图） | 支持（文生 3D） | 支持（文生视频） | +| 单张参考图 | 支持（图生图/编辑） | 支持（单图生 3D） | 支持（图生视频/首帧） | +| 多张参考图 | 部分支持（Kling 多图、Vidu 最多 14 张） | 支持（固定 4 视角，前/左/后/右） | 支持（参考生视频、首尾帧） | +| 视频输入 | 不支持 | 不支持 | 支持（视频编辑、超分等） | +| 音频输入 | 不支持 | 不支持 | 支持（数字人、对口型等） | + +## 适用场景建议 + +### 图像生成 + +适合需要**静态视觉产物**的场景：营销素材生成、电商商品图编辑、图像翻译保留排版、创意海报、虚拟模特试衣、涂鸦转成品图等。如果你的需求是快速批量出图且对延迟敏感，优先选择支持同步调用的模型（z-image-turbo、qwen-image、wan2.7-image）。如果需要复杂文本渲染和多段排版，Qwen-Image 系列是首选。创意工具类（虚拟模特、鞋靴模特等）适合垂直行业需求，但注意部分仅在北京地域提供。 + +### 3D 生成 + +适合需要**3D 资产**的场景：游戏道具快速建模、电商商品 3D 展示、建筑/工业设计预览等。如果对模型精度要求高，选 Tripo-H3.1（最高 200 万面，可设 `geometry_quality: ultra`）；如果追求速度，选 Tripo-P1.0（最高 2 万面）。输入方式的选择：有精确设计稿选单图生 3D，有多角度照片选多图生 3D（前/左/后/右四视角），只有文字描述则用文生 3D。注意此 API 仅限北京地域。 + +### 视频生成 + +适合需要**动态视频产物**的场景：广告短视频、影视素材生成、数字人播报、视频编辑与风格迁移、视频超分等。模型选择策略： + +- 通用文生/图生视频：万相 wan2.7 新版协议为推荐方案，功能最全（图生视频含首帧/首尾帧/续写、参考生视频含多主体+音色）。 +- 第三方模型：Kling 擅长智能分镜和视频编辑；PixVerse 支持 4K 超分和动作模仿；Vidu 适合广告参考生视频。 +- 人像/数字人场景：人像动画系列（animate-anyone、emo、liveportrait）适合舞蹈/唱演/播报；万相 `wan2.2-s2v` 适合单图+音频生成说话视频。 +- 需要音频同步：选支持 `audio` 参数的模型或使用专门的对口型接口。 + +## 技术选型速查 + +| 你的需求 | 推荐 API | 推荐模型 | +| --- | --- | --- | +| 快速文生图（低延迟） | 图像生成 | z-image-turbo 或 qwen-image（同步调用） | +| 高质量文生图+文本渲染 | 图像生成 | Qwen-Image 系列 | +| 图像编辑/改字/增删物体 | 图像生成 | Qwen-Image-Edit 或 wan2.7-image | +| 电商虚拟模特/试衣 | 图像生成 | wanx-virtualmodel / shoemodel-v1 | +| 3D 模型生成（高精度） | 3D 生成 | Tripo/Tripo-H3.1（ultra 精度） | +| 3D 模型生成（快速） | 3D 生成 | Tripo/Tripo-P1.0 | +| 文生视频/图生视频 | 视频生成 | wan2.7 系列（新版协议） | +| 视频编辑/风格迁移 | 视频生成 | wan2.7-videoedit 或 Kling | +| 数字人/对口型 | 视频生成 | wan2.2-s2v 或 PixVerse lipsync | +| 4K 视频超分 | 视频生成 | pixverse-upscale | + +## 被对比主题页 + +- [image generation](../api/image-generation.md) +- [3d generation](../api/3d-generation.md) +- [video generation api](../api/video-generation-api.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/image-video-3d-generation-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/image-video-3d-generation-comparison.md index c57c9b93..98518dbd 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/image-video-3d-generation-comparison.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/image-video-3d-generation-comparison.md @@ -1,73 +1,79 @@ -# 图像生成 vs 视频生成 vs 3D生成 +# 图像、视频与 3D 生成能力对比 -百炼平台同时提供图像生成、视频生成和 3D 模型生成三大视觉内容创作能力。三者在输入输出格式、模型生态、调用方式、计费模式和适用场景上存在显著差异。本文从开发者技术选型角度，对三类能力进行系统对比，帮助开发者根据业务需求选择最合适的方案。 +百炼平台提供图像生成、视频生成和 3D 模型生成三大多媒体内容生成能力，底层模型各有侧重。本页对三种生成能力的核心维度进行横向对比，帮助开发者在实际项目中快速做出技术选型决策。 ## 关键维度对比 -| 维度 | 图像生成 | 视频生成 | 3D生成 | -|------|----------|----------|--------| -| **输入格式** | 文本([prompt](../guides/prompt.md))、参考图片、涂鸦草图 | 文本([prompt](../guides/prompt.md))、首帧图片、参考图/视频、音频 | 文本([prompt](../guides/prompt.md))、单图、多图(4视角) | -| **输出格式** | PNG 图片(512x512 至 4K) | MP4 视频(5-10秒，1280x720等) | GLB 模型(带PBR材质贴图或无贴图) | -| **核心模型** | 千问(Qwen-Image)、万相(Wan)、Z-Image、可灵(Kling) | 万相(Wan 2.7)、HappyHorse、Pixverse、Vidu、Kling | Tripo-H3.1、Tripo-P1.0 | -| **调用方式** | 同步/异步均支持 | [异步调用](../concepts/async-invocation.md)(创建任务+轮询获取) | [异步调用](../concepts/async-invocation.md)(创建任务+轮询获取) | -| **典型响应时间** | 秒级至十秒级 | 1-5 分钟 | 分钟级(需轮询，建议间隔15秒) | -| **API端点** | [DashScope SDK](../concepts/dashscope-sdk.md) / HTTP | `/api/v1/services/aigc/video-generation/video-synthesis` | `/api/v1/services/aigc/video-generation/3d-generation` | -| **支持地域** | 多地域(部分模型仅北京) | 多地域(北京、新加坡等) | 仅华北2(北京) | -| **SDK兼容性** | [DashScope SDK](../concepts/dashscope-sdk.md)、HTTP | OpenAI兼容SDK、[DashScope SDK](../concepts/dashscope-sdk.md)、HTTP | 仅HTTP | -| **产物有效期** | 即时返回，无时效限制 | task_id 24小时有效 | 下载链接2小时有效，task_id 24小时有效 | -| **模型数量** | 20+ 模型(含创意工具) | 10+ 模型系列 | 2 个模型 | -| **编辑能力** | 局部重绘、风格迁移、扩图、超分 | 视频重绘、风格转换、口型替换 | 无编辑能力 | - -## 输入输出能力详细对比 - -| 能力 | 图像生成 | 视频生成 | 3D生成 | -|------|----------|----------|--------| -| 文生内容 | 支持(文生图) | 支持(文生视频) | 支持(文生3D) | -| 图生内容 | 支持(图像编辑、参考生图) | 支持(图生视频、参考生视频) | 支持(单图/多图生3D) | -| 多模态输入 | 多图参考、涂鸦 | 关键帧序列、音频驱动 | 4视角多图(前左后右) | -| 批量生成 | 单次1-9张 | 单次1条视频 | 单次1个模型 | -| 最大输出分辨率 | 4K(wan2.7-image-pro) | 1920x1080 | 最高200万面(H3.1) | - -## 计费与商业化对比 - -| 维度 | 图像生成 | 视频生成 | 3D生成 | -|------|----------|----------|--------| -| 计费单位 | 按张计费 | 按任务/时长计费 | 按任务计费 | -| 免费体验 | 部分模型有免费额度 | 部分模型有免费额度 | 需开通Tripo服务 | -| 商业化程度 | 大部分已商业化 | 主力模型已商业化 | 已商业化 | +| 维度 | 图像生成 | 视频生成 | 3D 生成 | +| --- | --- | --- | --- | +| **输入格式** | 文本提示词、参考图像（URL/Base64）、涂鸦图、分割掩码 | 文本提示词、首帧图像、首尾帧图像、参考图像/视频、音频 | 文本提示词、单张图像（JPEG/PNG）、多视角图像（4 张，前/左/后/右） | +| **输出格式** | PNG 图像，分辨率 512×512 ~ 4K，1-9 张 | MP4 视频，分辨率 720P ~ 4K，时长 3-10 秒（视模型而定） | GLB 三维模型（带/不带 PBR 贴图）+ 预览渲染图（PNG），有效期 2 小时 | +| **调用模式** | 同步（千问图像、万相 2.6+、Z-Image）或异步（可灵、Vidu、万相 2.5 及以下、创意工具类） | 全部异步（耗时 1~10 分钟） | 全部异步（必须携带 `X-DashScope-Async: enable`） | +| **主要模型** | 千问图像（Qwen-Image）、万相（Wan）图像系列、Z-Image、可灵图像、Vidu 图像 | 万相（Wan）视频系列、可灵（Kling）、Vidu、HappyHorse、爱诗（PixVerse）、人像动画系列 | Tripo/Tripo-H3.1（高精度）、Tripo/Tripo-P1.0（专业快速） | +| **API 端点（示例）** | `/api/v1/services/aigc/text2image/image-synthesis`（异步文生图）、`/api/v1/services/aigc/multimodal-generation/generation`（同步） | `/api/v1/services/aigc/video-generation/video-synthesis`（主流）、`/api/v1/services/aigc/image2video/video-synthesis`（旧版首尾帧） | `/api/v1/services/aigc/video-generation/3d-generation` | +| **地域支持** | 多地域（华北2北京为主，部分模型支持新加坡、弗吉尼亚、法兰克福） | 多地域（HappyHorse 覆盖最广：北京/新加坡/弗吉尼亚/法兰克福；可灵、Vidu、PixVerse 仅限北京） | 仅华北2（北京），API Key 也必须为北京地域 | +| **结果 URL 有效期** | 24 小时（异步任务的图像 URL） | 24 小时（task_id），视频文件 URL 视厂商而定 | task_id 有效期 24 小时；模型文件及渲染图 URL 仅 **2 小时**，需及时下载 | +| **轮询建议间隔** | 2~5 秒（图像生成通常较快） | 5~15 秒（视频生成耗时 1~10 分钟） | 15 秒（官方建议） | +| **典型生成时长** | 同步：秒级；异步：10~60 秒 | 1~5 分钟（普通视频）；5~10 分钟（视频编辑统一模型） | 数十秒到数分钟（取决于精度档位） | +| **最大精度/质量** | 4K 分辨率（可灵/Vidu 图像），PNG 无损 | 4K 分辨率（PixVerse 超清固定 4K） | 200 万面（Tripo-H3.1 ultra 模式） | +| **特色能力** | 图像翻译、虚拟模特、AI 试衣、创意海报、背景生成、画面扩展、擦除补全、人像分割、写真、创意文字等 20+ 细分功能 | 数字人（说话/唱歌）、视频口型替换、动作迁移、多镜头分镜、视频风格重绘、视频超清放大 | 多视角精准还原、PBR 材质、无贴图裸模式 | +| **计费方式** | 按模型分级：部分免费体验（额度有限），主力模型付费按量 | 按模型分级付费，无免费额度（需开通对应服务） | 按模型付费（需先在控制台开通 Tripo 服务） | + +## 调用流程对比 + +| 步骤 | 图像生成（同步） | 图像生成（异步） | 视频生成 | 3D 生成 | +| --- | --- | --- | --- | --- | +| 1 | POST 请求（无特殊 Header） | POST + `X-DashScope-Async: enable` | POST + `X-DashScope-Async: enable` | POST + `X-DashScope-Async: enable` | +| 2 | 直接返回图像 URL | 返回 `task_id` | 返回 `task_id` | 返回 `task_id` | +| 3 | — | GET 轮询 `/api/v1/tasks/{task_id}` | GET 轮询 `/api/v1/tasks/{task_id}` | GET 轮询 `/api/v1/tasks/{task_id}` | +| 4 | — | 状态 `SUCCEEDED` 后取图像 URL | 状态 `SUCCEEDED` 后取视频 URL | 状态 `SUCCEEDED` 后取 GLB URL（2 小时内下载） | ## 适用场景建议 -**选择图像生成的场景**： - -- 电商商品图、营销海报、社交媒体配图等静态视觉内容制作 -- 需要精确文字渲染(如广告文案嵌入图片) -- 图像局部编辑、风格转换、AI试衣等垂直场景 -- 对响应速度要求高(秒级出图) -- 需要批量生成多张候选图供筛选 - -**选择视频生成的场景**： - -- 短视频内容创作、广告视频制作 -- 数字人驱动(音频/文本驱动说话、唱歌) -- 服装展示、舞蹈动作等动态展示 -- 需要关键帧精确控制镜头运动 -- 已有视频的风格转换或口型替换 - -**选择3D生成的场景**： - -- 游戏资产、AR/VR场景中的3D物体快速原型 -- 电商3D商品展示(可旋转查看) -- 建筑/工业设计的概念验证模型 -- 需要标准PBR材质的可渲染模型 - -## 技术选型决策参考 - -1. **内容维度**：静态画面选图像生成；需要动态表现选视频生成；需要空间立体展示选3D生成。 -2. **时效要求**：图像生成响应最快(秒级)，适合实时交互；视频和3D均为分钟级异步任务，适合离线批处理。 -3. **生态成熟度**：图像生成模型最丰富、功能最全面；视频生成处于快速发展期，模型迭代频繁；3D生成目前模型较少，但输出质量已达可用水平。 -4. **地域限制**：3D生成仅限北京地域，部分图像创意工具同样限北京；视频生成地域覆盖较广。 -5. **集成复杂度**：图像生成支持同步调用，集成最简单；视频和3D需要实现异步轮询逻辑，建议封装任务状态管理层。 +### 选择图像生成 + +- 需要快速产出静态素材：商品图、海报、插画、概念图。 +- 有图像编辑诉求：局部重绘、扩图、去水印、风格迁移。 +- 电商场景：虚拟模特换装（`virtualmodel`）、AI 试衣（`aitryon`）、背景替换。 +- 需要图像内文字翻译且保持排版（`qwen-mt-image`）。 +- 需要同步低延迟接口（千问图像、万相 2.6+、Z-Image）。 + +### 选择视频生成 + +- 需要动态内容：产品展示视频、短片、宣传片。 +- 人像动画场景：数字人口播（`wan2.2-s2v`/EMO）、舞蹈驱动（AnimateAnyone）、表情包（Emoji）。 +- 已有静态图片，需赋予动态效果（图生视频）。 +- 视频内容再加工：口型替换（`videoretalk`/PixVerse lipsync）、风格重绘（`video-style-transform`）、视频超清（PixVerse upscale）。 +- 需要分镜叙事（可灵 `multi_shot`、万相2.7 多镜头、PixVerse c1 系列）。 + +### 选择 3D 生成 + +- 需要可直接用于游戏引擎、AR/VR、渲染管线的三维资产（GLB 格式）。 +- 从概念图或参考照片快速原型化 3D 模型。 +- 需要 PBR 材质的工业级资产（`Tripo-H3.1` ultra 模式，最高 200 万面）。 +- 对速度要求高于精度时选 `Tripo-P1.0`（速度更快，最高 2 万面）。 +- **注意**：项目必须能接受仅部署在北京地域这一硬性约束。 + +## 技术选型参考 + +| 需求特征 | 推荐方案 | 关键理由 | +| --- | --- | --- | +| 秒级出图、低延迟 | 图像生成（同步：千问图像 / 万相 2.6+ / Z-Image） | 同步接口，无需轮询 | +| 高质量静态素材批量生产 | 图像生成（可灵 / Vidu，异步 4K） | 支持 1-9 张批量，分辨率最高 4K | +| 动态内容、视频营销 | 视频生成（万相2.7 / 可灵 v3 / HappyHorse） | 内容丰富，支持多镜头分镜 | +| 跨地域部署 | 视频生成（HappyHorse，4 地域） | 覆盖北京/新加坡/弗吉尼亚/法兰克福 | +| 数字人 / 口播视频 | 视频生成（wan2.2-s2v / EMO / LivePortrait） | 专为人像动画设计，音频驱动 | +| 游戏/AR 三维资产 | 3D 生成（Tripo-H3.1） | 输出标准 GLB，支持 PBR，高达 200 万面 | +| 快速 3D 原型 | 3D 生成（Tripo-P1.0） | 速度优先，兼顾质量（2 万面上限） | +| 预算敏感、先验证可行性 | 图像生成（免费体验模型：virtualmodel / 海报 / 擦除等） | 部分功能有免费额度，适合 PoC | + +## 注意事项 + +1. **3D 生成地域唯一性**：3D 生成目前仅限华北2（北京），且 API Key 必须是北京地域的 Key，与多地域支持的图像/视频生成有本质区别，上线前务必确认部署地域。 +2. **3D 结果 URL 仅 2 小时有效**：3D 生成的模型文件和渲染图 URL 有效期极短（2 小时），需在业务层做及时下载和持久化存储，不能依赖 URL 直接分发。 +3. **视频生成全部异步**：视频生成没有同步接口，最短也需 1 分钟左右，不适合实时/低延迟场景。 +4. **图像生成的免费模型限额**：wanx-virtualmodel、image-erase-completion 等免费体验模型额度用完后不可调用且不支持付费续购，生产环境应优先使用付费的千问图像编辑或万相 2.1 图像编辑。 +5. **视频生成旧版协议不兼容**：万相 2.6 及以下版本使用旧版 API 协议，与 wan2.7 的新版协议不兼容，迁移时需修改请求结构，不能直接替换模型名称。 ## 被对比主题页 diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/image-vs-video-generation.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/image-vs-video-generation.md new file mode 100644 index 00000000..9f8aa872 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/image-vs-video-generation.md @@ -0,0 +1,75 @@ +# 图像生成与视频生成 API 对比 + +百炼平台同时提供图像生成与视频生成两大类视觉内容生成 API。两者共享部分模型家族（万相、可灵、Vidu），但在调用模式、输入输出格式、任务耗时和适用场景上存在显著差异。本页从开发者技术选型角度出发，对两类 API 的关键维度进行系统对比，帮助开发者根据业务需求快速决策。 + +## 关键维度对比 + +| 维度 | 图像生成 API | 视频生成 API | +| --- | --- | --- | +| 调用模式 | 同步（千问图像、万相2.6+、Z-Image）+ 异步（万相2.5及以下、可灵、Vidu、创意工具类） | 统一异步（创建任务 → 轮询结果） | +| 典型耗时 | 同步调用秒级返回；[异步调用](../concepts/async-invocation.md)通常数十秒 | 通常 1–5 分钟，部分模型（wanx2.1-vace）约 5–10 分钟 | +| 输出格式 | PNG 图像，分辨率 512×512 ~ 4K，单次 1–9 张 | 视频文件（URL 形式返回），支持多种宽高比与时长参数 | +| 主要模型家族 | 千问图像（Qwen-Image）、万相（Wan）、Z-Image、可灵（Kling）、Vidu | 万相（Wan 2.1–2.7）、HappyHorse、爱诗（PixVerse）、Vidu、可灵（Kling） | +| 核心任务类型 | 文生图、图像编辑、局部重绘、涂鸦作画、图像翻译、虚拟模特、试衣、海报生成、背景生成、擦除补全、画面扩展、人物分割、写真生成、创意文字 | 文生视频、图生视频（首帧/首尾帧）、参考生视频、视频编辑、风格重绘、视频超清、对口型、动作模仿、数字人/人像动画 | +| 输入形式 | 文本提示词 + 可选参考图像（多图输入支持融合编辑） | 文本提示词 + 图像/视频/音频等[多模态](../concepts/multimodal.md)输入（万相2.7支持文本/图像/音频/视频混合输入） | +| 同步 API 端点 | `POST /api/v1/services/aigc/multimodal-generation/generation`（新版模型） | 无同步端点 | +| 异步 API 端点 | 按功能区分：`/text2image/image-synthesis`、`/image2image/image-synthesis`、`/image-generation/generation` 等 | 统一异步两步流：创建任务获取 task_id → 轮询任务状态获取结果 | +| 异步请求头要求 | `X-DashScope-Async: enable`（万相2.5及以下、可灵、Vidu 等） | 同样需要异步请求头 | +| 结果有效期 | 异步任务 task_id 有效期 24 小时，图像 URL 有效期 24 小时 | 任务结果 URL 同样有时效限制，需及时下载转存 | +| 特色参数 | `size`、`n`（张数）、`style`（风格预设）、`negative_prompt`、`prompt_extend`、`thinking_mode` | `duration`（时长）、`aspect_ratio`（宽高比）、`mode`（std/pro）、`audio`（音频输入）、`shot_type`/多镜头描述、`style_level` | +| 第三方模型开通 | 可灵、Vidu 需在控制台开通 | 可灵、Vidu、PixVerse 需在控制台搜索并开通 | +| 计费方式 | 按生成图像张数计费；部分创意工具模型仅免费体验（额度用完不可调用） | 按生成视频时长/次数计费；部分人像动画模型支持后付费与独立部署 | + +## 能力覆盖差异 + +| 能力 | 图像生成 | 视频生成 | +| --- | --- | --- | +| 文本到视觉内容 | 文生图（多风格、多分辨率） | 文生视频（支持多镜头叙事描述） | +| 基于参考素材生成 | 参考图生图、多图融合、线稿生图 | 图生视频（首帧/首尾帧）、参考生视频（多主体一致性） | +| 内容编辑 | 局部重绘、擦除补全、扩图、去水印、超分、风格迁移 | 视频编辑（指令编辑、风格转换）、视频超清（4K）、风格重绘（8种预设风格） | +| 人像相关 | 虚拟模特、AI试衣、人物写真、人物分割 | 数字人（语音驱动）、动作模仿、对口型、视频换人、表情包生成 | +| 文字渲染 | 图像翻译（保留排版）、创意文字、中英文精准渲染 | 不涉及 | +| 音频驱动 | 不涉及 | 音频驱动口型/动作/唱演（wan2.2-s2v、EMO、LivePortrait 等） | + +## 适用场景建议 + +### 选择图像生成 API 的场景 + +- **电商素材制作**：商品主图、海报、背景替换、虚拟模特试穿等，需要快速批量产出静态素材。 +- **内容创作与设计**：插画生成、风格迁移、涂鸦转图像，追求秒级响应的交互式体验。 +- **文档与本地化处理**：图像翻译（保留原始排版）、文字精确修改。 +- **图像修复与增强**：擦除补全、画面扩展、超分辨率。 +- **对延迟敏感的应用**：同步调用模式（千问图像、万相2.6+、Z-Image）可在秒级返回结果，适合实时交互场景。 + +### 选择视频生成 API 的场景 + +- **短视频与营销内容**：文生视频、图生视频用于广告片、产品宣传视频制作。 +- **数字人与虚拟主播**：单张照片 + 音频生成说话/唱歌视频（wan2.2-s2v）、播报视频（LivePortrait）。 +- **影视与动画制作**：首尾帧控制、多镜头叙事、视频续写，适合有分镜脚本的专业制作流程。 +- **视频后期处理**：风格重绘、超清增强、对口型替换。 +- **角色一致性需求**：参考生视频（wan2.7-r2v）保持角色形象与音色一致，适合系列内容制作。 + +### 组合使用建议 + +在实际业务中，图像生成与视频生成常配合使用： + +1. **先图后视频**：用图像生成 API 产出高质量关键帧（如千问图像生成精确构图的画面），再作为首帧/首尾帧输入视频生成 API 产出动态内容。 +2. **素材加工流水线**：图像编辑完成素材准备（抠图、背景替换）→ 视频生成完成动态化 → 视频编辑完成风格统一。 +3. **选型优先级**：若业务只需静态视觉内容且对响应速度有要求，优先选择支持同步调用的图像生成模型；若需要动态叙事或音频驱动，则必须使用视频生成 API（全部为异步模式，需在架构中设计任务轮询与回调机制）。 + +## 技术选型要点 + +| 考量因素 | 建议 | +| --- | --- | +| 响应延迟 | 图像生成同步模式秒级返回；视频生成必须异步，需实现轮询或消息通知机制 | +| 架构复杂度 | 图像生成同步调用接入简单；视频生成需处理任务生命周期管理（创建、轮询、超时、失败重试） | +| 模型选择 | 追求文字渲染精度选千问图像/PixVerse；追求[多模态](../concepts/multimodal.md)综合能力选万相2.7；需要第三方风格选可灵/Vidu | +| 成本 | 图像按张计费成本较低；视频按时长计费成本较高，建议先用图像生成验证创意再投入视频生成 | +| 免费额度 | 部分图像创意工具仅有免费体验额度（用完即止）；视频类部分模型支持后付费与独立部署 | + +## 被对比主题页 + +- [image generation](../api/image-generation.md) +- [video generation api](../api/video-generation-api.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/image-vs-video-vs-3d-generation.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/image-vs-video-vs-3d-generation.md index bfc49c97..bb80941c 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/image-vs-video-vs-3d-generation.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/image-vs-video-vs-3d-generation.md @@ -1,47 +1,68 @@ -# 图像、视频与 3D 生成对比 +# 图像、视频与 3D 生成 API 对比 -阿里云百炼平台在 DashScope 网关上提供了图像、视频与 3D 三大类视觉内容生成能力。三者虽同属「生成式媒体」范畴、共用同一套鉴权与任务模型，但在输入输出格式、可用模型、调用协议、地域限制与产物形态上差异明显。本文面向开发者，横向梳理三类方案的关键维度，帮助在技术选型时快速判断该用哪一类 API。 +百炼平台提供三大视觉内容生成能力：图像生成、视频生成和 3D 模型生成。三者在输入方式、调用协议、模型生态和适用场景上各有侧重。本页从开发者技术选型角度，对三类 API 的关键维度进行横向对比，帮助开发者根据业务需求选择合适的接入方案。 ## 关键维度对比 | 维度 | 图像生成 | 视频生成 | 3D 生成 | | --- | --- | --- | --- | -| 主要能力 | 文生图、图像编辑、图像翻译、垂直创意工具（虚拟模特、扩图、擦除补全、海报等） | 文生视频、图生视频（首帧/首尾帧）、参考生视频、视频编辑、数字人、人像驱动、超清、对口型 | 文生 3D、单图生 3D、多图生 3D | -| 典型输入 | `prompt` / `negative_prompt`；编辑类传 `image_url`/`images`/`mask_image_url`；新版用 `messages` 多模态结构 | 文生用 `prompt`；图生/参考/编辑用 `media` 数组或 `image_url`/`video_url`/`audio_url`/`first_frame`/`last_frame` 等 | `prompt`（≤1024 字符）/ `image`（单图）/ `images`（4 元数组：前左后右），三者互斥 | -| 输出格式 | 图像 URL（有效期 24 小时），分辨率随模型而异（如 512×512~2048×2048、1K/2K/4K） | 视频 URL，分辨率 480P/540P/720P/1080P、可设 `duration` | 带贴图 PBR 材质 GLB（`pbr_model_url`）或无贴图基础模型（`base_model_url`）+ 预览渲染图，下载链接**仅 2 小时** | -| 代表模型 | Qwen-Image、万相 Wan/WanX（t2i/imageedit）、Z-Image、可灵、Vidu 等 | 万相 Wan 2.1-2.7、PixVerse、Vidu、可灵 Kling、HappyHorse、EMO/LivePortrait 等 | `Tripo/Tripo-H3.1`（最高 200 万面）、`Tripo/Tripo-P1.0`（最高 2 万面，速度更快） | -| 调用协议 | 异步（主流）+ **HTTP 同步**（仅 wan2.6/2.7、z-image 等新版） | **全部异步**（创建任务 → 轮询），无同步 | **全部异步**（创建任务 → 轮询），无同步 | -| 典型端点 | `text2image/image-synthesis`、`image2image/image-synthesis`、`multimodal-generation/generation`、`virtualmodel/generation` 等多种 | `video-generation/video-synthesis`（主）、部分数字人用 `image2video/video-synthesis` | `video-generation/3d-generation`（单一端点） | -| 生成耗时 | 通常 1-2 分钟 | 通常 1-5 分钟（统一编辑约 5-10 分钟） | 较长，建议轮询间隔约 15 秒 | -| 地域可用性 | 华北2（北京）为主，部分模型也支持新加坡/美国；大量创意工具**仅北京** | 华北2（北京）为主，第三方模型多**仅北京**，万相等部分支持新加坡/美国/德国 | **仅华北2（北京）** | -| 计费方式 | 仅对成功输出图片计费，含免费额度（通常 500 张/90 天），主子账号共享 | 按成功任务计费，随模型与分辨率/时长而异 | 仅对成功结果计数，`usage` 记录任务类型/数量/质量 | -| 关键请求头 | `Authorization`、`Content-Type`；异步需 `X-DashScope-Async: enable` | 同左，异步必带 `X-DashScope-Async: enable` | 同左，异步必带 `X-DashScope-Async: enable` | -| 任务状态 | `PENDING`/`RUNNING`/`SUSPENDED`/`SUCCEEDED`/`FAILED` | 同类异步枚举，过期返回 `UNKNOWN` | `PENDING`/`RUNNING`/`SUCCEEDED`/`FAILED`/`CANCELED`/`UNKNOWN` | +| 核心任务 | 文生图、图像编辑、图像翻译、局部重绘、虚拟模特、试衣、海报/背景生成、擦除补全、画面扩展等 | 文生视频、图生视频（首帧/首尾帧）、参考生视频、视频编辑、风格重绘、超清、对口型、数字人、动作模仿等 | 文生3D、单图生3D、多图生3D | +| 输出格式 | PNG 图像（512×512 ~ 4K） | 视频文件（支持多种分辨率与时长） | GLB 三维模型（带/不带贴图，最高 200 万面）+ 预览渲染图 | +| 调用模式 | 同步（千问图像、万相2.6+、Z-Image）+ 异步（万相2.5及以下、可灵、Vidu、创意工具类） | 仅异步（创建任务 → 轮询结果） | 仅异步（创建任务 → 轮询结果） | +| 主要模型系列 | 千问图像（Qwen-Image）、万相（Wan）、Z-Image、可灵（Kling）、Vidu | 万相（Wan 2.1–2.7）、HappyHorse、爱诗（PixVerse）、Vidu、可灵（Kling） | Tripo（Tripo-H3.1、Tripo-P1.0） | +| API 端点 | 同步：`/api/v1/services/aigc/multimodal-generation/generation`；异步因功能而异（text2image / image2image / image-generation 等） | 各模型独立端点，统一异步协议 | `/api/v1/services/aigc/video-generation/3d-generation` | +| 异步请求头 | [异步调用](../concepts/async-invocation.md)需 `X-DashScope-Async: enable` | 需要 | 必须携带，否则报错 | +| 请求体结构 | 同步模型采用 `model` + `input.messages`（[多模态](../concepts/multimodal.md)对话格式）+ `parameters`；异步模型采用 `model` + `input` + `parameters` | `model` + `input`（含 `media` [多模态](../concepts/multimodal.md)引用）+ `parameters` | `model` + `input`（[prompt](../guides/prompt.md) / image / images 三选一）+ `parameters` | +| 结果有效期 | 异步任务 task_id 24 小时有效，图像 URL 24 小时有效 | task_id 24 小时有效，结果 URL 有时效限制 | task_id 24 小时有效；模型文件与渲染图 URL 仅 2 小时有效 | +| 地域限制 | 无特殊限制 | 无特殊限制 | 仅华北2（北京）地域，必须使用北京地域 [API Key](../concepts/api-key.md) | +| 开通要求 | 千问/万相系列默认可用；可灵、Vidu 需控制台开通 | PixVerse、Vidu 需控制台搜索开通 | 需在控制台搜索"Tripo"并开通 | +| 计费方式 | 按生成图片张数计费；部分创意工具模型仅免费体验（额度用完不可付费续用） | 按生成视频时长/次数计费 | 按生成任务计费 | +| 典型耗时 | 秒级～数十秒 | 分钟级（建议轮询间隔参考各模型文档） | 分钟级（建议每 15 秒轮询） | -## 共性与差异要点 +## 输入能力对比 -**共性**：三者都基于 DashScope 网关，共用 `Authorization: Bearer $DASHSCOPE_API_KEY` 鉴权、`model`/`input`/`parameters` 的请求体结构，异步模式均为「创建任务拿 `task_id` → 轮询查询」，且 `task_id` 有效期统一为 24 小时、创建异步任务必须携带 `X-DashScope-Async: enable`。地域隔离规则也一致：模型、Endpoint 与 API Key 必须同地域，跨地域会鉴权失败。 +| 输入类型 | 图像生成 | 视频生成 | 3D 生成 | +| --- | --- | --- | --- | +| 纯文本提示词 | 支持（文生图） | 支持（文生视频，万相2.7支持多镜头叙事描述） | 支持（文生3D，最大 1024 字符） | +| 单图输入 | 支持（图生图/编辑） | 支持（首帧生视频） | 支持（单图生3D，JPEG/PNG，≤20MB） | +| 多图输入 | 支持（千问图像编辑多图融合） | 支持（参考生视频多主体参考图） | 支持（固定 4 张数组：前/左/后/右视角） | +| 音频输入 | 不支持 | 支持（数字人、对口型、参考生视频） | 不支持 | +| 视频输入 | 不支持 | 支持（视频编辑、视频续写、动作迁移） | 不支持 | +| 首尾帧控制 | 不适用 | 支持（万相2.2-kf2v、PixVerse kf2v、Vidu） | 不适用 | + +## 适用场景建议 -**核心差异**： +### 选择图像生成的场景 -- **协议丰富度**：只有图像生成的部分新版模型（wan2.6/2.7、z-image）支持一次请求返回结果的 HTTP 同步调用；视频与 3D **全部只能异步**。 -- **地域自由度**：图像与视频在北京之外还有一定跨地域支持，而 3D 生成**只有华北2（北京）**可用，选型时需特别注意。 -- **产物时效**：图像/视频 URL 有效期 24 小时，而 3D 模型下载链接**仅 2 小时**，需在生成后尽快下载并转存。 -- **输入约束**：3D 的 `prompt`/`image`/`images` 三者互斥，多图必须是固定 4 元数组（前左后右）；图像与视频则允许更灵活的多模态、多图输入组合。 +- **电商营销素材**：商品主图、海报、背景替换、虚拟模特试穿等，推荐使用千问图像或万相文生图。 +- **内容创作与设计**：插画、概念图、风格化创作，万相系列支持多种艺术风格预设。 +- **图像后期处理**：局部重绘、擦除补全、画面扩展、超分辨率等编辑需求，推荐千问图像编辑或万相图像编辑。 +- **多语言素材本地化**：图像翻译（qwen-mt-image）可保留原始排版精准翻译图中文字。 +- **低延迟交互场景**：千问图像和 Z-Image 支持同步调用，适合需要即时响应的应用。 -## 适用场景建议 +### 选择视频生成的场景 -- **图像生成**：适合海报、电商主图、创意配图、虚拟模特试衣、图像翻译/编辑等静态视觉需求。追求低延迟、希望一次请求出结果时，优先选支持 HTTP 同步的新版模型（wan2.6/2.7、z-image-turbo）；需要精细编辑（改文字、局部重绘、扩图、去水印）则用千问/万相编辑系列。 -- **视频生成**：适合短视频、广告片、数字人播报、口播/对口型、人像驱动等动态内容。需明确任务类型（文生/图生/首尾帧/参考/编辑）选择对应模型，注意多镜头控制方式在不同模型间不一致（万相 2.7、PixVerse-c1 用自然语言 `prompt`，旧版万相 2.6 需显式 `shot_type: "multi"` + `prompt_extend: true`）。 -- **3D 生成**：适合游戏/电商/XR 场景的 3D 资产快速建模。需要高精度、高面数选 `Tripo/Tripo-H3.1`（可用 `geometry_quality: ultra` 达 200 万面）；追求速度、面数需求不高选 `Tripo/Tripo-P1.0`。务必在北京地域开通 Tripo 服务并及时下载 2 小时时效的产物。 +- **短视频与广告制作**：文生视频或图生视频快速产出营销视频，万相2.7 支持多镜头叙事控制。 +- **角色一致性内容**：参考生视频（wan2.7-r2v）保持角色形象与音色一致，适合系列内容制作。 +- **数字人与虚拟主播**：wan2.2-s2v 单图+音频生成说话视频，LivePortrait 快速生成播报视频。 +- **视频后期增强**：风格重绘、4K 超清（PixVerse upscale）、对口型替换等。 +- **动作迁移与换人**：animate-move 迁移参考动作，animate-mix 替换视频主角。 -## 技术选型参考 +### 选择 3D 生成的场景 -1. **先按产物形态定类别**：要静态图片 → 图像；要动态视频 → 视频；要可交互 3D 模型（GLB） → 3D。 -2. **再评估延迟要求**：对响应速度敏感的图像场景可用同步协议；视频与 3D 必须做好异步轮询与任务状态处理（含 `FAILED`/`UNKNOWN`）。 -3. **确认地域与开通**：3D 与多数第三方视频/图像创意模型只在北京可用，需保证模型、Endpoint、API Key 同地域，并提前在控制台开通授权（如 Tripo）。 -4. **规划产物存储**：所有产出均为限时 URL，建议生成后立即转存至 OSS，其中 3D 仅 2 小时窗口最需注意。 -5. **统一工程实现**：三类 API 共用鉴权、请求头与异步模型，可复用同一套任务提交/轮询/重试封装，仅按 `model` 与端点差异做分支。 +- **游戏与虚拟世界资产**：快速生成带 PBR 材质的 GLB 模型，Tripo-H3.1 最高支持 200 万面。 +- **电商 3D 展示**：商品单图/多角度照片生成可交互 3D 模型。 +- **快速原型与概念验证**：文本描述直接生成三维模型，加速设计迭代。 +- **轻量级应用场景**：Tripo-P1.0 输出 2 万面模型，生成速度更快，适合移动端或 Web 端实时渲染。 + +## 技术选型要点 + +1. **时效性要求**：如果业务需要秒级响应，优先选择支持同步调用的图像生成模型（千问图像、万相2.6+、Z-Image）；视频和 3D 生成均为异步任务，需要设计任务轮询与回调机制。 +2. **结果持久化**：3D 生成的模型文件 URL 仅 2 小时有效，务必在任务完成后立即下载存储；图像异步结果 URL 有效期 24 小时；视频结果同样有时效限制，建议统一建立产物下载落盘流程。 +3. **[多模态](../concepts/multimodal.md)输入需求**：如果输入涉及音频或视频素材，只有视频生成 API 支持；图像和 3D 生成仅接受文本与图像输入。 +4. **地域与开通**：3D 生成仅限北京地域 [API Key](../concepts/api-key.md)，且需在控制台单独开通 Tripo 服务；PixVerse 和 Vidu 系列（图像/视频）同样需要提前开通。 +5. **模型生态丰富度**：图像生成模型系列最多（千问、万相、Z-Image、可灵、Vidu），功能覆盖最广；视频生成在万相2.7 新版协议下能力最全面；3D 生成当前仅有 Tripo 两个模型，功能聚焦。 +6. **免费额度注意**：图像生成中部分创意工具模型（虚拟模特、擦除补全等）仅提供免费体验额度，用完后无法付费续用，生产环境应评估替代方案（千问图像编辑或万相2.1图像编辑）。 ## 被对比主题页 @@ -50,3 +71,4 @@ - [3d generation](../api/3d-generation.md) + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-base-vs-memory.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-base-vs-memory.md new file mode 100644 index 00000000..9de9c216 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-base-vs-memory.md @@ -0,0 +1,101 @@ +# 知识库与长期记忆方案对比 + +在构建 AI 应用时，开发者经常需要为大模型补充"外部信息"。百炼平台提供了两类截然不同的外部信息方案：**知识库**（Knowledge Base）和**长期记忆**（Long-term Memory / Memory Library）。两者虽都能扩充模型可访问的信息范围，但设计目标、数据来源、检索机制与适用场景差异显著。本文从技术选型角度系统梳理两者的区别，帮助开发者快速做出决策。 + +--- + +## 核心定位 + +- **知识库**：基于 RAG（检索增强生成）技术，将企业私有文档、图片、音视频等静态语料库索引化，供模型在回答时实时召回。强调**文档级**、**结构化**知识管理，面向所有用户提供统一的"企业大脑"。 +- **长期记忆**：以**用户为粒度**，从对话中自动提取关键信息（记忆片段 + 用户画像），持久化存储并在后续对话中语义召回，解决大模型跨会话上下文丢失的问题。面向单一用户构建"个人记忆空间"。 + +--- + +## 关键维度对比 + +| 维度 | 知识库（Knowledge Base） | 长期记忆 / 记忆库（Long-term Memory） | +|------|--------------------------|---------------------------------------| +| **核心目标** | 企业私有文档的 RAG 检索增强 | 跨会话用户记忆的持久化与个性化召回 | +| **数据来源** | 开发者主动上传（文档/图片/音视频/Excel） | 对话自动提取（或 `custom_content` 直写） | +| **数据粒度** | 文档切片（最大 6000 Token/片） | 记忆片段（事件摘要）+ 结构化用户画像 | +| **用户隔离** | 无原生用户维度，整个知识库共享 | 以 `user_id` 严格隔离，不同用户完全独立 | +| **检索方式** | 向量检索 + 关键词检索 + Rerank 排序 | 语义相似度检索，可选 Rerank / Query 改写 | +| **写入方式** | 手动上传文件 → 解析 → 切片 → 建索引 | 调用 `AddMemory` API，系统自动提炼存储 | +| **更新机制** | 重新上传或编辑切片；索引需重建 | 写入时自动去重、合并或覆盖旧记忆 | +| **有效期** | 知识库存续期间长期有效 | 可配置 7/30/180 天或永不过期（默认规则 180 天） | +| **API 端点** | `bailian.cn-beijing.aliyuncs.com` | `dashscope.aliyuncs.com/api/v2/apps/memory/` | +| **计费方式** | 标准版 0.03 元/小时；旗舰版 0.2 元/RCU/小时 + Rerank Token 费用 | 目前按调用量计费（QPM 限额：总计 3000 QPM） | +| **地域限制** | 仅支持中国站华北 2（北京） | 通过 DashScope 网关，不限地域 | +| **管理入口** | 百炼控制台知识库模块 + API | 百炼控制台记忆库模块 + HTTP API + OpenClaw 插件 | +| **[多模态](../concepts/multimodal.md)支持** | 支持图片问答、音视频搜索（专用向量模型） | 仅支持文本记忆片段 | +| **零侵入接入** | 不支持，需主动调用检索 API 或绑定应用 | 支持（OpenClaw 插件 `before_agent_start` / `agent_end` 钩子） | +| **典型规模** | 单库最大 9999 GB（旗舰版），支持 15 库联检 | 按用户数据量弹性扩展，单账号 3000 QPM | + +--- + +## 各方案适用场景 + +### 知识库适用场景 + +- **企业内部问答**：员工手册、操作规程、合规文档的自然语言检索，内容由管理员统一维护。 +- **产品支持 / 客服机器人**：将产品文档、FAQ 建库，对所有用户提供一致的权威回答。 +- **富文本理解**：需要解析 PDF 图表、发票、合同等带版面信息的文档（视觉理解场景）。 +- **结构化数据查询**：Excel/CSV 数值型数据的自然语言查询（数据查询类知识库）。 +- **多媒体内容搜索**：音视频文件的语音转写 + 剧情检索（音视频搜索类知识库）。 +- **对知识质量要求高**：内容由人工审核把关，需精确引用来源，不接受"幻觉"。 + +### 长期记忆适用场景 + +- **个性化智能助手**：记住用户的偏好、习惯、历史决策，跨会话保持连贯上下文（"你之前说过喜欢简洁风格"）。 +- **健康/情感陪伴类应用**：持续追踪用户的身体状况记录、情绪日志、目标进展。 +- **多轮任务跟进**：用户分多次完成的长期任务（如学习计划、项目进度），每次对话自动接续上下文。 +- **用户画像沉淀**：通过 `profile_schema` 从对话中结构化提取年龄、职业、偏好等属性，用于个性化推荐。 +- **OpenClaw Agent 零侵入记忆**：已有 OpenClaw Agent 且不想改动代码的场景，通过插件自动捕获/召回。 + +--- + +## 技术选型建议 + +### 选择知识库，如果你需要： + +1. **共享语料库**：一份知识面向多用户，内容由团队或系统管理员维护。 +2. **文档精确引用**：回答需附带"来源文件 + 页码"，要求可溯源。 +3. **大规模静态知识**：文档体量大（GB 级），内容相对稳定，更新频率低。 +4. **[多模态](../concepts/multimodal.md)文档**：PDF、图片、音视频等格式的内容检索。 + +### 选择长期记忆，如果你需要： + +1. **用户级个性化**：每个用户有自己独立的记忆空间，互不干扰。 +2. **对话驱动的知识积累**：无需手动维护，系统从每次对话自动提炼并更新记忆。 +3. **跨会话连续性**：用户今天说的事，下次对话时 Agent 仍然"记得"。 +4. **轻量快速接入**：无需建索引，直接调用 REST API 即可集成，或通过 OpenClaw 插件零侵入启用。 + +### 两者结合使用 + +知识库与长期记忆并不互斥，结合使用可实现更完整的 AI 应用： + +- **知识库负责"公共知识"**：产品文档、规程、FAQ — 对所有用户一致。 +- **长期记忆负责"用户上下文"**：该用户的历史偏好、个性化配置、历史对话摘要。 +- 在生成时同时注入两类信息，实现"公共知识 + 个人记忆"的双轨增强。 + +--- + +## 小结 + +| 选型问题 | 推荐方案 | +|----------|----------| +| 我的知识是企业共享的静态文档吗？ | 知识库 | +| 我需要按用户隔离、记住每个人的偏好吗？ | 长期记忆 | +| 知识由用户对话自动生成而非手动上传吗？ | 长期记忆 | +| 需要处理 PDF / 图片 / 音视频等[多模态](../concepts/multimodal.md)内容吗？ | 知识库 | +| 我的 Agent 需要跨会话保持连续性吗？ | 长期记忆 | +| 需要精确引用来源文件和段落吗？ | 知识库 | +| 想要最低侵入地为 OpenClaw Agent 加记忆？ | 长期记忆（OpenClaw 插件） | + +## 被对比主题页 + +- [knowledge base](../guides/knowledge-base.md) +- [long term memory new](../api/long-term-memory-new.md) +- [memory library overview](../guides/memory-library-overview.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-memory-data-connection.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-memory-data-connection.md new file mode 100644 index 00000000..dd44cacd --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-memory-data-connection.md @@ -0,0 +1,72 @@ +# [知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)、记忆库与数据连接对比 + +[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)、记忆库与数据连接都用于把外部或历史信息接入百炼应用，但它们解决的问题层级不同：[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)面向“把文档、图片、音视频等资料检索后交给大模型回答”的 RAG 场景；记忆库面向“跨会话保存用户偏好、历史事实和画像”的长期记忆场景；数据连接面向“统一管理外部数据源或平台托管数据，并供应用实时访问”的数据接入场景。开发者做技术选型时，应先判断数据是相对稳定的知识内容、随用户持续变化的个人记忆，还是需要连接和查询的业务数据源。 + +## 关键维度对比 + +| 维度 | 知识库 | 记忆库 | 数据连接 | +| --- | --- | --- | --- | +| 核心目标 | 将企业私有文档、业务资料和多媒体内容解析、切片、向量化，用于检索增强生成 | 从对话中提取并持久化用户相关信息，在后续会话中语义召回并注入 Prompt | 管理文件、表格、数据库、语雀、OSS 等外部或托管数据源，供应用安全访问 | +| 典型数据形态 | PDF、DOCX、Markdown、Excel、图片、音视频、FAQ、产品手册、参数表等 | 对话消息提炼出的记忆片段、自定义写入内容、用户画像结构化属性 | 文件、表格、MySQL、PostgreSQL、PolarDB-X 2.0、语雀知识库、OSS 对象 | +| 输入方式 | 控制台上传、本地文件导入、OSS 或数据连接器导入；文档搜索类知识库可通过 API 上传和建索引 | 调用 AddMemory 接口传入 `messages` 或 `custom_content`，也可通过 OpenClaw 插件自动捕获 | 控制台创建连接器并配置存储位置或连接信息；文件/表格可导入，数据库可从 DMS 导入或手动配置 | +| 输出或使用结果 | 返回检索切片，或在知识问答服务中生成带引用的自然语言答案 | 返回与当前对话语义相关的记忆片段或用户画像，用于补充 Prompt | 为应用提供可访问的数据源；数据库连接器在满足条件时可执行 SQL 查询，OSS 可配合检索工具搜索文件 | +| 检索/查询机制 | 向量检索、关键词检索、排序模型精排、阈值过滤、标签和元数据过滤、多知识库路由 | 基于用户标识隔离命名空间，按 `messages` 语义检索记忆，常用 `top_k` 控制返回条数 | 平台托管数据可导入后使用；流处理连接器保留原数据源并实时访问；数据库 SQL 查询依赖 DMS 导入方式 | +| 支持模型与算法 | 生成模型可挂载千问、DeepSeek、Llama 等；向量模型包括 `text-embedding-v4`、`text-embedding-v3`、`multimodal-embedding-v1`、`qwen3-vl-embedding`；排序模型包括 `qwen3-rerank`、`qwen3-vl-rerank` | 由长期记忆服务完成记忆提炼、向量化和语义召回；OpenClaw 插件可在 Agent 生命周期中自动捕获和召回 | 本身是数据源管理能力，不直接等同于生成模型或向量模型；部分文件、表格、OSS 场景可作为后续知识库或应用工具的数据入口 | +| API 端点或入口 | 控制台知识库、智能体挂载、工作流“知识库”节点、知识检索服务、知识问答服务；文档搜索类知识库支持 SDK 或签名 HTTP 请求 | `https://dashscope.aliyuncs.com/api/v2/apps/memory/*`，常用 `/add`、`/memory_nodes/search`；OpenClaw 插件也可接入 | 百炼控制台“数据连接”页面；连接器创建和管理主要在控制台完成，数据库连接器可配合 DMS、DTS、EventBridge 等能力 | +| 关键参数 | TopK、排序模型、相似度阈值、最大召回数量、知识库权重、标签过滤、结构化字段/元数据过滤、切片方式 | `user_id`、`memory_library_id`、`project_id`、`profile_schema`、`top_k`、`meta_data`；OpenClaw 中还有 `autoCapture`、`autoRecall`、`minScore` | 连接器类型、网络类型、数据库地址和端口、DMS 数据源、OSS Bucket 标签、文件类目、表结构、标签、描述信息 | +| 数据隔离与权限 | 按[业务空间](../concepts/workspace.md)和知识库管理；可通过标签、元数据和知识库绑定范围控制召回 | 按 `user_id` 隔离不同用户记忆命名空间，可指定记忆库或画像规则 | 依赖主账号/RAM 授权、[业务空间](../concepts/workspace.md)、数据源访问权限、OSS 标签、数据库账号权限等 | +| 生命周期与更新 | 文档导入、解析、切片、索引后发布；更新资料需重新导入或更新索引；部分能力需创建时配置 | 对话后持续新增、去重、更新；控制台规则可能配置 7/30/180 天或永不过期，API 场景需注意默认规则差异 | 数据连接器保存连接配置；平台托管数据可导入更新，流处理连接器通常读取原数据源最新状态 | +| 成本关注点 | 向量化、排序、检索召回片段数、生成模型上下文 Token、存储规格 | 记忆提炼、存储、召回和注入 Prompt 带来的 Token 与调用成本 | 平台存储额度、表格超额按量付费、数据库/OSS/网络侧资源成本，以及后续应用调用成本 | +| 主要限制 | 部分知识库能力仅限特定地域；知识库类型创建后不可改；文档搜索类 API 不应套用到所有类型；向量维度不可修改 | 统一用户标识很关键；OpenClaw 插件暂不支持按 Agent 独立配置；画像字段设计不当会影响提取质量 | 数据库仅 DMS 导入方式支持 SQL 查询；OSS 需开通向量检索服务并配置标签；表格结构创建后不可修改；语雀仅支持公网版本 | +| 最适合的问题 | “根据公司制度回答问题”“检索产品手册并给出处”“在多份文档中比较参数” | “记住用户偏好”“跨会话延续个人任务”“根据历史对话补充上下文” | “让应用访问某个数据库/OSS/语雀/文件集合”“统一管理数据源连接”“实时查询业务数据” | + +## 适用场景建议 + +### 选择知识库的场景 + +当核心需求是让大模型基于相对稳定的资料回答问题、总结内容或提供可追溯引用时，应优先选择知识库。典型场景包括企业制度问答、产品手册检索、售后 FAQ、合同或规章查询、多媒体资料检索等。知识库适合把非结构化或半结构化资料转化为可检索切片，再通过智能体、工作流、知识检索服务或知识问答服务接入应用。 + +如果业务已经有自己的提示词编排和生成链路，可以使用知识检索服务只取回切片，再自行控制上下文拼装和模型调用；如果希望快速获得端到端问答能力，可以使用知识问答服务。对于多知识库混合场景，应重点设计知识库路由、权重、TopK、排序模型、阈值、标签和元数据过滤，避免召回噪声过多或遗漏关键资料。 + +### 选择记忆库的场景 + +当核心需求是让智能体“记住用户”而不是“检索企业资料”时，应选择记忆库。它适用于个人助理、长期陪伴、客户偏好管理、跨会话任务延续、用户画像沉淀等场景。例如，系统需要记住某个用户每天的提醒习惯、常用地址、职业背景、沟通偏好，或从多次对话中逐步完善画像，就应使用记忆库。 + +开发者需要特别关注 `user_id` 的设计，因为它直接决定记忆命名空间的隔离方式。若接入自有应用，通常使用 AddMemory 在对话结束后写入，使用 SearchMemory 在下一轮对话前召回；若使用 OpenClaw Agent，可通过插件自动完成捕获和召回。对于用户画像，应设计清晰且互斥的字段，避免同义字段重复导致提取结果不稳定。 + +### 选择数据连接的场景 + +当核心需求是把外部数据源纳入百炼平台统一管理，或让应用访问仍保留在原系统中的业务数据时，应选择数据连接。它适合连接数据库、OSS、语雀、文件和表格等数据源，尤其适合需要实时访问或集中管理数据源权限的场景。例如，应用需要读取业务数据库中的订单状态、连接 OSS 中的文件、导入表格数据，或管理语雀知识库入口，都应先评估数据连接。 + +数据连接不等同于知识库，也不自动提供完整 RAG 效果。对于文件、表格、OSS 等内容，如果最终目标是面向自然语言问答，通常还需要结合知识库、工具调用或应用编排使用。对于数据库连接器，要注意只有从 DMS 导入的数据源支持执行 SQL 查询，手动创建的自定义数据源不能直接执行 SQL。 + +## 面向开发者的选型路径 + +1. **先判断数据是否属于用户长期上下文**：如果信息围绕单个用户随对话不断积累，并且需要跨会话召回，优先使用记忆库；不要把用户偏好写入通用知识库，否则容易造成隔离、更新和隐私管理问题。 +2. **再判断数据是否需要检索增强生成**：如果数据是企业文档、产品资料、FAQ、多媒体资料，并且希望模型基于切片回答，优先使用知识库；重点调优切片、向量模型、排序、阈值和过滤条件。 +3. **最后判断是否只是需要连接数据源**：如果目标是管理数据库、OSS、语雀、文件或表格入口，或者让应用实时访问原系统数据，优先使用数据连接；后续是否转成知识库或工具调用取决于应用形态。 +4. **组合使用时明确边界**：数据连接可以作为数据导入或实时访问入口，知识库负责把资料检索给模型，记忆库负责保存用户个性化上下文。三者可以在同一个智能体中协同，但不要让单一能力承担所有职责。 + +## 常见组合架构 + +| 架构目标 | 推荐组合 | 说明 | +| --- | --- | --- | +| 企业知识问答 | 数据连接或文件上传 + 知识库 + 知识问答服务 | 数据连接负责接入文件或 OSS，知识库负责解析、切片、检索，知识问答服务负责生成答案和引用 | +| 个性化客服助手 | 知识库 + 记忆库 | 知识库回答产品和政策问题，记忆库保存客户偏好、历史诉求和画像，生成时同时注入企业知识与用户上下文 | +| 业务数据查询助手 | 数据连接 + 工作流或工具调用 + 可选知识库 | 数据连接管理数据库或表格入口，工作流负责查询和生成；若有字段说明、指标口径文档，可另建知识库辅助解释 | +| [多模态](../concepts/multimodal.md)资料检索 | 知识库 + OSS 或文件连接器 | OSS 或文件连接器管理资料来源，知识库使用[多模态](../concepts/multimodal.md)向量和排序模型完成图片、富文本文档或音视频检索 | +| 长期 Agent | 记忆库 + 知识库 + 数据连接 | 记忆库提供跨会话个人上下文，知识库提供稳定业务知识，数据连接提供实时业务状态或外部系统数据 | + +## 技术选型结论 + +如果用一句话区分：知识库解决“从资料中找答案”，记忆库解决“让应用记住用户”，数据连接解决“让平台接上数据源”。知识库更偏 RAG 检索与问答效果，记忆库更偏用户级长期上下文，数据连接更偏数据源接入、权限和实时访问。 + +开发者落地时不应只按名称选择，而应围绕数据生命周期、隔离粒度、访问方式和生成链路来设计：稳定资料进入知识库，个体用户历史进入记忆库，外部系统入口进入数据连接。这样可以减少权限混淆、降低无关召回，并为后续调优、审计和成本控制留下清晰边界。 + +## 被对比主题页 + +- [knowledge base](../guides/knowledge-base.md) +- [memory library overview](../guides/memory-library-overview.md) +- [data connection overview](../guides/data-connection-overview.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-vs-memory-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-vs-memory-comparison.md new file mode 100644 index 00000000..40fb6492 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/knowledge-vs-memory-comparison.md @@ -0,0 +1,56 @@ +# 知识库与记忆体系对比 + +在构建百炼平台上的智能体应用时，开发者常常需要为大模型补充"模型本身不知道的信息"。这类信息大致分为两类：一类是**企业级共享知识**（产品手册、FAQ、业务文档等），对应**知识库（RAG）**；另一类是**用户级个性化上下文**（用户偏好、历史事件、画像属性等），对应**记忆体系**——包括控制台侧的**记忆库（Memory Library）**与开放接口侧的**长期记忆（新）API**。三者都通过"提取/切片 → 向量化 → 语义检索 → 注入 Prompt"的模式增强生成效果，但定位、数据粒度和接入方式差异显著。本文从技术选型角度对三者进行对比。 + +## 关键维度对比 + +| 维度 | 知识库（RAG） | 记忆库（控制台/插件） | 长期记忆（新）API | +| --- | --- | --- | --- | +| 核心定位 | 企业私有知识增强，回答前检索文档切片 | 跨会话个性化记忆的可视化管理与零侵入接入 | 记忆片段/用户画像的 RESTful 增删改查 | +| 数据来源 | PDF、DOCX、Markdown、Excel、图片、音视频等文档 | 用户与智能体的对话内容（自动提取）或 `custom_content` 直写 | 对话 `messages`（自动提取）或 `custom_content`（≤512 字符） | +| 数据粒度 | 文档解析后的切片（chunk） | 记忆片段 + 结构化用户画像 | 记忆片段（memory node）+ 画像模板/用户画像 | +| 数据隔离维度 | 知识库 / [业务空间](../concepts/workspace.md)（Workspace） | `userId` 记忆空间隔离 | `user_id`（最大 64 字符），不同 ID 完全隔离 | +| 输入格式 | 多格式文件（本地上传 / OSS / 数据连接器） | 对话 messages、custom_content、画像模板 | JSON 请求体：`messages` 与 `custom_content` 互斥 | +| 输出格式 | 检索切片（知识检索服务）或生成式回答（知识问答服务） | 召回的记忆条目（自动注入 Prompt） | `memory_nodes` 数组（含 `content`、`event`、`old_content`） | +| 检索方式 | 向量 + 关键词双路召回，可加排序模型精排 | 语义检索（服务端向量化），`topK`/`minScore` 控制 | 语义检索，支持 `top_k`（1~100）、`min_score`、`enable_rerank`、query 重写 | +| 向量/排序模型 | `text-embedding-v4/v3`、`multimodal-embedding-v1`、`qwen3-rerank` 等，可选可配 | 由百炼服务端托管，不可自选 | 由百炼服务端托管，不可自选 | +| API 端点 | 阿里云百炼 SDK / 签名 HTTP（租约上传 → 建索引 → 检索） | OpenClaw 插件或同长期记忆 API | `https://dashscope.aliyuncs.com/api/v2/apps/memory/*` | +| 认证方式 | AccessKey + `WORKSPACE_ID`（签名请求） | DashScope API Key（`sk-` 开头） | `Authorization: Bearer $DASHSCOPE_API_KEY` | +| 挂载/接入方式 | 智能体应用、工作流节点、知识检索/知识问答服务、API 集成 | 控制台可视化管理、HTTP API、OpenClaw 插件（自动捕获/召回） | 纯 HTTP API，Python 可用 `agentscope-runtime` 封装类 | +| 数据有效期 | 长期保存，随文档管理更新 | 规则可配 7/30/180 天或永不过期（默认规则 180 天） | 文档描述"暂无失效日期"（不指定 `project_id` 时走默认规则） | +| 主要限制 | 仅中国站华北2（北京）地域；单次控制台导入 ≤50 文件；检索/问答服务最多绑 15 个知识库 | 所有 OpenClaw Agent 共享同一记忆；不支持 Coding Plan API Key | 总[限流](../concepts/rate-limit.md) 3000 QPM；add 120 QPM；search 300 QPM；messages ≤50 条 | +| 计费关注点 | 规格（标准版/旗舰版）、向量化与排序模型调用、召回切片带来的 Token 消耗 | 记忆读写走长期记忆 API 计量 | 按 API 调用计量，受 QPM [限流](../concepts/rate-limit.md)约束 | +| 典型场景 | 企业文档问答、产品手册检索、FAQ、参数表查询 | 个人助理记住用户偏好、跨会话延续上下文 | 自研应用精细控制记忆的写入、检索、更新与删除 | + +## 能力侧重差异 + +- **知识库**面向"多人共享、以文档为中心"的知识：具备完整的数据处理流水线（解析 → 切片 → 向量化 → 索引），并提供命中测试、Meta 信息抽取、标签/元数据过滤、相似度阈值等丰富的调优手段，还支持 SLS 日志审计。它回答的是"这个问题在企业资料里怎么说"。 +- **记忆库**面向"单人专属、以对话为中心"的信息：自动从对话中提取记忆片段并去重、动态更新，配合用户画像模板持久化结构化属性。它回答的是"这个用户是谁、喜欢什么、之前说过什么"。 +- **长期记忆（新）API** 是记忆库能力的开放接口层：提供 AddMemory / SearchMemory / ListMemory / DeleteMemory / UpdateMemory 及画像模板系列接口，适合需要把记忆能力嵌入自有系统、并对读写时机做精细编排的开发者。 + +## 适用场景与选型建议 + +1. **企业知识问答、文档检索类需求 → 选知识库。** 数据是文档而非对话，需要多格式解析、多知识库联合检索、召回调优和引用溯源时，知识库是唯一合适的方案。若已有生成链路只需要检索切片，用知识检索服务；想要端到端问答，用知识问答服务。 +2. **智能体需要"记住用户" → 选记忆体系。** 用户偏好、历史约定、画像属性等以用户为维度的碎片信息，不适合建成文档知识库；用记忆库自动提取 + 语义召回更贴合。 +3. **使用 OpenClaw 或希望零改造接入 → 记忆库 + OpenClaw 插件。** `autoCapture` / `autoRecall` 钩子无需改动 Agent 代码即可获得跨会话记忆，注意所有 Agent 共享同一记忆空间。 +4. **自研应用需要完全掌控记忆生命周期 → 直接调长期记忆（新）API。** 可自行决定何时写入（`messages` 提炼或 `custom_content` 直写）、何时检索（`top_k`、`min_score`、重排序、query 重写）、何时更新或删除，并通过 `user_id` 做多租户隔离。设计时需预留 QPM [限流](../concepts/rate-limit.md)余量。 +5. **两者组合是常态。** 成熟的智能体通常同时挂载知识库（回答业务问题）与记忆体系（保持个性化上下文）：检索企业知识用 RAG，召回用户记忆用长期记忆 API，再一并注入 Prompt。 + +## 选型速判 + +| 问题 | 是 → 建议 | +| --- | --- | +| 数据是文档/手册/FAQ 等静态资料？ | 知识库 | +| 信息以单个用户为维度、来自对话？ | 记忆库 / 长期记忆 API | +| 需要控制台可视化管理与有效期规则？ | 记忆库 | +| 需要在自有后端精细控制记忆增删改查？ | 长期记忆（新）API | +| 用 OpenClaw 且不想写代码？ | 记忆库 OpenClaw 插件 | +| 既要答业务问题又要记住用户？ | 知识库 + 记忆体系组合 | + +## 被对比主题页 + +- [knowledge base](../guides/knowledge-base.md) +- [memory library overview](../guides/memory-library-overview.md) +- [long term memory new](../api/long-term-memory-new.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-api-vs-guide.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-api-vs-guide.md new file mode 100644 index 00000000..8f405aee --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-api-vs-guide.md @@ -0,0 +1,40 @@ +# 托管智能体 API 与指南对比 + +百炼平台的 Managed Agents（托管智能体）同时提供了面向开发者的 REST API 文档和面向使用者的操作指南。两者覆盖同一套运行时，但视角与粒度不同：API 文档聚焦资源模型、端点签名与字段级约定，适合集成开发；指南聚焦概念理解、控制台操作与最佳实践，适合上手与配置。本文从多个维度对比两者，帮助开发者根据需要选择参考来源。 + +## 关键维度对比 + +| 维度 | API 文档（managed-agents-api） | 操作指南（managed-agents） | +| --- | --- | --- | +| 目标读者 | 后端开发者、SDK 使用者 | 产品经理、应用配置者、初次使用者 | +| 内容核心 | 资源模型、REST 端点、请求/响应字段、版本管理 | 概念解释、控制台向导、工具能力、上下文挂载 | +| 资源覆盖 | Agent、Environment、Session、Event、Skill、File 六类 | Agent、Environment、Session、Event 四类核心，附带工具与 Skill 说明 | +| 端点列举 | 每类资源均给出完整端点表（含创建/获取/列出/更新/归档/删除等） | 仅在流程中点到关键端点（`POST /agents`、`POST /environments`、`POST /sessions`、事件相关端点） | +| 认证方式 | 明确给出基地址拼接规则、API Key 鉴权头、`x-request-id` 追踪 | 未单独说明，依赖 API 文档 | +| 地域支持 | 标注当前仅 `cn-beijing` | 未提及 | +| 工具说明 | 通过 Skill 资源间接涉及 | 直接列出 7 个内置工具（bash、read、write、edit、glob、grep、download_file）及 MCP 服务接入 | +| 文件处理 | 独立 File 资源，含上传/查询/删除端点与配额（20 MB / 100 GB / 30 天） | 提及文件挂载与 `/mnt/session/uploads` 路径约定，限制标注为 10 MB | +| 版本管理 | 强调 Agent 版本递增、会话锁定版本、Skill 挂载需指定版本号 | 未展开版本机制 | +| 状态机 | 会话状态 `idle → running → idle/terminated` | 提及中断续接与工具审批，指向会话状态机文档 | +| 快速开始 | 五步流程：创建 Agent → Environment → Session → 发送 Event → 接收 SSE | 四步流程：配置智能体 → 配置环境 → 发起会话 → 发送事件并接收响应 | +| 典型场景 | 端到端 API 集成、自动化编排、SDK 封装 | 控制台配置、预览调试、理解概念与能力边界 | +| 模型示例 | 未给具体模型名 | 控制台向导示例 `qwen3.7-plus`，API 示例 `qwen3-max`，提示以控制台下拉为准 | + +## 适用场景建议 + +- **正在做 API 集成或 SDK 开发**：以 API 文档为主。它提供完整的端点列表、字段定义、版本与配额约定，是编码实现的唯一权威来源。 +- **首次接触 Managed Agents 或使用控制台配置**：以指南为主。它用通俗语言解释智能体、运行环境、会话、事件四要素，并给出控制台操作路径与工具能力概览。 +- **需要理解工具能力与沙箱行为**：优先看指南。它明确列出内置工具、MCP 服务接入方式与文件路径约定；API 文档只在 Skill/File 资源层面间接涉及。 +- **需要确认认证、地域、版本管理等运维细节**：优先看 API 文档。鉴权头、基地址、地域限制、版本递增与锁定规则均在此处。 +- **排查字段不一致或端点差异**：API 文档已标注总览页与详情页之间的方法名差异（如 Agent 更新端点 `PATCH` vs `POST`），开发时应以各资源详情页为准；指南中模型名称不一致同样提示以控制台实际可选为准。 + +## 技术选型参考 + +两者并非替代关系，而是互补。推荐的使用方式是：用指南建立整体认知与控制台操作能力，用 API 文典驱动代码实现与字段级约定。在自动化编排、多会话管理、Skill 版本控制等深度场景下，API 文档不可或缺；在团队协作、原型验证、工具选型等场景下，指南更高效。 + +## 被对比主题页 + +- [managed agents api](../api/managed-agents-api.md) +- [managed agents](../guides/managed-agents.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-api-vs-guides.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-api-vs-guides.md new file mode 100644 index 00000000..b8e1bf5c --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-api-vs-guides.md @@ -0,0 +1,54 @@ +# 托管智能体 API 与使用指南对比 + +百炼平台的 Managed Agents（托管智能体）提供两套文档视角：**API 参考**面向程序化集成，定义完整的 REST 端点、请求格式与资源生命周期；**使用指南**面向产品使用者，描述控制台操作流程、内置工具与最佳实践。两者描述的是同一套运行时，但侧重点不同。本文从开发者技术选型角度，对比两者的覆盖范围与适用场景，帮助团队决定在哪个阶段参考哪份文档。 + +## 关键维度对比 + +| 维度 | API 参考（managed-agents-api） | 使用指南（managed-agents） | +|------|-------------------------------|---------------------------| +| 目标读者 | 后端/平台工程师，需程序化创建与管理智能体资源 | 应用开发者与运营人员，通过控制台或最小代码跑通流程 | +| 接入方式 | REST API + SDK，所有操作均有对应端点 | 控制台向导为主，辅以关键 API 示例 | +| 资源模型覆盖 | 六类资源（Agent、Environment、Session、Event、Skill、File）的完整 CRUD 端点 | 四个核心对象（Agent、Environment、Session、Event）的概念与流程说明 | +| 认证说明 | 详细描述 Endpoint 拼装规则（`{workspace_id}.{region}.maas.aliyuncs.com`）、Bearer Token 鉴权、`x-request-id` 排障 | 不涉及认证细节，默认由控制台处理 | +| 版本管理 | 明确 Agent 版本递增、乐观锁更新、Skill 版本锁定等机制 | 未展开版本管理细节 | +| 工具与沙箱 | 仅定义 Environment 的沙箱配置端点 | 列出 7 个内置工具（bash、read、write、edit、glob、grep、download_file）及 MCP 服务接入方式 | +| 文件上传限制 | 单文件 20 MB，工作空间总容量 100 GB，保留 30 天 | 单文件 10 MB（上下文挂载场景） | +| 状态机描述 | Session 状态机（idle → running → idle / terminated）及 Event SSE 流 | 会话中断与续接、工具审批的用户视角说明 | +| 调试手段 | 无（纯接口层） | 控制台预览调试标签页，支持按事件类型筛选 | +| 典型流程 | 五步：创建 Agent → 创建 Environment → 创建 Session → 发送 Event → SSE 接收 | 四步：配置智能体 → 配置运行环境 → 发起会话 → 发送事件并接收响应 | + +## 内容互补关系 + +两份文档并非替代关系，而是同一能力的不同切面： + +- **API 参考**回答"怎么调"：端点路径、请求体字段、状态码、配额限制、版本锁定规则。适合编写 SDK 封装、CI/CD 自动化、多租户资源管理。 +- **使用指南**回答"怎么用"：控制台操作路径、内置工具清单、挂载路径约定（`/mnt/session/uploads`）、调试技巧。适合快速验证想法、配置智能体行为、排查执行问题。 + +## 适用场景建议 + +| 场景 | 推荐参考 | +|------|---------| +| 首次接触 Managed Agents，需要 5 分钟跑通 Demo | 使用指南（控制台向导 + 快速开始） | +| 将智能体能力嵌入自有产品后端 | API 参考（完整端点 + 认证方式） | +| 需要自定义工具链（Skill zip 包上传与版本管理） | API 参考（Skill 端点 + 安全扫描流程） | +| 为智能体配置预装依赖与网络策略 | 两者结合：指南说明 `config.packages` 用途，API 参考给出 Environment 请求体结构 | +| 排查会话异常、查看工具调用过程 | 使用指南（预览调试 + 事件类型筛选），辅以 API 参考的 Event 列表端点做程序化日志采集 | +| 多会话文件共享与上下文管理 | 使用指南（挂载时机、路径约定、会话隔离机制） | +| 实现会话中断续接与工具审批流 | API 参考（Session 状态机 + Event 注入端点） | + +## 技术选型建议 + +1. **原型验证阶段**：优先跟随使用指南在控制台完成配置，利用预览调试快速迭代提示词与工具组合，无需编写代码。 +2. **生产集成阶段**：切换到 API 参考，程序化管理 Agent 版本与 Environment 配置；注意 Agent 更新采用全量替换 + 乐观锁，需在应用层维护版本号。 +3. **注意事项**： + - API 总览与资源详情页在更新端点的 HTTP 方法标注上存在差异（`PATCH` vs `POST`），以各资源详情页为准。 + - 文档示例中的模型名称不一致（`qwen3.7-plus` vs `qwen3-max`），以控制台模型下拉列表中实际可选的模型 ID 为准。 + - 文件上传限制在不同入口可能不同（API 侧 20 MB vs 上下文挂载 10 MB），集成时应以实际调用返回的错误信息为准。 + - 当前仅支持 `cn-beijing` 地域，Endpoint 按工作空间 ID 拼装。 + +## 被对比主题页 + +- [managed agents api](../api/managed-agents-api.md) +- [managed agents](../guides/managed-agents.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-comparison.md new file mode 100644 index 00000000..e5038dbf --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-comparison.md @@ -0,0 +1,43 @@ +# 托管智能体指南与 API 对比 + +本页对比百炼 Managed Agents 的两份核心文档：面向控制台使用者的「托管智能体指南」与面向开发者的「Managed Agents API」。两者描述的是同一套托管运行时，但视角、详略与关注点差异显著。开发者做技术选型时应先明确自身入口（控制台可视化操作 vs API/SDK 编程调用），再据此选择文档作为主参考。 + +## 关键维度对比 + +| 维度 | 托管智能体指南 | Managed Agents API | +| --- | --- | --- | +| 文档定位 | 概念讲解 + 控制台操作向导 | REST/SDK 接口参考 + 端到端调用流程 | +| 目标读者 | 平台使用者、低代码/半托管场景开发者 | 需编程集成的后端/应用开发者 | +| 核心资源对象 | 4 类：Agent、Environment、Session、Event | 6 类：Agent、Environment、Session、Event、Skill、File | +| 使用入口 | 控制台向导 + 预览调试标签页 | REST API + SDK，按地域拼装 endpoint | +| 认证方式 | 未展开，依赖控制台登录态 | `Authorization: Bearer `，单 Key 访问工作空间全部资源 | +| 地域支持 | 未显式说明 | 当前仅 `cn-beijing` | +| 默认工具集 | 7 个内置工具（bash/read/write/edit/glob/grep/download_file）全选，可取消 | 通过 Environment 配置沙箱类型与预装依赖，工具粒度未在 API 文档单独枚举 | +| 上下文/文件挂载 | 强调 `/mnt/session/uploads` 路径约定、会话隔离、运行时追加 | 抽象为 File 资源，经安全审核（checking → available/rejected/type_rejected）后挂载 | +| 单文件上传上限 | 10 MB | 20 MB | +| 工作空间配额 | 未提及 | 总容量 100 GB，保留期 30 天 | +| 版本管理 | 未展开 | Agent 每次更新自动递增 version，会话创建时锁定版本；Skill 挂载须指定具体版本号 | +| 状态机描述 | 提及 idle/中断/续接/工具审批，指向会话文档详述 | 明确 idle → running → idle/terminated，含归档与硬删除语义 | +| 安全扫描 | 未提及 Skill/File 审核流程 | Skill：checking → active/rejected；File：checking → available/rejected/type_rejected | +| 事件流 | 会话级 SSE，按事件类型筛选（User/Agent/Tool/Tool_output/Error/Model/System） | `GET /sessions/{id}/events/stream` 长连接 SSE，含分页列出历史 | +| 典型流程步数 | 4 步（配置智能体 → 配置环境 → 发起会话 → 发送事件接收响应） | 5 步（创建 Agent → 配置 Environment → 创建 Session → 发送 Event → 订阅 SSE） | +| API 端点示例 | 仅给出创建类端点（agents/environments/sessions） | 完整 CRUD + 版本/归档/下载端点表格，覆盖全部资源 | +| 已知不一致 | 控制台示例模型名 `qwen3.7-plus` 与 API 示例 `qwen3-max` 不一致 | Agent/Environment/Session 更新端点在总览页标 `PATCH`、详情页标 `POST` | + +## 适用场景建议 + +- **选「托管智能体指南」**：需要快速理解 Managed Agents 是什么、与无状态智能体应用的区别、以及如何用控制台向导跑通最小流程；适合概念入门、POC 验证、非编程使用者。 +- **选「Managed Agents API」**：需要把托管智能体嵌入自有产品、做编程化编排与运维；需要精确的端点、字段、版本与状态机语义；适合生产集成、自动化流水线、SDK 封装层开发。 + +## 技术选型参考 + +1. 若团队无后端开发资源、仅做内部工具或演示，优先走控制台向导，以指南为主参考。 +2. 若需要多智能体复用、版本快照、Skill 市场化分发、文件安全审核等企业级能力，必须基于 API 文档设计资源治理与生命周期策略。 +3. 两份文档在模型名、HTTP method 上存在不一致，落地时以控制台模型下拉列表与各资源详情页为准，必要时提工单附 `x-request-id` 确认。 + +## 被对比主题页 + +- [managed agents](../guides/managed-agents.md) +- [managed agents api](../api/managed-agents-api.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-guide-vs-api.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-guide-vs-api.md index d5b6d709..dc5ab2f4 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-guide-vs-api.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-guide-vs-api.md @@ -1,52 +1,64 @@ -# 托管智能体：指南与 API 对比 +# 托管智能体使用指南与 API 参考对比 -百炼平台的 Managed Agents（托管智能体）提供了两套面向不同使用者的文档视角：**使用指南**侧重在控制台向导中完成端到端配置，适合快速上手与业务验证；**API 参考**则给出完整的 REST 资源模型与端点定义，适合工程化集成与自动化。二者描述的是同一套托管运行时——平台在服务端托管会话状态、沙箱环境与工具执行——但抽象层级、覆盖范围与目标读者不同。本页从关键维度做对比，帮助开发者按阶段选择合适的入口。 +百炼平台的 Managed Agents（托管智能体）提供两类文档入口：面向使用者的**使用指南**（guides/managed-agents）和面向开发者的 **API 参考**（api/managed-agents-api）。两者描述的是同一套产品能力，但侧重点、抽象层次和适用阶段不同。本文从开发者技术选型角度，对两类文档的定位、覆盖范围与关键细节进行对比，帮助读者快速判断在不同阶段应参考哪份文档。 ## 关键维度对比 | 维度 | 使用指南（guides/managed-agents） | API 参考（api/managed-agents-api） | -| --- | --- | --- | -| 定位 | 概念讲解 + 控制台向导操作流程 | REST 资源模型 + 端点契约 | -| 目标读者 | 初次接入、业务验证、快速体验 | 后端集成、SDK 调用、自动化编排 | -| 核心对象 | Agent、Environment、Session、Event（4 类） | Agent、Environment、Session、Event、Skill、File（6 类） | -| 操作方式 | 控制台向导 + 预览调试标签页 | 直接调用 REST API / SDK | -| 认证方式 | 控制台登录（隐式） | HTTP Header 携带 `Authorization: Bearer ` | -| API 基地址 | 未强调 | `https://{workspace_id}.{region}.maas.aliyuncs.com/api/v1/agentstudio` | -| 支持地域 | 未强调 | 当前仅 `cn-beijing` | -| Agent 端点 | `POST /api/v1/agentstudio/agents`（创建） | 创建/获取/列出/更新/归档/列出版本 全套 | -| 版本管理 | 未展开 | 每次更新递增 `version`，会话锁定当时快照 | -| 事件流 | SSE 事件流（`GET /sessions/{id}/events/stream`） | SSE 长连接 + 事件历史分页（`GET /sessions/{id}/events`） | -| Skill 支持 | 提及可挂载预置工具组合 | 完整生命周期：上传/安全扫描/版本/下载/删除 | -| File 支持 | 上传挂载，路径 `/mnt/session/uploads` | 独立资源：上传/查询/列出/删除，审核状态机 | -| 文件大小上限 | 单文件 10 MB（上传挂载语境） | 单文件 20 MB，工作空间总量 100 GB，保留 30 天 | -| 状态机 | 会话状态、中断续接、工具审批（概念性描述） | `idle` → `running` → `idle` / `terminated`（显式定义） | -| 示例模型 | `qwen3-max`（向导示例另见 `qwen3.7-plus`，以下拉列表为准） | `qwen3-max` | -| 典型场景 | 多步工具调用、代码执行、文件处理等长时任务的快速搭建 | 生产环境集成、批量会话编排、CI/自动化流水线 | - -## 关键差异说明 - -- **抽象层级**：指南把流程压缩成"配置智能体 → 配置环境 → 发起会话 → 发送事件"四步，隐藏了大量端点细节；API 参考把每类资源的增删改查、版本、归档、软/硬删除逐一列清，是精确的契约。 -- **资源覆盖**：指南聚焦运行链路的 4 个核心对象（Agent / Environment / Session / Event），而 API 额外把 **Skill** 与 **File** 提升为一等资源，并明确了它们的安全扫描/审核状态机与挂载版本锁定规则。 -- **文件配额不一致**：指南上下文提到单文件 10 MB，API 参考给出 20 MB、工作空间 100 GB、保留 30 天。以 API 参考的配额为准，并注意实际以控制台/接口返回为准。 -- **端点方法差异**：API 总览页把部分更新端点标注为 `PATCH`，而各资源详情页标注为 `POST`（Agent / Environment / Session 均如此）。集成时以各资源详情页为准。 -- **版本与快照语义**：只有 API 参考明确了"会话创建时锁定 Agent 版本、Skill 挂载锁定具体 version、更新不影响已有会话"的重要约束，这对生产环境的稳定性至关重要。 +|------|----------------------------------|-----------------------------------| +| 目标读者 | 产品经理、初次使用者、需要快速上手的开发者 | 需要程序化集成的后端/平台开发者 | +| 操作方式 | 控制台向导 + API 均可完成 | 纯 REST API / SDK 调用 | +| 核心概念覆盖 | 四个对象：Agent、Environment、Session、Event | 六个资源：Agent、Environment、Session、Event、Skill、File | +| 端点信息 | 仅提及关键端点（如 `POST /agents`、`POST /sessions`） | 完整端点表，含全部 CRUD、归档、版本管理操作 | +| 认证说明 | 未展开 | 详细说明基地址拼装规则（`{workspace_id}.{region}.maas.aliyuncs.com`）、Bearer [Token](../concepts/token.md) 鉴权、`x-request-id` 排障 | +| 版本管理 | 未涉及 | Agent 更新自动递增版本号、会话创建时锁定版本、Skill 挂载需指定具体版本 | +| 状态机 | 提及会话状态机支持中断与续接 | 给出完整状态流转（`idle` → `running` → `idle` / `terminated`） | +| Skill 资源 | 仅说明 Skill 可挂载预置工具组合 | 完整生命周期：上传 zip → 安全扫描（`checking` → `active` / `rejected`）→ 挂载 → 版本更新 | +| File 资源 | 提及单文件不超过 10 MB（上传限制） | 独立 File 资源体系：单文件上限 20 MB、工作空间总量 100 GB、保留期 30 天、审核状态流转 | +| 上下文与资源挂载 | 详细说明挂载时机、路径约定（`/mnt/session/uploads`）、会话隔离机制 | 侧重 API 字段（`resources` 字段、`POST /sessions/{id}/resources`） | +| 调试手段 | 控制台预览调试标签页，按事件类型筛选查看 | 通过 `GET /sessions/{id}/events` 分页列出事件历史 + SSE 流式订阅 | +| 地域支持 | 未明确 | 明确当前仅支持 `cn-beijing` | + +## 内容互补关系 + +两份文档并非替代关系，而是互补： + +- **使用指南**解释了"为什么这样设计"：Managed Agents 与智能体应用的本质区别（有状态 vs 无状态、独立沙箱 vs 共享运行时）、会话级事件流与响应级[流式输出](../concepts/streaming.md)的差异、挂载资源的隔离语义等。这些概念是正确调用 API 的前提。 +- **API 参考**回答了"具体怎么调"：每个资源的完整端点列表、请求/响应结构、乐观锁（更新需带 `version`）、软归档与硬删除的区别、配额限制等工程细节。 ## 适用场景建议 -- **选择「使用指南」入口**： - - 第一次接触 Managed Agents，想快速理解 Agent / Environment / Session / Event 之间的关系。 - - 需要在控制台向导中拖拽配置、用预览调试标签页按事件类型（User、Agent、Tool、Tool_output 等）观察执行过程。 - - 做业务原型、Demo 或轻量验证，暂不需要写代码。 +### 优先阅读使用指南的场景 + +- 初次接触 Managed Agents，需要理解产品定位与核心概念 +- 评估 Managed Agents 与智能体应用的技术选型 +- 使用控制台向导完成快速验证（PoC） +- 需要理解沙箱隔离、文件挂载路径约定、会话中断续接等运行时行为 + +### 优先阅读 API 参考的场景 -- **选择「API 参考」入口**： - - 需要把托管智能体嵌入后端服务、批量创建/复用 Agent 与 Environment。 - - 关注鉴权、地域、版本锁定、状态机、配额等工程化细节。 - - 需要管理 Skill 上传/安全扫描/版本挂载，或对 File 做上传/审核/挂载的自动化处理。 - - 构建 CI/自动化流水线，依赖稳定的端点契约与 `x-request-id` 排障能力。 +- 将 Managed Agents 集成到自有产品或自动化流水线 +- 需要精细控制资源生命周期（版本管理、归档、删除策略） +- 开发 Skill 包并管理其发布与挂载流程 +- 需要处理文件上传审核状态、配额规划 +- 排查线上问题（利用 `x-request-id`、事件历史分页查询） -## 技术选型参考 +### 推荐的阅读顺序 -推荐路径是"**先指南、后 API**"：用指南在控制台跑通端到端流程、确认业务可行性，再切换到 API 参考完成工程化集成。两者并非二选一，而是覆盖同一运行时的不同生命周期阶段。集成落地时，务必以 API 参考为准处理认证（Bearer API Key）、地域限制（`cn-beijing`）、版本快照语义与端点方法差异，并按 API 侧的 20 MB / 100 GB / 30 天配额规划文件资源。 +1. 先通读使用指南，建立 Agent → Environment → Session → Event 的心智模型 +2. 按使用指南中的四步流程在控制台跑通一次端到端任务 +3. 切换到 API 参考，将控制台操作映射为对应的 REST 调用 +4. 针对 Skill 和 File 等进阶资源，以 API 参考为准进行程序化管理 + +## 已知文档差异 + +在使用两份文档时需注意以下不一致之处： + +| 差异点 | 使用指南 | API 参考 | 建议 | +|--------|----------|----------|------| +| 示例模型名称 | 控制台向导示例为 `qwen3.7-plus` | API 代码示例为 `qwen3-max` | 以控制台模型下拉列表中实际可选的模型 ID 为准 | +| 单文件大小限制 | 上传文件不超过 10 MB | File 资源单文件上限 20 MB | 10 MB 可能为控制台上传通道限制，API 直传上限为 20 MB；以实际调用返回为准 | +| Agent 更新端点 | 未涉及 | 总览页标注 `PATCH`，详情页标注 `POST` | 以资源详情页为准 | ## 被对比主题页 @@ -54,3 +66,4 @@ - [managed agents api](../api/managed-agents-api.md) + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-vs-api.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-vs-api.md new file mode 100644 index 00000000..ca1c6ebd --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-agents-vs-api.md @@ -0,0 +1,74 @@ +# 托管 Agent 使用方式对比：控制台 vs API + +百炼平台的 Managed Agents 支持两种使用方式：通过**控制台（可视化向导）**完成配置与调试，或通过 **REST API / SDK** 以编程方式全流程编排。两种方式底层共用同一套资源模型（Agent、Environment、Session、Event），但在操作入口、适用场景和集成深度上存在显著差异。本页帮助开发者根据团队技术栈和业务需求做出选型。 + +## 关键维度对比 + +| 维度 | 控制台 | API / SDK | +|------|--------|-----------| +| **操作入口** | 百炼控制台 Web UI，向导式交互 | REST API + 官方 SDK（Python/Node.js 等） | +| **认证方式** | 登录阿里云账号，控制台会话鉴权 | HTTP Header `Authorization: Bearer ` | +| **API Endpoint** | 不涉及（内部调用） | `https://{workspace_id}.{region}.maas.aliyuncs.com/api/v1/agentstudio` | +| **支持地域** | 控制台页面同步 | 当前仅支持 `cn-beijing` | +| **Agent 配置** | 填写表单：名称、模型、系统提示词、工具勾选 | `POST /agents`，JSON body 定义全部字段 | +| **Environment 配置** | 向导默认云端托管沙箱，可选预装依赖 | `POST /environments`，支持 `config.packages` 预装 apt/pip | +| **Session 创建** | 向导自动绑定 Agent ID + Environment ID | `POST /sessions`，手动指定 `agent_id` 与 `environment_id` | +| **事件交互** | 预览调试标签页直接对话，支持按事件类型筛选 | `POST /sessions/{id}/events` 发送消息；`GET /sessions/{id}/events/stream` SSE 订阅 | +| **版本管理** | 每次保存自动递增版本，UI 可查版本历史 | `GET /agents/{agent_id}/versions` 分页查询；更新时需携带 `version` 作乐观锁 | +| **文件挂载** | 直接上传本地文件到沙箱（≤10 MB/文件） | 先 `POST /files` 上传审核，审核通过后挂载到会话（≤20 MB/文件，工作空间总量 ≤100 GB） | +| **Skill 管理** | 控制台页面上传 zip 包、查看审核状态 | `POST /skills` 创建；`POST /skills/{id}/versions` 上传新版本；挂载时必须指定 `version` | +| **调试能力** | 内置预览调试标签页，事件流可视化过滤（User/Agent/Tool/Tool_output/Error/Model/System） | 需自行解析 SSE 事件流，适合集成到自动化测试或监控系统 | +| **自动化/CI 集成** | 不适合（手动操作为主） | 原生支持，可嵌入 CI/CD 流水线 | +| **资源软归档** | 控制台提供归档操作 | `POST /agents/{id}/archive`、`POST /sessions/{id}/archive` 等 | +| **错误追踪** | 控制台展示错误事件 | 每次响应携带 `x-request-id`，提工单时附上可加速定位 | + +## 四步典型流程对比 + +两种方式的核心流程完全对应，区别仅在操作载体： + +| 步骤 | 控制台 | API | +|------|--------|-----| +| 1. 配置智能体 | 填写名称、选择模型（如 `qwen3-max`）、输入系统提示词、勾选工具 | `POST /agents`，指定 `model`、`system_prompt`、`tools` | +| 2. 配置运行环境 | 向导选择默认沙箱，可选预装依赖 | `POST /environments`，`config.packages` 声明 apt/pip 依赖和网络策略 | +| 3. 发起会话 | 向导自动绑定，点击"开始"即创建 Session | `POST /sessions`，绑定 `agent_id` + `environment_id` | +| 4. 交互与接收响应 | 预览调试标签页直接输入消息，实时查看工具执行过程 | `POST /sessions/{id}/events` 发送消息；`GET /sessions/{id}/events/stream` 接收 SSE 事件 | + +## 文件配额差异 + +| 限制项 | 控制台上传 | API 上传 | +|--------|-----------|---------| +| 单文件大小 | ≤ 10 MB | ≤ 20 MB | +| 工作空间总容量 | 文档未单独说明 | ≤ 100 GB | +| 文件保留期 | 文档未单独说明 | 30 天 | +| 审核流程 | 自动（上传后即挂载） | 异步审核（`checking` → `available` / `rejected` / `type_rejected`），仅 `available` 可挂载 | + +## 适用场景建议 + +### 推荐使用控制台的场景 + +- **快速原型验证**：首次体验 Managed Agents，5 分钟内跑通端到端流程，无需写代码。 +- **提示词调优**：通过预览调试标签页反复迭代系统提示词，实时观察工具执行轨迹。 +- **非技术成员协作**：产品经理、运营人员需要配置或查看 Agent 行为，无需了解 API 细节。 +- **故障排查**：按事件类型（Tool_output、Error 等）筛选，快速定位执行异常。 + +### 推荐使用 API / SDK 的场景 + +- **生产环境集成**：将 Agent 调用嵌入业务系统（后端服务、数据处理流水线、自动化脚本）。 +- **CI/CD 自动化**：通过 API 创建/归档 Agent、批量跑测试用例、断言 SSE 事件流输出。 +- **动态资源挂载**：在会话运行中通过 `POST /sessions/{id}/resources` 实时追加文件，无需重启。 +- **版本管理与灰度**：精确控制 Agent `version`、Skill `version`，实现灰度发布与回滚。 +- **大规模并发**：程序化创建多个 Session 并行执行，控制台不支持批量操作。 + +## 技术选型建议 + +1. **探索期**：先用控制台完成 Agent 配置和调试，确认提示词和工具组合有效后，再通过 API 固化到代码。 +2. **生产期**：API 方式为首选；控制台保留用于监控和应急调试。 +3. **混用模式**：在控制台创建并调试 Agent（获得 `agent_id`），在代码中直接引用该 ID 创建 Session——控制台和 API 操作的是同一套资源，可自由混用。 +4. **注意模型名称一致性**：控制台向导示例中出现 `qwen3.7-plus`，而 API 代码示例使用 `qwen3-max`，请以控制台模型下拉列表中实际可选的模型 ID 为准。 + +## 被对比主题页 + +- [managed agents](../guides/managed-agents.md) +- [managed agents api](../api/managed-agents-api.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-vs-custom-agents.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-vs-custom-agents.md new file mode 100644 index 00000000..37e0ced6 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/managed-vs-custom-agents.md @@ -0,0 +1,73 @@ +# 托管 Agent 与自建应用调用对比 + +阿里云百炼平台针对不同复杂度的智能体场景提供了两条主要落地路径：一是由平台托管运行时的 **Managed Agents**，二是通过 **DashScope / Responses API 调用**已在控制台配置好的**智能体应用**或**工作流应用**。二者定位不同、编排位置不同、状态管理方式不同，直接决定了适用场景与工程成本。本文横向对比二者的关键维度，帮助开发者在技术选型时做出判断。 + +## 背景与定位 + +- **Managed Agents（托管智能体）**：面向多步工具调用、代码执行、文件处理等长时任务，平台在服务端托管**会话状态**、**独立沙箱**与**工具执行器**。智能体在云端容器中自主运行 shell、读写文件、安装依赖，事件历史持久化，支持中断与续接。 +- **自建应用调用（智能体应用 / 工作流应用）**：开发者在控制台编排应用（模型、提示词、知识库、插件、工作流节点等），再通过 API 从业务系统调用。运行时**无状态**，上下文由调用方或平台的 `session_id` 维护，适合问答、对话及可编排的工作流任务。 +- **调用协议**：自建应用可选 **DashScope API**（`Application.call` / `POST /apps/{APP_ID}/completion`）或 **OpenAI 兼容的 Responses API**（`POST /api/v2/apps/agent/{APP_ID}/compatible-mode/v1/responses`）。Managed Agents 走独立的 AgentStudio API（`/api/v1/agentstudio/*`）。 + +## 关键维度对比 + +| 维度 | Managed Agents（托管） | 智能体/工作流应用（DashScope API） | 智能体应用（Responses API，OpenAI 兼容） | +| --- | --- | --- | --- | +| 运行模式 | 服务端托管会话，长时任务，支持中断/续接 | 单次同步或流式调用，无状态 | 单次同步 / 流式 / 异步（`background=true`） | +| 状态管理 | 会话（Session）在服务端持久化，事件历史可回溯 | `session_id` 云端保存 1 小时、最多 50 轮；或客户端自维护 `messages` | 客户端在 `input` 数组中传完整消息历史 | +| 执行环境 | 独立云端沙箱容器，可预装 apt/pip 依赖、可配网络策略 | 平台共享运行时 | 平台共享运行时 | +| 输入格式 | 通过 `POST /sessions/{id}/events` 发送用户消息、工具结果、系统事件 | `{"input":{"prompt":"..."}, "parameters":{}, "biz_params":{...}}` | `input`：字符串或消息数组，支持 `input_image` / `input_file` [多模态](../concepts/multimodal.md) 类型 | +| 输出格式 | 会话级 SSE 事件流（User / Agent / Tool / Tool_output / Error / Model / System） | `{"output":{"text","finish_reason","session_id"}, "usage":{...}, "request_id":"..."}` | OpenAI Responses 标准结构（`response.output` 数组） | +| API 端点 | `/api/v1/agentstudio/agents`、`/environments`、`/sessions`、`/sessions/{id}/events`、`/sessions/{id}/events/stream` | `POST https://dashscope.aliyuncs.com/api/v1/apps/{APP_ID}/completion` | `POST https://dashscope.aliyuncs.com/api/v2/apps/agent/{APP_ID}/compatible-mode/v1/responses` | +| SDK 支持 | AgentStudio HTTP API（可搭配任意 HTTP 客户端） | [DashScope SDK](../concepts/dashscope-sdk.md)（Python / Java），或 HTTP（Node.js / curl 等） | OpenAI 官方 SDK（Python / Java 等） | +| 内置工具 | `bash`、`read`、`write`、`edit`、`glob`、`grep`、`download_file`，可挂载 MCP 服务与 Skill | 由应用编排决定（知识库检索、插件、工作流节点等） | 同 DashScope 侧，取决于应用配置 | +| [多模态](../concepts/multimodal.md) | 通过文件挂载到 `/mnt/session/uploads` 参与工具处理 | 依赖应用内模型能力 | 原生支持 `input_image`（VL 模型 + 自定义处理）与 `input_file`（全文引用 / 切片检索） | +| 多轮对话 | 天然多轮：会话即上下文，事件历史持久化 | `session_id`（1 小时、≤ 50 轮）或自维护 `messages` | `input` 数组显式携带完整对话历史 | +| [流式输出](../concepts/streaming.md) | 默认 SSE 事件流 | 支持流式（工作流需在结束/输出节点开启开关） | `stream=true`；异步任务暂不支持流式 | +| 异步执行 | 天然长时任务，会话可中断/续接 | 暂不支持（需自行轮询业务状态） | `background=true` 提交任务，通过 `responses.retrieve(task_id)` 轮询 | +| 自定义参数透传 | 通过 Agent/Environment/Session 配置及事件负载 | `biz_params.user_defined_params` 面向自定义插件 / 工作流插件节点 | 同 DashScope 应用侧的插件透传 | +| 支持模型 | 由 Agent 配置选择（示例：`qwen3-max`；以控制台可选模型为准） | 由应用编排决定 | 由智能体应用编排决定，[多模态](../concepts/multimodal.md) 场景需选通义千问 VL 系列 | +| 资源挂载 | `resources` 字段或运行时 `POST /sessions/{id}/resources` 挂载，路径统一在 `/mnt/session/uploads`，单文件 ≤ 10 MB，会话隔离 | 通过知识库 / 附件 / 插件参数传递 | 通过 `input_file` URL 或应用知识库 | +| 计费与配额 | 计费与沙箱运行时、模型调用相关；单文件 ≤ 10 MB | 按模型调用量计费；`session_id` 缓存 1 小时 | 按模型调用量计费；异步/流式受限见文档 | +| 可用地域 | 见 AgentStudio 文档 | 仅华北 2（北京） | 仅华北 2（北京） | +| 典型场景 | 代码执行、批量文件处理、多步工具编排、长时任务 | 客服问答、RAG 检索、工作流驱动的业务自动化 | 需要 OpenAI 生态兼容、多模态输入或异步长任务的交互 | + +## 各方案的适用场景建议 + +### 优先选 Managed Agents + +- 任务需要**执行 shell 命令**、**读写文件**、**运行/调试代码**、**安装依赖**等真实计算操作。 +- 单轮响应无法覆盖，需要**长时运行**、**中断续接**或**人工审批**工具调用。 +- 需要**独立沙箱**（例如运行不受信代码、隔离网络策略、需要预装的运行环境）。 +- 事件级可观测性要求高，需按 User / Agent / Tool / Model / Error 等类型审计执行过程。 +- 会话内文件产物需被隔离拷贝管理，避免污染原始资源。 + +### 优先选 DashScope 应用调用 + +- 已经在控制台完成**智能体应用**或**工作流**编排，业务系统只需一次调用即可拿到结果。 +- 需要使用**自定义插件**并通过 `biz_params.user_defined_params` 透传业务参数（如用户级鉴权、上下文字段）。 +- 多轮对话强度中等，愿意使用平台 `session_id` 或客户端自维护 `messages`。 +- 语言栈以 Python / Java 为主，倾向使用 [DashScope SDK](../concepts/dashscope-sdk.md) 的成熟示例。 + +### 优先选 Responses API（OpenAI 兼容） + +- 现有代码或团队约定使用 **OpenAI SDK/生态**，希望以最小改动接入百炼。 +- 需要**多模态输入**（`input_image` / `input_file`）或严格 OpenAI 输出结构。 +- 需要**异步长任务**（`background=true`）并轮询结果，避免请求超时。 +- 需要显式控制**完整对话历史**（在 `input` 中传消息数组）而不依赖 `session_id`。 + +## 技术选型参考 + +- **要不要托管运行时？** 若任务本质是"让 Agent 在容器里自己干活"，选 Managed Agents；若任务本质是"调一次已配置好的应用拿结果"，选自建应用调用。 +- **要不要 OpenAI 兼容？** 是则 Responses API；否，且要用工作流/自定义插件，则 DashScope API；两者都不满足（需沙箱/工具执行）则 Managed Agents。 +- **多轮对话怎么管？** 短会话选 `session_id`；需要精细控制或跨端同步选自维护 `messages` / Responses `input` 数组；长时有状态任务选 Managed Agents 会话。 +- **要不要异步？** 长时任务但希望"提交-轮询"模式选 Responses API `background=true`；真正需要 Agent 自主推进的长任务选 Managed Agents（会话持久化 + 中断续接）。 +- **模型与地域约束**：自建应用调用（两套 API）当前均限华北 2（北京）；模型 ID 需以控制台下拉列表为准，注意文档中 `qwen3-max` 与 `qwen3.7-plus` 等示例不一致的情况。 +- **迁移路径**：可以先用自建应用调用跑通业务闭环，当出现需要"沙箱执行 + 长时状态"的诉求时再迁移到 Managed Agents；两者可在同一账号 / [业务空间](../concepts/workspace.md) 内共存，通过应用 ID / Agent ID 区分。 + +## 被对比主题页 + +- [managed agents](../guides/managed-agents.md) +- [bailian application calling](../guides/bailian-application-calling.md) +- [application call](../api/application-call.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-comparison.md new file mode 100644 index 00000000..d5bc91c5 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-comparison.md @@ -0,0 +1,110 @@ +# 记忆能力对比：记忆库与长期记忆 + +百炼平台为解决大模型跨会话上下文丢失问题提供了完整的记忆能力体系。本文对比「记忆库概览」（guides/memory-library-overview）与「长期记忆（新）API」（api/long-term-memory-new）两个文档，帮助开发者理解两者在定位、覆盖范围和使用方式上的差异，以便在技术选型时快速定位所需信息。 + +## 对比维度 + +| 维度 | 记忆库概览（memory-library-overview） | 长期记忆（新）API（long-term-memory-new） | +| --- | --- | --- | +| 文档定位 | 概念指南，介绍记忆库整体架构、核心能力与接入方式 | API 参考手册，详细列出全部 RESTful 接口的端点、参数与返回值 | +| 目标读者 | 做技术选型、架构设计的开发者与产品经理 | 直接调用 API 进行集成的后端开发者 | +| 覆盖范围 | 涵盖 API 直连、OpenClaw 插件、控制台管理、记忆规则配置 | 聚焦于 HTTP API 层面，不涉及 OpenClaw 插件与控制台操作 | +| Base URL | `https://dashscope.aliyuncs.com/api/v2/apps/memory/*`（提及） | `https://dashscope.aliyuncs.com/api/v2/apps/memory/`（完整说明） | +| 认证方式 | `Authorization: Bearer $DASHSCOPE_API_KEY` | `Authorization: Bearer $DASHSCOPE_API_KEY`，并指引 [API Key](../concepts/api-key.md) 获取链接 | +| 核心能力 | 记忆片段 + 用户画像，含自动去重、动态更新 | 记忆片段 + 用户画像，含完整 CRUD 接口 | +| 接入方式 | API 直连 + OpenClaw 记忆插件（零侵入） | 仅 API 直连 | +| 接口列表 | 仅列举 AddMemory、SearchMemory 的关键参数 | 完整列出 11 个接口（增删改查记忆片段 + 画像模板全生命周期） | +| 使用限制 | 未提及 QPM 限制 | 明确标注：总 3000 QPM、add 120 QPM、search 300 QPM | +| 记忆有效期 | 说明差异：API 文档称暂无失效日期，控制台默认 180 天 | 明确标注：生成的记忆片段与用户画像暂无失效日期 | +| 检索高级参数 | 提及 top_k、memory_library_id | 详列 top_k、min_score、enable_rerank、enable_judge、enable_rewrite、project_ids 等 | +| OpenClaw 插件 | 完整覆盖：安装、配置项、生命周期钩子说明 | 不涉及 | +| 控制台管理 | 涉及记忆库与记忆规则管理说明 | 不涉及 | +| 示例代码 | cURL + Python（agentscope-runtime） | cURL 为主，含请求/返回体完整字段说明 | + +## 详细差异分析 + +### 文档定位与覆盖范围 + +**记忆库概览**是从产品视角出发的入门指南，回答"记忆库是什么、能做什么、怎么接入"的问题。它同时覆盖了三种接入路径——API 直连、OpenClaw 插件、控制台可视化管理——让读者快速建立整体认知。 + +**长期记忆（新）API**则是从工程视角出发的接口参考，回答"具体调哪个端点、传什么参数、返回什么结构"的问题。它将全部 11 个接口以表格形式列出，并逐一给出 HTTP 方法、路径与说明，适合直接查阅。 + +### 接口完整度 + +| 接口能力 | 记忆库概览 | 长期记忆（新）API | +| --- | --- | --- | +| AddMemory（添加记忆） | 有参数说明 | 有完整请求体/返回体说明 | +| SearchMemory（搜索记忆） | 有参数说明 | 有完整请求体/返回体说明 + 高级检索参数 | +| ListMemory（列出记忆） | 提及封装类 | 有独立接口条目 | +| DeleteMemory（删除记忆） | 未涉及 | 有独立接口条目 | +| UpdateMemory（更新记忆） | 未涉及 | 有独立接口条目 | +| CreateProfileSchema（创建画像模板） | 提及封装类 | 有独立接口条目 | +| ListProfileSchemas | 未涉及 | 有独立接口条目 | +| DeleteProfileSchema | 未涉及 | 有独立接口条目 | +| UpdateProfileSchema | 未涉及 | 有独立接口条目 | +| GetProfileSchema | 未涉及 | 有独立接口条目 | +| GetUserProfile | 提及封装类 | 有独立接口条目 | + +### 接入方式差异 + +记忆库概览介绍了两种接入方式： + +1. **API 直连**：通过 HTTPS 调用 `memory/*` 系列接口，适合需要灵活控制写入与检索流程的应用。 +2. **OpenClaw 记忆插件**：通过 `before_agent_start`（自动召回）和 `agent_end`（自动捕获）两个生命周期钩子实现零侵入接入，适合已使用 OpenClaw Agent 框架的场景。所有读写均由百炼服务端完成提炼、向量化和语义检索。 + +长期记忆（新）API 文档仅覆盖 API 直连方式，不涉及插件接入，但提供了更完整的接口集（包括记忆片段和画像模板的增删改查全生命周期）。 + +### 检索能力 + +在 SearchMemory 接口上，长期记忆（新）API 文档提供了更丰富的高级检索参数： + +| 高级参数 | 记忆库概览 | 长期记忆（新）API | +| --- | --- | --- | +| `top_k` | 有（建议 3–10） | 有（取值 1~100，默认 10） | +| `min_score` | 无 | 有（值域 [0,1]，默认 0.3） | +| `enable_rerank` | 无 | 有（搜索结果重排序） | +| `enable_judge` | 无 | 有（意图判别回调） | +| `enable_rewrite` | 无 | 有（query 重写） | +| `project_ids` | 无 | 有（多规则混合检索） | + +### 使用限制 + +长期记忆（新）API 文档明确给出了 QPM 限制： + +| 限制项 | 值 | +| --- | --- | +| 全部接口总计 | 3000 QPM | +| 记忆片段 add 接口 | 120 QPM | +| 记忆片段 search 接口 | 300 QPM | + +记忆库概览文档未提及这些限制，开发者需参考 API 参考文档了解限流策略。 + +## 适用场景建议 + +### 推荐阅读「记忆库概览」的场景 + +- **技术选型阶段**：需要了解百炼记忆能力整体架构，评估是否满足业务需求 +- **选择接入方式**：需要对比 API 直连与 OpenClaw 插件两种方案的优劣 +- **配置 OpenClaw 插件**：需要安装、配置记忆插件的详细步骤 +- **理解记忆有效期**：需要了解控制台记忆规则与 API 行为的差异 +- **控制台管理记忆库**：需要通过可视化界面管理记忆库和规则 + +### 推荐阅读「长期记忆（新）API」的场景 + +- **API 集成开发**：需要查阅具体接口的请求体、返回体字段定义 +- **实现记忆 CRUD**：需要调用 List/Delete/Update 等概览文档未覆盖的接口 +- **画像模板管理**：需要创建、更新、删除画像模板 +- **高级检索调优**：需要使用 rerank、judge、rewrite 等高级检索参数 +- **容量与限流评估**：需要了解 QPM 限制以做容量规划 +- **精确参数查阅**：需要确认 `top_k` 取值范围、`min_score` 默认值等细节 + +## 总结 + +两个文档并非竞争关系，而是互补关系。记忆库概览是**入门与选型的起点**，长期记忆（新）API 是**开发与集成的落点**。建议开发者先阅读记忆库概览建立整体认知、选定接入方式，再查阅长期记忆（新）API 获取具体接口实现细节。两者配合使用可以完成从技术评估到工程落地的完整闭环。 + +## 被对比主题页 + +- [memory library overview](../guides/memory-library-overview.md) +- [long term memory new](../api/long-term-memory-new.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-features-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-features-comparison.md new file mode 100644 index 00000000..30ece2ba --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-features-comparison.md @@ -0,0 +1,64 @@ +# 长期记忆 API 与记忆库对比 + +百炼平台围绕"跨会话记忆"提供了两个层次的文档视角：**长期记忆（新）API** 是底层 RESTful 接口规范，定义了记忆片段与用户画像的完整 CRUD 操作；**记忆库（Memory Library）** 则是面向用户的产品能力概述，涵盖控制台管理、API 接入和 OpenClaw 插件三种使用方式。二者本质上是同一套后端服务的不同入口——记忆库的 API 直连方式即为长期记忆（新）API。本文从开发者技术选型角度，对比两种接入路径的能力边界与适用场景。 + +## 关键维度对比 + +| 维度 | 长期记忆（新）API 直连 | 记忆库 + OpenClaw 插件 | +| --- | --- | --- | +| 接入方式 | HTTPS RESTful 调用，任意语言/框架 | OpenClaw 插件一键安装，零代码接入 | +| API 端点 | `https://dashscope.aliyuncs.com/api/v2/apps/memory/*`（11 个接口） | 插件内部封装同一组 API，无需直接调用 | +| 认证方式 | Header `Authorization: Bearer $DASHSCOPE_API_KEY` | 插件配置文件中填写 `apiKey`（`sk-` 开头） | +| 输入格式 | `messages`（对话列表，最多 50 条）或 `custom_content`（≤512 字符），二者互斥 | 由插件自动采集对话内容，无需手动构造 | +| 输出格式 | JSON：`memory_nodes` 数组（含 `memory_node_id`、`content`、`event`） | 自动注入 Agent 上下文，开发者无感知 | +| 记忆写入 | 手动调用 `AddMemory`，支持自动提取与自定义内容两种模式 | `autoCapture`（默认开启），对话结束自动提取存储 | +| 记忆召回 | 手动调用 `SearchMemory`，支持 `top_k`（1–100）、`min_score`、重排序、意图判别、query 重写 | `autoRecall`（默认开启），对话前自动检索注入，`topK` 默认 5 | +| 用户画像 | 支持完整生命周期：创建/更新/删除/查询画像模板 + 获取用户画像 | 通过 `profileSchema` 配置项关联画像模板 | +| 记忆库管理 | 通过 `memory_library_id` 指定目标记忆库 | 共享统一配置，所有 Agent 共用同一记忆空间 | +| 记忆规则 | 通过 `project_id` 指定片段规则（有效期、提取策略） | 使用默认记忆库的默认规则 | +| 多用户隔离 | `user_id` 字段（≤64 字符），不同 `user_id` 完全隔离 | 配置文件中 `userId` 字段，全局唯一 | +| 限流 | 全接口合计 3000 QPM；add 120 QPM；search 300 QPM（账号级） | 受同一账号级限流约束 | +| 有效期 | API 文档标注"暂无失效日期" | 控制台默认规则预置 180 天，可配置 7/30/180 天或永不过期 | +| SDK 支持 | Python `agentscope-runtime` 封装类（`AddMemory`、`SearchMemory` 等） | OpenClaw CLI 命令（`plugins install`、`modelstudio-memory stats`） | +| 控制台可视化 | 无（纯 API） | 支持在百炼控制台查看/管理记忆库与规则 | + +## 适用场景建议 + +### 选择 API 直连的场景 + +- **自定义应用集成**：需要在自有后端服务中精确控制记忆的写入时机、检索策略和 Prompt 注入逻辑。 +- **高级检索需求**：需要启用重排序（`enable_rerank`）、意图判别（`enable_judge`）、query 重写（`enable_rewrite`）等高级参数。 +- **多记忆库/多规则管理**：业务上需要为不同场景创建独立记忆库和提取规则（如客服记忆与营销记忆分离）。 +- **用户画像深度使用**：需要动态创建/更新画像模板，或在业务逻辑中直接消费结构化画像数据。 +- **高吞吐写入**：需要批量或高频写入记忆（注意 add 接口限流 120 QPM）。 + +### 选择 OpenClaw 插件的场景 + +- **OpenClaw Agent 快速接入**：已有 OpenClaw 框架的 Agent，希望零代码获得跨会话记忆能力。 +- **原型验证**：快速验证长期记忆对 Agent 效果的提升，无需编写集成代码。 +- **统一记忆管理**：单用户或少量用户场景，所有 Agent 共享记忆空间即可满足需求。 + +### 注意事项 + +1. **有效期差异**：API 直写且不指定 `project_id` 时使用默认规则；控制台默认规则有 180 天有效期。如需永久记忆，应在控制台将规则有效期设为"永不过期"，或创建自定义规则并在 API 调用时指定 `project_id`。 +2. **插件限制**：OpenClaw 插件为统一配置，不支持按 Agent 独立配置记忆空间；不支持百炼 Coding Plan 的 [API Key](../concepts/api-key.md)。 +3. **混合使用**：两种方式操作的是同一后端存储，可以先用插件快速接入，后续再迁移到 API 直连做精细化控制，已有记忆数据不会丢失。 + +## 选型决策速查 + +| 决策因素 | 推荐方案 | +| --- | --- | +| 需要精细控制写入/检索逻辑 | API 直连 | +| 使用 OpenClaw 框架且无定制需求 | OpenClaw 插件 | +| 需要多记忆库隔离 | API 直连 | +| 需要控制台可视化管理 | 记忆库控制台 + API 直连 | +| 快速 MVP 验证 | OpenClaw 插件 | +| 需要用户画像完整 CRUD | API 直连 | + +## 被对比主题页 + +- [long term memory new](../api/long-term-memory-new.md) +- [memory library overview](../guides/memory-library-overview.md) + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-solutions-vs.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-solutions-vs.md new file mode 100644 index 00000000..b99f81b2 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-solutions-vs.md @@ -0,0 +1,56 @@ +# 长期记忆与记忆库对比 + +百炼平台提供两套紧密关联但定位不同的记忆能力：「长期记忆（新）API」和「记忆库（Memory Library）」。前者是一组 RESTful 接口，聚焦于记忆片段与用户画像的增删改查；后者是更高层的产品概念，涵盖记忆库管理、记忆规则配置以及多种接入方式（API 直连、OpenClaw 插件、控制台可视化管理）。本文从开发者技术选型角度对两者进行对比，帮助你快速判断应使用哪一套能力。 + +## 关键维度对比 + +| 维度 | 长期记忆（新）API | 记忆库（Memory Library） | +| --- | --- | --- | +| 定位 | 底层 RESTful API 集合，提供记忆片段与用户画像的 CRUD 操作 | 产品级概念，包含记忆库、记忆规则、接入方式与控制台管理的完整体系 | +| 文档归属 | API 参考（application-api-reference） | 用户指南（application-user-guide） | +| Base URL | `https://dashscope.aliyuncs.com/api/v2/apps/memory/` | 同上（API 直连时使用相同端点） | +| 认证方式 | Header `Authorization: Bearer $DASHSCOPE_API_KEY` | 同左（API 直连）；OpenClaw 插件在配置中填 `apiKey` | +| 接入方式 | 仅 HTTP API 直连 | API 直连 + OpenClaw 记忆插件（零侵入）+ 控制台可视化管理 | +| 记忆类型 | 记忆片段（自动提取）+ 用户画像（profile schema） | 记忆片段 + 用户画像（同左，但额外支持记忆规则配置） | +| 记忆有效期 | 生成的记忆片段与用户画像暂无失效日期 | 控制台默认规则预置 180 天有效期，可配置 7/30/180 天或永不过期；API 直写且不指定 `project_id` 时使用默认规则 | +| 核心接口 | AddMemory、SearchMemory、ListMemory、DeleteMemory、UpdateMemory、CreateProfileSchema、ListProfileSchemas、GetUserProfile 等 | 复用长期记忆 API 全部接口，并叠加记忆库 / 记忆规则管理能力 | +| 限流 | 全部接口合计 3000 QPM；AddMemory 120 QPM；SearchMemory 300 QPM | 同左（底层共用同一 API 网关） | +| 编程语言支持 | cURL / 任意 HTTP 客户端 | cURL + Python `agentscope-runtime` 封装类 + OpenClaw 插件 | +| OpenClaw 集成 | 不直接提供 | 提供 `@modelstudio/modelstudio-memory-for-openclaw` 插件，通过 `before_agent_start` / `agent_end` 钩子实现自动捕获与召回 | +| 控制台管理 | 不涉及 | 支持在百炼控制台可视化管理记忆库、记忆规则与记忆内容 | +| 记忆库隔离 | 通过 `memory_library_id` 参数区分，不传则使用默认记忆库 | 每个账号自带一个不可删除的默认记忆库，可创建多个自定义记忆库 | +| 记忆规则 | 通过 `project_id` 指定规则，不传使用默认 | 支持自定义记忆片段规则（有效期、提取策略等），预置默认项目规则 | + +## 适用场景建议 + +### 选择长期记忆（新）API 的场景 + +- **已有自有 Agent 框架**，只需调用 HTTP 接口完成记忆存取，不需要控制台可视化管理。 +- **需要对记忆片段做精细 CRUD**，例如单独 Update、Delete 某条记忆，或管理画像模板的完整生命周期。 +- **需要自定义记忆有效期**为永不过期，且通过 API 直写而不依赖控制台规则配置。 +- **对限流有明确预期**，需要按 120/300 QPM 的接口级限流做容量规划。 + +### 选择记忆库（Memory Library）的场景 + +- **希望零侵入接入**，使用 OpenClaw Agent 且不愿手动编排记忆读写逻辑——安装插件后自动完成捕获与召回。 +- **需要在控制台可视化管理**记忆库与记忆规则，包括调整有效期（7/30/180 天或永不过期）、查看记忆内容、配置提取策略。 +- **使用 Python 开发**，希望用 `agentscope-runtime` 封装类简化调用，而非手写 HTTP 请求。 +- **需要多记忆库隔离**，希望按业务线或场景创建不同记忆库并分别配置规则。 +- **团队协作场景**，需要非开发人员通过控制台查看和管理记忆数据。 + +## 技术选型建议 + +两套能力并非互斥，而是分层关系：记忆库是产品层概念，长期记忆 API 是其底层接口层。实际开发中常见的做法是： + +1. **快速验证阶段**：先用 API 直连方式（即长期记忆 API）跑通 AddMemory → SearchMemory 的核心链路，验证记忆提取与召回效果。 +2. **生产化阶段**：在控制台创建专属记忆库并配置记忆规则（有效期、提取策略），通过 `memory_library_id` 和 `project_id` 参数将 API 调用绑定到对应规则。 +3. **OpenClaw 场景**：如果 Agent 基于 OpenClaw 构建，直接安装记忆插件即可获得完整的自动捕获/召回能力，无需手动编排 API 调用时序。 + +无论选择哪种路径，底层的记忆存储、语义检索和画像提取引擎是同一套，差异仅在接入方式和管理粒度上。根据团队技术栈和运维需求选择即可。 + +## 被对比主题页 + +- [long term memory new](../api/long-term-memory-new.md) +- [memory library overview](../guides/memory-library-overview.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-solutions.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-solutions.md new file mode 100644 index 00000000..3eb08704 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-solutions.md @@ -0,0 +1,61 @@ +# 长期记忆与记忆库对比 + +百炼平台为解决大模型跨会话上下文丢失的问题，提供了长期记忆能力。该能力以"长期记忆（新）API"和"记忆库"两种形态对外服务：前者是一组 RESTful 接口，开发者可自行编排写入与检索流程；后者在 API 之外还提供控制台可视化管理与 OpenClaw 插件零侵入接入。本文从输入格式、接入方式、API 端点、[计费](../concepts/billing.md)、典型场景等维度进行对比，帮助开发者在技术选型时做出合适选择。 + +## 关键维度对比 + +| 维度 | 长期记忆（新）API | 记忆库（Memory Library） | +| --- | --- | --- | +| 定位 | 纯接口层，提供记忆片段与用户画像的 CRUD | 能力集合 + 管理入口，包含 API、控制台、OpenClaw 插件三种接入 | +| 接入方式 | HTTPS 直调 `https://dashscope.aliyuncs.com/api/v2/apps/memory/*` | API 直连（同左）、控制台可视化管理、OpenClaw 插件零侵入 | +| 认证方式 | `Authorization: Bearer $DASHSCOPE_API_KEY` | 同 API；OpenClaw 插件在 `openclaw.json` 配置 `apiKey` | +| 输入格式 | `messages`（对话数组，最多 50 条）或 `custom_content`（自定义文本，最大 512 字符），二者互斥 | 同 API；OpenClaw 插件通过生命周期钩子自动捕获，无需手动传参 | +| 输出格式 | 返回 `memory_nodes`（含 `memory_node_id` / `content` / `event` / `old_content`）与 `request_id` | API 返回同左；控制台提供可视化列表；插件自动注入 Prompt | +| 核心接口 | AddMemory、SearchMemory、ListMemory、DeleteMemory、UpdateMemory、CreateProfileSchema、ListProfileSchemas、DeleteProfileSchema、UpdateProfileSchema、GetProfileSchema、GetUserProfile | 包含上述全部接口；额外提供控制台记忆规则管理与插件配置 | +| 记忆片段有效期 | 暂无失效日期 | 控制台默认规则预置 180 天有效期，支持按规则配置 7/30/180 天或永不过期；API 直写不指定 `project_id` 时使用默认规则 | +| 用户画像 | 通过 `profile_schema` 传入画像模板 ID，API 自动提取结构化属性 | 同 API；控制台可可视化管理画像模板 | +| 检索能力 | 语义相似度搜索，支持 `top_k`、`min_score`、`enable_rerank`、`enable_judge`、`enable_rewrite`、`project_ids` 混合检索 | API 检索同左；OpenClaw 插件通过 `topK`、`minScore`、`autoRecall` 配置自动召回 | +| 限流 | 全部接口合计 3000 QPM；AddMemory 120 QPM；SearchMemory 300 QPM | 同 API 限流规则 | +| [计费](../concepts/billing.md)方式 | 按 API 调用计量 | 同 API；OpenClaw 插件底层仍调用同一组接口 | +| 管理界面 | 无，纯接口 | 控制台记忆库详情页，可管理记忆片段规则、画像模板、查看记忆内容 | +| OpenClaw 支持 | 需自行对接 | 原生支持 `@modelstudio/modelstudio-memory-for-openclaw` 插件，`before_agent_start` 自动召回、`agent_end` 自动捕获 | +| 多 Agent 隔离 | 通过 `user_id` 隔离不同用户记忆空间 | OpenClaw 插件为统一配置，所有 Agent 共享同一记忆，暂不支持按 Agent 独立配置 | +| 语言 SDK | 原生 HTTP；Python 可用 `agentscope-runtime` 封装类 | 同 API；OpenClaw 提供独立插件 CLI | + +## 适用场景建议 + +### 选择长期记忆（新）API 的场景 + +- **需要精细控制写入与检索流程**：开发者希望在对话结束后自行决定何时调用 `AddMemory`、何时调用 `SearchMemory`，并将召回结果以自定义方式注入 Prompt。 +- **已有自研 Agent 框架**：不依赖 OpenClaw 生态，希望以纯 HTTP 方式集成到现有 Python / Java / Go 服务中。 +- **需要混合检索与高级参数**：如开启重排序（`enable_rerank`）、意图判别（`enable_judge`）、query 重写（`enable_rewrite`）、多规则检索（`project_ids`）等。 +- **需要自定义 `custom_content` 直写**：跳过对话提炼，直接写入指定文本。 + +### 选择记忆库（Memory Library）的场景 + +- **需要控制台可视化管理**：希望在百炼控制台直接查看、管理记忆片段规则、画像模板和已存储的记忆内容，而不仅通过 API。 +- **使用 OpenClaw Agent 并希望零侵入接入**：通过插件实现自动捕获 / 自动召回，无需在业务代码中手动调用记忆接口。 +- **需要灵活配置记忆有效期**：控制台规则支持 7/30/180 天或永不过期，适合对记忆保留周期有明确要求的业务。 +- **快速原型验证**：控制台 + 插件组合可最快跑通"写入 → 检索 → 注入"闭环，降低初期集成成本。 + +### 二者组合使用 + +记忆库的 API 直连方式与长期记忆（新）API 面向同一组后端接口，二者并非互斥。常见组合是：生产环境用长期记忆（新）API 做精细控制，同时用控制台做运营管理与规则调优；若部分 Agent 基于 OpenClaw 构建，则这些 Agent 走插件通道，其余自研服务走 API。 + +## 技术选型参考 + +| 选型要点 | 推荐 | +| --- | --- | +| 自研框架、需要全链路控制 | 长期记忆（新）API | +| 基于 OpenClaw 的 Agent、希望最少代码改动 | 记忆库 + OpenClaw 插件 | +| 需要可视化管理记忆规则与内容 | 记忆库（控制台） | +| 对记忆有效期有差异化要求 | 记忆库（控制台规则配置） | +| 需要重排序、意图判别、query 重写等高级检索 | 长期记忆（新）API | +| 多用户、需要 `user_id` 隔离 | 两者均支持，按接入方式选择 | + +## 被对比主题页 + +- [long term memory new](../api/long-term-memory-new.md) +- [memory library overview](../guides/memory-library-overview.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-vs-knowledge-base.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-vs-knowledge-base.md new file mode 100644 index 00000000..f714f293 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/memory-vs-knowledge-base.md @@ -0,0 +1,78 @@ +# 长期记忆与知识库对比 + +百炼平台同时提供**长期记忆（Memory Library）** 与**知识库（Knowledge Base，RAG）** 两类"给大模型补充上下文"的能力。二者常被同时提起，但设计目标、数据形态、检索链路和计费模型均不同：长期记忆解决的是**跨会话保留用户级信息**的问题（谁、偏好什么、说过什么），知识库解决的是**为模型注入企业私域文档与最新信息**的问题（产品手册、FAQ、图文表格）。本文从技术选型视角对两者做系统对比，帮助开发者按场景挑选合适的方案，也可以在同一 Agent 中组合使用。 + +## 一句话定位 + +- **长期记忆**：面向**用户对话历史**的自动提炼与语义召回，输出"用户画像 + 记忆片段"，用于跨会话个性化。 +- **知识库**：面向**企业私有文档 / 结构化数据 / 音视频**的解析、切片、向量化与检索，用于让模型引用可信来源作答（RAG）。 + +## 关键维度对比 + +| 维度 | 长期记忆（Memory Library） | 知识库（Knowledge Base / RAG） | +| --- | --- | --- | +| 主要用途 | 跨会话保留用户偏好、历史事件、画像属性 | 为大模型补充私域文档、最新信息、结构化数据 | +| 数据来源 | 对话消息 `messages`，或自定义写入 `custom_content` | PDF/DOCX/PPT/TXT/MD/HTML、图片、音视频、表格、RDS 数据源等 | +| 数据形态 | 自然语言短片段（记忆片段） + 结构化画像（profile schema） | 解析后的文档切片（Text/图片/表格），支持 Meta 属性 | +| 提取方式 | 服务端从对话自动提取关键信息、自动去重与更新 | 用户上传原文件，服务端按选定 parser 解析 + 切片 | +| 隔离粒度 | 按 `user_id` 天然隔离，同 user_id 共享命名空间 | 按知识库 ID 隔离，可打 tag / 归入类目实现分组过滤 | +| 检索方式 | 语义检索（向量），可选 rerank / query 重写 / 意图判别 | 向量 + 关键词双路召回 → Rerank 重排 → 相似度阈值过滤 → TopK | +| 关键调参 | `top_k`（1–100，默认 10）、`min_score`（默认 0.3）、`enable_rerank/judge/rewrite` | 初步向量/关键词 TopK（1–100，默认 50）、相似度阈值（0.01–1.0）、TopK ≤ 20、Rerank 权重 | +| 输入接口 | `AddMemory` / `SearchMemory` / `ListMemory` / `UpdateMemory` / `DeleteMemory` + Profile Schema CRUD | `ApplyFileUploadLease` → `AddFile` → `CreateIndex` → `SubmitIndexJob`；检索用知识检索 / 知识问答服务 | +| Base URL / 端点 | `https://dashscope.aliyuncs.com/api/v2/apps/memory/*` | `bailian.cn-beijing.aliyuncs.com`（阿里云百炼 SDK） | +| 认证方式 | DashScope `Authorization: Bearer $DASHSCOPE_API_KEY` | 阿里云 AK/SK（`ALIBABA_CLOUD_ACCESS_KEY_ID/SECRET`） + `WORKSPACE_ID` | +| 地域限制 | 全网可用（DashScope 网关） | 仅中国站**华北 2（北京）** | +| 零侵入接入 | OpenClaw 记忆插件（`before_agent_start` 自动召回、`agent_end` 自动捕获） | 挂载给百炼智能体 / 工作流应用，作为上下文源 | +| SDK 支持 | Python `agentscope-runtime`（`AddMemory`、`SearchMemory` 等封装类） | 阿里云百炼 SDK（Java/Python），控制台可视化建库 | +| 支持模型 | 与调用侧模型无关（服务端自带提取/向量化模型） | 覆盖主流预置模型（QwQ/Max/Plus/Turbo/Coder 等）、VL 模型、DeepSeek/Llama3/Yi、Qwen3/2.5/2 及其自定义微调模型 | +| 生命周期 | API 直写"暂无失效日期"；控制台记忆规则可配 7/30/180 天或永不过期 | 与知识库存活期一致，可按文件删除 / 更新切片 | +| 更新语义 | 自动去重 + 事件流（`ADD`/`UPDATE`/`DELETE`） | 增量上传新文件、重新切片；单切片可编辑（10–6,000 字符） | +| 主要限流 / 配额 | 总计 3000 QPM；`add` 120 QPM；`search` 300 QPM | 标准版 1 QPS 固定；旗舰版 50–10,000 QPS（1–200 RCU） | +| 存储 / 数量限制 | 单条 `messages` ≤ 50；`custom_content` ≤ 512 字符；`user_id` ≤ 64 字符 | 平台存储：标准版 ≤ 100 GB / 旗舰版 ≤ 9,999 GB；类目 500、文件 100,000、数据表 1,000；单切片 ≤ 6,000 Token | +| 计费方式 | 按 API 调用计费（DashScope 记忆 API 费用） | 规格费 + 模型调用费；标准版 0.03 元/知识库/小时，旗舰版 0.2 元/RCU/小时；向量化 / Rerank 按 Token 计费；2026-01-04 起正式计费 | +| 免费额度 | — | 一次性 720 小时，仅抵扣标准版（老用户至 2026-02-03，新用户开通 30 天内） | +| 典型输出 | 记忆片段列表（`memory_node_id`、`content`、`event`） + 用户画像 JSON | 命中切片列表 + 来源文件；知识问答服务可直接返回带引用的自然语言答案 | + +## 典型场景对比 + +| 场景 | 推荐方案 | 说明 | +| --- | --- | --- | +| "记住用户偏好、日程、习惯，下次自然唤起" | 长期记忆 | 自动从对话提取事件，`user_id` 天然多租户 | +| 结构化用户画像（年龄、职业、地区、行业偏好） | 长期记忆 + Profile Schema | 用画像模板固化属性字段 | +| 让智能体回答"我们公司产品手册第 3 章说什么" | 知识库 | 挂载 PDF/DOCX，走 RAG 检索链路 | +| 图文并茂的技术白皮书、含公式/图表 PDF | 知识库（视觉理解类型 + 大模型文档解析） | 强制切换 `qwen3-vl-embedding`，可命中图片切片 | +| FAQ / 参数表极低延时问答 | 知识库（极速问答类型） | 后端针对低延时优化，不支持图片 | +| 音视频课程 / 会议纪要检索 | 知识库（音视频类型） | 语音识别 + 视频帧 + 剧情解析 | +| 表格 / RDS 数据问答 | 知识库（数据查询类型） | 使用专用向量模型 | +| Agent 需要同时"记住用户"和"引用私域文档" | 长期记忆 + 知识库 组合 | 前者提供个性化上下文，后者提供事实依据 | +| 需要多知识库并行召回 + 路由 | 知识检索服务（最多 15 个知识库） | 配置混排模型 / 混排模式 | +| 需要直接生成带引用、可拒答的答复 | 知识问答服务 | 支持极速模式与多轮智能（Agentic）模式 | +| 快速把 OpenClaw Agent 接入长期记忆 | OpenClaw 记忆插件 | 零侵入，通过生命周期钩子自动捕获 / 召回 | + +## 技术选型建议 + +1. **看数据主体**：如果要注入的是"用户历史 / 偏好"，选长期记忆；如果要注入的是"文档 / 数据 / 音视频"，选知识库。 +2. **看隔离维度**：长期记忆按 `user_id` 隔离，天然适合 To C 场景；知识库按空间 / 类目 / tag 隔离，天然适合 To B / 企业内部知识。 +3. **看运维成本**：长期记忆无需上传文件、无需建索引，"调用即用"；知识库需要选择解析器 / 切片策略 / Meta 抽取，且 **Meta 抽取和多轮改写只能在创建时开启，不可事后追加**。 +4. **看时延与并发**：知识库标准版固定 1 QPS，高并发必须上旗舰版并购买 RCU；长期记忆按 QPM 计流，`add` 120 QPM、`search` 300 QPM，超限需分片写入或降频。 +5. **看成本结构**：长期记忆是纯 API 调用费；知识库额外有"规格费（按小时）+ 模型调用费（向量化 + Rerank，按 Token）"，**多知识库下模型调用量按知识库数倍增**，精度要求不高可关闭 Rerank 降本。 +6. **看模型覆盖**：知识库需要挂载给具体大模型，选型时确认使用的是预置文本 / 视觉 / DeepSeek / Qwen 系列并与知识库类型匹配；长期记忆与调用侧模型解耦，仅通过 DashScope API 使用。 +7. **看地域**：知识库当前只在**华北 2（北京）**开放，跨地域业务需评估网络与合规；长期记忆全网可用。 +8. **组合使用**：推荐架构是"用户请求 → 从长期记忆召回画像 + 相关记忆片段 → 从知识库召回文档切片 → 合并入 Prompt → 模型生成 → 会话结束后写回长期记忆"。OpenClaw 用户可通过记忆插件把长期记忆写入 / 召回做成零侵入，把知识库挂载在智能体侧做上下文源。 + +## 常见误区 + +- **不要用知识库来存"用户偏好"**：会导致跨用户污染，且切片粒度不适配短句事件。 +- **不要用长期记忆来存"产品文档"**：`custom_content` 上限 512 字符，无法承载长文本、图片和表格。 +- **知识库创建后**不能补配 Meta 抽取和多轮改写，评估阶段务必先开好。 +- **长期记忆有效期在 API 与控制台不一致**：API 文档写"暂无失效日期"，控制台默认记忆规则预置 180 天。以控制台规则为准，或显式指定 `project_id`。 +- **DashScope API Key ≠ 阿里云 AK/SK**：长期记忆用前者，知识库 API 用后者，切勿混用。 +- **OpenClaw 记忆插件不支持 Coding Plan 的 API Key**，且所有 Agent 共享同一记忆空间，不支持按 Agent 隔离。 + +## 被对比主题页 + +- [long term memory new](../api/long-term-memory-new.md) +- [memory library overview](../guides/memory-library-overview.md) +- [knowledge base](../guides/knowledge-base.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-deployment-inference.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-deployment-inference.md new file mode 100644 index 00000000..cb5400aa --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-deployment-inference.md @@ -0,0 +1,61 @@ +# 模型部署与推理对比 + +本文对比百炼平台的两个核心推理能力方向：**模型部署**（将预置或调优后的模型部署为专属推理服务）与**高速推理**（通过 TPM 预留和快速模式提升吞吐量与输出速度）。两者在定位上有重叠（都涉及专属容量与预付费），但解决的问题和适用场景不同。本文帮助开发者根据业务需求做出技术选型。 + +## 关键维度对比 + +| 维度 | 模型部署 | 高速推理 | +| --- | --- | --- | +| 核心目标 | 将模型部署为资源专享的推理服务，支持调优后模型上线 | 提升推理吞吐量（TPM 预留）和输出速度（快速模式） | +| 主要能力 | 预置吞吐（PTU）、模型单元（MU）、按 [Token](../concepts/token.md) 计费、模型导入（LoRA）、API 部署 | TPM 预留（容量保障）、快速模式 Fast mode（速度提升） | +| 计费方式 | 三种：PTU（按天预付费）、模型单元（预付费）、按 [Token](../concepts/token.md) 用量（后付费） | TPM 预留（按 kTPM 预付费）；快速模式（按 token 计费，与标准 API 一致） | +| 资源独占性 | PTU 和模型单元均为专属资源，不与他人共享 | TPM 预留为专属容量；快速模式使用共享资源 | +| 支持模型 | 部分预置模型 + 所有 LoRA 调优后模型 | TPM 预留覆盖千问3.7-Max/Plus、千问3.6-Flash、GLM-5.2/5.1、DeepSeek-v4 等；快速模式支持 glm-5.2-fast-preview 等 | +| 模型导入 | 支持 LoRA 模型从 OSS 导入（千问3/2.5 系列、VL 系列） | 不涉及模型导入，仅使用平台预置模型 | +| 部署方式 | 控制台 + API（`POST /api/v1/deployments`），API 方式仅华北2 | TPM 预留：控制台创建；快速模式：指定 model ID 即可 | +| 接入方式 | `model` 参数设为 `deployed_model`（专属服务 ID） | TPM 预留：`model` 替换为专属 model code；快速模式：`model` 指定为 fast 模型 ID | +| API 端点 | DashScope 标准 endpoint（`dashscope.aliyuncs.com`） | 快速模式需用专属域名：`{workspace_id}.cn-beijing.maas.aliyuncs.com` | +| 扩缩容 | PTU 自助增减吞吐量；模型单元自助增减副本；按 [Token](../concepts/token.md) 需控制台申请 | TPM 预留支持调整输入/输出 TPM（可归 0 保留 code）；快速模式无扩缩容概念 | +| 超额处理 | PTU：自动溢出按量或返回 429；输入超上限自动转按量计费 | TPM 预留：自动溢出按量或返回 429；快速模式：超出 TPM 进入排队队列，不立即限流 | +| 退费规则 | PTU 预付费按天计费，无法提前退费；模型单元首月内退订日单价 1.2 倍 | TPM 预留缩容/退订按 1.5 倍系数结算 | +| 地域支持 | 控制台全地域；API 部署仅华北2（北京） | TPM 预留支持华北2（北京）和新加坡；快速模式华北2（北京） | +| 成熟度 | 正式可用 | TPM 预留正式可用；快速模式处于 preview 阶段 | +| 长输入/缓存 | PTU 部分模型支持长输入（最高 256K）和前缀缓存折扣 | TPM 预留部分模型支持长输入阶梯系数和缓存折扣 | + +## 能力关系说明 + +模型部署和高速推理并非完全互斥，而是从不同角度解决推理性能问题： + +- **PTU（模型部署）与 TPM 预留（高速推理）** 在"专属容量保障"这一诉求上高度相似，均通过预付费锁定专属吞吐量、支持自动溢出策略。区别在于 PTU 是模型部署体系下的计费方式之一，与模型单元、按 Token 计费并列；TPM 预留则是独立的高速推理加速能力，侧重于对已有预置模型的吞吐量锁定。 +- **模型单元** 是模型部署独有的能力，支持自定义性能指标、PD 分离模式，且是部署 LoRA 调优后模型的唯一途径。 +- **快速模式** 是高速推理独有的能力，无需预付费、无需部署，仅通过指定 fast 模型 ID 即可获得 1.5~2 倍的输出 TPS，适合对速度敏感但不需要专属资源的场景。 + +## 适用场景建议 + +| 场景特征 | 推荐方案 | 理由 | +| --- | --- | --- | +| 需要部署 LoRA 调优后的专属模型 | 模型部署（模型单元或按 Token 计费） | 仅模型部署支持模型导入和调优后模型上线 | +| 高负载生产环境，需稳定吞吐保障 | 模型部署（PTU）或高速推理（TPM 预留） | 两者均提供专属容量保障，PTU 偏向完整部署体系，TPM 预留更轻量 | +| 对输出速度敏感（AI 编程、Agent 多步推理） | 高速推理（快速模式） | 1.5~2 倍 TPS 提升，按 token 计费无需预付，接入零改动 | +| 需要自定义性能指标、PD 分离部署 | 模型部署（模型单元） | 模型单元支持部署规格、副本数、推理模式等细粒度配置 | +| 效果验证阶段，不使用不计费 | 模型部署（按 Token 计费） | 按量计费，一个月不使用自动释放 | +| 短期高峰需锁定容量，到期可释放 | 高速推理（TPM 预留） | 按天预付费，支持 1~30 天短周期，到期自动续费可选 | +| 自动化集成部署（CI/CD） | 模型部署（API 方式） | 提供完整的部署/查询/删除 API，适合自动化编排 | + +## 选型决策流程 + +1. **是否需要部署调优后的模型？** 是 → 选择模型部署（模型单元）。 +2. **是否对输出速度（TPS）有明确要求？** 是且无需专属资源 → 选择快速模式。 +3. **是否需要专属容量保障？** 是 → 在 PTU（模型部署）和 TPM 预留（高速推理）间选择： + - 需要完整部署体系（API 管理、模型单元配置、PD 分离）→ PTU + - 仅需锁定预置模型的吞吐量，轻量接入 → TPM 预留 +4. **是否仅用于效果验证？** 是 → 模型部署（按 Token 计费）。 + +> 注意：计费方式在服务创建后无法更改，切换需先下线已部署的模型。快速模式处于 preview 阶段，规格可能随版本调整，生产环境使用前请确认最新状态。 + +## 被对比主题页 + +- [model deployment 1](../guides/model-deployment-1.md) +- [model high speed inference](../guides/model-high-speed-inference.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-eval-vs-app-eval.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-eval-vs-app-eval.md new file mode 100644 index 00000000..b353acd0 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-eval-vs-app-eval.md @@ -0,0 +1,78 @@ +# 模型[评测](../concepts/evaluation.md)与应用[评测](../concepts/evaluation.md)对比 + +百炼平台提供两套独立的[评测](../concepts/evaluation.md)体系：**模型评测**面向基础模型（当前仅文本生成类）的推理能力评估，**应用评测**面向已发布的智能体应用和工作流应用的端到端输出质量评估。两者在评测对象、数据来源、评分机制和适用阶段上存在本质差异。本文从开发者技术选型角度，对两套体系进行系统对比，帮助判断在模型选型、调优验证、应用迭代等不同阶段应使用哪套工具。 + +## 关键维度对比 + +| 维度 | 模型评测 | 应用评测 | +|------|----------|----------| +| 评测对象 | 文本生成类基础模型（含调优后模型） | 已发布的智能体应用、工作流应用 | +| 核心目标 | 模型能力对比与选型、调优效果验证 | 应用输出质量评估、RAG 链路归因与调优 | +| 评测方式 | 自定义评测（自有数据集）+ 基线评测（公开标准数据集） | 自动评测（基于知识库自动生成评测集）+ 手动评测（人工构建评测集） | +| 数据来源 | 评测数据集（Prompt + Completion）或推理结果集 | 知识库自动生成、手动上传（.xls/.xlsx/.jsonl）、应用观测导入 | +| 评分机制 | 大模型评估（裁判模型打分）、规则评估（ROUGE/BLEU/Cosine/字符串匹配）、人工评估 | LLM 评估器、Code 评估器（Python 规则）、预置模板评估器、人工标签标注 | +| 裁判/评估模型 | 推荐千问-Max，可自选 | 仅支持 qwen-max 和 qwen-plus | +| 多对象对比 | 支持多模型同任务对比，可参与排行榜 | 多应用横向评测（最多 8 个应用或同一应用不同版本） | +| 前提条件 | 无特殊前提，有数据集即可 | 应用须已发布且关联知识库；须开通应用观测；需相应权限 | +| 归因分析 | 无（仅提供逐条评分和综合统计） | 有（BadCase 归因到 RAG 流程环节，给出 Prompt/检索/切片优化建议） | +| 地域限制 | 基线评测仅北京地域；自定义评测无限制 | 无明确地域限制 | +| 版本体系 | 单一版本 | 新旧两套系统并存（新版：评测任务 + 评估器 + 标签） | +| API/SDK 支持 | 不支持，仅控制台操作 | 不支持，仅控制台操作 | +| 计费构成 | 被评测模型推理费 + 裁判模型评分费 | 评估模型 Token 费（自动评测）；Code 评估器无额外费用 | + +## 评分能力细分对比 + +| 评分能力 | 模型评测 | 应用评测（新版） | +|----------|----------|------------------| +| 语义理解评分 | 大模型评估（数值型 0-5 / 分类型 Pass-Fail） | LLM 评估器（相关性、有害性、幻觉检测等） | +| 确定性规则评分 | 规则评估：ROUGE、BLEU、Cosine、字符串匹配 | Code 评估器：Python 代码自定义规则 | +| 人工标注 | 人工评估（Pass/Fail 标注） | 标签管理（分类/布尔/数字/文本四种标签类型） | +| 评分器复用 | 评测维度模板可跨任务复用 | 评估器可跨任务复用，且可从历史标注结果自动生成新评估器 | +| 单任务评估器上限 | 无明确上限（可关联多个维度） | 最多 10 个评估器 | + +## 适用场景建议 + +### 选择模型评测的场景 + +- **模型选型**：在多个候选模型间做能力对比，尤其是使用 C-Eval、MMLU、GSM8K 等标准基线快速横评。 +- **调优效果验证**：微调模型上线前，用同一数据集对比调优前后得分变化。 +- **确定性任务质量把控**：翻译、摘要、Function Calling 等有标准答案的场景，用 ROUGE/BLEU/字符串匹配做低成本自动化评测。 +- **纯模型层能力评估**：不涉及知识库检索、工具调用等应用层逻辑，只关注模型本身的生成质量。 + +### 选择应用评测的场景 + +- **RAG 应用质量迭代**：智能体应用上线后，需要定位 BadCase 是检索问题、Prompt 问题还是知识库切片问题。 +- **多版本 A/B 对比**：同一应用的不同 Prompt 版本或配置版本间做横向对比（最多 8 个）。 +- **端到端输出验收**：评测对象包含知识库检索、工具调用等完整链路，模型评测无法覆盖。 +- **人工标注经验固化**：通过新版标签体系积累标注数据，并自动抽象为 LLM 评估器实现评测自动化。 + +### 两者结合使用 + +典型工作流：先用**模型评测**完成基座模型选型和调优验证，确定最优模型后部署到应用；再用**应用评测**对完整应用链路做持续质量监控和归因优化。两套体系互补而非互斥——模型评测回答"哪个模型更好"，应用评测回答"应用哪里出了问题"。 + +## 选型决策速查 + +| 你的需求 | 推荐方案 | +|----------|----------| +| 对比 3 个候选模型在自有业务数据上的表现 | 模型评测（自定义评测） | +| 快速了解模型在标准学术基准上的能力 | 模型评测（基线评测） | +| 验证微调模型是否比基座模型有提升 | 模型评测（自定义评测 + 推理结果集复用） | +| 智能体应用回答质量下降，需定位原因 | 应用评测（自动评测 + 归因分析） | +| 对比同一应用新旧版本效果 | 应用评测（多应用横向评测） | +| 工作流应用输出格式校验 | 应用评测（新版评测任务 + Code 评估器） | +| 建立持续的人工标注质检流程 | 应用评测（新版标签管理 + 评测任务） | + +## 注意事项 + +- 两套体系均不支持 API/SDK 编程化调用，如需自动化评测流水线，模型评测可参考 PAI Judge Model API。 +- 模型评测中 1-3% 的分差通常为评测噪声，不建议仅凭微小分差做选型决策。 +- 应用评测的自动评测要求应用已发布且关联知识库，开发调试阶段无法使用。 +- 应用评测存在新旧两套系统，建议新项目直接使用新版评测体系（评测任务 + 评估器 + 标签），旧版手动评测功能后续可能被替代。 +- LLM 评分器（两套体系均涉及）存在位置偏差和自我偏好偏差，建议定期人工抽查校准评分一致性。 + +## 被对比主题页 + +- [model evaluation introduction](../guides/model-evaluation-introduction.md) +- [application evaluation](../guides/application-evaluation.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-monitoring-vs-app-monitoring.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-monitoring-vs-app-monitoring.md new file mode 100644 index 00000000..ec9eebca --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-monitoring-vs-app-monitoring.md @@ -0,0 +1,67 @@ +# 模型监控与应用监控对比 + +百炼平台提供两套互补的可观测能力：**模型监控**（含模型用量统计与运行监控）和**应用观测**（应用监控）。前者聚焦于模型调用层面的用量、性能与错误指标，后者聚焦于应用内部的端到端调用链路追踪。两者定位不同、粒度不同，开发者需要根据观测目标选择合适的工具，或组合使用。本文从监控对象、数据粒度、开通方式、指标体系等维度进行对比，帮助开发者做出技术选型。 + +## 关键维度对比 + +| 维度 | 模型监控 | 应用观测 | +| --- | --- | --- | +| 监控对象 | 模型调用（按「模型 + [业务空间](../concepts/workspace.md)」维度） | 应用内部调用链路（智能体应用、工作流应用、高代码应用） | +| 观测粒度 | 单次模型 API 调用 | 应用级 Trace / Span（含 LLM、检索、工具调用等节点） | +| 数据延迟 | 用量统计约 1 小时；普通监控小时级；高级监控分钟级 | 分钟级 | +| 开通方式 | 自动采集，无需额外开通（高级监控/推理日志需手动开启） | 需手动开通：授权 OpenTelemetry 服务角色 → 开通服务 → 初始化 LogStore | +| 覆盖范围 | 主账号下所有[业务空间](../concepts/workspace.md)的全部模型调用 | 仅已添加观测且已发布的应用 | +| 核心指标 | 调用次数、失败率、调用时长、首 Token 延时、RPM/TPM、Token 用量、安全拦截次数 | 调用次数与失败率、Token 总量（输入/输出）、平均首 Token 耗时、平均调用时长、各节点延时 | +| 链路追踪 | 不支持（仅单次调用维度） | 支持完整调用链路，节点嵌套展开（CHAIN/AGENT/LLM/RETRIEVER/TOOL 等） | +| 请求/响应内容查看 | 需开通推理日志，部分模型支持 | 默认记录 Prompt 与输出内容 | +| 告警能力 | 支持（短信/邮件/电话/钉钉/企微/Webhook，4 级告警） | 不支持 | +| 外部集成 | 高级监控支持 Prometheus HTTP API，可接入 Grafana 或自建应用 | 无 API，仅控制台操作；支持导出 JSONL/EXCEL | +| 数据标注与[评测](../concepts/evaluation.md) | 不支持 | 支持 Span 标注（布尔/分类/数字/文本）、支持将数据添加到[评测](../concepts/evaluation.md)集 | +| 用量/费用管理 | 支持（用量统计、免费额度管理、费用概览与费用告警） | 不支持（费用管理请使用模型用量页面） | +| 数据保留 | 用量统计最近 30 天；更早数据在费用与成本页面查询 | 最长 30 天调用记录 | +| 计费 | 免费 | 功能本身免费，观测数据存储费用由 OpenTelemetry 服务收取 | +| 地域限制 | 高级监控/告警/推理日志仅支持北京、上海、新加坡、弗吉尼亚（部分能力） | 无明确地域限制说明 | + +## 适用场景建议 + +### 选择模型监控的场景 + +- **成本管控**：需要按[业务空间](../concepts/workspace.md)、API-KEY、模型维度统计 Token 消耗与调用量，管理免费额度，设置费用告警。 +- **模型服务质量监控**：关注模型级别的失败率、限流（429）、调用时长、首 Token 延时等 SLA 指标。 +- **主动告警需求**：需要对超时、Token 突增、失败率飙升等异常进行实时告警通知（电话/短信/钉钉等）。 +- **接入自有监控体系**：已有 Grafana 或 Prometheus 监控栈，希望通过标准 HTTP API 拉取模型指标并自定义看板。 +- **内容安全审计**：需要查看内容安全拦截次数，或通过推理日志审计请求/响应内容。 + +### 选择应用观测的场景 + +- **端到端链路排查**：应用包含多步调用（检索 → 重排 → LLM → 工具），需要定位具体哪个环节出错或耗时过长。 +- **RAG 流程优化**：需要观察 RETRIEVER、REWRITER、RERANKER、EMBEDDING 等节点的表现，优化检索召回效果。 +- **Prompt 调试与质量分析**：需要查看每次调用的完整输入输出，按状态、延时、Token 量等条件筛选特定请求。 +- **[评测](../concepts/evaluation.md)数据积累**：希望将线上真实调用标注后导入评测集，构建持续评测闭环。 +- **工作流调试**：工作流应用包含条件判断、脚本转换、函数计算等节点，需要验证执行路径是否符合预期。 + +### 组合使用建议 + +| 场景 | 推荐组合 | +| --- | --- | +| 生产环境全面运维 | 模型监控（告警 + Grafana 看板）+ 应用观测（链路排查） | +| 成本优化 | 模型监控为主（用量统计 + 免费额度管理），应用观测辅助定位高 Token 消耗的节点 | +| 应用开发调试阶段 | 应用观测为主（Span 筛选 + 数据标注），模型监控辅助确认模型侧无异常 | +| 仅有 API 直调模型（无应用） | 仅模型监控即可满足需求 | + +## 选型要点总结 + +1. **看什么**：看模型整体健康度与成本选模型监控；看应用内部执行细节选应用观测。 +2. **要告警**：只有模型监控提供告警能力，应用观测无告警功能。 +3. **要集成**：模型监控的高级监控可接入 Grafana/Prometheus 生态；应用观测仅支持控制台操作和数据导出。 +4. **要评测**：应用观测支持标注和评测集导入，是构建数据飞轮的关键入口。 +5. **开通成本**：模型监控自动采集、零配置；应用观测需要开通 OpenTelemetry 服务，且存储产生额外费用。 + +两者并非互斥关系——模型监控回答"模型调用是否正常、花了多少钱"，应用观测回答"应用内部哪一步出了问题、为什么慢"。生产环境建议同时启用，形成从模型层到应用层的完整可观测体系。 + +## 被对比主题页 + +- [model monitoring](../guides/model-monitoring.md) +- [application monitoring](../guides/application-monitoring.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-approaches.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-approaches.md new file mode 100644 index 00000000..ce1c8d0b --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-approaches.md @@ -0,0 +1,119 @@ +# 模型微调、压缩与高速推理对比 + +在百炼平台的模型生产链路中，模型微调（Fine-tuning）、模型压缩（量化）与高速推理（TPM 预留 / 快速模式）分别面向"效果定制"、"部署降本"与"吞吐/延迟优化"三个正交问题。三者不冲突，可按 `微调 → 压缩（可选）→ 部署 → 高速推理接入` 的顺序组合使用。本文汇总它们的能力边界、计费方式、接入方式与典型场景，便于开发者在做技术选型时对号入座。 + +## 定位速览 + +- **模型微调**：改变模型**参数本身**，用来把领域知识、指令遵循、人类偏好或音色/风格写进模型，解决"模型不会做"、"做得不好"、"不符合偏好"等**效果问题**。 +- **模型压缩**：把已经微调好的**全精度模型**量化为**低精度模型**，在效果基本可控的前提下**降低部署 MU 规格**，解决"部署太贵"的**成本问题**。 +- **高速推理**：不改动模型本身，通过 **TPM 预留**锁定专属吞吐容量、或使用 **快速模式**（`-fast-preview` 系列）加速单请求 TPS，解决"被限流"、"输出太慢"的**推理性能问题**。 + +## 关键维度对比 + +| 维度 | 模型微调（Fine-tuning） | 模型压缩（量化） | 高速推理 · TPM 预留 | 高速推理 · 快速模式（Fast mode） | +| --- | --- | --- | --- | --- | +| 解决的问题 | 效果定制：知识注入、指令遵循、偏好对齐、音色/风格 | 部署降本：降低 MU 规格 | 容量保障：锁定专属吞吐，避免公共池限流 | 输出加速：提升单请求 TPS | +| 是否修改模型参数 | 是（写入领域知识/偏好） | 是（降低参数精度，不可逆） | 否 | 否 | +| 支持模型范围 | 千问 3.x/2.5 文本、Qwen-VL、万相图像/视频、CosyVoice 等，各模型支持 CPT/SFT/DPO 的组合不同 | 仅百炼平台微调产出的自定义模型（如 qwen3.5-flash-2026-02-23）；不支持基础模型/第三方模型 | 千问 3.6/3.7、GLM-5.1/5.2、DeepSeek-v4-Pro/Flash、Kimi-K2.6 等，以控制台开放范围为准 | 已上线 `glm-5.2-fast-preview` 等 `-fast-preview` 系列，preview 阶段动态调整 | +| 输入形态 | 训练数据集：SFT 用 ChatML、DPO 追加 chosen/rejected、CPT 用纯文本、VL 打包 ZIP、CosyVoice 用 `train/*.wav` + text | 全精度微调模型 + 可选校准数据集（≤5 个，语义贴近推理场景） | 预付费订单参数（输入 kTPM、输出 kTPM、时长、溢出策略） | 与标准 API 相同的 `messages` 请求体 | +| 输出形态 | 新的微调模型（`finetuned_output`），可部署后调用 | 低精度量化模型（新的模型 ID，带自定义后缀） | 一个**专属 model code**，作为 `model` 参数使用 | 直接返回 `chat/completions` 响应 | +| 主要接入点 / API 端点 | `POST /api/v1/files`（上传数据）→ `POST /api/v1/fine-tunes`（创建任务）→ 轮询 → `POST /api/v1/deployments` 部署 | 控制台：模型 → 模型训练 → 模型压缩 → 创建压缩任务（暂以控制台为主） | 控制台创建 TPM 预留，业务侧沿用 `dashscope.aliyuncs.com/compatible-mode/v1`，仅替换 `model` 为专属 code | 使用业务空间专属域名 `https://{workspace_id}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` + `-fast-preview` 系列 model ID | +| 训练/构建成本 | 高：数据准备、训练耗时、Token/训练单元计费 | 压缩任务本身**限时免费**（截止时间以控制台公告为准） | 无训练成本，仅预付吞吐容量费 | 无训练成本 | +| 运行/调用计费 | 部署后按 MU 规格或按 Token 计费 | 部署阶段按压缩后**更小的 MU 规格**计费（示例：MU1\*2 → MU8\*1，节省约 56%） | 预付费按 **kTPM/天**，输入/输出分别计价；支持长输入阶梯系数与缓存折扣；超额可自动溢出到按 Token 付费或 429 | 与标准 API 一致的**按 Token 计费**，输入/输出/缓存命中分别计价 | +| 地域可用性 | 仅华北2（北京） | 仅华北2（北京） | 华北2 与新加坡各有独立价目 | 华北2、新加坡分别定价 | +| 生效方式 | 训练完成 → 部署 → 用 `deployed_model` 调用 | 压缩完成 → 用压缩后模型 ID 部署调用 | 创建成功即生成专属 model code；短时冲高需预热 | 更换 `model` + 域名即刻生效 | +| 可逆性 / 组合性 | 支持多轮训练与迭代；推荐 `CPT → SFT → DPO` 递进 | **不可逆**：压缩模型不支持继续微调或二次压缩，须回到全精度重压 | 支持扩缩容（含归 0 保留 code）、续费、退订；退订后 code 立即失效 | preview 阶段能力可能调整；与 TPM 预留可叠加使用 | +| 典型场景 | 深度定制：安全合规对齐、专业领域问答、专属音色/画风 | 已微调模型上线阶段进一步压缩部署成本 | 流量可预估、对限流零容忍的生产接口 | AI 编程助手、Agent 多步推理、实时对话等对 TPS 敏感的场景 | + +## 各方案适用场景建议 + +### 模型微调 + +优先选择微调的场景： + +- Prompt 工程 + 插件调用 + RAG 已尝试但效果仍不达标。 +- 需要注入**大量领域知识**或专属**知识体系**（CPT，1000 万+ Token）。 +- 需要模型稳定遵循**特定输出格式或流程**（SFT，1000+ ChatML 样本）。 +- 需要对齐特定**用户偏好或安全策略**（DPO，100+ chosen/rejected 对）。 +- 视觉理解、图像/视频生成、语音合成场景的**风格/音色定制**。 + +不建议微调的场景： + +- 数据量极小（<1000 条 SFT 样本，且无法扩充）——先做 Prompt 工程或少样本示例。 +- 只是想让模型"更快"或"更便宜"——那是压缩或高速推理要解决的问题。 +- 需求变化极快、每周都要迭代业务规则——微调迭代成本高，优先在 Prompt/知识库层解决。 + +### 模型压缩 + +优先选择压缩的场景： + +- 已有微调模型将进入**规模化生产部署**，MU 成本压力大。 +- 业务对**推理精度**的容忍度较高，可以用业务测试集验证不同量化模板的实际效果。 +- 处于免费期，希望在同一微调模型上并行尝试多个量化模板挑最优。 + +不建议压缩的场景： + +- 使用**基础模型或第三方模型**——不在支持范围。 +- 精度极度敏感、SLA 要求严格（对齐质量、事实性极高要求）——量化不可逆，风险较大。 +- 模型仍在快速迭代微调阶段——先稳定微调模型，再进入压缩环节。 + +### 高速推理 · TPM 预留 + +优先选择 TPM 预留的场景： + +- 已有稳定的日/周流量画像，可估算 RPM 与输入/输出平均长度。 +- 面向 **C 端高峰、金融交易、客服 SLA** 等对公共池限流零容忍的场景。 +- 需要固定预算 / 财务上更希望**预付费**而非按量波动。 +- 命中大量**缓存**的场景：不同模型的缓存折扣（8%~25%）可显著放大等效容量。 + +配合 `自动溢出至按 token 付费` 使用可获得容量保底 + 弹性溢出的组合；对成本严格封顶的场景则选择 `仅使用预留容量` + HTTP 429。 + +### 高速推理 · 快速模式 + +优先选择快速模式的场景： + +- **AI 编程助手 / IDE 补全**：需要 80~100 TPS 的[流式输出](../concepts/streaming.md)。 +- **Agent 多步推理**：单轮延迟直接影响整条 chain 的响应时间。 +- **实时对话 / 语音助手**：用户对首 token 与后续 token 速度都敏感。 +- 需要**按量计费**弹性，接受 preview 阶段可能的能力调整。 + +注意事项： + +- 需要使用业务空间维度的**专属域名**，且模型 ID 必须为 `-fast-preview` 系列。 +- 超出 TPM 会**排队**而非立即拒绝——业务侧仍要做超时与重试策略。 +- `glm-5.2-fast-preview` 默认返回 `reasoning_content` 思考字段，前端展示逻辑需相应适配。 + +## 组合与技术选型参考 + +从产品生命周期角度，三条能力线通常按下面的方式协同： + +1. **能力打磨阶段**：先用 Prompt 工程 + 知识库；确认能力缺口后走 **模型微调**（`CPT?→SFT→DPO?`），产出自定义模型。 +2. **降本阶段**：微调模型进入准生产后，若 MU 成本吃紧，评估 **模型压缩**——先在免费期用不同量化模板做业务集回归，再挑选精度损失可接受、MU 规格最小的模板部署。 +3. **性能扩展阶段**：稳定上线后，根据流量特征选择高速推理方案—— + - 流量可预估、对限流零容忍 → **TPM 预留**。 + - 关注单请求 TPS、面向 IDE/Agent/实时对话 → **快速模式**。 + - 两者不冲突：若目标模型同时开放 TPM 预留与 fast 域名（以控制台开放范围为准），可先购买 TPM 预留、再走 fast 域名调用。 + +一图表述的决策路径： + +- 想让模型"**做得对 / 做得像**"？→ 模型微调。 +- 微调已经完成，想"**部署更便宜**"？→ 模型压缩（不可逆，需回归精度）。 +- 想"**不被限流**、有容量兜底"？→ TPM 预留。 +- 想"**输出更快**"？→ 快速模式（`-fast-preview` + 专属域名）。 +- 想"**又不被限流、又输出更快**"？→ 在支持的模型上叠加使用 TPM 预留 + 快速模式。 + +## 常见误区 + +- **把压缩当作微调的替代**：压缩改的是精度不是能力，能力缺口只能通过微调补齐。 +- **把高速推理当作降本手段**：TPM 预留是预付费容量，快速模式按 Token 计费与标准 API 一致，两者都不会直接降低单位 Token 单价，真正降本要靠模型压缩换 MU 规格或缓存折扣。 +- **忽略地域约束**：微调与压缩**仅华北2（北京）可用**且必须使用该地域 API Key；高速推理华北2 与新加坡分别定价与开放。 +- **忽略压缩的不可逆性**：压缩后模型不能继续微调或二次压缩；如需迭代必须从上游全精度微调模型重跑。 +- **忽略快速模式的域名切换**：Fast mode 必须使用业务空间专属域名 `{workspace_id}.cn-beijing.maas.aliyuncs.com`，直接沿用 `dashscope.aliyuncs.com` 不会生效。 + +## 被对比主题页 + +- [fine tuning](../guides/fine-tuning.md) +- [model compression](../guides/model-compression.md) +- [model high speed inference](../guides/model-high-speed-inference.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-comparison.md index 47044f8e..7eded9fe 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-comparison.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-comparison.md @@ -1,85 +1,55 @@ -# 模型微调、压缩与部署对比 +# 模型微调、压缩与高速推理对比 -在百炼平台上，将一个基础模型转化为可在生产环境中使用的定制化推理服务，通常涉及三个阶段：**模型微调（Fine-tuning）**、**模型压缩（Quantization）** 和 **模型部署（Deployment）**。三者在模型生产链路中依次衔接（调优 → 压缩（可选）→ 部署），各自解决不同层面的问题。本文从功能定位、适用范围、操作方式和成本等维度进行系统对比，帮助开发者在技术选型时做出合理决策。 - -## 功能定位与链路关系 - -模型微调、压缩与部署构成一条完整的模型生产流水线： - -- **模型微调**处于链路上游，目标是基于自有数据定制模型能力，使模型在特定领域或任务上表现更优。 -- **模型压缩**处于链路中间，是可选环节，通过量化技术降低模型参数精度，从而减小部署所需的算力规格和推理成本。 -- **模型部署**处于链路下游，将训练好（或压缩后）的模型发布为在线推理服务，供应用调用。 - -三者缺一不可地覆盖了"训练 → 优化 → 上线"的完整生命周期，但各自的输入输出、关注点和约束条件差异显著。 +在百炼平台上，当基础模型加 Prompt 工程无法满足业务需求时，开发者通常会进入"模型优化"环节。平台提供了三类定位不同但可串联使用的能力：**模型微调**（fine-tuning）解决"模型能力不够/不对口"的问题；**模型压缩**（量化）解决"微调模型部署成本过高"的问题；**高速推理**（TPM 预留 / 快速模式）解决"推理容量不确定或输出速度不足"的问题。三者分别作用于模型生产链路的不同阶段：微调产出定制模型 → 压缩降低部署规格 → 高速推理保障线上调用体验。本文从技术选型角度对比三者的关键差异，帮助开发者判断该在哪个环节投入。 ## 关键维度对比 -| 维度 | 模型微调（Fine-tuning） | 模型压缩（Quantization） | 模型部署（Deployment） | -|------|------------------------|------------------------|----------------------| -| **核心目标** | 基于自有数据定制模型能力，提升特定场景表现 | 降低模型参数精度，减小部署规格与推理成本 | 将模型发布为在线推理服务，供应用调用 | -| **在链路中的位置** | 上游（第一步） | 中间（可选） | 下游（最后一步） | -| **输入** | 基础模型 + 训练数据集（JSONL / ZIP 等） | 微调产出的自定义模型 | 预置模型 / 微调模型 / 压缩后模型 | -| **输出** | 微调后的自定义模型 | 低精度量化模型 | 可调用的在线推理服务（API 端点） | -| **支持的模型范围** | 广泛：Qwen3.6/3.5/3/2.5 系列文本模型、Qwen-VL 视觉模型、Wan 图像/视频模型、CosyVoice 语音模型 | 较窄：仅支持百炼平台微调产出的特定自定义模型（如 qwen3.5-flash） | 最广泛：预置模型（Qwen、DeepSeek、GLM 等）+ 微调模型 + 压缩模型 + OSS 导入的 LoRA 模型 | -| **支持的模态** | 文本生成、视觉理解、图像生成、视频生成、语音合成 | 仅文本生成 | 文本生成、[多模态](../concepts/multimodal.md)、语音合成等 | -| **操作方式** | 控制台 + API | 仅控制台 | 控制台 + API | -| **是否可逆** | 可重复训练和迭代 | 不可逆：压缩后不支持继续微调或二次压缩 | 可随时上线/下线，支持重新部署 | -| **典型耗时** | 数小时至数天（取决于数据量和训练轮次） | 平台自动完成，排队后运行 | 分钟级（状态变为"运行中"即可调用） | -| **计费方式** | 按 Token 用量计费：训练数据 Token 数 x 训练轮次 x 训练单价 | 压缩任务本身限时免费 | 三种模式：预置吞吐（PTU）、模型单元（MU，按时长）、Token 用量（按调用量） | -| **地域限制** | 无特殊限制 | 仅华北2（北京） | 无特殊限制 | - -## 训练方法与压缩模板选择 - -### 微调方法 - -百炼平台提供三种递进式微调方法，推荐按 CPT → SFT → DPO 的顺序使用： - -| 方法 | 目标 | 数据要求 | 典型场景 | -|------|------|----------|----------| -| CPT（继续预训练） | 注入领域知识 | 1000万+ Token 无标签文本 | 金融/医疗/法律等垂直领域适配 | -| SFT（监督微调） | 学会遵循指令 | 1000+ 条高质量问答对 | 客服、代码助手、Agent 工具调用 | -| DPO（直接偏好优化） | 对齐人类偏好 | 100+ 组正负样本对 | 安全合规强化、降低幻觉 | - -每种方法支持全参训练和高效训练（LoRA）两种模式。全参训练效果更好但耗时更长、成本更高；LoRA 训练收敛快、成本低，适合快速验证。 - -### 压缩量化模板 - -量化模板决定压缩后的部署规格，MU 编号越大表示部署规格越小、成本越低，但精度损失可能越大。可选配校准数据以提升量化精度，建议选择与推理场景语义相近的数据集。 +| 维度 | 模型微调 | 模型压缩 | 高速推理（TPM 预留 / 快速模式） | +| --- | --- | --- | --- | +| 解决的核心问题 | 补充领域知识、遵循指令、对齐偏好、增强推理（CPT/SFT/DPO/RL） | 通过量化降低微调模型的部署 MU 规格与推理成本 | TPM 预留：锁定专属吞吐容量，不受公共[限流](../concepts/rate-limit.md)影响；快速模式：输出 TPS 提升至 1.5~2 倍（80~100 TPS） | +| 所处链路阶段 | 模型生产（训练） | 模型生产（训练后、部署前，可选环节） | 模型调用（推理） | +| 输入 | 数据集：CPT 纯文本 `{"text":...}`、SFT ChatML 问答对、DPO chosen/rejected 对、RL 含 `rollout_extra`；[多模态](../concepts/multimodal.md)需 ZIP 包 | 已微调成功的自定义模型 + 量化模板（可选校准数据集，最多 5 个） | 标准推理请求（messages 等），无需改造请求体 | +| 输出/产物 | 新的微调模型（`finetuned_output`），需部署后调用 | 压缩后模型（源模型名 + 自定义后缀），部署规格更小 | 无新模型产物；TPM 预留生成专属模型 code，快速模式使用 fast 模型 ID | +| 支持模型 | Qwen3.x 商业版/开源系列、Qwen-VL、wan 图像/视频、cosyvoice-v3-flash 等；RL 限 qwen3.5-9b 等且需商务开通 | 仅百炼平台微调产出的自定义模型（如 qwen3.5-flash-2026-02-23 微调版），不支持基础模型/第三方模型 | TPM 预留：千问3.7-Max/Plus、GLM-5.2/5.1、DeepSeek-v4 等已开放模型；快速模式：当前仅 glm-5.2-fast-preview | +| 使用入口 | 控制台零代码 或 DashScope API（/files、/fine-tunes、/deployments） | 控制台（模型训练 > 模型压缩），也可通过 API 管理任务 | TPM 预留：控制台创建 + 标准域名调用；快速模式：专属 workspace 域名 + fast 模型 ID | +| 接入改动 | 需上传数据、训练、部署，调用时改为新模型名 | 无调用侧改动，部署压缩后模型即可 | TPM 预留：仅替换 `model` 为专属 code；快速模式：换专属域名 `{workspace_id}.cn-beijing.maas.aliyuncs.com` + fast 模型 ID | +| 计费方式 | 按 Token 训练计费或模型训练单元 MTU（API 仅 Token，RL 仅 MTU）；部署另计费 | 压缩任务限时免费；压缩后模型部署按（更小的）MU 规格计费 | TPM 预留：按 kTPM 预付费（按天，1~365 天）；快速模式：按 token 计费，与标准 API 一致 | +| 地域限制 | 仅华北2（北京） | 仅华北2（北京） | 按模型/控制台为准；快速模式支持北京与新加坡 | +| 成熟度 | 正式（RL 需商务开通） | 正式（任务限时免费） | TPM 预留正式；快速模式 preview | +| 主要风险/限制 | 数据质量要求高；可能过拟合；需关注 Loss 曲线调参 | 单向不可逆：不能继续微调、不能二次压缩；量化可能有精度损失 | TPM 预留退订按 1.5 倍系数结算违约金；快速模式规格可能变化、支持模型少 | -## 部署计费方式对比 +## 典型场景与适用建议 -| 计费方式 | 计费公式 | 适用场景 | 特点 | -|---------|---------|---------|------| -| 预置吞吐（PTU） | 使用时长 x (输入 TPM 单价 x 输入 TPM + 输出 TPM 单价 x 输出 TPM) | 流量稳定的高负载生产环境 | 保障吞吐额度内不限速，超额自动切换按量计费；TPS 约为按量的 1.5~2.0 倍 | -| 模型单元（MU） | 使用时长（小时）x 模型单元数量 x 模型单元单价 | 需要资源独占和自定义性能指标 | 支持 PD 分离计算模式，可降低首 Token 延迟 | -| Token 用量 | 输入 Token 数 x 输入单价 + 输出 Token 数 x 输出单价 | 调用量不稳定、用量较少的场景 | 仅支持部分 LoRA 调优模型，不使用不计费 | +### 模型微调：能力定制 -## 适用场景建议 +- **何时使用**：Prompt 工程、RAG 等轻量手段无法满足需求，且拥有足量高质量领域数据时。 +- **典型场景**：金融/医疗/法律领域适应（CPT）、客服流程与 Agent 工具调用（SFT）、抑制幻觉与安全对齐（DPO）、数学推理与代码生成（RL）。 +- **建议**：按 `CPT（可选）→ SFT → DPO（可选）→ RL（可选）` 递进；小数据快速验证用 LoRA 高效训练，追求效果且模型支持时优先全参训练（两者费用相同）。 -**场景一：快速验证模型定制效果** -推荐路径：SFT 高效训练（LoRA）→ 直接部署（Token 用量计费）。跳过压缩环节，以最低成本快速上线验证。 +### 模型压缩：部署降本 -**场景二：生产环境成本敏感** -推荐路径：SFT 全参训练 → 模型压缩 → 部署（MU 计费）。通过压缩降低部署规格（如从 MU1x2 降至 MU8x1，成本节省约 56%），适合长期运行的在线服务。 +- **何时使用**：微调模型已验证效果，但部署 MU 规格费用偏高（示例中从 ¥108/小时降至 ¥47/小时，节省约 56%）。 +- **典型场景**：微调模型长期在线部署、对推理成本敏感的规模化服务。 +- **建议**：压缩不可逆，务必保留上游全精度微调模型；在免费期内尝试多个量化模板，分别部署后用业务测试集对比精度与成本；校准数据选与推理场景语义相近的样本可提升量化精度。 -**场景三：高并发低延迟的核心业务** -推荐路径：微调（按需）→ 部署（PTU 计费）。PTU 模式提供预留吞吐保障，TPS 提升约 1.5~2.0 倍，适合流量稳定且对延迟敏感的场景。 +### 高速推理：调用侧保障 -**场景四：[多模态](../concepts/multimodal.md)模型定制** -推荐路径：视觉/图像/视频/语音 SFT → 直接部署（MU 计费）。当前压缩功能仅支持文本模型，[多模态](../concepts/multimodal.md)微调模型需直接部署。 +- **TPM 预留**：流量可预估、业务高峰不能接受公共[限流](../concepts/rate-limit.md)（429）时使用，如大促、批量任务窗口期。创建前用容量计算器估算 kTPM；默认"自动溢出至按量计费"可避免业务中断。 +- **快速模式**：对输出速度敏感的交互场景，如 AI 编程助手、Agent 多步推理、实时对话。注意其为 preview 阶段，且必须走专属 workspace 域名，与 TPM 预留的"标准域名 + 专属 code"方式不可混用。 +- **建议**：追求更高吞吐可进一步考虑 PTU 专属部署；流量波动大或短期使用则维持按量付费并叠加资源包优化费用。 -## 技术选型要点 +## 选型决策参考 -1. **是否需要微调**：如果预置模型已满足需求，可直接部署，无需微调。当模型在特定领域表现不佳、需要定制输出格式或降低幻觉时，再考虑微调。 -2. **是否需要压缩**：压缩可显著降低部署成本，但会带来一定精度损失且不可逆。建议在免费期内对同一微调模型尝试多个量化模板，分别部署后用业务测试集验证推理效果，选择最优方案。 -3. **如何选择部署计费方式**：流量稳定选 PTU，需要资源独占和灵活配置选 MU，用量少且不稳定选 Token 用量。注意 Token 用量模式仅支持部分 LoRA 模型。 -4. **地域约束**：模型压缩当前仅在华北2（北京）地域可用，规划链路时需考虑地域一致性。 -5. **不可逆操作提醒**：压缩后的模型不支持继续微调或二次压缩，务必保留上游全精度微调模型以备后续迭代。 +1. **效果不达标** → 先做微调（优先 SFT，再视情况叠加 DPO/RL），而不是直接买容量——容量和速度解决不了"答得不对"的问题。 +2. **微调模型部署太贵** → 做模型压缩，用更小 MU 规格部署；接受可能的轻微精度损失并用测试集验证。 +3. **调用被[限流](../concepts/rate-limit.md) / 高峰不稳** → 购买 TPM 预留，替换专属模型 code 即可，代码改动最小。 +4. **输出速度不够快** → 使用快速模式（当前限 glm-5.2-fast-preview）。 +5. **组合使用**：三者可串联——微调产出定制模型、压缩降低部署成本、TPM 预留/快速模式保障调用体验；但注意微调与压缩仅在华北2（北京）地域可用，规划时需统一地域与 API Key。 ## 被对比主题页 - [fine tuning](../guides/fine-tuning.md) - [model compression](../guides/model-compression.md) -- [model deployment 1](../guides/model-deployment-1.md) +- [model high speed inference](../guides/model-high-speed-inference.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-lifecycle.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-lifecycle.md new file mode 100644 index 00000000..d85b50f9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-lifecycle.md @@ -0,0 +1,60 @@ +# 模型微调、压缩与部署对比 + +模型微调、模型压缩和模型部署分别对应百炼自定义模型生命周期中的“效果定制、成本优化、服务上线”三个阶段。三者不是互斥方案，而是顺序衔接的生产链路：通常先通过微调让模型适配业务任务，再视成本与性能目标选择是否压缩，最后通过部署形成可调用、可计费、可监控的推理服务。 + +## 关键维度对比 + +| 维度 | 模型微调 | 模型压缩 | 模型部署 | +|---|---|---|---| +| 生命周期定位 | 训练/调优阶段，生成自定义模型 | 微调后、部署前的可选优化阶段 | 上线推理阶段，将模型发布为可调用服务 | +| 核心目标 | 提升业务任务效果、对齐领域知识/偏好/风格/音色 | 通过量化降低部署规格和推理成本 | 获取稳定吞吐、低延迟或资源独占的推理能力 | +| 输入对象 | 基础模型、已有调优模型、训练集/验证集 | 百炼平台微调产出的全精度自定义模型 | 预置模型、调优后模型、导入的 LoRA 模型或压缩后模型 | +| 输出结果 | 新的调优模型 ID 或调优产物 | 低精度量化后的自定义模型 | `deployed_model` 专属服务 ID，可用于后续推理和管理 | +| 典型数据格式 | 文本 SFT 使用 ChatML JSONL；DPO 使用 `messages/chosen/rejected`；CPT 使用 `{"text":"..."}`；VL 可含 image/video；图像/视频/语音有各自数据集要求 | 可选校准数据，建议与目标推理场景语义相近；最多选择数量以控制台为准 | 部署配置本身不是训练数据；API 创建时主要配置 `plan`、规格、容量、上下文、限流等参数 | +| 支持模型/产物 | 文本生成、千问 VL、万相图像/视频、CosyVoice 等，支持项依模型和地域而异 | 仅支持百炼平台微调产出的可压缩自定义模型；不支持基础模型或第三方模型 | PTU 支持部分预置模型；模型单元支持部分预置模型与所有调优后模型；按 Token 仅支持部分 LoRA 调优模型 | +| 主要接口/入口 | 文本/[多模态](../concepts/multimodal.md)/万相等调优任务通常通过控制台或 `/api/v1/fine-tunes` 等接口创建，具体随模型类型变化 | 控制台：模型训练 → 模型压缩 → 创建压缩任务 | API：`POST /api/v1/deployments` 创建，`GET /api/v1/deployments/{deployed_model}` 查询；也可通过控制台管理 | +| 地域限制 | 多个能力强调华北2（北京），部分 OSS 挂载还支持新加坡；具体以模型能力为准 | 仅华北2（北京） | API 部署文档限定华北2（北京）；不同计费和模型能力以控制台为准 | +| 计费方式 | 训练任务与相关资源按具体调优能力计费；需结合训练时长、模型规模和平台规则确认 | 压缩任务本身限时免费；压缩后模型按部署阶段 MU 规格计费 | PTU、模型单元、按 Token 用量三类；创建后即可能开始计费，计费方式创建后不可更改 | +| 可逆性/后续操作 | 可基于目标模型支持项继续实验或重新训练；不同训练方式有顺序建议 | 不可逆；压缩后不支持继续微调或二次压缩，需从上游全精度微调模型重新压缩 | 可扩缩容、监控、下线、删除；如需切换计费方式通常要下线后重建 | +| 风险点 | 数据格式不合规、训练方式与模型不匹配、过拟合、思考模型格式错误、模态资源限制 | 精度损失、校准数据不匹配、模板选择过激、无法回退 | 容量不足、成本失控、服务创建后计费方式不可改、[业务空间](../concepts/workspace.md)/[API Key](../concepts/api-key.md) 不匹配 | +| 典型场景 | 行业问答、客服流程、安全合规回答、固定 IP/画风/动作、专属音色 | 已有微调模型效果满足要求，但 MU 成本偏高，需要用量化降低推理成本 | 高并发生产、稳定流量保障、独占资源推理、微调模型上线验证与生产服务 | + +## 适用场景建议 + +### 选择模型微调：优先解决“效果不够好” + +当 Prompt 工程、插件调用、RAG 或后处理仍不能满足业务效果时，应优先考虑微调。文本模型适合行业知识适配、固定任务流程、安全合规与偏好对齐；图像和视频模型适合固定人物、IP、画风、动作或特效复现；CosyVoice 调优适合同一发音人的高还原度单音色定制。 + +开发者需要重点评估三件事：目标模型是否支持所需训练方式，训练数据是否满足格式与质量要求，以及调优收益是否足以覆盖训练和后续部署成本。若只是请求级风格控制、一次性音色生成或基础模型本身不支持的语种/能力，通常不应通过微调解决。 + +### 选择模型压缩：优先解决“部署太贵” + +模型压缩适用于“微调模型效果已经可接受，但部署规格或小时成本过高”的场景。它在百炼中主要指量化，不包括剪枝或知识蒸馏。压缩前应准备与线上推理相近的校准数据，并对不同量化模板分别进行[评测](../concepts/evaluation.md)，避免仅按成本最低选择模板。 + +由于压缩不可逆，且压缩后模型不能继续微调或二次压缩，建议保留全精度微调模型作为上游基线。若后续还要追加训练或调整能力，应先回到微调阶段生成新的全精度模型，再重新压缩。 + +### 选择模型部署：优先解决“如何稳定上线” + +部署是把模型变成可调用服务的必要步骤。若业务流量稳定且要求吞吐保障，可选择预置吞吐（PTU）；若需要资源独占、自定义性能指标或部署调优后模型，优先评估模型单元；若只是效果验证且并发和延迟要求不高，可在支持的 LoRA 调优模型上使用按 Token 用量计费。 + +部署前要明确计费方式、容量、上下文长度、RPM/TPM 限制、是否开启思考模式以及[业务空间](../concepts/workspace.md)绑定关系。计费方式创建后无法直接更改，部署成功后即可能开始计费，因此开发者应先用测试环境验证，再上线生产规格。 + +## 技术选型参考 + +1. **从业务目标倒推链路**：效果不足选微调，成本过高选压缩，需要稳定服务选部署。不要在没有效果基线的情况下先做压缩，也不要把部署当成调优手段。 +2. **推荐生产链路**：基础模型评估 → 数据准备 → 微调 → 离线[评测](../concepts/evaluation.md) → 可选压缩 → 压缩模型[评测](../concepts/evaluation.md) → 部署 → 线上监控与扩缩容。 +3. **保留可回退资产**：微调后的全精度模型是后续压缩和重新实验的关键资产；压缩模型适合上线降本，但不适合作为继续训练的起点。 +4. **成本评估要覆盖全阶段**：微调关注训练投入，压缩关注部署规格下降，部署关注 PTU/MU/Token 的持续费用。最终应以端到端效果、延迟、吞吐和单位请求成本共同决策。 +5. **优先小规模验证**：先用代表性数据集验证微调收益，再用线上相似数据验证压缩损失，最后用真实流量参数计算部署容量，避免一次性创建高规格服务导致成本浪费。 + +## 一句话总结 + +模型微调用来“把模型调到适合业务”，模型压缩用来“把可用模型部署得更省”，模型部署用来“把模型稳定地提供给线上系统调用”。三者组合使用时，开发者应以效果为前提、以成本为约束、以部署稳定性为最终交付标准。 + +## 被对比主题页 + +- [fine tuning](../guides/fine-tuning.md) +- [model compression](../guides/model-compression.md) +- [model deployment 1](../guides/model-deployment-1.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-methods.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-methods.md new file mode 100644 index 00000000..238bdb8d --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-methods.md @@ -0,0 +1,62 @@ +# 模型优化方式对比 + +百炼平台为开发者提供了多种模型优化手段，分别从不同维度提升模型在业务场景中的表现。本文对比三种主要的模型优化方式——**模型调优（Fine-Tuning）**、**模型压缩**和**高速推理（TPM 预留 / 快速模式）**——帮助开发者根据自身需求做出技术选型。 + +三者定位不同：模型调优关注"让模型更适合你的数据"，模型压缩关注"让模型部署更便宜"，高速推理关注"让模型跑得更快、更稳"。 + +## 关键维度对比 + +| 维度 | 模型调优（Fine-Tuning） | 模型压缩 | 高速推理 | +| --- | --- | --- | --- | +| **核心目标** | 通过 CPT / SFT / DPO 让模型学习领域知识、遵循指令、对齐偏好 | 将全精度微调模型量化为低精度版本，降低部署规格和成本 | TPM 预留锁定专属吞吐容量；快速模式提升单请求输出速度 | +| **适用阶段** | 模型生产链路起点（调优） | 模型调优之后、部署之前（可选） | 模型部署后的推理调用阶段 | +| **输入（对象）** | 基础模型 + 训练数据集 | 百炼微调产出的全精度自定义模型 | 已部署的模型 + API 调用 | +| **输出（产物）** | 自定义微调模型 | 量化后的低精度模型 | 专属模型 code（TPM）或 fast 模型 ID | +| **支持模型** | 文本生成、视觉理解、图像生成、视频生成、语音合成 | 百炼微调产出的自定义模型（如 qwen3.5-flash-2026-02-23） | 千问、GLM、DeepSeek、Kimi 等系列（TPM 预留）；glm-5.2-fast-preview（快速模式） | +| **操作入口** | 控制台「模型调优」或 HTTP API | 控制台「模型压缩」 | 控制台「TPM 预留」或替换 API 的 `model` 参数 | +| **数据要求** | CPT 需 1000 万+ Token；SFT 需 1000+ 条；DPO 需 100+ 组 | 校准数据集（条件选填，最多 5 个） | 无需训练数据，仅需配置容量参数 | +| **[计费](../concepts/billing.md)方式** | 按 Token [计费](../concepts/billing.md)（API）或训练单元（控制台） | 压缩任务限时免费；部署按 MU 规格[计费](../concepts/billing.md) | TPM 预留按 kTPM 预付费（输入/输出分别计价）；快速模式按 Token 计费 | +| **地域限制** | 仅华北2（北京） | 仅华北2（北京） | TPM 预留：华北2（北京）+ 新加坡；快速模式：华北2（北京）+ 新加坡 | +| **可逆性** | 可多次调优、组合 CPT→SFT→DPO | 不可逆，压缩后不可继续微调或二次压缩 | 可扩缩容、续费、退订 | +| **成熟度** | 正式商用 | 正式商用 | TPM 预留正式商用；快速模式为 preview 阶段 | + +## 三者关系 + +模型调优、模型压缩和高速推理不是互斥的，而是可串联的模型生产链路： + +``` +基础模型 → [模型调优 CPT/SFT/DPO] → 微调模型 → [模型压缩（可选）] → 量化模型 → 部署 → [高速推理 TPM 预留 / 快速模式] → 线上调用 +``` + +- 先调优获得定制化模型 +- 再通过压缩降低部署成本 +- 最后通过高速推理保障线上性能 + +## 适用场景建议 + +| 场景特征 | 推荐方式 | 说明 | +| --- | --- | --- | +| 模型在领域知识、指令遵循或偏好对齐上不满足要求 | **模型调优** | 用业务数据定制模型行为，解决"效果不够好"的问题 | +| 微调模型部署成本过高（MU 规格 / 每小时费用） | **模型压缩** | 量化后部署规格显著降低，如 qwen3.5-flash 压缩后成本节省约 56% | +| 业务有明确峰值流量，不能接受公共资源限流（429） | **TPM 预留** | 锁定专属吞吐容量，刚性兑付 | +| 对单请求输出延迟敏感（AI 编程助手、Agent 多步推理、实时对话） | **快速模式** | 输出 TPS 提升至标准 API 的 1.5~2 倍 | +| 既要专属容量又要更快输出 | **TPM 预留 + 快速模式** | 但需确认支持模型重叠情况，快速模式目前仅支持 glm-5.2-fast-preview | +| 成本敏感、数据量较小、快速验证 | **调优选 LoRA 高效训练** | 文本生成模型全参与高效训练费用相同，百炼推荐优先全参训练 | +| 追求全局效果最优、对成本不敏感 | **调优选全参训练** | 更新全部参数，效果更优 | +| 已有微调模型，想在免费期内测试多种部署规格 | **模型压缩** | 建议在免费期对同一微调模型尝试多个量化模板，部署后用业务测试集验证 | + +## 选型决策参考 + +1. **先确定优化方向**：是模型能力不足（调优）、部署太贵（压缩）还是推理不够快/稳（高速推理）？三者解决的是不同阶段的问题。 +2. **评估数据储备**：调优需要大量高质量训练数据（CPT 尤其需千万级 Token），数据不足时优先考虑压缩或高速推理。 +3. **成本视角**：调优按 Token / 训练单元计费；压缩任务本身限时免费，收益体现在后续部署成本下降；TPM 预留为预付费模式，适合流量可预估的业务。 +4. **地域约束**：调优和压缩仅华北2（北京）可用；高速推理在北京和新加坡均开放。若业务部署在非北京地域，需优先考虑高速推理方案。 +5. **可逆性**：压缩不可逆，操作前务必确认上游微调模型可用；调优和高速推理均支持灵活调整。 + +## 被对比主题页 + +- [fine tuning](../guides/fine-tuning.md) +- [model compression](../guides/model-compression.md) +- [model high speed inference](../guides/model-high-speed-inference.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-vs.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-vs.md new file mode 100644 index 00000000..5bc9c587 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization-vs.md @@ -0,0 +1,59 @@ +# 微调与模型压缩对比 + +## 背景与目的 + +微调（Fine-tuning）与模型压缩（Model Compression）是百炼平台模型生产链路中两个不同阶段的可选能力。微调位于链路前段，通过训练将领域知识、任务能力或人类偏好写入模型参数，解决"效果不够"的问题；模型压缩位于链路后段，通过量化将全精度模型转为低精度版本，解决"部署太贵"的问题。两者并非互斥替代，而是先后关系：先微调得到自定义模型，再按需压缩以降低推理成本。本文从多个维度对比两者的技术特征、适用场景与限制，帮助开发者在技术选型时做出合理判断。 + +## 关键维度对比 + +| 维度 | 微调（Fine-tuning） | 模型压缩（Model Compression） | +| --- | --- | --- | +| **链路位置** | 模型调优阶段（链路前段） | 模型压缩阶段（链路后段，位于微调与部署之间） | +| **核心目标** | 把领域知识、任务能力、人类偏好或特定音色/风格写入模型参数 | 将全精度模型量化为低精度版本，降低部署 MU 规格与推理成本 | +| **技术手段** | CPT（持续预训练）、SFT（监督微调）、DPO（直接偏好优化）；全参训练或 LoRA 高效训练 | 量化（特指低比特量化，不涉及结构剪枝或知识蒸馏） | +| **输入要求** | 训练数据集（ChatML 问答对、偏好对、纯文本等），支持文件上传或 OSS 挂载 | 已微调完成的自定义模型作为源模型；部分量化模板需校准数据集（最多 5 个） | +| **支持模型范围** | 广泛：千问文本系列、Qwen-VL 视觉系列、万相图像/视频、CosyVoice 语音合成 | 有限：仅支持百炼平台微调产出的自定义模型，不支持基础模型或第三方模型 | +| **训练方式差异** | 不同模型支持的方式不同（CPT/SFT/DPO，全参/LoRA），需按模型查表 | 无训练方式概念，仅选择量化模板；MU 编号越大规格越小、成本越低、但精度损失可能越大 | +| **操作入口** | 控制台"模型调优"页面，或 API/命令行（`POST /api/v1/fine-tunes` 等四步流程） | 控制台"模型训练 → 模型压缩 → 创建压缩任务" | +| **计费方式** | 按 [Token](../concepts/token.md) 计费（API 创建的任务）或使用模型训练单元预付费/后付费（仅控制台创建） | 压缩任务本身限时免费（截止时间以控制台公告为准）；压缩后模型在部署阶段按 MU 规格计费 | +| **地域限制** | 仅华北2（北京）地域可用，需使用该地域的 API Key | 仅华北2（北京）地域可用 | +| **可逆性** | 可对同一模型多次微调（如 CPT → SFT → DPO 递进组合） | 不可逆：压缩后模型不支持继续微调，也不支持二次压缩；如需调整须从上游全精度模型重新压缩 | +| **产出物** | 全精度自定义微调模型，可继续微调或压缩 | 低精度量化模型，直接用于部署 | +| **成本示例** | 取决于数据量、训练方式、epoch 等参数，成本较高 | 示例：qwen3.5-flash-2026-02-23 压缩前 MU1*2（108 元/小时）→ 压缩后 MU8*1（47 元/小时），节省约 56% | +| **任务状态** | API 轮询 `status`：从创建到 `SUCCEEDED` | 7 种状态：PENDING → QUEUING → RUNNING → SUCCEEDED/FAILED/CANCELING/CANCELED | +| **关键超参数** | `learning_rate`、`n_epochs`、`batch_size`、`max_length`、`lora_rank` 等 | 任务名称、量化产出模型名后缀、量化模板、校准数据 | +| **数据格式** | SFT 用 ChatML JSONL；DPO 追加 chosen/rejected；CPT 用纯文本；视觉需打包 ZIP | 校准数据为已有数据集（无需特殊格式化），仅当模板含校准输入参数时需提供 | + +## 适用场景建议 + +### 优先选择微调的场景 + +- **Prompt 工程已到瓶颈**：已有 [prompt](../guides/prompt.md) 优化、插件调用等手段无法达到效果，需要将领域知识或任务能力深度写入模型参数。 +- **需要[多模态](../concepts/multimodal.md)定制**：视觉理解（Qwen-VL）、图像/视频生成（万相）、语音合成（CosyVoice）等模态需要特定风格或音色。 +- **需要对齐人类偏好**：通过 DPO 让模型回答更符合业务安全和质量要求（如安全合规对齐）。 +- **需要递进式优化**：可按 CPT（补知识）→ SFT（学指令）→ DPO（对齐偏好）逐步组合使用。 +- **需要灵活迭代**：微调后仍可继续微调或压缩，保留后续优化空间。 + +### 优先选择模型压缩的场景 + +- **微调模型已满足效果但部署成本偏高**：模型能力达标，推理费用过高成为瓶颈。 +- **对推理延迟或吞吐有更高要求**：低精度模型在同等硬件下推理更快、并发更高。 +- **已确定不再需要继续微调**：模型效果已经稳定，压缩的不可逆性不构成障碍。 +- **希望在免费期内多方案验证**：压缩任务限时免费，可对同一微调模型尝试多个量化模板，分别部署后用业务测试集选最优方案。 + +### 典型组合流程 + +两者最常见的组合是**串联使用**： + +1. 在控制台或通过 API 完成微调训练（如 SFT 或 CPT → SFT → DPO），得到全精度自定义模型。 +2. 在模型压缩页面选择该微调模型作为源模型，选取合适的量化模板（权衡成本与精度损失），按需配置校准数据。 +3. 压缩成功后部署量化模型，按 MU 规格计费。 + +> **注意**：压缩是不可逆操作。建议在压缩前确认微调模型效果已达预期，或在免费期内对多个量化模板做 A/B 验证后再正式上线。 + +## 被对比主题页 + +- [fine tuning](../guides/fine-tuning.md) +- [model compression](../guides/model-compression.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization.md new file mode 100644 index 00000000..091b6469 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-optimization.md @@ -0,0 +1,57 @@ +# 模型微调与压缩对比 + +本页对比百炼平台的**模型微调（Fine-Tuning）**与**模型压缩（Compression/量化）**两项能力。二者同属模型生产链路，但定位不同：微调用于在基础模型上注入领域知识、对齐指令或偏好，产出全精度定制模型；压缩则在微调之后将模型量化为低精度版本，以更低的 MU 规格部署、降低推理成本。完整的模型生产链路为：模型调优 → 模型压缩（可选）→ 模型部署。理解二者差异，有助于在不同阶段做出正确的技术选型。 + +## 关键维度对比 + +| 维度 | 模型微调（Fine-Tuning） | 模型压缩（Compression） | +| --- | --- | --- | +| **核心目标** | 注入领域知识、学会遵循指令、对齐人类偏好 | 将全精度微调模型量化为低精度版本，降低部署规格与推理成本 | +| **在链路中的位置** | 模型生产链路的上游环节 | 模型调优与模型部署之间的可选环节 | +| **技术手段** | CPT（持续预训练）、SFT（监督微调）、DPO（直接偏好优化）；训练模式分全参训练与高效训练（LoRA） | 量化（Quantization），不涉及结构剪枝或知识蒸馏 | +| **输入数据** | 需要训练数据集（纯文本/ChatML/ChatML + chosen-rejected），通过文件上传或 OSS 挂载加载 | 输入是已完成的全精度微调模型；部分量化模板需要校准数据集（最多 5 个） | +| **产物** | 全精度自定义模型（`finetuned_output`） | 低精度量化模型（独立部署模型名） | +| **支持的模型范围** | 文本生成（Qwen3.x/Qwen2.5 等）、视觉理解（千问 VL）、图像生成（万相）、视频生成（万相）、语音合成（CosyVoice） | 仅支持百炼平台微调产出的自定义模型，不支持基础模型或第三方模型；具体可压缩模型以控制台展示为准 | +| **训练方式/模式限制** | 文本生成支持全参训练与高效训练（推荐全参）；图像生成、视频生成、语音合成仅支持高效训练（`efficient_sft`） | 不涉及训练方式选择；通过选择量化模板决定部署规格，MU 编号越大规格越小、成本越低、精度损失可能越大 | +| **操作入口** | 控制台（模型调优页面）或 HTTP API（上传数据集 → 创建任务 → 查询状态 → 部署 → 调用） | 仅控制台（模型 → 模型训练 → 模型压缩 → 创建压缩任务），无独立 API | +| **计费方式** | API 创建的任务仅按 [Token](../concepts/token.md) 计费，不支持训练单元；控制台创建可使用预付费/后付费训练单元 | 压缩任务本身限时免费（截止时间以控制台公告为准）；压缩后模型按 MU 规格计费 | +| **成本示例** | 视训练方式与数据量而定 | 如 qwen3.5-flash-2026-02-23 微调模型，压缩前 MU1*2（108 元/小时），压缩后 MU8*1（47 元/小时），节省约 56% | +| **可逆性** | 微调模型可继续微调或作为压缩源模型 | 不可逆：压缩后模型不支持继续微调或二次压缩，需从上游全精度微调模型重新压缩 | +| **地域限制** | 仅华北2（北京）可用 | 仅华北2（北京）可用 | +| **任务配置可修改性** | 训练任务创建后超参数等配置不可改 | 压缩任务创建后不可修改配置 | +| **任务状态** | 创建后轮询 `status` 至 `SUCCEEDED`，再部署 | 7 种状态：PENDING → QUEUING → RUNNING → SUCCEEDED/FAILED/CANCELING → CANCELED | + +## 适用场景建议 + +### 选择模型微调的场景 + +- **需要注入领域知识**：拥有大量领域语料（1000 万+ [Token](../concepts/token.md)），希望通过 CPT 让模型掌握行业背景。 +- **需要定制指令遵循行为**：有标注好的指令-回答对（1000+ 条），通过 SFT 让模型按特定格式或风格响应。 +- **需要对齐人类偏好**：拥有 chosen/rejected 对（100+ 组），通过 DPO 降低不良输出、提升回答质量。 +- **[多模态](../concepts/multimodal.md)定制**：需要对视觉理解、图像生成、视频生成或语音合成模型进行定制化训练。 +- **快速验证想法**：数据量较小、成本敏感时，可先用高效训练（LoRA）快速迭代，再视需要升级全参训练。 + +### 选择模型压缩的场景 + +- **降低推理成本**：微调模型部署规格偏高，希望在保持能力的前提下降低 MU 规格，如将 MU1*2 降至 MU8*1。 +- **上线前做成本与精度的权衡**：微调已完成，建议在免费期内对同一微调模型尝试多个量化模板，分别部署后用业务测试集验证，选最优方案再正式上线。 +- **生产环境对成本敏感**：推理量大、部署时间长，量化带来的小时费率下降能产生可观的总成本节省。 + +### 组合使用 + +二者并非互斥，而是链路上的先后关系。典型路径为：先用微调得到满足效果的全精度定制模型，再视成本需求对该微调模型执行压缩，最终部署量化版本。需要注意压缩不可逆，压缩后的模型不能再微调，因此应在微调效果稳定后再进入压缩环节。 + +## 技术选型参考 + +- **是否需要新知识/新行为**：是 → 微调（CPT/SFT/DPO）；否，只需降本 → 压缩。 +- **是否已有满意的微调模型**：是 → 可考虑压缩进一步降本；否 → 先做微调。 +- **是否需要继续迭代模型**：是 → 暂不压缩（压缩后不可再微调）；否，且追求低成本部署 → 压缩。 +- **是否需要 API 发起**：微调支持 HTTP API 全流程；压缩仅支持控制台操作。 +- **是否使用训练单元**：如需预付费/后付费训练单元，微调必须通过控制台创建；API 创建的微调任务仅按 [Token](../concepts/token.md) 计费。 + +## 被对比主题页 + +- [fine tuning](../guides/fine-tuning.md) +- [model compression](../guides/model-compression.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-vs-application-evaluation.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-vs-application-evaluation.md new file mode 100644 index 00000000..8b5e40b9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-vs-application-evaluation.md @@ -0,0 +1,63 @@ +# 模型评测与应用评测对比 + +在百炼平台上，"评测"分为两条独立的产品线：**模型评测**面向文本生成类模型本身，衡量模型的原始推理能力；**应用评测**面向基于模型构建的智能体应用和工作流应用，衡量端到端的业务输出质量。两者在评测对象、数据组织、评分机制和前提条件上差异明显。本页面帮助开发者厘清两套体系的边界，在模型选型、调优验证与应用迭代等不同阶段选对工具。 + +## 关键维度对比 + +| 维度 | 模型评测 | 应用评测 | +| --- | --- | --- | +| 评测对象 | 文本生成类模型（预置模型、调优后模型） | 已发布的智能体应用、工作流应用 | +| 评测方式 | 自定义评测、基线评测（公开数据集） | 自动评测（单应用 / 多应用横向）、手动评测；新版评测任务 | +| 数据来源 | 评测集（Prompt + Completion）或推理结果集 | 基于知识库自动生成评测集，或手动上传；新版支持从应用观测导入 | +| 评分机制 | 5 种评测维度：大模型评估（数值型/分类型）、规则评估（文本相似度/字符串匹配）、人工评估 | 评估器（预置模板 / LLM 评估器 / Code 评估器）+ 标签人工标注 | +| 裁判/评估模型 | 推荐千问-Max | 仅支持 qwen-max、qwen-plus | +| 横向对比能力 | 排行榜（得分 0-100，绑定评测维度） | 多应用横向评测（最多 8 个应用或版本） | +| 前提条件 | 数据管理模块上传评测集并发布版本 | 应用已发布且配置知识库；开通应用观测并加入观测列表 | +| 权限要求 | 控制台模型评测模块权限 | 子账号需管理员或"应用评测-操作"权限 | +| API / SDK | 不提供公开 API/SDK，仅控制台操作 | 以控制台操作为主 | +| 地域限制 | 基线评测仅北京地域，自定义评测全地域 | 无基线评测概念 | +| 计费方式 | 被评测模型推理费 + 裁判模型评分费（按 Token）；推理结果集不产生推理费 | LLM 评估器产生 Token 费用；Code 评估器无额外费用 | +| 报告产出 | 综合得分、通过率、分数分布、数据明细，结果可下载（基线评测除外） | 评测报告含总正确率（≥4 分占比）、BadCase 分析、归因分析与调优建议 | +| 版本体系 | 单一体系 | 存在新旧两套系统，新版升级了评测任务、评估器和标签体系 | + +## 评分组件对比 + +| 对比项 | 模型评测的评测维度 | 应用评测的评估器 | +| --- | --- | --- | +| LLM 打分 | 大模型评估-数值型（默认 0~5 分）/ 分类型（Pass/Fail） | LLM 评估器（相关性、有害性、幻觉检测） | +| 规则打分 | 文本相似度（ROUGE/BLEU/Cosine 等 7 种算法）、字符串匹配 | 文本匹配、文本相似度、格式校验模板；Code 评估器（Python 规则） | +| 人工标注 | 人工评估-分类型（Pass/Fail 逐条标注） | 标签体系（分类/布尔/数字/文本四种类型） | +| 复用机制 | 维度模板可被多个任务复用；类型创建后不可更改 | 可基于历史评测任务的标注结果抽象为新的 LLM 评估器 | +| 数量限制 | 无明确上限 | 每个评测任务最多 10 个评估器，建议组合 3-5 个 | + +## 适用场景建议 + +**选择模型评测，当你需要：** + +- 在多个候选模型间做**选型对比**，或验证微调/调优后模型相对基座的提升； +- 用 C-Eval、MMLU、GSM8K 等**公开基准**量化模型通用能力（基线评测，仅北京地域）； +- 评估的是模型的原始输入输出能力，不涉及知识库检索、工具调用等应用层逻辑； +- 复用推理结果集反复打分以控制成本（不重复产生推理费用）。 + +**选择应用评测，当你需要：** + +- 验证智能体或工作流应用的**端到端效果**，包括 RAG 检索、Prompt 编排等整体链路； +- 在同一评测基准下横向对比最多 8 个应用或同一应用的不同版本，支撑**版本迭代决策**； +- 获得**归因分析与调优建议**（Prompt、检索配置、知识库切片层面的优化方向）； +- 结合应用观测数据做持续质量监控，并通过标签体系沉淀人工标注经验。 + +**组合使用建议：** 典型落地路径是先用模型评测完成基座模型选型与调优验证，再基于选定模型构建应用，用应用评测持续验证 RAG 配置与 Prompt 迭代效果。两者的评测集互不通用（模型评测使用数据管理模块的 EvaluationSet，应用评测使用自建或知识库生成的评测集），需分别准备数据。 + +## 选型注意事项 + +- 模型评测仅支持文本生成类模型，且无公开 API/SDK；如需程序化评测可参考 PAI Judge Model API 作为替代。 +- 应用自动评测强依赖知识库与应用观测：未配置知识库或未加入观测列表的应用无法发起自动评测；多应用横向评测要求所有应用关联至少一个相同知识库。 +- 两者的 LLM 打分均存在位置偏差与自我偏好偏差，重要决策建议辅以人工抽查；1-3% 的分差通常为评测噪声。 +- 应用评测的手动评测属旧版功能，新建流程建议直接采用新版"评测任务 + 评估器 + 标签"体系。 + +## 被对比主题页 + +- [model evaluation introduction](../guides/model-evaluation-introduction.md) +- [application evaluation](../guides/application-evaluation.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/model-vs-application-monitoring.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-vs-application-monitoring.md new file mode 100644 index 00000000..c4831a57 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/model-vs-application-monitoring.md @@ -0,0 +1,64 @@ +# 模型监控与应用监控对比 + +阿里云百炼平台提供了两套面向不同层次的观测能力：**模型监控**（含模型用量统计）聚焦于单个模型的调用指标、Token 消耗、日志与告警；**应用观测**（应用监控）则面向智能体应用、工作流应用和高代码应用，提供端到端的调用链路追踪与性能统计。本文对比两者的定位、能力边界与适用场景，帮助开发者在成本管理、性能观测、故障排查等场景中正确选型和组合使用。 + +## 核心定位差异 + +- **模型监控**：以"模型 + [业务空间](../concepts/workspace.md)"为维度，回答"某个模型被调用得怎么样"——调用量、Token 消耗、延时、失败率、[限流](../concepts/rate-limit.md)情况，并支持主动告警与 Prometheus 数据导出。 +- **应用观测**：以"应用"为维度，回答"一次应用请求内部发生了什么"——通过 CHAIN/AGENT/RETRIEVER/LLM 等节点还原完整调用链路，查看每个环节的输入输出、延时与 Token 量。 + +## 关键维度对比 + +| 维度 | 模型监控 | 应用观测（应用监控） | +| --- | --- | --- | +| 观测对象 | 模型列表中的所有模型（含调优后的自定义模型） | 智能体应用、工作流应用、高代码应用（不支持 Assistant API 创建的智能体应用） | +| 观测粒度 | 模型 + [业务空间](../concepts/workspace.md)维度的聚合指标与单次调用日志 | 应用维度的调用链路（Trace/Span），可下钻到 CHAIN、LLM、RETRIEVER 等节点 | +| 核心指标 | 调用量、Token 消耗、调用时长、首 Token 延时、RPM/TPM、失败率、[限流](../concepts/rate-limit.md)错误、内容安全拦截 | 调用次数与失败率、Token 总量/输入/输出、平均单次请求 Token 量、平均首 Token 耗时、平均调用时长 | +| 数据延迟 | 普通监控为小时级；高级监控为分钟级（收费）；推理日志为分钟级 | 分钟级 | +| 数据保留 | 用量数据仅保留 30 天，更早需导出账单 | 调用记录最长可查 30 天 | +| 日志/明细 | 需开通审计日志与推理日志，可查看单次调用的请求/响应内容（仅部分模型、部分地域支持） | 自动追踪 Prompt、输出、延时、Token 量，支持按 Request ID / Trace ID / Span ID 检索 | +| 链路追踪 | 无（仅单模型调用维度） | 支持节点级链路追踪与嵌套展开（高代码应用仅能观测入口 CHAIN 节点） | +| 告警能力 | 支持主动告警（需开启高级监控），通知方式含短信、邮件、电话、钉钉、企业微信、Webhook | 无内置告警能力 | +| 开放接口 | 高级监控数据支持标准 Prometheus HTTP API，可接入 Grafana 或自建应用 | 无 API，仅支持控制台操作；数据可导出为 JSONL / EXCEL | +| 开通前提 | 用量与普通监控开箱即用；高级监控、推理日志需在模型监控配置中开通 | 首次使用需授权 OpenTelemetry 服务角色、开通 OpenTelemetry 服务并初始化 LogStore | +| 地域限制 | 高级监控限北京、上海、新加坡、弗吉尼亚；告警限北京、新加坡、弗吉尼亚；推理日志限部分地域部分模型 | 无明确地域限制说明，取决于 OpenTelemetry 服务可用性 | +| 计费方式 | 普通监控免费；高级监控为收费功能 | 功能本身不收费，观测数据的存储费用由 OpenTelemetry 服务收取 | +| 数据标注与评测联动 | 不支持 | 支持对 Span 添加标签，并可将 Span 数据直接加入评测集作为评测样本 | +| 典型场景 | 模型成本管理、[限流](../concepts/rate-limit.md)与失败排查、性能基线监控、异常消耗告警 | 应用链路排障、RAG 检索效果分析、Prompt 优化、线上数据回流评测 | + +## 适用场景建议 + +### 优先选择模型监控 + +- **成本管理与用量核算**：按模型、API-KEY、推理类型（实时/批量）统计 Token 消耗，配合免费额度管理与"用完即停"开关控制成本。 +- **模型层故障排查**：定位 429 限流、内容安全拦截、调用失败等模型侧错误，通过失败详情下钻原因。 +- **性能基线与告警**：关注首 Token 延时、非首 Token 延时、RPM/TPM 等指标，对超时、Token 突增等静默失败设置主动告警。 +- **接入现有监控体系**：需要把指标接入 Grafana 或自建监控平台时，使用高级监控的 Prometheus HTTP API。 + +### 优先选择应用观测 + +- **应用内部链路排障**：一次应用调用涉及改写（REWRITER）、检索（RETRIEVER）、重排（RERANKER）、模型推理（LLM）、插件（TOOL）等多个环节，需要定位是哪个节点慢或出错。 +- **RAG / 智能体效果分析**：查看知识库检索返回的切片、模型思考过程与最终输出，评估检索与生成质量。 +- **线上数据回流评测**：通过数据标注和"添加到评测集"，将真实线上调用沉淀为评测样本，持续优化应用。 + +### 组合使用建议 + +两者互补而非互斥。典型实践是： + +1. 用**应用观测**定位一次慢请求发生在哪个节点（如 LLM 节点延时高）。 +2. 再用**模型监控**验证该模型在同时段的整体延时、限流与失败率，判断是模型侧全局问题还是应用侧个例。 +3. 对模型层面的异常（限流、消耗突增）配置**模型告警**，实现主动发现；应用层面的质量问题则通过应用观测的标注与评测集闭环改进。 + +## 限制与注意事项 + +- 模型监控的普通监控用量汇总为小时级延迟，高峰期可能达 1-2 小时；推理日志仅记录开通之后的调用，无法补录历史。 +- 模型监控的高级监控与告警存在地域限制，且高级监控为收费功能。 +- 应用观测没有 API，只能通过控制台操作；高代码应用目前不支持追踪内部链路，需在代码中使用 AgentScope-AI 的 Tracing 模块并在部署时添加 `--telemetry enable` 参数才能上报数据。 +- 应用观测暂不支持观测长期记忆的检索过程，TextRetriever / VectorRetriever 默认返回 100 个切片且不可调整。 + +## 被对比主题页 + +- [model monitoring](../guides/model-monitoring.md) +- [application monitoring](../guides/application-monitoring.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/monitoring-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/monitoring-comparison.md index 3e3d1202..3ffde1f0 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/monitoring-comparison.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/monitoring-comparison.md @@ -1,53 +1,47 @@ # 应用监控与模型监控对比 -百炼平台提供两套互补的监控体系：**应用观测**和**模型监控**。应用观测侧重于端到端追踪应用内部的调用链路与节点级性能，帮助开发者定位应用层面的延时与逻辑问题；模型监控则聚焦于模型维度的调用性能、Token 消耗、费用趋势与告警，帮助开发者管控模型使用成本与稳定性。理解两者的定位差异，有助于开发者建立完整的可观测性方案。 - -## 核心维度对比 - -| 维度 | 应用观测 | 模型监控 | -|------|----------|----------| -| **监控对象** | 应用（智能体应用、工作流应用、高代码应用） | 模型（含自定义模型和调优模型） | -| **观测粒度** | 节点级（CHAIN、LLM、RETRIEVER、TOOL 等多种节点类型） | 模型级（按模型名称、API-KEY、推理类型筛选） | -| **核心指标** | 调用延时、Token 用量（输入/输出）、首 Token 耗时、调用次数与失败率 | 调用时长、首 Token 延时、RPM、TPM、Token 消耗、失败率、限流错误、内容安全错误 | -| **数据更新频率** | 分钟级 | 普通监控为小时级；高级监控为分钟级 | -| **数据保留时长** | 最长 30 天 | 用量统计不支持查看 30 天以前的数据 | -| **告警能力** | 无 | 支持告警规则配置，通知方式含短信、邮件、电话、钉钉机器人、企业微信机器人、Webhook | -| **用量统计** | 无独立用量统计，通过监控统计图表查看 Token 总量 | 按[业务空间](../concepts/workspace.md)维度统计，支持免费额度管理与用完即停 | -| **数据导出** | 支持导出为 JSONL 或 EXCEL | 高级监控数据可通过 Prometheus HTTP API 接入 Grafana 等外部系统 | -| **数据标注** | 支持对 Span 添加标签（布尔值/分类/数字/文本），可关联评测集 | 不支持 | -| **日志查看** | 通过 Trace 详情查看 Prompt 输入输出与原始数据 | 开通推理日志后可查看每次调用的输入、输出及 Token 消耗（仅华北2北京地域部分模型） | -| **地域限制** | 无特殊地域限制 | 高级监控仅支持北京、新加坡、弗吉尼亚；告警仅支持北京、新加坡 | -| **计费** | 功能免费，观测数据存储由 OpenTelemetry 服务收费 | 功能免费，高级监控数据存储在私有 Prometheus 实例中 | -| **操作方式** | 仅控制台操作，无 API | 控制台操作 + Prometheus API 接入 | - -## 开通与前置条件对比 - -| 维度 | 应用观测 | 模型监控 | -|------|----------|----------| -| **开通步骤** | 授权 OpenTelemetry 服务角色权限 → 开通 OpenTelemetry 服务 → 初始化 LogStore | 普通监控默认可用；高级监控需在模型监控配置中手动开启 | -| **子账号权限** | 需要 AliyunBailianFullAccess + 应用观测页面权限 + ram:CreateServiceLinkedRole 策略 | 标准百炼控制台权限即可 | -| **推荐操作账号** | 主账号 | 无特殊要求 | +百炼平台提供两种层次的监控能力：**应用监控**（应用观测）关注[业务空间](../concepts/workspace.md)内智能体应用、工作流应用、高代码应用的端到端调用链路与运行指标；**模型监控**关注底层模型的用量统计、运行性能、推理日志与告警。两者观测对象不同但互补——应用监控自上而下追踪应用内部各节点的调用链，模型监控自下而上汇总所有[业务空间](../concepts/workspace.md)的模型调用量与性能。本文从观测对象、数据维度、延迟粒度、告警能力、接入方式、[计费](../concepts/billing.md)等关键维度做对比，帮助开发者根据需求选择合适的监控手段，或组合使用以构建全链路可观测体系。 + +## 关键维度对比 + +| 维度 | 应用监控（应用观测） | 模型监控 | +| --- | --- | --- | +| **观测对象** | 智能体应用、工作流应用、高代码应用的完整调用链路 | 主账号下所有[业务空间](../concepts/workspace.md)的模型调用（按"模型 + 业务空间"维度） | +| **监控粒度** | Span/节点级（CHAIN、AGENT、RETRIEVER、LLM、TOOL 等），支持嵌套展开 | 模型级（按模型 Code 列表），详情页分安全、成本、性能、错误四类指标 | +| **数据延迟** | 分钟级 | 用量统计约 1 小时；普通监控小时级；高级监控（Prometheus）分钟级 | +| **关键指标** | 调用次数（含失败）、Token 总量（输入/输出）、平均单次 Token、平均首 Token 耗时、平均调用时长 | 调用次数/失败率、Token 消耗、调用时长、首 Token 延时、RPM/TPM、限流(429)、内容安全拦截等 | +| **调用链追踪** | 支持，CHAIN 节点表示一次完整应用调用，可展开查看各子节点详情 | 不追踪应用内部链路，仅记录单次模型调用的请求/响应（推理日志） | +| **数据筛选** | 按 Span 模式（Root/All/Model Span）、状态、Span Name、延时、Token、标签等多字段组合 | 按 API-KEY、推理类型（实时/批量）、模型名称、时间范围与精度筛选 | +| **推理日志** | 可查看最长 30 天调用记录（Prompt、输出、延时、Token），支持按 Request ID/Trace ID/Span ID 检索 | 需开通推理日志后记录输入/输出/耗时，非所有模型支持请求响应内容记录 | +| **告警能力** | 无主动告警功能 | 支持告警规则，通知方式含短信、邮件、电话、钉钉/企业微信机器人、Webhook；告警等级分紧急/错误/警告/普通 | +| **数据导出** | 支持 JSONL、EXCEL 格式导出 | 支持接入 Grafana 或自建应用（Prometheus HTTP API） | +| **外部接入** | 无 API，仅控制台操作 | 支持 Prometheus HTTP API 接入 Grafana，提供 model_call_count、model_call_duration 等标准指标 | +| **[计费](../concepts/billing.md)方式** | 功能本身免费，数据存储在 OpenTelemetry 服务，存储费用由该服务收取 | 用量统计与普通监控免费；高级监控（Prometheus）与推理日志可能产生额外费用 | +| **适用地域** | 随应用所在业务空间 | 高级监控与告警限北京、上海、新加坡、弗吉尼亚等部分地域 | +| **数据标注与评测** | 支持对 Span 添加标签（布尔/分类/数字/文本），可将 Span 直接加入评测集 | 不支持数据标注，但 Token 消耗数据可作为成本评测的参考 | ## 适用场景建议 -### 应用观测适合以下场景 +### 选用应用监控的场景 -- **调用链路排查**：应用响应慢或出错时，需要逐节点定位瓶颈，例如区分是检索环节还是模型推理环节导致延时过高。 -- **Prompt 调试**：查看每次调用的完整输入输出，对比不同 Prompt 的效果。 -- **数据质量管理**：通过 Span 筛选与标注功能，对线上真实调用数据进行质量打分，并将优质样本导入评测集。 -- **工作流应用调试**：工作流包含多种节点类型（意图分类、脚本转换、条件判断等），需要观察每个节点的执行情况。 +- 需要追踪智能体应用或工作流应用内部各节点（检索、重写、[向量化](../concepts/embedding.md)、大模型推理、插件调用等）的调用链路和耗时分布 +- 需要查看应用级 Prompt 内容、输出结果、Token 消耗明细，排查单次调用问题 +- 需要将真实线上调用数据加入评测集，对应用做端到端质量评测 +- 需要对调用数据做标签标注，用于分类筛选或后续分析 +- 排查应用级错误（如某个节点超时、检索未命中、插件调用失败） -### 模型监控适合以下场景 +### 选用模型监控的场景 -- **成本管控**：按[业务空间](../concepts/workspace.md)统计模型用量与费用，配合免费额度管理控制预算。 -- **稳定性保障**：配置告警规则，在失败率上升或限流异常时及时收到通知。 -- **性能基线建立**：通过 RPM、TPM、首 Token 延时等指标建立性能基线，持续跟踪模型表现。 -- **多模型对比**：对比不同模型在相同业务场景下的调用时长、Token 消耗等指标，辅助模型选型。 -- **外部可视化集成**：将监控数据接入 Grafana 等系统，构建统一的运维大盘。 +- 需要掌握全账号或各业务空间的模型 Token 消耗总量与趋势，做成本管控 +- 需要监控模型性能（调用时长、首 Token 延时、RPM/TPM）与失败率，保障模型服务稳定性 +- 需要设置阈值告警，在 Token 突增、超时、失败率上升时及时收到通知 +- 需要查看推理日志做内容审计或故障排查（请求/响应原文） +- 需要将监控数据接入 Grafana 或自建运维大盘，做自定义可视化与告警编排 +- 管理免费额度，设置"免费额度用完即停"控制成本 -### 建议组合使用 +### 组合使用 -在生产环境中，推荐同时启用两套监控：用模型监控建立全局的成本与稳定性视图并配置告警，用应用观测在出现异常时深入排查具体调用链路。两者从不同维度覆盖可观测性需求，互为补充而非替代。 +当应用出现性能或成本问题时，推荐先用应用监控定位到具体异常节点（如 LLM 节点延时过高、检索节点返回过多切片），再通过模型监控查看对应模型的性能指标与失败率，交叉分析应用级调用链与模型级指标，实现全链路可观测。 ## 被对比主题页 diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/monitoring-vs.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/monitoring-vs.md new file mode 100644 index 00000000..81850c01 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/monitoring-vs.md @@ -0,0 +1,52 @@ +# 应用监控与模型监控对比 + +百炼平台提供了两种不同层次的监控能力：**应用监控**（应用观测）关注应用层端到端调用链路，**模型监控**关注模型层用量与运行指标。两者面向不同的观测粒度，开发者在排障、成本优化和性能调优时需要选择合适的监控视角。本页对比两者的关键维度，帮助开发者做技术选型。 + +## 关键维度对比 + +| 维度 | 应用监控（应用观测） | 模型监控 | +| --- | --- | --- | +| 监控对象 | 应用级（智能体应用、工作流应用、高代码应用） | 模型级（按模型 + [业务空间](../concepts/workspace.md)维度） | +| 观测粒度 | 端到端调用链路（CHAIN → AGENT/RETRIEVER/LLM/TOOL 等嵌套节点） | 模型调用记录（调用量、[Token](../concepts/token.md)、延时、失败率等指标） | +| 数据来源 | 可观测链路 OpenTelemetry 服务 | 平台自动采集 + Prometheus（高级监控） | +| 数据延迟 | 分钟级 | 用量统计约 1 小时；普通监控小时级；高级监控/推理日志分钟级 | +| 接入方式 | 仅控制台操作，无 API | 控制台 + 标准 Prometheus HTTP API（可接 Grafana / 自建应用） | +| 支持范围 | 智能体应用、工作流应用、高代码应用（不支持 Assistant API 创建的智能体应用；高代码应用仅入口 CHAIN 节点） | 主账号下所有[业务空间](../concepts/workspace.md)的模型调用，新模型首次同步后自动加入 | +| 关键指标 | 延时（调用时长）、[Token](../concepts/token.md) 量（输入+输出）、调用次数、失败率、平均首 [Token](../concepts/token.md) 耗时 | 调用次数、Token 用量、调用时长、首 Token 延时、RPM/TPM、失败次数/失败率、限流错误次数（429） | +| 调用链路追踪 | 支持，可展开 CHAIN 内部嵌套节点查看 Prompt、输出、延时等 | 不追踪应用内部链路，仅记录模型调用级别的请求/响应 | +| 数据导出 | 支持 JSONL / EXCEL 格式导出 Trace 数据 | 推理日志页签查看请求/响应；Prometheus HTTP API 拉取指标 | +| 筛选维度 | 状态、Span Name、输入/输出、延时、Token、标签；支持 Root/All/Model Span 模式 | API-KEY、推理类型、时间范围及精度；告警历史按规则/等级/状态筛选 | +| 告警能力 | 无主动告警 | 支持短信、邮件、电话、钉钉/企业微信机器人、Webhook；四级告警等级（紧急/错误/警告/普通） | +| 历史数据保留 | 最长 30 天调用记录 | 用量统计不支持查看 30 天以前数据；推理日志仅记录开通后的调用 | +| 数据标注 | 支持（布尔值/分类/数字/文本四种标签类型），与评测集共享标签管理 | 不支持数据标注 | +| 评测联动 | 支持将 Span 数据直接加入评测集 | 不支持评测联动 | +| 计费 | 功能本身不收费，数据存储费用由 OpenTelemetry 服务收取 | 功能本身不收费；Prometheus 实例存储费用由云监控收取 | +| 地域限制 | 无特殊地域限制 | 高级监控/告警/推理日志仅部分地域支持（北京、上海、新加坡、弗吉尼亚等） | + +## 适用场景建议 + +### 选择应用监控的场景 + +- **应用排障**：需要追踪智能体应用或工作流应用内部完整调用链路（检索 → 重写 → 向量化 → 重排 → LLM → 插件），定位具体节点的问题。 +- **性能优化**：需要查看每个节点的延时与 Token 消耗，分析瓶颈在检索、Embedding 还是 LLM 推理。 +- **评测样本采集**：需要将线上真实调用直接作为评测集样本，利用数据标注功能标记异常案例。 +- **Prompt 调试**：需要查看具体 Prompt 内容、模型输出和思考过程，按 Request ID / Trace ID / Span ID 精确检索。 + +### 选择模型监控的场景 + +- **成本管理**：需要按[业务空间](../concepts/workspace.md)维度汇总 Token 消耗，设置 Token 消耗阈值告警，掌握免费额度使用情况。 +- **模型选型**：需要对比不同模型的调用量、失败率、平均调用时长、首 Token 延时等指标，辅助选择最优模型。 +- **故障检测**：需要主动告警发现静默失败（超时、Token 突增、限流等），支持多渠道通知。 +- **Grafana 集成**：需要将监控数据接入 Grafana 看板或自建运维系统，通过 Prometheus HTTP API 灵活查询自定义指标。 +- **推理日志审计**：需要记录每次调用的输入/输出及耗时，用于内容审计与事后排查。 + +### 两者结合使用 + +大多数生产环境建议两者结合：模型监控提供宏观的用量趋势、成本统计和主动告警；应用监控提供微观的调用链路分析和节点级排障能力。当模型监控发现异常（如某模型失败率突增）时，可切换到应用观测深入追踪具体应用链路，定位根因节点。 + +## 被对比主题页 + +- [application monitoring](../guides/application-monitoring.md) +- [model monitoring](../guides/model-monitoring.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-api-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-api-comparison.md index a7e7444d..c8642870 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-api-comparison.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-api-comparison.md @@ -1,76 +1,88 @@ -# [多模态](../concepts/multimodal.md)生成 API 对比（图像/视频/3D） +# [多模态](../concepts/multimodal.md)生成 API 对比 -百炼平台提供图像、视频、3D 三类[多模态](../concepts/multimodal.md)生成 API，分别面向不同的内容产出形态。三者都通过 DashScope HTTP 接口调用，统一使用 API Key 鉴权，并遵循"创建任务 → 轮询结果"的异步任务模式（部分图像模型支持同步调用）。本页从输入格式、输出格式、支持模型、API 端点、调用模式、[计费](../concepts/billing.md)与典型场景等维度做横向对比，帮助开发者根据产出目标与技术约束做选型。 +百炼平台提供三大类[多模态](../concepts/multimodal.md)生成能力：图像生成、3D 资产生成和视频生成。三者均通过 DashScope 网关以 REST / SDK 方式调用，且由于生成耗时较长，大多采用 `X-DashScope-Async: enable` 异步"创建任务 → 轮询 task_id"的两步流程。本文从输入格式、输出格式、支持模型、API 端点、调用模式、计费与典型场景等维度对三者进行对比，帮助开发者根据业务需求做技术选型。 ## 关键维度对比 -| 维度 | 图像生成 | 视频生成 | 3D 生成 | +| 维度 | 图像生成 | 3D 生成 | 视频生成 | | --- | --- | --- | --- | -| 产出形态 | 静态图片（PNG） | 视频文件 | GLB 模型 + 预览渲染图 | -| 输入格式 | 文本、图像（图生图/编辑）、参考图 | 文本、图像（首帧/首尾帧）、参考图、视频、音频 | 文本、单图、多图（前/左/后/右 4 视角，固定数组长度 4） | -| 输出格式 | PNG，1–6 张或多图组图 | 视频 URL | PBR 材质 GLB（`pbr_model_url`）或无贴图基础模型（`base_model_url`），含 1 张预览渲染图 | -| 调用模式 | 同步（千问/万相2.6+/Z-Image 等新版）或异步（V1 及部分编辑/创意类） | 仅异步 | 仅异步 | -| API 端点 | 同步：`POST /api/v1/services/aigc/multimodal-generation/generation`；异步轮询：`GET /api/v1/tasks/{task_id}` | `POST /api/v1/services/aigc/video-generation/video-synthesis`（部分走 `image2video/video-synthesis`）；轮询：`GET /api/v1/tasks/{task_id}` | `POST /api/v1/services/aigc/video-generation/3d-generation`；轮询：`GET /api/v1/tasks/{task_id}` | -| 必需请求头 | `Authorization`；异步需 `X-DashScope-Async: enable` | `Content-Type`、`Authorization`、`X-DashScope-Async: enable` | `X-DashScope-Async: enable`（缺少报 `current user api does not support synchronous calls`） | -| 典型耗时 | 同步秒级返回；异步 1–2 分钟 | 1–5 分钟，万相2.1 视频编辑 5–10 分钟 | 较长，轮询建议间隔约 15 秒 | -| task_id 有效期 | 24 小时 | 24 小时 | 24 小时，超时返回 `UNKNOWN` | -| 产物下载链接有效期 | 随接口返回 | 随接口返回 | 2 小时，需及时下载 | -| 支持模型系列 | 千问图像、万相（Wan/wanx）、Z-Image、可灵 | 万相（HappyHorse/Wan/wanx）、爱诗 PixVerse、Vidu、可灵 | Tripo（`Tripo/Tripo-H3.1` 高精度、`Tripo/Tripo-P1.0` 专业快速） | -| 地域可用性 | 北京/新加坡/弗吉尼亚等多地域，地域独立鉴权不可混用；千问-图像翻译仅北京 | 同地域约束，模型/Endpoint/API Key 必须同地域；PixVerse、Vidu 仅北京 | 仅华北2（北京） | -| 业务空间专属域名 | 支持（`{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` 等） | 支持（北京 `{WorkspaceId}.cn-beijing.maas.aliyuncs.com` 等） | 走默认 dashscope 域名 | -| SDK 支持 | 部分模型支持 DashScope SDK（Python/Java） | HTTP 为主 | HTTP | -| [计费](../concepts/billing.md)方式 | 按张数/模型[计费](../concepts/billing.md) | 按任务/时长计费 | 按任务计费（`usage` 记录任务类型与生成数量） | -| 典型场景 | 文生图、图生图、图像编辑、虚拟模特、试衣、海报、背景生成、擦除补全、画面扩展、人物写真 | 文生视频、图生视频、参考生视频、视频编辑、视频换人、数字人、肖像动态视频 | 文生 3D、单图生 3D、多图生 3D，游戏/电商/工业设计资产 | - -## 调用模式差异 - -三类 API 在调用流程上高度一致，均采用"创建任务 → 轮询查询"模式，但图像 API 额外提供**同步调用**能力： - -- **图像生成**：千问图像系列（qwen-image-2.0-pro/max/plus）、万相 2.6/2.7 文生图与编辑、Z-Image 等新版模型支持一次请求即返回结果的同步调用，走 `multimodal-generation/generation` 端点；V1 版及部分编辑/创意类模型仍需异步。同步模式流程更简单，适合交互式场景。 -- **视频生成 / 3D 生成**：因耗时较长（视频 1–10 分钟，3D 资产更久），统一仅支持异步。请求必须携带 `X-DashScope-Async: enable`，缺少该头会报错 `current user api does not support synchronous calls`。 - -三者都强调"请勿重复创建任务"，`task_id` 有效期 24 小时，直接轮询即可。 - -## 输入能力对比 - -| 输入方式 | 图像 | 视频 | 3D | +| **输入格式** | 文本 [prompt](../guides/prompt.md)（文生图）；参考图 URL + [prompt](../guides/prompt.md)（图生图 / 图像编辑）；多图参考（部分模型支持） | 三选一：文本 [prompt](../guides/prompt.md)（文生 3D）；单图 URL（单图生 3D）；4 视角图片数组（多图生 3D） | 文本 prompt（文生视频）；首帧/首尾帧图像 URL（图生视频）；参考图 + prompt（参考生视频）；视频 URL（视频编辑/超分等） | +| **输出格式** | 静态图像（JPEG/PNG），支持 1~6 张 | GLB 格式 3D 模型（带 PBR 材质或无贴图）+ 预览渲染图 | 视频文件（MP4 等），含分辨率、时长、可选音频 | +| **支持模型** | Qwen-Image / Qwen-Image-Edit、万相 wan2.7/2.6/2.5/2.2/v1、Z-Image、Kling、Vidu、创意工具系列 | Tripo/Tripo-H3.1（高精度）、Tripo/Tripo-P1.0（专业快速） | 万相 wan2.7/2.6/2.2、HappyHorse、PixVerse、Vidu、Kling、人像动画系列 | +| **API 端点** | 文生图/编辑：`/api/v1/services/aigc/text2image/image-synthesis`（异步）或 HTTP 同步（wan2.7/2.6/z-image/qwen-image）；查询：`/api/v1/tasks/{task_id}` | 创建：`/api/v1/services/aigc/video-generation/3d-generation`；查询：`/api/v1/tasks/{task_id}` | 创建：`/api/v1/services/aigc/video-generation/video-synthesis`；查询：`/api/v1/tasks/{task_id}` | +| **调用模式** | 异步为主（`X-DashScope-Async: enable`）；wan2.7/2.6/z-image/qwen-image 支持同步 | 仅异步（`X-DashScope-Async: enable` 必选） | 仅异步（`X-DashScope-Async: enable` 必选） | +| **地域限制** | 部分模型仅北京地域（wanx-v1、创意工具系列） | 仅华北2（北京） | 需模型、Endpoint、[API Key](../concepts/api-key.md) 同一地域（北京/新加坡/美国/德国） | +| **输出分辨率** | 512×512 ~ 2048×2048（部分 4K） | 面数最高 200 万（H3.1）/2 万（P1.0） | 480P ~ 1080P，部分支持 4K | +| **产物有效期** | — | GLB 下载链接 2 小时 | task_id 有效期 24 小时 | +| **task_id 有效期** | 24 小时 | 24 小时 | 24 小时 | +| **典型计费** | 按张计费（如 wanx-v1 0.16 元/张）；部分创意工具免费体验 | 按任务类型计费 | 按视频时长/分辨率计费 | +| **典型耗时** | 秒级（同步）至数十秒（异步） | 分钟级 | 分钟级 | + +## 调用流程对比 + +三者均遵循异步两步流程，但端点路径和请求头要求有所不同： + +| 步骤 | 图像生成 | 3D 生成 | 视频生成 | | --- | --- | --- | --- | -| 纯文本 | 支持，复杂文字渲染能力强（千问系列） | 支持（文生视频） | 支持，中英文等多语言，最大 1024 字符 | -| 单图输入 | 支持（图生图、图像编辑） | 支持（首帧生视频） | 支持，JPEG/PNG，宽高 [20,6000]，≤20MB | -| 多图输入 | 部分编辑模型支持多图输入/输出 | 支持（参考生、首尾帧） | 支持，固定 4 视角（前/左/后/右），有效 2–4 张 | -| 视频输入 | 不适用 | 支持（视频编辑、参考生视频） | 不适用 | -| 音频输入 | 不适用 | 万相2.7 支持[多模态](../concepts/multimodal.md)输入含音频 | 不适用 | +| 创建任务 | POST 图像合成端点 | POST `/api/v1/services/aigc/video-generation/3d-generation` | POST `/api/v1/services/aigc/video-generation/video-synthesis` | +| 异步头 | 大多数需要 `X-DashScope-Async: enable` | **必选** | **必选** | +| 轮询查询 | GET `/api/v1/tasks/{task_id}` | GET `/api/v1/tasks/{task_id}` | GET `/api/v1/tasks/{task_id}` | +| 轮询建议 | — | 约 15 秒间隔 | — | +| 查询 RPS | — | 默认 20 | — | +| 状态枚举 | PENDING/RUNNING/SUCCEEDED/FAILED | PENDING/RUNNING/SUCCEEDED/FAILED/CANCELED/UNKNOWN | PENDING/RUNNING/SUCCEEDED/FAILED/UNKNOWN | -3D 生成的多图输入有严格的视角顺序约束（前/左/后/右），不需要的视角传空对象 `{}`，这与图像/视频的"多图作为参考"语义不同。 +## 输入参数对比 -## 产物与质量参数 - -| 项 | 图像 | 视频 | 3D | +| 参数类型 | 图像生成 | 3D 生成 | 视频生成 | | --- | --- | --- | --- | -| 输出规格 | 总像素 512×512~2048×2048，宽高比 1:4~4:1，1–6 张；万相2.7 支持 4K | 视频文件 URL | 面数：H3.1 最高 200 万面，P1.0 最高 2 万面 | -| 质量参数 | 分辨率、张数、宽高比 | 分辨率、时长、镜头叙事（`shot_type: multi`） | `texture_quality`（标清/高清）、`geometry_quality`（standard/ultra）、`pbr`、`texture` | -| 一致性能力 | 千问编辑支持角色一致性 | 万相2.7 参考生支持角色形象与音色一致性 | 多图视角约束保证几何一致性 | -| 预览能力 | 直接返回图片 | 直接返回视频 | 额外返回 `rendered_image_url` 预览渲染图 | +| **prompt** | 支持，中英文文本 | 支持，最大 1024 字符 | 支持，中英文 | +| **图像输入** | 参考图 URL（图生图/编辑） | 单图 URL 或 4 视角图片数组 | 首帧/尾帧 URL（图生视频）；参考图（参考生视频） | +| **[多模态](../concepts/multimodal.md)数组** | 部分模型支持多图参考 | images 数组（固定 4 元素，视角顺序前/左/后/右） | media 数组（first_frame/last_frame/image_url/video/audio_url/reference_image） | +| **分辨率参数** | 宽高自由设置 | texture_quality / geometry_quality | resolution / size（480P~4K） | +| **数量参数** | 1~6 张 | count 固定 1 | duration（秒）、seed、watermark 等 | ## 适用场景建议 -- **选图像生成 API**：需要静态视觉产出，强调文字渲染、风格化、精确编辑（增删移动物体、改动作）、虚拟模特/试衣/海报等电商与营销场景。优先用同步调用模型（千问图像、万相2.6+/2.7、Z-Image）以简化流程；批量或创意类任务再用异步。 -- **选视频生成 API**：需要动态叙事、数字人、肖像动态视频、视频编辑/换人。文生视频、图生视频（首帧/首尾帧）、参考生视频均可，万相2.7 是推荐的新版协议，支持多模态输入与角色/音色一致性。注意 PixVerse、Vidu 仅北京地域可用且需单独开通。 -- **选 3D 生成 API**：需要可直接导入引擎/3D 软件的 GLB 资产，适用于游戏、电商商品 3D 展示、工业设计。仅北京地域可用，需开通 Tripo。高精度选 `Tripo/Tripo-H3.1`（最高 200 万面），追求速度选 `Tripo/Tripo-P1.0`。 +### 图像生成 + +- **需要高文本渲染质量**（海报、配图含文字排版）→ 优先选择 Qwen-Image 系列。 +- **追求高性价比的通用文生图** → Z-Image-Turbo 或万相 wan2.6 系列。 +- **图像编辑需求**（改字、增删物体、风格迁移）→ Qwen-Image-Edit 或万相 wan2.7-image 编辑能力。 +- **创意工具场景**（虚拟模特、涂鸦作画、创意海报）→ 对应专项工具模型，但注意部分仅北京地域或免费体验。 +- **需要 HTTP 同步调用简化集成** → 选择 wan2.7 / wan2.6 / z-image / qwen-image 等支持同步的模型。 + +### 3D 生成 + +- **游戏/影视高精度 3D 资产** → Tripo/Tripo-H3.1，最高 200 万面，支持 PBR 材质。 +- **快速原型/预览** → Tripo/Tripo-P1.0，面数较低但速度更快。 +- **无贴图基础模型需求** → 同时将 `texture` 和 `pbr` 设为 `false`，获取 `base_model_url`。 +- **注意**：仅限北京地域 [API Key](../concepts/api-key.md)，开通前需在控制台模型市场搜索 Tripo 并完成授权。 -## 技术选型参考 +### 视频生成 -1. **产出形态决定大类**：图片→图像 API；视频→视频 API；3D 模型→3D API。三者端点不同，不可混用。 -2. **延迟敏感优先同步**：仅图像 API 提供同步调用，适合交互式产品；视频与 3D 必须异步，需在业务侧实现轮询或配置异步任务回调（3D 查询接口默认 RPS 20）。 -3. **地域与鉴权**：三类均要求模型、Endpoint、API Key 同地域。3D 仅北京可用；千问-图像翻译、PixVerse、Vidu 也仅北京。建议迁移到业务空间专属域名以获得更好性能与稳定性。 -4. **任务复用**：`task_id` 24 小时有效，三类都要求轮询而非重复创建任务；3D 产物下载链接仅 2 小时，需及时落盘。 -5. **输入约束**：3D 多图必须按前/左/后/右 4 视角顺序；图像图文混排需开启 `enable_interleave=true` 并配合 SSE 流式；视频首尾帧、参考生有专属模型变体。 -6. **模型开通**：可灵、PixVerse、Vidu、Tripo 均需先在控制台搜索并开通授权，再调用 API。 +- **最新推荐体验** → 万相 wan2.7（新版协议），支持图生视频首帧/首尾帧/续写、文生视频、参考生视频及视频编辑。 +- **旧版项目兼容** → wan2.6 及早期模型走旧版协议（图生视频仅首帧）。 +- **需要 4K 超清或对口型/动作模仿** → PixVerse 系列（pixverse-c1/v6/upscale/lipsync/motioncontrol）。 +- **智能分镜、多主体参考** → 可灵 Kling（kling-v3 / kling-v3-omni）。 +- **人像动画/数字人** → animate-anyone / emo / liveportrait / wan2.2-s2v 等系列。 +- **注意路径差异**：万相图生动作（wan2.2-animate-move）、视频换人（wan2.2-animate-mix）、数字人（wan2.2-s2v）使用 `image2video/video-synthesis` 路径，而非 `video-generation/video-synthesis`。 + +## 技术选型总结 + +| 选型场景 | 推荐方案 | 关键考量 | +| --- | --- | --- | +| 静态营销素材 / 产品配图 | 图像生成（Qwen-Image / Z-Image） | 文本渲染质量、输出张数、同步调用可用性 | +| 电商虚拟模特 / 创意海报 | 图像生成创意工具（wanx-virtualmodel / wanx-poster） | 地域限制、部分免费体验 | +| 游戏/影视 3D 资产 | 3D 生成（Tripo-H3.1） | 面数精度、PBR 材质、仅北京地域 | +| 快速 3D 原型验证 | 3D 生成（Tripo-P1.0） | 速度优先、面数较低 | +| 短视频/广告素材 | 视频生成（wan2.7 / Kling） | 分辨率、时长、多镜头分镜 | +| 人像动画 / 数字人 | 视频生成（animate-anyone / wan2.2-s2v） | 口型同步、动作模仿、路径差异 | +| 超清/对口型视频 | 视频生成（PixVerse upscale / lipsync） | 4K 超分、对口型能力 | ## 被对比主题页 - [image generation](../api/image-generation.md) -- [video generation api](../api/video-generation-api.md) - [3d generation](../api/3d-generation.md) +- [video generation api](../api/video-generation-api.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-api.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-api.md new file mode 100644 index 00000000..38716f9f --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-api.md @@ -0,0 +1,65 @@ +# [多模态](../concepts/multimodal.md)生成 API 对比 + +百炼平台提供图像、3D 和视频三类[多模态](../concepts/multimodal.md)生成能力，均通过 DashScope 网关以 REST / SDK 方式调用。三者共享「异步创建任务 → 轮询 task_id」的基础调用模式，但在输入模态、输出产物、支持模型矩阵、地域限制和计费粒度上存在显著差异。本页旨在帮助开发者快速理解三者的技术边界，为方案选型提供参考。 + +## 关键维度对比 + +| 对比维度 | 图像生成（Image Generation） | 3D 生成（3D Generation） | 视频生成（Video Generation） | +| --- | --- | --- | --- | +| **输入格式** | 文本提示词；参考图（图生图 / 图像编辑）；多图参考（部分模型支持最多 14 张） | `prompt`（文生 3D，≤1024 字符）；`image`（单图生 3D）；`images`（多图生 3D，固定 4 视角：前/左/后/右） | `prompt` 文本；`media` [多模态](../concepts/multimodal.md)数组（首帧/尾帧/参考图/视频/音频）；旧版协议直接 URL 字段 | +| **输出格式** | 单张或多张图像（1~6 张），JPEG/PNG 格式 | PBR 材质 GLB 模型或无贴图基础 GLB 模型 + 预览渲染图 | 视频文件，支持 480P~4K 分辨率，时长常见 5/8 秒，部分模型可生成音频 | +| **支持模型** | Qwen-Image / Qwen-Image-Edit、万相 wan2.7/wan2.6/wan2.5/wanx-v1、Z-Image、Kling、Vidu 等多家系列 + 多项创意工具 | Tripo-H3.1（高精度，最高 200 万面）、Tripo-P1.0（专业，速度更快，最高 2 万面） | 万相 wan2.7/wan2.6/早期版本、HappyHorse、PixVerse、Vidu、Kling、人像动画系列、万相动作/换人/数字人 | +| **API 端点** | 各模型独立端点，多数为异步 `X-DashScope-Async: enable`；wan2.6/2.7/z-image/qwen-image 部分支持 HTTP 同步调用 | `POST /api/v1/services/aigc/video-generation/3d-generation`（创建）+ `GET /api/v1/tasks/{task_id}`（轮询），**仅异步** | `POST /api/v1/services/aigc/video-generation/video-synthesis`（创建）+ `GET /api/v1/tasks/{task_id}`（轮询），**仅异步**；部分模型用 `image2video/video-synthesis` 路径 | +| **地域限制** | 多数模型多地可用；部分创意工具（图像翻译、涂鸦作画、虚拟模特等）为北京地域独占 | **仅限华北2（北京）**，必须使用北京地域 [API Key](../concepts/api-key.md) | 需模型/Endpoint URL/[API Key](../concepts/api-key.md) 同一地域；北京/新加坡/美国/德国多地域支持，跨地域调用失败 | +| **调用模式** | 异步为主，部分支持同步 | 纯异步（`X-DashScope-Async: enable` 必选） | 纯异步（`X-DashScope-Async: enable` 必选） | +| **轮询间隔建议** | 取决于具体模型，多数秒级完成 | 约 15 秒，查询接口默认 RPS 20 | 取决于模型和时长，任务有效期 24 小时 | +| **产物有效期** | 视具体接口而定 | 下载链接有效期 **2 小时** | 下载链接有效期视具体接口而定 | +| **典型场景** | 营销素材、创意海报、虚拟模特、图像编辑、翻译、局部重绘 | 游戏资产、电商 3D 展示、工业设计原型 | 短视频创作、广告视频、数字人播报、视频编辑、对口型/动作模仿 | + +## 调用模式对比 + +| 特性 | 图像生成 | 3D 生成 | 视频生成 | +| --- | --- | --- | --- | +| 同步调用 | 部分支持（wan2.6/2.7、z-image、qwen-image 等） | 不支持 | 不支持 | +| 异步创建 + 轮询 | 大多数模型 | 唯一方式 | 唯一方式 | +| `X-DashScope-Async` 必选 | 仅异步模型 | 必选 | 必选 | +| task_id 有效期 | 视接口而定 | 24 小时 | 24 小时 | + +## 输入灵活性对比 + +| 能力 | 图像生成 | 3D 生成 | 视频生成 | +| --- | --- | --- | --- | +| 纯文本输入 | 支持 | 支持（≤1024 字符） | 支持 | +| 单图输入 | 支持（图生图 / 图像编辑） | 支持（单图生 3D，JPEG/PNG，≤20MB） | 支持（首帧 / 尾帧） | +| 多图输入 | 支持（最多 14 张参考图，Vidu 等） | 支持（固定 4 视角：前/左/后/右） | 支持（多参考图、首尾帧组合） | +| 视频输入 | 不支持 | 不支持 | 支持（部分模型接收视频 URL 作为输入） | +| 音频输入 | 不支持 | 不支持 | 支持（数字人、对口型等模型） | +| 多输出 | 支持（1~6 张） | 支持（PBR 模型 + 渲染图） | 单个视频文件 | + +## 适用场景建议 + +- **图像生成**：适合需要快速产出高质量静态视觉内容的场景，如电商商品图、营销海报、虚拟模特展示、图像编辑与翻译。当需要文本渲染能力强（Qwen-Image）、高性价比批量出图（Z-Image Turbo）、或多图参考融合（Kling / Vidu）时优先选用。部分创意工具（涂鸦作画、局部重绘、人像风格重绘）适合特定创意需求。 +- **3D 生成**：适合需要将文本或少量参考图转化为可直接使用的 3D 模型资产的场景，如游戏道具、电商 3D 展示、工业设计快速原型。**仅限北京地域**，需提前开通 Tripo 服务。追求精度选 Tripo-H3.1（可设 `ultra` 几何精度，最高 200 万面），追求速度选 Tripo-P1.0。 +- **视频生成**：适合需要动态视觉内容的场景，如短视频创作、广告视频、数字人播报与口型同步。万相 wan2.7 为新版协议首选（支持多主体参考生视频 + 音色），需要 4K 超清或动作模仿选 PixVerse，需要智能分镜或首尾帧控制选 Kling。人像动画系列（animate-anyone / emo / liveportrait）适合舞蹈、唱演、播报等人物驱动场景。 + +## 选型决策流程 + +1. **确定输出模态**：静态图像 → 图像生成；3D 模型 → 3D 生成；动态视频 → 视频生成。 +2. **确认地域**：3D 生成仅限北京；视频生成需确保模型/Endpoint/[API Key](../concepts/api-key.md) 同地域；图像生成多数多地可用但部分创意工具限北京。 +3. **评估输入素材**：是否有参考图、多视角图、首尾帧或音视频素材，据此筛选支持的模型。 +4. **权衡同步/异步**：如需快速同步响应，图像生成中的 wan2.6/2.7、z-image、qwen-image 系列支持 HTTP 同步；3D 和视频生成均为异步，需规划轮询或回调机制。 +5. **关注产物有效期**：3D 生成下载链接仅 2 小时有效，视频和图像产物有效期视接口而定，均需及时下载。 + +## 来源文档 + +- [image generation](../api/image-generation.md)（api/image-generation.md） +- [3d generation](../api/3d-generation.md)（api/3d-generation.md） +- [video generation api](../api/video-generation-api.md)（api/video-generation-api.md） + +## 被对比主题页 + +- [image generation](../api/image-generation.md) +- [3d generation](../api/3d-generation.md) +- [video generation api](../api/video-generation-api.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-apis.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-apis.md new file mode 100644 index 00000000..8465bbaa --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-apis.md @@ -0,0 +1,71 @@ +# 图像、视频与 3D 生成能力对比 + +百炼平台的[多模态](../concepts/multimodal.md)生成能力覆盖图像、视频和 3D 三类核心资产生产链路。三者都可以从文本或视觉素材出发，但在输入约束、调用方式、结果形态、任务耗时、地域要求和工程集成重点上差异明显。本文面向开发者梳理关键维度，帮助在营销创意、内容生产、数字人、商品展示、游戏与工业资产等场景中做技术选型。 + +## 关键维度对比 + +| 维度 | 图像生成与编辑 | 视频生成 API | 3D 生成 | +| --- | --- | --- | --- | +| 主要目标 | 生成或编辑二维图片，覆盖文生图、图生图、局部重绘、图像翻译、虚拟模特、试衣、海报、背景、擦除补全等 | 生成或编辑动态视频，覆盖文生视频、图生视频、首尾帧生视频、参考生视频、视频编辑、数字人、动作迁移、口型同步、超清与风格重绘等 | 生成可下载的三维模型，支持文生 3D、单图生 3D、多图生 3D | +| 典型输入 | 文本提示词、单张或多张图片、遮罩或涂抹区域、风格参数、翻译目标语言等 | 文本提示词、首帧/尾帧图片、参考图片、参考视频、音频、[多模态](../concepts/multimodal.md)素材等 | `prompt`、`image` 或 `images` 三选一；多图模式按前、左、后、右传入固定长度数组 | +| 典型输出 | PNG 图片，通常为 1～9 张，分辨率可到 1K/2K/4K 或 512×512～2048×2048，依模型而定 | 视频文件 URL，常伴随封面、音频或中间状态信息，时长、分辨率和模式依模型而定 | GLB 模型文件 URL，包括 PBR 材质模型或无贴图基础模型，并返回预览渲染图 | +| 代表模型 | 千问图像、万相图像、Z-Image、可灵图像、Vidu 图像、千问图像翻译、万相创意工具类模型 | 万相 2.7/2.6/2.5/2.2、HappyHorse、PixVerse、Vidu、Kling 等 | `Tripo/Tripo-H3.1`、`Tripo/Tripo-P1.0` | +| 调用方式 | 新版千问图像、万相 2.6+、Z-Image 等支持同步调用；可灵、Vidu、万相旧版和多数工具类模型多为异步 | 以异步为主，创建任务后使用 `task_id` 轮询结果 | 强制异步，必须创建任务后轮询查询 | +| 主要 Endpoint | 同步[多模态](../concepts/multimodal.md)生成：`/api/v1/services/aigc/multimodal-generation/generation`；异步图像常见为 `/text2image/image-synthesis`、`/image2image/image-synthesis`、`/image-generation/generation` 等 | 统一视频生成入口：`/api/v1/services/aigc/video-generation/video-synthesis`，北京和新加坡地域常用[业务空间](../concepts/workspace.md)专属域名 | `/api/v1/services/aigc/video-generation/3d-generation` | +| 异步请求头 | 异步模型需 `X-DashScope-Async: enable`；同步模型不需要 | 通常必须带 `X-DashScope-Async: enable` | 必须带 `X-DashScope-Async: enable`，否则会报不支持同步调用 | +| 地域要求 | 依模型和[业务空间](../concepts/workspace.md)而定；新版接口通常使用[业务空间](../concepts/workspace.md)专属域名 | 模型、Endpoint 和 [API Key](../concepts/api-key.md) 必须同地域；北京、新加坡较常见，部分旧模型有其他地域或旧域名 | 仅支持华北 2（北京），[API Key](../concepts/api-key.md) 也必须为北京地域 | +| 结果时效 | 异步图像结果 URL 通常有有效期，常见为 24 小时；需按文档及时保存 | `task_id` 通常 24 小时有效，视频结果 URL 也应及时下载 | `task_id` 24 小时有效；模型文件和渲染图 URL 仅 2 小时有效 | +| 关键参数 | `prompt`、`negative_prompt`、`size`、`n`、`style`、`watermark`、`prompt_extend`、图像 URL、遮罩、编辑指令等 | `prompt`、`media`、分辨率、时长、水印、音频、风格、模式、随机种子等；不同模型的媒体字段差异较大 | `texture_quality`、`geometry_quality`、`pbr`、`texture`；`geometry_quality=ultra` 仅适用于 Tripo-H3.1 | +| 工程关注点 | 模型版本差异、同步/异步分流、批量图片数量、素材 URL 可访问性、提示词与图像编辑参数适配 | 任务生命周期管理、退避轮询、地域一致性、素材格式和顺序、长耗时与失败重试 | 输入互斥校验、多视角图片顺序、北京地域校验、结果文件快速下载和资产入库 | +| 适合场景 | 电商图、广告海报、图像局部修改、商品背景、人像写真、视觉素材翻译、UI/图表还原 | 短视频广告、分镜生成、角色一致视频、商品动态展示、数字人、视频续写和风格化 | 游戏/AR/VR 资产、商品 3D 展示、原型模型、三维素材快速生成 | + +## 适用场景建议 + +### 优先选择图像生成的场景 + +当最终交付物是静态视觉资产，或者需要高频、低延迟地产出多版创意时，应优先使用图像生成与编辑能力。图像接口覆盖面最广，既有同步模型适合交互式应用，也有异步模型适合批处理或高质量生成。典型场景包括商品主图、海报、头像、虚拟模特、试衣、背景替换、局部修图、图片翻译和创意文字。 + +如果业务需要在已有图片上做精确修改，应优先评估千问图像编辑或万相图像编辑，而不是重新文生图；如果目标是 UI、图表、带文字海报或跨语言图片本地化，则应重点关注文字渲染和图像翻译模型的能力边界。 + +### 优先选择视频生成的场景 + +当业务目标包含时间维度、镜头运动、角色表演或动态广告时，应选择视频生成 API。视频生成的工程复杂度高于图像：多数模型异步执行，单任务耗时通常以分钟计，素材字段和参数协议也随模型变化较大。因此开发者应把视频生成视为任务系统，而不是普通同步推理接口。 + +如果输入只有文本，可从文生视频模型开始；如果已有商品图、角色图或关键帧，应选择图生视频、首尾帧生视频或参考生视频。需要角色一致性、多主体互动或广告展示时，参考生视频更合适。新项目不建议直接复用旧版万相参数，应以目标模型的新版文档为准，尤其注意万相 2.7 与 2.1～2.6 的协议差异。 + +### 优先选择 3D 生成的场景 + +当最终交付物需要进入三维渲染、游戏引擎、AR/VR、商品 3D 展示或后续建模流程时，应选择 3D 生成。3D 生成不像图像或视频那样主要输出可直接播放的媒体，而是输出 GLB 模型文件，因此后续通常还需要下载、存储、预览、质检、压缩或导入资产管线。 + +文生 3D 适合快速探索概念；单图生 3D 适合从商品图或角色图生成模型；多图生 3D 适合对形体一致性要求更高的场景。若需要更高面数和几何细节，可评估 `Tripo/Tripo-H3.1`；若更关注速度和常规专业资产，可评估 `Tripo/Tripo-P1.0`。 + +## 开发者选型参考 + +1. **先按交付形态选型**：交付静态图选图像，交付动态内容选视频，交付可复用三维资产选 3D。不要仅因为输入都是文本就混用接口，输出形态决定后续工程链路。 +2. **再按调用模式设计架构**：图像既有同步也有异步；视频和 3D 基本按异步任务处理。只要涉及异步，就应统一封装创建任务、保存 `task_id`、轮询、超时、失败重试和结果下载。 +3. **严格校验地域和 Endpoint**：视频和 3D 对地域更敏感。3D 仅支持北京地域；视频必须保证模型、[API Key](../concepts/api-key.md) 与 Endpoint 同地域。生产环境建议在配置层绑定模型与地域，避免运行时混用。 +4. **按模型文档适配输入协议**：视频各模型对 `media` 类型、URL 字段和素材数量要求差异明显；3D 的 `prompt`、`image`、`images` 互斥；图像编辑也会因模型系列不同而有不同参数。通用封装可以抽象任务生命周期，但不要把具体参数强行统一。 +5. **及时持久化生成结果**：异步结果 URL 通常有有效期，3D 模型 URL 仅 2 小时。对于需要长期使用的资产，应在任务成功后立即下载到自有存储，并记录生成参数、模型版本和源素材。 +6. **用成本和时延约束模型选择**：图像适合高频迭代和 A/B 创意；视频单次成本和等待时间更高，应减少重复创建任务；3D 生成还要考虑后处理和资产管线成本。 + +## 推荐组合方案 + +| 业务目标 | 推荐能力组合 | 说明 | +| --- | --- | --- | +| 电商商品内容生产 | 图像生成/编辑 + 视频图生视频 | 先生成或优化商品图，再用首帧或参考图生成短视频广告 | +| 品牌海报与多语言投放 | 图像生成 + 图像翻译 | 先生成主视觉，再对图中文字做本地化并保留排版 | +| 角色一致短视频 | 参考生视频 + 图像编辑 | 先用图像编辑稳定角色或素材，再作为参考生成视频 | +| 3D 商品展示 | 单图/多图生 3D + 图像渲染预览 | 使用商品多视角图生成 GLB，并将预览图用于审核或展示 | +| 游戏或虚拟场景原型 | 文生图 + 文生 3D | 先用图像探索概念风格，再生成可导入的三维模型 | + +## 小结 + +图像生成能力覆盖最广、交互性最好，适合静态视觉资产和快速创意迭代；视频生成强调时间维度和多模态素材控制，适合动态内容生产，但需要更完善的异步任务治理；3D 生成面向三维资产管线，地域和结果时效约束更强，适合需要 GLB 模型的场景。实际选型时，应以交付物形态为第一判断标准，再结合模型协议、地域、任务耗时、结果有效期和后处理成本做工程设计。 + +## 被对比主题页 + +- [image generation](../api/image-generation.md) +- [video generation api](../api/video-generation-api.md) +- [3d generation](../api/3d-generation.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-comparison.md index 9ac91dd6..8f600c8b 100644 --- a/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-comparison.md +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/multimodal-generation-comparison.md @@ -1,66 +1,46 @@ -# 图像生成、视频生成与3D生成对比 +# 图像、视频与3D生成对比 -百炼平台提供图像生成、视频生成和3D生成三大[多模态](../concepts/multimodal.md)内容创作能力。三者在输入输出格式、模型生态、调用方式和适用场景上各有侧重。本文从开发者技术选型角度，对这三类生成能力进行系统对比，帮助快速定位最适合业务需求的方案。 +百炼平台在[多模态](../concepts/multimodal.md)内容生成方向提供了三大类 API：**图像生成与编辑**、**视频生成**和 **3D 模型生成**。三者在底层模型生态、输入输出形态、调用模式（同步/异步）、接口路径与计费方式上差异明显。本页从开发者技术选型的角度，对这三类能力做横向对比，帮助你快速判断应接入哪类 API、选择哪个模型系列，以及集成时需要注意哪些约束。 ## 关键维度对比 -| 维度 | 图像生成 | 视频生成 | 3D生成 | -|------|---------|---------|--------| -| **输入格式** | 文本提示词、参考图像（单张/多张）、涂鸦草图 | 文本提示词、首帧/首尾帧图像、参考图像/视频/音频 | 文本提示词、单张图像、多图（4视角：前/左/后/右） | -| **输出格式** | PNG 图像（512x512 至 4K） | MP4 视频 | GLB 模型（PBR 材质或无贴图基础模型）+ 预览渲染图 | -| **调用方式** | 同步调用为主，部分模型支持异步 | 全部异步（创建任务 → 轮询结果） | 全部异步（创建任务 → 轮询结果） | -| **主要模型系列** | 千问-图像、万相(Wan/Wanx)、Z-Image、可灵(Kling) | 万相(Wan)、HappyHorse、爱诗(PixVerse)、Vidu、可灵(Kling) | Tripo（H3.1 / P1.0） | -| **模型数量** | 20+ 款模型覆盖各类场景 | 6 大模型家族，任务类型丰富 | 2 款模型（高精度 / 专业快速） | -| **可用地域** | 部分全地域，部分仅华北2（北京） | 多数仅华北2（北京），HappyHorse 支持海外地域 | 仅华北2（北京） | -| **批量输出** | 单次可生成 1-9 张图像 | 单次生成 1 条视频 | 单次生成 1 个3D模型 | -| **产物有效期** | 即时返回，URL 有时效 | task_id 有效期 24 小时 | task_id 有效期 24 小时，下载链接有效期 2 小时 | -| **典型生成耗时** | 秒级至十秒级 | 分钟级 | 分钟级（耗时较长） | - -## 能力覆盖对比 - -| 能力 | 图像生成 | 视频生成 | 3D生成 | -|------|:-------:|:-------:|:-----:| -| 文本生成 | 支持 | 支持 | 支持 | -| 图像/图片参考生成 | 支持 | 支持（首帧/首尾帧） | 支持（单图/多图） | -| 内容编辑 | 支持（局部重绘、风格迁移、扩图等） | 支持（指令编辑、视频迁移） | 不支持 | -| [多模态](../concepts/multimodal.md)混合输入 | 支持（文+图） | 支持（文+图+视频+音频） | 不支持 | -| 中文文字渲染 | 支持（千问、Z-Image 等） | 不适用 | 不适用 | -| 人像/人物专项 | 支持（人像风格重绘、AI试衣） | 支持（数字人、舞动人像、悦动人像等） | 不适用 | -| PBR 材质输出 | 不适用 | 不适用 | 支持 | - -## 计费方式差异 - -- **图像生成**：按张计费，不同模型单价不同（如扩图 0.18 元/张）。部分创意工具仅提供免费体验额度，用完不可付费续用。 -- **视频生成**：按任务计费，费用与视频时长、分辨率、模型版本相关。 -- **3D生成**：按任务计费，费用与贴图质量（standard/detailed）和几何精度（standard/ultra）相关。 +| 维度 | 图像生成 | 视频生成 | 3D 生成 | +| --- | --- | --- | --- | +| 核心任务 | 文生图、图像编辑、图像翻译、局部重绘、虚拟模特、试衣、海报、扩图、擦除补全等 | 文生视频、图生视频（首帧/首尾帧）、参考生视频、视频编辑、风格重绘、超清、对口型、数字人/人像动画 | 文生3D、单图生3D、多图生3D | +| 模型生态 | 千问图像（Qwen-Image）、万相（Wan）、Z-Image、可灵（Kling）、Vidu 及多种创意工具类模型 | 万相（wan2.1–2.7）、HappyHorse、爱诗（PixVerse）、Vidu、可灵（Kling）、AnimateAnyone/EMO/LivePortrait 等人像动画模型 | Tripo 系列（Tripo-H3.1、Tripo-P1.0） | +| 输入格式 | 文本提示词、图像（单图/多图）、涂抹区域、手绘草图等 | 文本提示词、图像（首帧/首尾帧/参考图）、音频、视频（[多模态](../concepts/multimodal.md)组合） | 文本提示词，或单图（JPEG/PNG），或固定 4 视角多图（前/左/后/右，`prompt`/`image`/`images` 三选一互斥） | +| 输出格式 | PNG 图像，1–9 张，分辨率 512×512~2048×2048 或 1K/2K/4K | 视频文件（各系列时长、分辨率、镜头能力不同，PixVerse 可 4K 超清） | GLB 三维模型（PBR 材质或无贴图）+ 预览渲染图 | +| 调用模式 | 同步（千问图像、万相 2.6+、Z-Image 推荐）与异步（万相 2.5 及以下、可灵、Vidu、创意工具类）并存 | 统一异步（创建任务 → 轮询结果） | 强制异步，必须携带 `X-DashScope-Async: enable` | +| 典型 API 端点 | 同步：`/services/aigc/multimodal-generation/generation`；异步：`/services/aigc/text2image/image-synthesis`、`/services/aigc/image2image/image-synthesis` 等（按功能细分） | 各模型家族对应的视频生成任务端点（异步创建 + `GET /api/v1/tasks/{task_id}` 轮询） | `POST /services/aigc/video-generation/3d-generation` + `GET /api/v1/tasks/{task_id}` | +| 地域限制 | 常规地域可用 | 部分三方模型（PixVerse、Vidu、Kling）需在控制台先开通 | 仅华北2（北京）地域，必须使用北京地域 API Key | +| 结果有效期 | 图像 URL 有效期 24 小时（异步 task_id 24 小时） | task_id 24 小时，结果 URL 需及时下载 | task_id 24 小时；模型文件/渲染图 URL 仅 2 小时 | +| 计费方式 | 按张计费；部分创意工具类模型仅免费体验（额度用完不可付费续用） | 按任务/时长计费，不同模型家族价格不同；三方模型需单独开通 | 按任务计费，Tripo 服务需在控制台开通 | +| 关键参数 | `prompt`、`negative_prompt`、`size`、`n`、`style`、`watermark`、`prompt_extend` | `prompt`、首帧/首尾帧图像、`mode`、`aspect_ratio`、`duration`、`audio`、`function`（旧版统一编辑模型） | `texture_quality`、`geometry_quality`（仅 H3.1）、`pbr`、`texture` | + +## 能力与复杂度差异 + +- **图像生成**能力矩阵最宽：除基础文生图/编辑外，还覆盖图像翻译、虚拟模特、AI 试衣、海报生成等垂直场景，且新版模型（千问图像、万相 2.6+、Z-Image）支持同步调用，集成最简单、延迟最低。 +- **视频生成**模型家族最多、任务类型最细：从通用文生/图生视频到参考生视频、视频编辑、对口型、动作迁移、数字人均有覆盖，但全部走异步任务模式，需要实现轮询逻辑；部分能力（如万相数字人 wan2.2-s2v）还有前置检测步骤。 +- **3D 生成**接口最收敛：仅 Tripo 两个模型、三种输入方式，但限制最多——地域锁定北京、强制异步、产物 URL 仅 2 小时有效，工程上必须做好即时下载与持久化。 ## 适用场景建议 -**选择图像生成的场景：** -- 电商商品图、营销海报、社交媒体配图等静态视觉内容 -- 需要精细文字渲染或图文混排的场景（如带中文的宣传图） -- 图像编辑与风格迁移（如局部重绘、背景替换、AI试衣） -- 对生成速度要求高、需要批量出图的场景 - -**选择视频生成的场景：** -- 短视频创作、广告片制作、动态内容营销 -- 人像动画（数字人播报、舞蹈视频、唱演视频） -- 需要多镜头叙事或多角色互动的复杂视频 -- 视频风格转换和口型替换等后期编辑 - -**选择3D生成的场景：** -- 游戏资产、AR/VR 场景中的3D模型快速原型 -- 电商3D商品展示 -- 需要 PBR 材质的高精度3D资产生产（最高 200 万面） -- 从多视角图片重建3D物体 - -## 技术选型要点 - -1. **生成速度**：图像生成最快（秒级），视频和3D生成均需分钟级等待，且必须使用[异步调用](../concepts/async-invocation.md)模式。 -2. **模型生态丰富度**：图像生成和视频生成均拥有多个模型家族可选，3D生成目前仅有 Tripo 系列。 -3. **地域限制**：3D生成仅限北京地域；视频和图像生成的部分模型也有地域限制，选型前需确认目标地域的模型可用性。 -4. **输出后处理**：视频和3D的产物下载链接有时效限制（3D仅 2 小时），需在业务流程中及时下载存储。 -5. **开通流程**：3D生成和部分视频/图像模型需在百炼控制台额外搜索并开通服务，不是默认可用。 +| 场景 | 推荐方案 | +| --- | --- | +| 电商主图、海报、社媒配图、文字渲染 | 图像生成（千问图像擅长复杂文字渲染，万相支持多种艺术风格） | +| 已有图片的修改、多图融合、去水印、扩图、超分 | 图像编辑（千问图像编辑 / 万相图像编辑） | +| 短视频创作、广告分镜、单/多镜头叙事 | 视频生成（wan2.7 文生视频支持自然语言分镜，可灵 v3-omni 支持智能分镜） | +| 角色一致性的系列视频、IP 内容 | 参考生视频（wan2.7-r2v、pixverse-c1-r2v、vidu reference2video） | +| 数字人播报、口播、唱演 | wan2.2-s2v、EMO、LivePortrait、pixverse-lipsync 等人像动画类模型 | +| 游戏资产、AR/VR、商品三维展示 | 3D 生成（高精度选 Tripo-H3.1 ultra 模式，追求速度选 Tripo-P1.0） | + +## 选型要点 + +1. **延迟敏感选同步图像生成**：只有图像类新版模型提供同步调用；视频与 3D 一律异步，需按 15 秒左右间隔轮询任务状态。 +2. **统一轮询逻辑可复用**：视频与 3D 的异步协议一致（创建任务拿 `task_id` → `GET /api/v1/tasks/{task_id}`，状态 PENDING → RUNNING → SUCCEEDED/FAILED），可以抽象成同一套任务轮询组件。 +3. **注意产物时效**：三类结果 URL 都有有效期，3D 的 2 小时最短，务必在回调/轮询成功后立即转存到自有存储（如 OSS）。 +4. **确认开通与地域**：三方视频模型（PixVerse、Vidu、Kling）和 Tripo 均需先在控制台开通；3D 生成还要求北京地域 API Key，跨地域调用会直接失败。 +5. **免费体验模型不适合生产**：图像类的部分创意工具模型（虚拟模特、海报生成、擦除补全等）仅提供免费额度且不可付费续用，生产环境应优先选择千问图像编辑或万相 2.1+ 图像编辑等付费模型。 ## 被对比主题页 diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/omni-realtime-vs-realtime-user-guide.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/omni-realtime-vs-realtime-user-guide.md new file mode 100644 index 00000000..b46dc8d0 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/omni-realtime-vs-realtime-user-guide.md @@ -0,0 +1,53 @@ +# Omni 实时 API 与实时 API 使用指南对比 + +百炼平台提供两套实时交互相关的文档体系：**Omni 实时 API**（Qwen-Omni-Realtime）聚焦于全模态实时对话模型的事件协议与 SDK 用法，属于 API 参考层；**Realtime API 使用指南**则从接入方案视角出发，覆盖 WebSocket、WebRTC、AOQ 三种传输协议的选型、鉴权、SDK 集成与接入流程。两者面向的开发决策不同——前者回答"用什么模型、发什么事件"，后者回答"用什么协议连、怎么集成到端侧"。本页从技术选型角度对二者进行关键维度对比。 + +## 关键维度对比 + +| 维度 | Omni 实时 API | Realtime API 使用指南 | +|------|--------------|----------------------| +| 文档定位 | API 参考（事件协议、参数、模型能力） | 接入指南（协议选型、鉴权、端侧集成） | +| 传输协议 | WebSocket（DashScope SDK 封装） | WebSocket / WebRTC / AOQ 三选一 | +| 支持模型 | qwen3.5-omni-plus/flash-realtime、qwen3-omni-flash-realtime、qwen-omni-turbo-realtime | 上述全模态模型 + qwen3.5-livetranslate-flash-realtime、multimodal-dialog、FunASR 系列、CosyVoice 系列、qwen-audio-3.0-realtime 等 | +| 输入模态 | 音频（16 kHz PCM）+ 图像（JPG Base64 ≤256KB）+ 文本 | 音频 + 视频流（由端侧 SDK 采集编码） | +| 输出模态 | 文本、音频（24 kHz PCM）或两者组合 | 取决于所选模型/应用类型 | +| 交互模式 | server_vad / semantic_vad / Manual（手动提交） | WebSocket 支持全部模式；WebRTC 仅支持 VAD 模式 | +| 鉴权方式 | API Key（随 WebSocket 建连） | API Key；AOQ 需 AppServer 代理换取临时 Token，Key 不下发端侧 | +| SDK | DashScope Python SDK（≥1.25.17）/ Java SDK | DashScope SDK（WebSocket）；AOQ Client SDK v1.0.1（Android .aar / iOS .framework / HarmonyOS .har）；WebRTC 使用浏览器原生 API | +| 弱网对抗 | 无（WebSocket 层面） | WebSocket 差 / WebRTC 良好 / AOQ 极致 | +| 回声消除与降噪 | 无，需业务自行处理 | WebRTC 和 AOQ 内置 | +| 高级能力 | 工具调用（tools）、联网搜索（enable_search）、语义 VAD、静默引导（idle_timeout_ms）、声音复刻、输入转录（含语言/情绪识别） | 侧重连接层能力：媒体门控、SDP 交换、Token 鉴权流程 | +| 端侧平台 | 全平台（任何支持 WebSocket 的环境） | WebSocket 全平台；WebRTC 浏览器/移动端；AOQ 仅 Android / iOS / HarmonyOS | + +## 适用场景建议 + +### 选择 Omni 实时 API 文档的场景 + +- **服务端语音助手 / 电话客服**：通过 Python/Java SDK 在服务端建立 WebSocket 连接，精细控制事件流（如工具调用、联网搜索）。 +- **需要图像理解的[多模态](../concepts/multimodal.md)对话**：需发送视频帧/图像并结合语音交互（如实时视频问答），Omni API 提供 `input_image_buffer.append` 等图像事件。 +- **需要 Manual 模式（按下即说）**：WebRTC 协议不支持手动模式，若业务需要显式控制响应触发，应使用 WebSocket + Omni 实时 API。 +- **需要工具调用或联网搜索**：仅 Qwen3.5-Omni-Realtime 系列通过 Omni API 的 `tools` / `enable_search` 参数支持。 + +### 选择 Realtime API 使用指南的场景 + +- **浏览器端实时对话**：使用 WebRTC 协议，利用内置回声消除和降噪，无需自建音频前处理管线。 +- **移动端原生应用（弱网环境）**：AOQ 协议提供极致弱网对抗，适合户外、地铁等网络不稳定场景；但需部署 AppServer 做 Token 代理。 +- **仅需 ASR 或 TTS 单能力**：FunASR（实时语音识别）和 CosyVoice（实时语音合成）仅通过 WebSocket 协议接入，参见使用指南的模型支持表。 +- **已有 RTC 基础设施**：团队已有 WebRTC 媒体服务器或 SFU 架构，可直接复用现有管线。 + +## 技术选型决策参考 + +1. **确定模型需求**：若需要全模态（音频+图像+文本）对话能力，核心模型为 qwen3.5-omni 系列，两套文档均覆盖；若仅需 ASR/TTS，只看使用指南。 +2. **确定部署形态**：纯服务端 → WebSocket + Omni API 事件协议；浏览器 → WebRTC；移动原生 → AOQ。 +3. **确定交互模式**：需要 Manual 模式或工具调用 → 必须走 WebSocket（Omni API 协议层）；仅 VAD 自动对话 → 三种协议均可。 +4. **确定音频前处理责任**：WebSocket 方案需业务自行实现回声消除与降噪；WebRTC/AOQ 内置该能力。 +5. **安全合规**：移动端产品不应在客户端持有 API Key，AOQ 的 AppServer Token 代理模式是推荐做法；WebSocket/WebRTC 方案需自行在服务端代理建连请求。 + +两套文档互为补充：使用指南解决"怎么连"，Omni 实时 API 参考解决"连上之后说什么"。实际项目中通常需要同时参考两者——先按使用指南完成协议选型与建连，再按 Omni API 事件协议实现业务逻辑。 + +## 被对比主题页 + +- [omni realtime api](../api/omni-realtime-api.md) +- [realtime api user guide](../guides/realtime-api-user-guide.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/platform-extension.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/platform-extension.md new file mode 100644 index 00000000..28c62b6e --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/platform-extension.md @@ -0,0 +1,52 @@ +# 插件与 MCP 对比 + +在百炼平台上，**插件（Plug-in）** 与 **MCP（Model Context Protocol）** 都是为大模型扩展外部工具能力的主要手段，但两者在技术协议、服务来源、部署方式和集成模式上有显著差异。插件是百炼平台原生的工具集成机制，适合快速接入官方/三方 API；MCP 则是基于 Anthropic 开源标准协议的统一信息通道，适合需要跨平台互通、自定义部署复杂工具链的场景。本页帮助开发者根据业务需求做出技术选型。 + +## 关键维度对比 + +| 对比维度 | 插件（Plug-in） | MCP（Model Context Protocol） | +| --- | --- | --- | +| **技术协议** | 百炼平台自有接口规范，工具以 RESTful API（GET/POST）形式定义 | 基于 Anthropic 提出的 MCP 开源标准协议，支持 SSE 与 Streamable HTTP 两种传输方式 | +| **服务来源** | 官方插件、三方插件（云市场）、自定义插件 | 官方 MCP 服务、自定义 MCP 服务（脚本部署/从 AI 网关导入/从阿里云 OpenAPI 导入） | +| **工具定义方式** | 在插件下配置 API 路径、请求参数、返回数据，工具路径指向插件 URL 的相对路径 | 遵循 MCP 协议的代码包，托管到函数计算 FC；或通过 AI 网关/OpenAPI 升级为 MCP 服务 | +| **支持模型** | 通义千问-Turbo/Plus/Max、通义千问VL-Max/Plus 等（兼容性以控制台执行结果为准） | 须集成在智能体或[工作流](../concepts/workflow.md)应用中使用，不限制具体模型；无法在直连千问 API 时接入 | +| **集成方式** | [智能体应用](../concepts/agent-application.md)、[工作流](../concepts/workflow.md)应用、Assistant API | [智能体应用](../concepts/agent-application.md)、[工作流](../concepts/workflow.md)应用、外部调用（第三方应用 + 个人项目 SDK） | +| **数量限制** | 每个[智能体应用](../concepts/agent-application.md)最多添加 10 个工具 | 单个智能体最多同时添加 5 个 MCP 服务 | +| **调用机制** | 智能体/Assistant API 由模型自动判断调用；工作流中插件作为节点按编排执行 | 智能体由模型自动判断调用；工作流中每个 MCP 节点只能使用一个工具，需手动指定输入并传递输出 | +| **鉴权方式** | 支持服务级和用户级鉴权；鉴权类型含 basic、bearer、appcode；信息可放 Header 或 Query | 敏感信息使用 KMS 凭据加密管理；外部调用鉴权使用 `Authorization: Bearer ` | +| **部署与运行** | 自定义插件通过百炼平台在线配置 API；三方插件经云市场开通后直接使用 | 自定义 MCP 服务托管在函数计算 FC，分基础模式（按调用时长计费，有冷启动）和极速模式（部署费 + 调用费，常驻在线） | +| **计费方式** | 官方插件多为免费或限时免费（需申请开通） | 云部署限时免部署费，部分服务按第三方费率收取（如联网搜索 2000 次免费，之后 29 元/千次）；自定义部署按调用时长/部署时长计费 | +| **本地/远程资源** | 无特殊限制（取决于 API 本身） | FC 无固定出口公网 IP，访问远程资源需配置 IP 白名单或 VPC 网络；不支持访问本地资源（本地资源建议本地部署） | +| **跨平台互通** | 仅限百炼平台内部使用 | 支持 Cherry Studio、Cursor 等第三方客户端一键配置，也可通过 SDK 集成到个人项目 | +| **首次访问授权** | 需授权服务关联角色 `AliyunServiceRoleForSFMAccessCloudAPI`（主账号直接授权，RAM 子账号需创建权限策略） | 需在 MCP 广场开通或部署服务；已开通用户需从旧版 SSE 升级到 Streamable HTTP 协议 | +| **[Token](../concepts/token.md) 消耗** | 工具返回结果作为上下文传入模型 | MCP 返回内容同样作为上下文传入模型，间接增加输入/输出 [Token](../concepts/token.md) | +| **版本更新** | 平台管理更新 | npx/uvx 部署的服务在源版本更新后不自动更新，需手动重新部署；私有 npm 仓库暂不支持 | + +## 各方案适用场景建议 + +### 选择插件（Plug-in）当： + +- **快速接入现成 API**：官方插件（如代码解释器、计算器、图片生成、夸克搜索等）开通即用，无需编写代码或部署基础设施。 +- **简单 RESTful API 工具化**：已有 HTTP 接口，只需配置路径、参数、返回格式即可让模型调用，适合快速验证 MVP。 +- **仅百炼内部使用**：工具只需在智能体应用或工作流应用中使用，不涉及跨平台或外部集成。 +- **对成本敏感且调用量不大**：大量官方插件免费或限时免费，适合预算有限的场景。 +- **工作流编排中的确定性节点**：需要按固定逻辑链路执行工具调用（非模型自主决策），插件节点方式更直观。 + +### 选择 MCP 当： + +- **需要跨平台工具复用**：同一套工具需同时服务于百炼智能体、Cherry Studio、Cursor 等不同客户端，MCP 协议提供统一接口。 +- **工具链复杂、需自定义部署**：有现成的 MCP 协议代码包（Node.js/Python），或需把已有 RESTful API 通过 AI 网关升级为 MCP 服务。 +- **操作阿里云资源**：需通过 OpenAPI 操作 OSS、ECS 等阿里云产品，从 OpenAPI 开发者门户直接发布为 MCP 服务最为便捷。 +- **长时间在线、调用频繁**：极速模式下服务常驻运行，无冷启动延迟，适合生产环境高频调用。 +- **外部应用集成**：需要把百炼的 MCP 服务集成到第三方应用或个人项目，通过 SDK 灵活编码调用。 + +### 混合使用 + +两者并非互斥。实际项目中常见做法是：用 MCP 接入跨平台复用的核心工具链，同时用官方插件补充搜索、图片生成等百炼原生能力，在智能体应用中统一编排。选型时优先考虑工具的来源生态、部署复杂度、跨平台需求和长期维护成本。 + +## 被对比主题页 + +- [plug in](../guides/plug-in.md) +- [model context protocol](../guides/model-context-protocol.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/realtime-api-vs-omni-realtime.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/realtime-api-vs-omni-realtime.md new file mode 100644 index 00000000..d3008a21 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/realtime-api-vs-omni-realtime.md @@ -0,0 +1,60 @@ +# 实时 API 与 Omni Realtime API 对比 + +百炼平台中的“实时 API”与“Omni Realtime API”都面向低延迟语音、视频和[多模态](../concepts/multimodal.md)交互，但两者定位层级不同：实时 API 更像一套实时交互接入方案，覆盖 WebSocket、WebRTC、AOQ 等多种传输协议与多类实时模型；Omni Realtime API 则是其中面向 Qwen-Omni-Realtime 模型的事件驱动 WebSocket 接口规范，重点描述会话参数、客户端事件、服务端事件和模型交互流程。开发者选型时，应先判断业务需要的是“端到端接入通道与端侧能力”，还是“直接控制 Omni 实时模型的消息协议”。 + +## 关键维度对比 + +| 维度 | 实时 API | Omni Realtime API | 选型提示 | +|---|---|---|---| +| 定位 | 面向实时交互场景的总体接入方案，覆盖多协议、多端侧和多模型类型 | 面向 Qwen-Omni-Realtime 系列模型的实时[多模态](../concepts/multimodal.md)对话 API | 若需要选择接入协议、端侧 SDK 或弱网方案，优先看实时 API；若已确定使用 Omni 实时模型并需要事件协议细节，查看 Omni Realtime API | +| 传输协议 | 支持 WebSocket、WebRTC、AOQ 三类接入方式 | 主要基于 WebSocket 长连接的事件驱动协议 | 浏览器/移动端实时媒体链路可考虑 WebRTC/AOQ；服务端直接对接模型事件流可使用 Omni Realtime API | +| 端侧适配 | WebSocket 适合全平台服务端集成；WebRTC 适合浏览器和已有 RTC 基础设施；AOQ 适合 Android、iOS、HarmonyOS 等移动端原生应用 | 更偏向客户端或服务端按事件协议发送音频、图像、会话配置和响应控制事件 | 需要端侧 SDK、媒体采集播放、回声消除、降噪和弱网对抗时，实时 API 的 WebRTC/AOQ 更完整 | +| 输入格式 | 取决于协议和模型类型，可覆盖音频、视频、[多模态](../concepts/multimodal.md)输入，以及 FunASR、CosyVoice、音频对话等实时能力 | 通过事件提交输入，典型包括 `input_audio_buffer.append`、`input_image_buffer.append`、`session.update` 等；音频输入固定为 16 kHz 单声道 PCM | 若需要精细控制音频缓冲区、图片输入、会话参数和轮次提交，Omni Realtime API 更直接 | +| 输出格式 | 取决于所选协议和模型；可通过媒体流、DataChannel 或 WebSocket 事件获得模型响应 | 服务端返回事件，如 `response.audio.delta`、`response.audio_transcript.delta`、转录增量、工具调用参数和 `response.done` | 需要逐事件处理转录、音频增量、工具调用和错误信息时，Omni Realtime API 的事件模型更清晰 | +| 支持模型 | 覆盖 qwen3.5-omni 实时全模态、livetranslate、多模态开发套件、FunASR、CosyVoice、qwen-audio 实时语音对话等 | 聚焦 `qwen3.5-omni-plus-realtime`、`qwen3.5-omni-flash-realtime`、`qwen3-omni-flash-realtime`、`qwen-omni-turbo-realtime` 等 Omni 实时模型 | 如果业务不只用 Omni 模型，还涉及 ASR、TTS、实时翻译或音频对话，应从实时 API 的支持矩阵开始选型 | +| API 端点 | WebRTC/AOQ 常见端点形态为 `https://{endpoint}/api/v1/webrtc/realtime?model=<模型名>`；WebSocket 走对应 DashScope 接入 | 通过 WebSocket 长连接与事件协议交互，围绕会话、音频缓冲区、响应和对话项事件展开 | 端点与建连方式随协议不同而变化，接入前需确认目标协议和模型是否兼容 | +| 鉴权方式 | 三种协议均使用 [API Key](../concepts/api-key.md) 建连鉴权；AOQ 推荐 AppServer 使用 [API Key](../concepts/api-key.md) 换取临时 Token 下发给客户端 | 通常通过 [API Key](../concepts/api-key.md) 建立 WebSocket 连接，后续在连接内用事件更新会话 | 端侧应用尤其应避免将 API Key 写入客户端；移动端原生场景优先采用 AOQ 的临时 Token 模式 | +| 会话控制 | WebRTC 有媒体门控和 `session.created` 后恢复媒体发送等要求；AOQ 建连前需关闭媒体发送，`session.updated` 后再开启 | 通过 `session.update` 配置 `modalities`、`voice`、`turn_detection`、`tools`、`enable_search`、采样参数等 | 需要大量动态调整模型行为、VAD、工具或联网搜索时，应重点参考 Omni Realtime API 的会话参数 | +| VAD / 手动模式 | WebRTC 仅支持 VAD 模式；不同协议对媒体提交方式有差异 | 支持 `server_vad`、部分模型支持 `semantic_vad`；`turn_detection: null` 可启用 Manual 模式 | “按下即说”等明确轮次控制场景更适合使用支持 Manual 模式的 Omni Realtime API WebSocket 事件协议 | +| 工具调用与联网搜索 | 实时 API 作为接入方案层面描述较少，具体能力取决于模型和 API | 支持 Function Calling、联网搜索等；`tools` 与 `enable_search` 互斥 | 如果业务需要工具调用闭环、回传 `function_call_output`，应按 Omni Realtime API 的事件流程实现 | +| 弱网、回声消除、降噪 | WebSocket 弱网能力较弱且无内置 AEC/降噪；WebRTC 良好且内置；AOQ 弱网对抗最强且内置 | API 侧主要定义事件交互，端侧音频处理与网络对抗能力不作为核心协议能力 | 移动端语音/视频通话、复杂网络环境或需要内置音频处理时，优先选择实时 API 的 AOQ/WebRTC 方案 | +| SDK 与集成复杂度 | WebSocket 可用 [DashScope SDK](../concepts/dashscope-sdk.md)；AOQ 需集成 Android/iOS/HarmonyOS SDK 及音频插件；WebRTC 需处理 SDP、DataChannel、媒体轨道 | 需要按事件协议自行管理音频编码、缓冲区提交、响应取消、工具调用和错误处理 | 快速服务端原型可用 WebSocket；高质量端侧体验需投入 WebRTC/AOQ 集成；精细模型控制需实现 Omni 事件协议 | +| 计费方式 | 文档切片未给出独立计费细节，通常应以所选模型、协议和百炼计费说明为准 | 文档切片未给出独立计费细节，通常按 Omni 实时模型用量和官方计费规则执行 | 不应仅凭 API 名称判断成本，正式上线前需核对目标模型的最新计费规则、音频/文本/多模态计量口径 | +| 典型场景 | 实时语音/视频对话、移动端弱网通话、浏览器实时互动、ASR/TTS、实时翻译、多模态开发套件接入 | Omni 实时多模态对话、语音 VAD/Manual 轮次控制、工具调用、联网搜索、声音与会话参数控制 | 若目标是“搭建实时交互产品”，先选实时 API 协议；若目标是“驱动 Omni 模型完成对话与工具调用”，重点实现 Omni Realtime API | + +## 适用场景建议 + +### 优先选择实时 API 的场景 + +- 需要在 WebSocket、WebRTC、AOQ 之间做协议选型，或者同一业务需要覆盖服务端、浏览器端和移动端原生应用。 +- 业务重点是实时音视频链路质量，例如弱网对抗、回声消除、降噪、媒体采集播放、端侧 SDK 集成等。 +- 移动端原生应用不希望暴露 API Key，需要通过 AppServer 换取临时 Token，并使用 AOQ SDK 建连。 +- 除 Omni 实时模型外，还需要接入实时 ASR、实时 TTS、实时语音对话、实时翻译或多模态开发套件。 +- 团队更关心“如何把实时能力接入产品端”，而不是逐个事件控制模型交互细节。 + +### 优先选择 Omni Realtime API 的场景 + +- 已明确使用 Qwen-Omni-Realtime 系列模型，需要按 WebSocket 事件协议直接控制会话和响应。 +- 需要精细处理 `session.update`、音频缓冲区追加/提交/清空、响应创建/取消、对话项创建等事件。 +- 需要实现 Function Calling，并根据 `response.function_call_arguments.done` 执行业务工具后回传结果。 +- 需要使用 `server_vad`、`semantic_vad` 或 Manual 模式控制对话轮次，尤其是“按下即说”一类交互。 +- 需要启用联网搜索、配置音色、调整输出文本/音频模态、处理转录增量和音频增量等模型级能力。 + +## 开发者选型路径 + +1. **先确定产品形态**：如果是移动端实时通话、浏览器实时对话或多端音视频产品，先从实时 API 的协议能力开始；如果是服务端或客户端直接驱动 Omni 实时模型，则从 Omni Realtime API 的事件协议开始。 +2. **再确定模型范围**：如果只使用 Qwen-Omni-Realtime 系列，Omni Realtime API 能提供更细的参数和事件说明；如果还涉及 ASR、TTS、实时翻译或 qwen-audio，实时 API 的模型矩阵更适合作为入口。 +3. **评估网络与音频处理要求**：弱网、AEC、降噪和端侧媒体链路是 WebRTC/AOQ 的优势；纯 WebSocket 事件接入更适合服务端集成和精细协议控制。 +4. **评估安全边界**：任何客户端方案都不应硬编码 API Key。移动端原生优先采用 AppServer 换取 Token 的 AOQ 模式；浏览器 WebRTC 的 SDP 交换也建议由 AppServer 代理。 +5. **评估交互控制复杂度**：如果需要工具调用、联网搜索、Manual 模式、响应取消和细粒度错误处理，应按 Omni Realtime API 的事件流设计状态机。 + +## 总结 + +实时 API 是“接入实时能力的方案层入口”，帮助开发者在协议、端侧平台、弱网能力和模型类型之间做选择；Omni Realtime API 是“驱动 Omni 实时模型的协议层接口”，强调会话参数、事件类型、VAD/Manual 轮次、工具调用和响应流处理。实际项目中两者并非互斥：可以先用实时 API 选择合适的传输方案，再按 Omni Realtime API 实现具体模型交互。对于追求端侧体验和网络鲁棒性的应用，优先关注 WebRTC/AOQ；对于追求模型事件控制和业务编排能力的应用，优先关注 Omni Realtime API。 + +## 被对比主题页 + +- [realtime api user guide](../guides/realtime-api-user-guide.md) +- [omni realtime api](../api/omni-realtime-api.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/comparisons/realtime-apis-comparison.md b/skills/bailian-docs-llm-wiki/wiki/comparisons/realtime-apis-comparison.md new file mode 100644 index 00000000..1977044b --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/comparisons/realtime-apis-comparison.md @@ -0,0 +1,57 @@ +# Omni Realtime API 与 Realtime API 使用指南对比 + +百炼平台提供了两套面向实时[多模态](../concepts/multimodal.md)交互的接入体系：**Omni Realtime API**（Qwen-Omni-Realtime 系列模型的 WebSocket 事件协议接口）和 **Realtime API 使用指南**（覆盖 WebSocket / WebRTC / AOQ 三种传输协议的整体接入方案）。两者定位不同——前者是模型级 API 参考，聚焦事件协议与参数细节；后者是平台级接入指南，聚焦传输协议选型、端侧 SDK 集成与鉴权流程。本页从开发者技术选型角度梳理两者的关键差异与适用场景。 + +## 关键维度对比 + +| 维度 | Omni Realtime API | Realtime API 使用指南 | +| --- | --- | --- | +| 文档定位 | 模型 API 参考（事件协议、参数、SDK 方法） | 平台接入指南（协议选型、端侧集成、鉴权） | +| 传输协议 | 仅 WebSocket | WebSocket、WebRTC、AOQ（AI over QUIC） | +| 支持模型 | qwen3.5-omni-realtime 系列、qwen3-omni-flash-realtime、qwen-omni-turbo-realtime | 上述模型 + qwen3.5-livetranslate-flash-realtime、multimodal-dialog、FunASR 系列、CosyVoice 系列、qwen-audio-3.0-realtime 等 | +| 输入模态 | 音频（16 kHz PCM）、图像（JPG Base64 ≤256KB）、文本 | 音频、视频流（含摄像头实时画面）、文本 | +| 输出模态 | 文本、音频（24 kHz PCM），可配置 modalities | 文本、音频（内置 Opus 编解码可选） | +| 交互模式 | server_vad / semantic_vad / Manual（手动提交） | WebSocket 支持全部模式；WebRTC 仅支持 VAD 模式 | +| 鉴权方式 | WebSocket 建连时 [API Key](../concepts/api-key.md) 鉴权 | WebSocket/WebRTC 使用 [API Key](../concepts/api-key.md)；AOQ 通过 AppServer 换取临时 [Token](../concepts/token.md)，[API Key](../concepts/api-key.md) 不下发端侧 | +| 端侧 SDK | DashScope Python/Java SDK（OmniRealtimeConversation 类） | [DashScope SDK](../concepts/dashscope-sdk.md)（WebSocket）+ AOQ Client SDK v1.0.1（Android .aar / iOS .framework / HarmonyOS .har） | +| 弱网对抗 | 无（依赖 WebSocket 本身） | WebSocket 差；WebRTC 良好；AOQ 极致 | +| 回声消除/降噪 | 无，需业务侧自行处理 | WebRTC 和 AOQ 内置；WebSocket 无 | +| 工具调用 | 支持（仅 qwen3.5 系列，含 function_call_output 回传流程） | 未涉及工具调用细节，依赖底层模型能力 | +| 联网搜索 | enable_search（仅 qwen3.5 系列，与 tools 互斥） | 未涉及 | +| 声音复刻 | 支持 | 未涉及 | +| 采样参数配置 | 详细（temperature / top_p / top_k / presence_penalty / seed 等，各模型默认值不同） | 未涉及参数细节 | +| 目标平台 | 服务端（Python/Java 后端集成） | 全平台：服务端、浏览器、Android / iOS / HarmonyOS 原生应用 | + +## 适用场景建议 + +### 选择 Omni Realtime API 的场景 + +- **服务端语音助手 / 电话客服**：后端通过 Python/Java SDK 建立 WebSocket 长连接，精细控制会话参数（VAD 阈值、采样参数、音色）。 +- **需要工具调用或联网搜索**：仅 Omni Realtime API 文档描述了完整的 function call 事件流与回传协议。 +- **图像+音频[多模态](../concepts/multimodal.md)理解**：需要在对话中发送视频帧截图（Base64 图像）进行实时分析。 +- **需要声音复刻**：自定义音色能力仅在此 API 层面有说明。 +- **快速原型验证**：单协议（WebSocket）、单 SDK 依赖，上手最快。 + +### 选择 Realtime API 使用指南的场景 + +- **浏览器端实时对话**：通过 WebRTC 协议直接在浏览器建立音视频通道，内置回声消除与降噪，无需服务端中转音频流。 +- **移动端原生应用（弱网环境）**：AOQ 协议针对移动弱网深度优化，SDK 提供 Android / iOS / HarmonyOS 原生组件，含 Opus 编解码插件。 +- **安全合规要求高**：AOQ 方案的 [Token](../concepts/token.md) 鉴权机制确保 [API Key](../concepts/api-key.md) 不暴露到客户端，适合正式发布的 C 端产品。 +- **仅需流式 ASR 或 TTS**：FunASR / CosyVoice 系列仅通过 Realtime API 指南中的 WebSocket 方案接入，不属于 Omni Realtime API。 +- **实时同传**：qwen3.5-livetranslate-flash-realtime 模型在 Realtime API 指南中有明确的协议支持矩阵。 + +## 技术选型参考 + +1. **协议优先级**：移动端弱网 → AOQ；浏览器 → WebRTC；服务端集成/快速原型 → WebSocket。三者共享同一套事件语义（session.created / session.updated 等），切换协议时业务逻辑迁移成本低。 +2. **模型选择**：需要工具调用、联网搜索、semantic_vad 等高级能力时，必须选用 qwen3.5-omni-realtime 系列（plus 或 flash）；仅需基础语音对话可考虑 qwen-omni-turbo-realtime（注意其采样参数不可修改）。 +3. **鉴权架构**：正式产品建议统一走 AppServer 代理模式——即使使用 WebSocket/WebRTC，也避免在客户端硬编码 [API Key](../concepts/api-key.md)。 +4. **音频格式注意**：Omni Realtime API 输入固定 16 kHz PCM、输出固定 24 kHz PCM，不支持自定义采样率；AOQ/WebRTC 方案由 SDK 处理编解码（Opus），开发者无需手动重采样。 +5. **组合使用**：两者并非互斥。典型架构为——移动端通过 AOQ SDK 建连（Realtime API 指南），会话内使用 Omni Realtime API 定义的事件协议（session.update、response.create 等）进行交互控制。 + +## 被对比主题页 + +- [omni realtime api](../api/omni-realtime-api.md) +- [realtime api user guide](../guides/realtime-api-user-guide.md) + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/agent-application.md b/skills/bailian-docs-llm-wiki/wiki/concepts/agent-application.md index 5b2f9be9..cfa9354f 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/agent-application.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/agent-application.md @@ -1,45 +1,51 @@ # 智能体应用 -智能体应用（Agent）是阿里云百炼平台的核心应用构建模式之一，通过自然语言零代码配置，让大模型基于角色设定自主决策、动态规划并调用知识库、MCP、Skill 等工具来完成任务。相较于流程固定的工作流应用，智能体强调 AI 的自主性，适合意图开放、需要动态编排的对话与轻量任务场景。 +智能体应用（Agent Application）是百炼平台的核心应用形态之一，通过自然语言配置（零代码）构建，由大模型根据系统提示词自主规划工具调用顺序，实现知识问答、任务处理等场景。开发者无需编写编排逻辑，只需定义角色、挂载工具与知识库，模型即可基于 ReAct 循环完成多步推理与执行。 -## 版本演进：Agent 1.0 与 Agent 2.0 +## 两代架构 -百炼提供两代技术架构不同的智能体，**不支持直接升级或版本切换**，迁移需重新创建： +百炼提供两代智能体应用，**推荐优先使用 Agent 2.0**： -- **新版智能体（Agent 2.0）**：2025 年 12 月 26 日上线。将知识库、MCP 等能力统一抽象为「工具」，由智能体自主规划调用时机与顺序，并完整展示「规划-执行-反思」链路。无旧版依赖时推荐使用。**仅支持 API 调用，不支持任何分享渠道**（魔笔/UI、钉钉、微信、组件、音视频互动）。 -- **旧版智能体（Agent 1.0）**：通过知识库（RAG）+ 插件扩展能力，先检索知识再决策是否调用工具，适合意图单一、流程固定的简单任务。自定义插件有 **5 秒超时限制**。分享渠道均为 1.0 功能。 +- **Agent 2.0**：将知识库、MCP 等统一为工具，由模型基于 ReAct 循环自主规划调用时机与顺序，完整暴露"思考-执行-反思"链路。2025 年 12 月 26 日上线。 +- **Agent 1.0**：先命中知识库再决策是否调用其他工具，仅展示最终结果。 -## 核心能力配置（Agent 2.0） +> 两代基于不同技术架构，**不支持直接升级或迁移**，切换需重新创建应用。 -- **模型选择**：推荐具备强工具调用能力的模型（如千问-Max 系列）；可配置最长回复长度、`temperature`、`enable_thinking`（思考模式）等参数。 -- **提示词（System Prompt）**：定义角色、行为指令与能力边界，支持自定义变量嵌入。 -- **内置工具**：沙箱环境下的 `bash`、`write`、`read`、`edit`、`glob`、`grep`、`download_file`，默认关闭需按需开启。 -- **知识库**：作为工具由智能体自主调用，支持标签过滤限定查询范围。 -- **MCP**：外部工具以 MCP 协议接入，支持动态非固定顺序调用。 -- **Skill**：可扩展能力包，智能体在对话中自动识别匹配任务并调用，无需额外编码。 -- **记忆**：短期记忆支持 0-30 轮上下文；长期记忆暂未支持。 -- **ReAct 最大轮次**：取值 1-50，限制单次会话内工具调用最大次数。 +## 关键配置 -## 文件问答 +| 配置项 | 说明 | +| --- | --- | +| 模型 | 推荐 `千问-Max` 系列以保障多步工具调用效果；支持 `enable_thinking`（仅思考模型可配）、`temperature`、最长回复长度 | +| 系统提示词 | 定义角色、输出格式、约束和工具使用指引；支持自定义变量（通过 `/` 引用） | +| 知识库（RAG） | 作为工具由 Agent 2.0 自主调用；支持按标签过滤检索范围；可开启"展示回答来源"以角标形式返回 | +| MCP 与插件 | Agent 2.0 全部外部工具统一走 MCP 协议；插件可一键转换为 MCP | +| 内置工具 | `bash` / `write` / `read` / `edit` / `glob` / `grep` / `download_file`，沙箱内运行，默认关闭 | +| 预解析文件 | 控制上传文件是否由平台预置解析器提取文本；关闭时文件 URL 作为上下文透传 | +| ReAct 最大轮次 | 1–50，超限后强制生成最终回复 | +| 短期记忆 | 0–30 轮上下文 | -智能体应用支持上传文件进行问答，提供三种处理模式： +## 文件问答 -| 模式 | 适用场景 | 特点 | -|------|---------|------| -| 全文引用 | 文档总结、全文翻译 | 简单直接，受上下文长度限制 | -| 切片检索（RAG） | 长文档问答、知识库检索 | 能处理超长文件，效果依赖检索策略 | -| 自定义处理 | 图片转换、视频分析等 | 功能灵活，依赖配置的工具 | +智能体支持三种文件处理模式： -限制：单会话最多 10 个文件，单文件不超过 10 MB（超出需用文件上传 API）。支持文档、图片、视频、音频等格式。 +| 模式 | 机制 | 适用场景 | +| --- | --- | --- | +| 全文引用 | 平台解析后将文件整体注入 Prompt（受上下文长度限制） | 全文总结、翻译、润色 | +| 切片检索（RAG） | 平台解析并切片，按相关性召回若干片段 | 长文档问答、来源定位 | +| 自定义处理 | 只把文件 URL 或原始内容交给模型，由模型调用工具处理 | 图像风格转换、视频分析 | -## 发布与调用 +- 单会话上传上限 **10 个文件**，单文件 **≤ 10 MB**。 +- 页面上传的文件仅当前会话有效；生产场景推荐通过文件上传 API 换取 `session_file_id`（有效期约 24 小时）。 +- API 调用时**无法**动态切换处理模式，取决于应用发布时的配置。 +- 文件问答共享所属智能体应用的限流：**默认 100 次/分钟**。 -所有应用类型均需先发布才能通过 API 集成，核心步骤：在应用配置页点击「发布」→ 在「发布渠道」查看调用方式。RAM 账号发布前需拥有 `ram:CreateServiceLinkedRole` 权限。 +## 调用方式 -API 调用与工作流应用完全一致，通过 `Application.call` / `POST /apps/{app_id}/completion` 触发： +通过 DashScope SDK 或 HTTP API 调用，接口为 `POST /apps/{APP_ID}/completion`： ```python import os +from http import HTTPStatus from dashscope import Application response = Application.call( @@ -47,35 +53,46 @@ response = Application.call( app_id='YOUR_APP_ID', prompt='你是谁？' ) -print(response.output.text) +if response.status_code == HTTPStatus.OK: + print(response.output.text) ``` -响应结构为 `{"output": {"finish_reason", "session_id", "text"}, "usage": {...}, "request_id": "..."}`，业务侧主要消费 `output.text`。 +响应结构为 `{"output": {"finish_reason", "session_id", "text"}, "usage": {...}, "request_id": "..."}`，业务侧主要消费 `output.text`。2025 年 11 月起还支持通过 Responses API 调用，提供同步与异步两种模式。 + +## 发布与分享 -## 分享与组件化（仅 Agent 1.0） +- **Agent 1.0** 支持多种分享渠道：UI 应用/魔笔、钉钉、微信公众号、组件化发布、音视频实时互动。临时体验二维码与 UI 体验链接有效期均为 24 小时。 +- **Agent 2.0** 仅支持通过 API 调用，不支持上述分享渠道。 -- **分享渠道**：UI 应用/魔笔、钉钉、微信公众号、音视频实时互动。UI 体验链接与音视频临时二维码有效期均为 **24 小时**。分享产生的费用由应用创建者 UID 账号承担。 -- **组件化**：智能体或工作流可发布为模块化组件供其他应用复用。接入智能体时组件作为工具，大模型据「组件描述」自动判断调用；预设系统参数 `query`、`imageList` 无法删除。注意避免嵌套调用（A↔B）和多级调用（A→B→C 易超时）。 +智能体或[工作流](workflow.md)应用可发布为模块化组件供其他应用复用，组件预设系统参数 `query` 和 `imageList`（不可删除，可隐藏）。接入方支持智能体应用（作为工具，模型自动调用）和[工作流](workflow.md)应用（作为组件节点，手动传参）。 -## 与 Managed Agents 的区别 +## 评测与监控 -Managed Agents 是服务端托管运行时，与无状态的智能体应用不同：它在独立云端沙箱容器中维护会话状态，支持中断与续接、事件历史持久化，面向多步工具调用、代码执行、文件处理等长时运行任务。 +- **评测**：支持自动评测（基于知识库自动生成评测集并评分）和手动评测（人工标注）。单应用评测深度评估表现，多应用横向评测最多对比 8 个应用。新版评测系统通过评估器（LLM / Code）和标签管理构建多维评测闭环。 +- **监控**：应用观测功能可端到端追踪调用链路，提供延时、[Token](token.md) 量等分钟级指标。支持 Root Span / All Span / Model Span 筛选模式，可将 Span 数据直接加入评测集。观测覆盖智能体应用、[工作流](workflow.md)应用和高代码应用。 -## 计费说明 +## 与其他形态的对比 -- **模型调用**：按模型类型和 Token 用量计费。 -- **知识库**：按量付费，召回的文本切片会增加输入 Token；自 2026 年 1 月 4 日起正式计费。 -- **MCP/插件**：部分官方 MCP 按调用计费，第三方 MCP 由第三方收取。 +| 维度 | 智能体应用 | 工作流应用 | 高代码应用 | +| --- | --- | --- | --- | +| 开发方式 | 自然语言配置（零代码） | 可视化节点编排（低代码） | Python 项目（专业代码） | +| 控制方式 | 模型自主规划 | 预定义节点顺序 | 代码完全控制 | +| 适合人群 | 业务/产品/运营 | 业务分析师/IT 实施 | AI 工程师 | +| 典型场景 | 客服、知识问答、任务助理 | 报告生成、审批流、数据标注 | 私有算法、复杂系统集成 | -百炼提供限时免费额度，可在模型广场查看。 +与 Managed Agents 的区别：智能体应用为无状态调用（应用侧维护上下文），Managed Agents 由服务端托管会话状态、沙箱环境与工具执行，支持中断与续接，适合长时运行的多步工具调用任务。 ## 关联主题页 +- [start using](../guides/start-using.md) - [llm application](../guides/llm-application.md) -- [bailian application calling](../guides/bailian-application-calling.md) -- [managed agents](../guides/managed-agents.md) - [application publishing and sharing](../guides/application-publishing-and-sharing.md) -- [start using](../guides/start-using.md) +- [bailian application calling](../guides/bailian-application-calling.md) +- [application evaluation](../guides/application-evaluation.md) +- [application monitoring](../guides/application-monitoring.md) +- [application permission management](../guides/application-permission-management.md) +- [application use cases](../guides/application-use-cases.md) - [skill](../guides/skill.md) +- [managed agents](../guides/managed-agents.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/api-key.md b/skills/bailian-docs-llm-wiki/wiki/concepts/api-key.md index a6884fa8..fa1a3d50 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/api-key.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/api-key.md @@ -1,80 +1,82 @@ # API Key 鉴权 -API Key 是调用阿里云百炼平台模型与应用的核心鉴权凭证，以 `Authorization: Bearer ` 的形式随请求携带，用于标识调用者身份并界定其可访问的模型、应用与业务空间范围。 +API Key 是调用阿里云百炼模型与应用服务的核心鉴权凭证，所有 API 请求（OpenAI 兼容、Anthropic 兼容、DashScope 原生协议）均通过它验证调用者身份并确定权限范围。API Key 的调用权限由其归属的[业务空间](workspace.md)决定，并与地域、计费方案严格绑定。 -## 在百炼平台的使用场景 +## 类型与前缀 -API Key 贯穿百炼的各类调用入口，不同场景下的获取与使用方式略有差异： +百炼平台存在三套相互隔离的 API Key 体系，必须与对应的 Base URL 配套使用，混用会导致 401/403 鉴权失败或意外扣费： -- **模型 API 直连**：调用文本、图像、视频、语音、向量等模型前，先在控制台对应地域的 **API Key** 页面创建 Key，配合服务端点 `base_url` 发起请求。 -- **应用与知识库调用**：智能体应用、工作流应用、知识检索/问答接口（DashScope 应用网关）均通过 API Key Bearer 鉴权，Base URL 常需拼接业务空间 ID，如 `https://{workspaceId}.cn-beijing.maas.aliyuncs.com`。 -- **框架集成**：LlamaIndex（RAG）、Spring AI Alibaba（智能体/工作流/知识库检索）均以 API Key 鉴权复用百炼能力，通常通过环境变量注入。 -- **百炼 CLI（`bl`）**：支持控制台登录、`--api-key`、环境变量、配置文件、临时传入等多种认证方式，可组合使用。 -- **不可信环境（浏览器/移动端）**：应由后端生成临时 API Key，避免永久 Key 泄露。 -- **子业务空间**：需使用该空间自身的 API Key，并预先为其授予相应模型的调用权限。 +| 类型 | 前缀 | 配套 Base URL | 适用场景 | +| --- | --- | --- | --- | +| 按量付费 Key | `sk-ws`（新建）/ `sk-`（存量） | `dashscope.aliyuncs.com` 或 `{WorkspaceId}.{region}.maas.aliyuncs.com` | 常规模型与应用调用、后端服务 | +| Token Plan 专属 Key | `sk-sp-` | `token-plan.cn-beijing.maas.aliyuncs.com` | Claude Code、Cursor 等 AI 工具订阅式使用 | +| 临时 API Key | `st-` | 与生成它的永久 Key 相同 | 浏览器、移动端等不可信环境 | -## Key 的类型与前缀 +## 获取与管理 -百炼的 API Key 因计费方式不同而彼此隔离，前缀是重要区分标识： +- **创建入口**：百炼控制台右上角选择地域（华北2（北京）、新加坡、日本（东京）、德国（法兰克福）；美国（弗吉尼亚）走独立控制台），进入 API Key 页面创建。需主账号或具备 `管理员` / `API-Key` 页面权限的子账号。 +- **仅展示一次**：新 Key 明文仅在创建时展示一次，关闭弹窗后无法再次查看，丢失需重置或重建。 +- **地域隔离**：各地域的 API Key 相互独立，不能跨地域使用，请求时需搭配对应地域的 Endpoint。 +- **控制台操作**：支持编辑、删除、禁用、重置（禁用/重置在美国（弗吉尼亚）地域不可用）。 +- **编程管理**：可通过 OpenAPI（CreateApiKey / ListApiKeys / ResetApiKey 等）管理，这类管控接口使用阿里云 AccessKey 签名鉴权，而非 API Key 本身。 -| 类型 | 前缀 | 说明 | -| --- | --- | --- | -| 按量付费（安全升级后） | `sk-ws` | 仅创建时展示一次明文，关闭弹窗后无法再查看，务必立即保存 | -| 按量付费（升级前旧 Key） | `sk-` | 仍可正常使用 | -| Token Plan / Coding Plan 专属 | `sk-sp-` | 套餐专属，与通用 Key 不可混用 | -| 临时 API Key | `st-` | 由后端接口生成、限时有效 | +## 权限模型 -> API Key 与其配套的 Base URL 必须成对使用。按量付费、Token Plan、Coding Plan 三者的 Key 与 Base URL 混用会导致意外扣费或返回 401/403 鉴权失败。 +- **归属[业务空间](workspace.md)**：Key 的调用权限完全由归属[业务空间](workspace.md)决定，同一空间内权限相同，无需按模型类型创建多个 Key。 +- **子空间隔离**：子业务空间的 Key 只能调用已授权模型和该空间内的应用；在百炼上调优后的模型仅能用其所在空间的 Key 调用。 +- **自定义权限**：创建时可选 `全部` 或 `自定义`——支持 IP 白名单（最多 20 个 IPv4/IPv6 地址或网段）与可访问模型范围限制（美国（弗吉尼亚）地域不支持自定义配置）。 -## 创建时的关键选项 +## 典型使用场景 -- **归属业务空间**：决定 Key 的调用权限。默认空间的 Key 可调用所有标准模型及默认空间应用；子空间的 Key 仅能调用已授权模型及本空间应用。同一空间内的 Key 权限相同，无需按模态分别创建。 -- **权限**：可选「全部」或「自定义」。自定义可配置 IP 白名单（最多 20 个 IPv4/IPv6 地址或网段）及可访问的模型/应用范围。 -- **创建主体**：需主账号，或具备「管理员」/「API-Key」页面权限的子账号。RAM 子账号在订阅套餐前需主账号授予 `AliyunBailianFullAccess` 等权限。 +### 模型 / 应用 API 调用 -## 配置与使用 +推荐将 Key 配置为环境变量 `DASHSCOPE_API_KEY`，避免硬编码泄露： -推荐将 API Key 写入环境变量，避免硬编码泄漏： +```bash +export DASHSCOPE_API_KEY='YOUR_KEY' # Linux / macOS +setx DASHSCOPE_API_KEY "YOUR_KEY" # Windows CMD +``` -- 通用约定使用 `DASHSCOPE_API_KEY`。 -- Spring AI Alibaba 应用集成用 `DASHSCOPE_API_KEY`，知识库检索用 `AI_DASHSCOPE_API_KEY`（约定不一致，关键是 `application.yml` 中占位符与实际变量名匹配）。 +SDK 调用时从环境变量读取，并搭配正确的 `base_url`： -调用时除 API Key 外，还需指定与地域、协议匹配的 `base_url`（OpenAI 兼容协议与 Anthropic 兼容协议不同）。 +```python +from openai import OpenAI +import os -## 临时 API Key +client = OpenAI( + api_key=os.getenv("DASHSCOPE_API_KEY"), + base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", +) +``` -面向浏览器、移动端等不可信环境，通过后端接口用永久 Key 换取限时凭证： +HTTP 直调时通过请求头 `Authorization: Bearer $DASHSCOPE_API_KEY` 传递。向量化、排序、图像视频生成等所有模型 API 均使用同一鉴权方式。 -``` -curl -X POST "https://dashscope.aliyuncs.com/api/v1/tokens?expire_in_seconds=1800" \ - -H "Authorization: Bearer $DASHSCOPE_API_KEY" -``` +### 不可信环境：临时 API Key -关键参数与特性： +浏览器、移动 App 中不应暴露永久 Key，应由后端调用 `POST https://dashscope.aliyuncs.com/api/v1/tokens?expire_in_seconds=` 生成临时 Key： -- `expire_in_seconds`：有效期（TTL），单位秒，范围 `[1, 1800]`，默认 60 秒。 -- 返回体中 `token`（`st-` 前缀）为临时 Key，`expires_at` 为过期 UNIX 时间戳。 -- 临时 Key 继承永久 Key 的全部权限（含模型/知识库访问限制），到期自动失效，无法提前删除。 -- 各地域（北京 / 新加坡 / 弗吉尼亚）的 Key 与 Endpoint 不互通，需配套使用。 +- `expire_in_seconds` 范围 `[1, 1800]` 秒，默认 60 秒。 +- 临时 Key 继承永久 Key 的全部权限（含模型/知识库访问限制），到期自动失效，无法手动删除。 -## 编程化管理 +### 百炼 CLI -除控制台外，百炼提供 OpenAPI（`CreateApiKey` / `GetApiKey` / `ListApiKeys` / `UpdateApiKey` / `DeleteApiKey` / `EnableApiKey` / `DisableApiKey` / `ResetApiKey`）以编程方式管理 Key。这些接口使用阿里云账号 AccessKey 签名认证，并需具备相应 RAM 权限，与业务调用所用的 API Key Bearer 鉴权不同。 +`bl` CLI 支持多种认证方式：`bl auth login --console`（浏览器登录，推荐）、`bl auth login --api-key sk-xxx`、环境变量、`bl config set --key api_key`，或单次 `--api-key` 临时传入。可用 `bl auth status --output json` 验证。 -## 开发者要点 +## 安全最佳实践 -- 优先用环境变量注入，切勿硬编码或提交到代码仓库。 -- 升级后的 `sk-ws` Key 只显示一次，创建后立即保存。 -- 明确当前场景使用的是按量付费、Token Plan、Coding Plan 中的哪种 Key，并搭配对应 Base URL。 -- 不可信环境一律走临时 API Key，生产环境的文件存储不要依赖临时 URL。 +- 使用环境变量或密钥管理服务注入，切勿将真实 Key 写入代码仓库、日志或公开渠道。 +- 面向前端的场景一律使用临时 API Key。 +- 按业务线划分子业务空间，用各自空间的 Key 实现权限隔离与分账。 +- 需要更细粒度控制时启用自定义权限（IP 白名单 + 模型范围）。 +- Key 疑似泄露时立即在控制台重置或禁用。 ## 关联主题页 - [preparations](../api/preparations.md) -- [more](../api/more.md) -- [frameworks](../api/frameworks.md) +- [get started with models](../guides/get-started-with-models.md) - [more about models](../api/more-about-models.md) +- [more](../api/more.md) +- [vector and sort](../api/vector-and-sort.md) - [token plan guide](../guides/token-plan-guide.md) -- [knowledge](../api/knowledge.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/async-invocation.md b/skills/bailian-docs-llm-wiki/wiki/concepts/async-invocation.md index 8e78a250..56e3d086 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/async-invocation.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/async-invocation.md @@ -1,52 +1,66 @@ -# 异步调用与任务轮询 +# 异步调用 -异步调用是百炼平台为耗时较长的模型任务（图像生成、视频生成、3D 生成等）设计的调用模式：客户端先提交任务拿到 `task_id`，再通过轮询或事件通知获取最终结果，从而避免长连接等待与请求超时。 +异步调用是百炼平台针对耗时较长的模型任务（如图像、视频、3D 生成）提供的一种调用机制：客户端先提交任务获得 `task_id`，再通过轮询或事件通知获取最终结果，从而避免长时间占用连接导致的请求超时。 -## 核心流程：创建任务 → 轮询获取 +## 基本流程 -异步调用统一分为两步： +异步调用统一遵循"创建任务 → 查询结果"两步： -1. **创建任务**：向对应能力的生成端点发起 `POST` 请求，请求头必须携带 `X-DashScope-Async: enable`，否则会报错 `current user api does not support synchronous calls`。请求成功后返回一个 `task_id`。 -2. **轮询查询结果**：用该 `task_id` 发起 `GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}`，读取 `output.task_status` 直到任务进入终态（`SUCCEEDED` / `FAILED`），再从结果中取回产物 URL。 +1. **创建任务**：向业务接口发起 POST 请求，HTTP 请求头中添加 `X-DashScope-Async: enable`（部分仅支持异步的接口若缺少该头会报错 `current user api does not support synchronous calls`）。响应返回 `output.task_id` 和初始状态 `PENDING`。 +2. **查询结果**：调用通用任务查询接口获取状态与产物： -任务状态的典型流转为：`PENDING`（排队中）→ `RUNNING`（处理中）→ `SUCCEEDED`（成功）/ `FAILED`（失败）。不同能力还可能出现 `SUSPENDED`（挂起）、`CANCELED`（已取消）、`UNKNOWN`（任务不存在或已过期）等状态。 +``` +GET https://dashscope.aliyuncs.com/api/v1/tasks/{task_id} +``` + +任务状态流转为： + +``` +PENDING → RUNNING → SUCCEEDED / FAILED +``` + +其他状态包括 `CANCELED`（已取消）和 `UNKNOWN`（任务不存在或已超过保留期）。建议轮询间隔 15 秒左右，查询接口默认[限流](rate-limit.md)约 20 QPS/RPS。 ## 在不同场景中的使用 -- **3D 生成（Tripo）**：仅支持异步，端点为 `.../aigc/video-generation/3d-generation`，轮询建议间隔约 15 秒，生成耗时较长，产物下载链接有效期仅 2 小时。 -- **视频生成**：所有厂商模型（万相、PixVerse、Vidu、可灵等）统一走异步，端点通常为 `.../aigc/video-generation/video-synthesis`（部分数字人/换人类模型使用 `.../aigc/image2video/video-synthesis`）。单次任务通常耗时 1-5 分钟，个别统一编辑模型约 5-10 分钟。 -- **图像生成**：多数传统模型仅支持异步（`text2image` / `image2image` 等端点），生成通常需 1-2 分钟；而新版模型（wan2.6 / wan2.7、z-image-turbo 等）走 `multimodal-generation/generation` 端点，支持 HTTP 同步一次拿结果。请勿把同步协议用在旧模型上。 -- **应用调用（智能体/工作流）**：Responses API（OpenAI 兼容）通过设置 `background=true` 开启异步，创建任务后返回任务 ID 再轮询；DashScope 应用调用接口目前暂不支持异步。 +- **图像生成**：千问图像、万相 2.6+、Z-Image 等新版模型推荐同步调用；万相 2.5 及以下、可灵（Kling）、Vidu 等图像模型采用异步模式。 +- **视频生成**：文生视频、图生视频、参考生视频、视频编辑、数字人等所有视频类 API 统一采用异步任务模式（万相、HappyHorse、PixVerse、Vidu 等系列均是如此）。 +- **3D 生成**：Tripo 系列（文生3D/单图生3D/多图生3D）仅支持异步调用，且仅限华北2（北京）地域；`task_id` 有效期 24 小时，产物 URL 有效期仅 2 小时，需及时下载。 +- **应用调用**：OpenAI 兼容的 Responses API 通过设置 `background=true` 开启异步执行，适合生成报告、多步骤工具调用等长耗时任务；DashScope 应用 API 暂不支持异步。 + +## 任务管理 API -## 关键参数与配置 +百炼提供三个通用的异步任务管理接口（均[限流](rate-limit.md) 20 QPS，主账号维度）： -- **请求头**：创建任务必须带 `X-DashScope-Async: enable`；同时需 `Authorization: Bearer $DASHSCOPE_API_KEY`、`Content-Type: application/json`。 -- **`task_id` 有效期**：一般为 **24 小时**，过期查询返回 `UNKNOWN`。请勿重复创建任务，轮询即可。 -- **`background`（应用 Responses API）**：布尔值，默认 `false`，设为 `true` 开启异步执行。 -- **查询限流**：任务查询接口默认约 20 QPS / RPS（主账号维度）。 +| 操作 | 接口 | 说明 | +| --- | --- | --- | +| 查询单个任务 | `GET /api/v1/tasks/{task_id}` | 返回 `output.task_status` 及结果；已完成任务通常保留 24 小时 | +| 批量查询 | `GET /api/v1/tasks/?start_time=...&end_time=...&status=...` | 支持按时间范围、模型、状态过滤，单次跨度不超过 24 小时 | +| 取消任务 | `POST /api/v1/tasks/{task_id}/cancel` | 仅可取消 `PENDING` 状态的任务 | -## 异步任务管理与完成通知 +## 任务完成通知（替代轮询） -百炼提供三个通用的异步任务管理接口： +频繁轮询浪费资源且易触发[限流](rate-limit.md)。百炼已接入阿里云事件总线 EventBridge，任务完成后可主动推送通知： -- **查询单个任务**：`GET .../api/v1/tasks/{task_id}`。 -- **批量查询**：`GET .../api/v1/tasks/?start_time=xxx&end_time=xxx&status=xxx`，单次时间跨度不超过 24 小时。 -- **取消任务**：`POST .../api/v1/tasks/{task_id}/cancel`，仅能取消 `PENDING` 状态的任务。 +| 方案 | 适用场景 | 特点 | +| --- | --- | --- | +| HTTP 回调 URL | 通用场景 | 需公网或 VPC 可达的 HTTP 接口，接入简单 | +| RocketMQ | 消息可靠性要求高 | 无丢失、支持失败重试，需开通 RocketMQ 实例 | -为避免频繁轮询浪费资源并触发限流，百炼已接入阿里云事件总线 EventBridge，支持任务完成后主动推送通知（HTTP 回调 URL 或 RocketMQ 两种方案）。事件源为 `acs.dashscope`，事件类型为 `dashscope:System:AsyncTaskFinish`，收到通知后只需调用一次查询接口即可获取结果。 +事件源为 `acs.dashscope`，事件类型为 `dashscope:System:AsyncTaskFinish`；收到通知后根据事件体中的 `data.task_id` 调用一次查询接口即可获取结果。 -## 开发者建议 +## 关键注意事项 -- 妥善保存 `task_id`，切勿因未及时轮询到结果而重复创建任务。 -- 轮询间隔不宜过密（如约 15 秒），高频或生产场景优先使用异步完成通知。 -- 及时下载产物：图像/视频 URL 有效期通常为 24 小时，3D 模型下载链接有效期仅 2 小时。 -- 处理失败时读取响应中的 `code` 与 `message`，对照百炼错误码文档排查。 +- 创建任务后请保存 `task_id`，切勿因未及时收到结果而重复创建任务。 +- 任务及产物有保留期限：任务记录一般保留 24 小时，生成的文件 URL（如 3D 模型、渲染图）可能仅 2 小时有效，应尽快下载转存。 +- 部分接口需替换 Endpoint 中的 `{WorkspaceId}`，并使用对应地域的 API Key。 +- 控制轮询频率，或优先采用 EventBridge 通知方案，避免触发限流。 ## 关联主题页 -- [3d generation](../api/3d-generation.md) -- [video generation api](../api/video-generation-api.md) - [image generation](../api/image-generation.md) +- [video generation api](../api/video-generation-api.md) +- [3d generation](../api/3d-generation.md) - [more about models](../api/more-about-models.md) - [application call](../api/application-call.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/billing.md b/skills/bailian-docs-llm-wiki/wiki/concepts/billing.md index 6b73c92e..2ba2cea8 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/billing.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/billing.md @@ -1,85 +1,88 @@ # 计费 -计费是百炼平台围绕模型推理、模型训练、模型部署及增值服务所建立的费用体系，涵盖按量付费、订阅制（[Token](token.md) Plan / Coding Plan）、预留容量（TPM 预留 / PTU）等多种方式，开发者可根据用量规模和业务场景灵活组合，实现成本最优。 +计费是阿里云百炼平台对模型调用、训练、部署及附加功能进行费用计量与结算的机制，默认采用按量后付费（分钟级出账、按月结算），同时提供预付费的节省计划、资源包和订阅套餐等多种优惠方案。 -## 计费场景与方式 +## 费用构成 -百炼平台的计费覆盖以下主要场景： +百炼的模型相关费用分为以下几类： -| 场景 | 计费方式 | 说明 | -|------|---------|------| -| 模型推理调用 | 按量付费（[Token](token.md) 计价） | 按输入/输出 [Token](token.md) 分别计价，部分模型支持阶梯计费 | -| 模型训练 | 按训练 Token 计费 | （训练数据 Token + 混合训练数据 Token）× 循环次数 × 单价 | -| 模型部署 | PTU / 模型单元 / Token 用量 | 三种方式创建后不可更改，需下线重建才能切换 | -| 知识库 | 按规格计费 | 标准版 0.03 元/小时，旗舰版 0.2 元/RCU/小时 | -| TPM 预留 | 按 kTPM 预付费 | 按天计费，输入/输出分别定价 | -| Token Plan 团队版 | 坐席订阅制 | 按 Credits 抵扣，198 元/坐席/月起 | -| Coding Plan | 月度订阅制 | 按模型调用次数限额，200 元/月 | +| 费用类型 | 计费方式 | 说明 | +| --- | --- | --- | +| 模型调用（实时推理） | 按输入/输出 Token 计费 | 默认按量付费，部分模型支持阶梯计价、上下文缓存折扣、Batch 半价 | +| 模型训练 | 按训练 Token 计费 | 千问文本、万相图像/视频等均有独立单价 | +| 模型部署 | 按使用时长计费 | 预置吞吐（TPM）或模型单元两种模式，支持后付费与预付费 | +| 知识库规格 | 按运行时长计费 | 标准版 0.03 元/知识库/小时，旗舰版 0.2 元/RCU/小时 | +| 附加功能 | 按次计费 | 联网搜索、MCP 广场等独立计费，不在节省计划抵扣范围 | -## 费用抵扣顺序 +抵扣优先级为：**免费额度 > 资源包 > 其他模型节省计划 > AI 通用型节省计划 > 按量付费**。 -多种付费方式并存时，系统按以下优先级自动抵扣： +## 不同场景下的计费 -**免费额度 → 资源包 → 其他模型节省计划 → AI 通用型节省计划 → 按量付费** +### 模型调用 -Token Plan 和 Coding Plan 的套餐额度独立于按量计费体系，不参与上述抵扣链路。 +模型调用默认按量计费，按输入 Token 和输出 Token 分别计价，单价为每百万 Token 的人民币金额。部分模型实行阶梯计费：单次请求的输入 Token 总量决定所属阶梯，该请求所有 Token 均按对应阶梯单价结算。 -## 免费额度 +折扣机制： -首次开通百炼时自动发放新人免费额度，有效期 30~90 天。关键限制： +- **Batch 调用**：支持 Batch 的模型，输入和输出 Token 单价均按实时推理价格的 50% 计费。 +- **上下文缓存**：命中缓存的输入 Token 享有折扣（如显式缓存命中按标准输入的 10%），显式缓存创建可能按 125% 计费。Batch 半价与缓存折扣不能同时生效。 +- **限时活动**：部分模型有限时折扣。 -- 仅适用于华北2（北京）地域、中国内地服务部署范围的实时推理 -- 不支持抵扣 Batch 调用、模型调优、模型部署、自定义模型 -- 主账号与 RAM 子账号共享 -- 建议开启"免费额度用完即停"功能，防止额度耗尽后自动扣费 +跨地域价格差异较大，例如 qwen3-max 在新加坡的输入单价远高于中国内地。 -## 成本优化方案 +### 模型训练 -### AI 通用型节省计划 +按训练 Token 计费，公式因模型类型而异： -承诺每月消费金额获取阶梯折扣，最高可享 5.3 折。覆盖阿里直供全部模型，承诺金额 1,000 元起，支持 3/6/12/24 个月周期。当月未用完的额度自动清零，不可累积。 +- **千问文本模型**：费用 =（训练数据 Token 总数 + 混合训练数据 Token 总数）× 循环次数 × 训练单价 +- **万相图像生成**：费用 = 训练 Token 总量 × 训练单价 +- **万相视频生成**：费用 = 训练 Token 总量 × 训练单价（Token 总量与视频计费时长、像素、轮次相关） -### 资源包 +### 模型部署 -预先购买具体 Token 数量，用于抵扣特定模型的实时推理用量。适合用量明确且集中在单一模型的场景。 +部署提供三种计费方式： -### Batch 调用优惠 +| 计费方式 | 适用场景 | 特点 | +| --- | --- | --- | +| 预置吞吐（PTU） | 高负载生产环境，流量稳定可预估 | 按使用时长 × TPM 单价计费，保障额度内不限速 | +| 模型单元 | 大规模推理业务，需自定义性能 | 按使用时长 × 模型单元数量 × 单价计费，资源独占 | +| 按 Token 用量 | 效果验证，对并发和延迟要求不高 | 按输入/输出 Token 计费，不使用不计费 | -支持 Batch 调用的模型，输入和输出 Token 单价均按实时推理价格的 50% 计费。 +部署计费的关键规则： -### 上下文缓存折扣 +- 计费方式在服务创建后无法更改，切换需先下线再重新部署。 +- 部署成功后即开始计费，即使未发起任何调用。 +- PTU 部署支持长输入阶梯系数和前缀缓存折扣；超出购买 TPM 时按溢出策略处理（自动溢出转按量付费，或仅使用 PTU 容量返回 429）。 +- 后付费部署在账户欠费后仍保留资源并计费 24 小时，超 24 小时停止计费但底层资源将被删除。 -部分模型支持上下文缓存，命中缓存的输入 Token 按折扣系数消耗额度（例如 deepseek-v4-pro 为 0.08，即按 8% 折算）。Batch 调用与上下文缓存不能同时生效。 +### TPM 预留 -## 多地域定价差异 +TPM 预留为指定模型锁定专属推理吞吐量，按输入 TPM 和输出 TPM 分别计价，单位为「每 10,000 TPM 的日预付费」。预留容量内的调用不受公共资源限流影响，部分模型支持长输入阶梯系数和缓存折扣。溢出策略可选自动溢出按量或仅预留容量返回 429。 -同一模型在不同地域的价格可能不同。华北2（北京）使用人民币定价，新加坡等国际地域按国际价格结算，通常高于国内价格。[API Key](api-key.md) 必须与 Base URL 同一地域，否则会报 `401` 错误。 +### 订阅套餐 -## 账单查询 +Token Plan 是以 Credits 统一计量的订阅服务，分为个人版和团队版： -- **费用概览**：控制台"用量 & 费用 > 费用概览"查看当月总消费，支持按模型或 [API Key](api-key.md) 筛选 -- **模型用量**：按[业务空间](workspace.md)维度统计，数据延迟约 1 小时，不支持查看 30 天以前的数据 -- **出账时间**：大模型推理分钟级出账（2~10 分钟），批量推理和训练小时级出账 -- **分账管理**：通过[业务空间](workspace.md)标签按部门或项目归集费用，T+1 天生效 +- **个人版**：三个档位（Lite/Standard/Pro），采用 5 小时和 7 天两层滑动窗口限额，额度用完后调用被阻断，不会转为按量计费。 +- **团队版**：按月度总额度制，无滑动窗口限额，另可购买共享用量包作为跨坐席弹性补充。 +- **Coding Plan**：独立的订阅产品，按模型调用次数计费（非 Credits），与 Token Plan 之间无法迁移或升级。 -## 欠费与停止计费 +Token Plan 使用以 `sk-sp-` 开头的专属 API Key，与百炼通用 API Key（`sk-` 开头）格式不同，两者的 API Key 和 Base URL 完全隔离，必须配套使用，不可混用。 -账户可用额度小于 0 时视为欠费。Token Plan 和 Coding Plan 的套餐额度独立于账户余额，欠费期间可继续使用。停止计费的方式： +## 关键参数与配置 -- **模型推理**：停止 API 调用，删除不再使用的 [API Key](api-key.md) -- **模型部署**：在控制台下线已部署模型；包月预付费需额外退订实例 -- **模型训练**：无进行中的训练任务即不产生费用 -- **订阅制**：关闭自动续费，到期自动停止 +- **开通条件**：阿里云账户余额不小于 0 元；服务需分地域开通，开通后暂不支持关闭。 +- **免费额度**：首次开通百炼时自动发放（通常 100 万 Token），有效期 90 天，仅抵扣实时推理费用。仅华北2（北京）地域且服务部署范围为中国内地的模型享有免费额度。免费额度用完即停，可在控制台逐模型开启。 +- **节省计划**：AI 通用型节省计划承诺在 3/6/12/24 个月内每月消费一定金额（1000 元起），享受阶梯式折扣，最高 5.3 折，覆盖模型调用、上下文缓存、批量推理等，但不抵扣模型调优、模型部署、联网搜索等。 +- **账单查询**：费用明细与发票申请通过阿里云费用与成本控制台操作。 +- **万相会员**与百炼 API 计费体系相互独立，会员权益不适用于 API 调用。 ## 关联主题页 - [test 1](../guides/test-1.md) - [token plan guide](../guides/token-plan-guide.md) -- [model deployment 1](../guides/model-deployment-1.md) -- [model monitoring](../guides/model-monitoring.md) - [support](../guides/support.md) -- [knowledge base](../guides/knowledge-base.md) -- [use chat client or development tool](../guides/use-chat-client-or-development-tool.md) +- [model deployment 1](../guides/model-deployment-1.md) - [model high speed inference](../guides/model-high-speed-inference.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/dashscope-sdk.md b/skills/bailian-docs-llm-wiki/wiki/concepts/dashscope-sdk.md index de46d3bd..f820a7ed 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/dashscope-sdk.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/dashscope-sdk.md @@ -1,67 +1,36 @@ # DashScope SDK -DashScope SDK 是阿里云百炼平台官方提供的软件开发工具包，封装了模型与应用调用的原生（DashScope）接口，让开发者用少量代码即可接入通义千问、万相、Qwen-MT、Qwen-OCR 等模型能力以及智能体/工作流应用。相比 HTTP 直连和 [OpenAI 兼容接口](openai-compatible-interface.md)，DashScope 原生接口暴露的参数最完整、功能集最丰富。 +DashScope SDK 是阿里云百炼平台的原生官方 SDK，提供 Python 和 Java 两种语言实现，用于调用百炼平台的全部模型能力与应用调用接口。它是百炼功能覆盖最完整的接入方式，当 [OpenAI 兼容接口](openai-compatible-interface.md)无法满足需求时（如特定模型仅支持 DashScope 协议），DashScope SDK 是唯一选择。 -## 适用语言与安装 +## 安装 -DashScope SDK 主要提供 Python 与 Java 两种官方实现，部分场景也可用 HTTP（如 Node.js 借助 `axios`）替代： +| 语言 | 安装方式 | 版本要求 | +|------|----------|----------| +| Python | `pip install -U dashscope` | Python >= 3.8 | +| Java | Maven/Gradle 添加 `com.alibaba:dashscope-sdk-java` | 建议 >= 2.12.0 | -- **Python**：`python3 -m pip install -U dashscope` -- **Java**：通过 Maven / Gradle 引入 `com.alibaba:dashscope-sdk-java`，建议版本 `>= 2.12.0` -- **Node.js / 其他语言**：目前无官方 SDK，直接走 HTTP API（发起 POST 请求） +> **注意**：Node.js 没有官方 DashScope SDK，Node.js 开发者可通过 [OpenAI 兼容接口](openai-compatible-interface.md)或 HTTP 直接调用。 -> 注意：不同能力对 SDK 语言与地域的支持存在差异。例如 `qwen-deep-research` **仅支持 Python DashScope SDK，且仅限华北2（北京）地域**，暂不支持 Java SDK 与 [OpenAI 兼容接口](openai-compatible-interface.md)。 +## 认证配置 -## 在不同场景中的使用 +SDK 通过 [API Key](api-key.md) 鉴权，推荐将 [API Key](api-key.md) 配置到环境变量 `DASHSCOPE_API_KEY`，SDK 会自动读取该变量，避免在代码中硬编码： -### 1. 调用文本生成模型（Qwen 系列） - -Qwen 系列可通过 OpenAI 兼容、Anthropic 兼容或 DashScope 原生三类接口调用。其中 DashScope 是百炼原生接口，**功能集最完整、参数支持最丰富**；当需要使用最全的采样参数、插件或业务字段而兼容接口未暴露时，应改用 DashScope 原生接口。 - -### 2. 调用专用模型（[more](../api/more.md) models） - -法律、意图理解、翻译、OCR 等专用模型大多支持 OpenAI 兼容或 DashScope 两种方式调用： - -- `farui-plus`（法律大模型）：通过 DashScope SDK（Python / Java）调用 -- `tongyi-intent-detect-v3` / `qwen-mt-plus` / `qwen3.5-ocr`：OpenAI 兼容或 DashScope 均可 -- `qwen-deep-research`（深度研究）：仅 Python DashScope SDK - -### 3. 调用图像生成与编辑模型 - -千问-图像（Qwen-Image）、万相（Wan/Wanx）、Z-Image 等图像模型均可通过 HTTP 或 DashScope SDK 调用，覆盖文生图、图像编辑、图像翻译、风格迁移等能力。 - -### 4. 调用智能体应用与工作流应用 - -已创建并发布的智能体应用、工作流应用可通过 DashScope SDK 集成到业务系统，二者调用方式一致： - -- Python：`from dashscope import Application`，调用 `Application.call(...)` -- Java：构造 `ApplicationParam` 后调用 `application.call(param)` -- HTTP 等价接口：`POST https://dashscope.aliyuncs.com/api/v1/apps/{APP_ID}/completion` - -响应统一为 `{"output": {...}, "usage": {...}, "request_id": "..."}` 结构，业务侧主要消费 `output.text`。 - -## 关键参数与配置 - -### 鉴权（API Key） +```bash +# Linux / macOS +export DASHSCOPE_API_KEY="YOUR_DASHSCOPE_API_KEY" +``` -- SDK 通过 [API Key 鉴权](api-key.md)，**推荐将密钥写入环境变量 `DASHSCOPE_API_KEY`**，SDK 会自动读取，避免在代码中硬编码。 -- 调用特定地域（如华北2/北京）或子业务空间下的模型/应用时，需使用对应地域的 API Key，并按需提供 Workspace ID。 +也可在调用时通过 `api_key` 参数显式传入。 -### 通用请求参数 +## 使用场景 -- `model`（string，必选）：目标模型名称。 -- `messages`（array）：对话消息列表，按顺序排列，需由调用方维护上下文。 -- `app_id`（应用调用）：目标应用 ID，从控制台应用卡片复制。 -- `prompt` / `input`：用户输入内容。 -- `stream`（bool，可选）：是否[流式输出](streaming.md)；如 `qwen-deep-research` 的反问阶段需设为 `true`。 -- `session_id`（应用多轮对话）：由云端维护上下文，免去手动拼接历史。 +### 调用文本生成模型 -### 模型专属参数示例 +DashScope 原生接口提供百炼平台最完整的参数集和功能支持。适合需要使用平台全部能力的场景，如特殊参数配置、平台独有功能等。若仅需标准对话能力且已有 OpenAI 生态代码，可优先考虑 [OpenAI 兼容接口](openai-compatible-interface.md)以降低迁移成本。 -- **Qwen-MT（翻译）**：通过 `translation_options`（OpenAI SDK 中放入 `extra_body`）控制 `source_lang`、`target_lang`、`terms`（术语干预）、`tm_list`（翻译记忆）、`domain_prompt`（领域提示）。 -- **Qwen-OCR**：`messages.content` 为[多模态](multimodal.md)数组，可设 `min_pixels` / `max_pixels` 控制图像像素阈值。 +### 调用百炼应用（智能体 / 工作流） -## Python 快速示例（应用调用） +通过 `Application.call`（Python）或 `ApplicationParam` + `Application`（Java）调用已发布的智能体应用和工作流应用： ```python import os @@ -80,20 +49,74 @@ else: print(response.output.text) ``` -## 使用建议 +```java +import com.alibaba.dashscope.app.*; + +ApplicationParam param = ApplicationParam.builder() + .apiKey(System.getenv("DASHSCOPE_API_KEY")) + .appId("YOUR_APP_ID") + .prompt("你是谁？") + .build(); + +Application application = new Application(); +ApplicationResult result = application.call(param); +System.out.printf("text: %s%n", result.getOutput().getText()); +``` + +对应的 HTTP 端点为 `POST https://dashscope.aliyuncs.com/api/v1/apps/{APP_ID}/completion`。 -- 优先使用环境变量管理 API Key，区分不同地域的密钥。 -- 需要最完整功能与参数时选 DashScope 原生接口；追求生态兼容、迁移成本最低时可选 [OpenAI 兼容接口](openai-compatible-interface.md)。 -- 接入前先对照具体模型的 API 参考，确认其支持的 SDK 语言、协议与地域。 +### 调用仅支持 DashScope 的专用模型 + +部分模型只能通过 DashScope SDK 调用，不支持 [OpenAI 兼容接口](openai-compatible-interface.md)： + +- `farui-plus`（法律大模型）：支持 Python/Java DashScope SDK。 +- `qwen-deep-research`（深度研究）：**仅支持 Python DashScope SDK**，不支持 Java SDK 和 [OpenAI 兼容接口](openai-compatible-interface.md)，且仅限华北2（北京）地域。 + +其他模型如 `tongyi-intent-detect-v3`、`qwen-mt-plus`、`qwen3.5-ocr` 同时支持 OpenAI 兼容和 DashScope 两种调用方式。 + +## 关键参数 + +### 模型调用通用参数 + +- `model`（string，必选）：模型名称。 +- `messages`（array，必选）：对话消息列表。 +- `stream`（bool，可选）：是否启用[流式输出](streaming.md)。 + +### 应用调用参数 + +- `app_id`（string，必选）：应用 ID，从控制台应用卡片获取。 +- `prompt`（string）：用户输入。 +- `session_id`（string，可选）：多轮对话时传入上一次响应返回的 session_id 以维护上下文。 +- `biz_params`（dict，可选）：自定义参数，透传到应用内部。 + +### 响应结构 + +应用调用响应统一为 `{"output": {"finish_reason", "session_id", "text"}, "usage": {...}, "request_id": "..."}` 结构，业务侧主要消费 `output.text`。 + +## 与 [OpenAI 兼容接口](openai-compatible-interface.md)的选择建议 + +| 场景 | 推荐方式 | +|------|----------| +| 已有 OpenAI 生态代码，追求最低迁移成本 | OpenAI 兼容接口 | +| 需要平台最完整的参数和功能 | DashScope SDK | +| 调用百炼应用（智能体/工作流） | DashScope SDK 或 Responses API | +| 模型仅支持 DashScope 协议（如 qwen-deep-research） | DashScope SDK（唯一选择） | +| Node.js / Go 项目 | OpenAI 兼容接口或 HTTP 调用 | + +## 注意事项 + +- DashScope 域名（`dashscope.aliyuncs.com`）为存量兼容域名，请求超时 600 秒；生产环境建议迁移至[业务空间](workspace.md)专属域名（`{WorkspaceId}.{region}.maas.aliyuncs.com`），超时 3600 秒且提供 99.9% SLA。 +- [API Key](api-key.md) 必须与 Base URL 所属地域匹配，不同地域的 Key 和端点不能混用，否则报 401 错误。 +- 跨接口迁移时（DashScope 与 OpenAI 兼容之间），需核对参数映射差异，两者参数集合并不完全一致。 ## 关联主题页 -- [image generation](../api/image-generation.md) +- [qwen api reference](../api/qwen-api-reference.md) - [more models](../api/more-models.md) - [bailian application calling](../guides/bailian-application-calling.md) -- [qwen api reference](../api/qwen-api-reference.md) +- [get started with models](../guides/get-started-with-models.md) +- [preparations](../api/preparations.md) - [application call](../api/application-call.md) - diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/embedding.md b/skills/bailian-docs-llm-wiki/wiki/concepts/embedding.md index 1334d698..98b985aa 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/embedding.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/embedding.md @@ -1,54 +1,72 @@ -# 向量嵌入 +# 向量化 -向量嵌入（Embedding）是将文本、图像、视频等非结构化数据转换为固定维度的数值向量，使语义相近的内容在向量空间中距离也相近。百炼平台提供多类嵌入模型，支撑语义检索、RAG 召回、跨模态搜索、聚类推荐等下游任务。 +向量化（Embedding）是将文本、图片、视频等非结构化数据通过模型映射为高维数值向量的过程。在百炼平台中，向量化是语义检索、推荐、聚类、分类与 RAG 检索的核心基础——向量模型将不同模态的数据编码到同一语义空间，使机器能够基于向量距离衡量语义相似度。 -## 在百炼平台的使用场景 +## 在百炼平台中的使用场景 -- **RAG 知识库召回**：知识库创建时选择向量模型，对导入文档切片做嵌入入库；查询时对 Query 做同款嵌入，再与切片向量做相似度匹配，作为 RAG 流程的第一段召回。文档搜索、数据查询、音视频搜索类知识库支持 `text-embedding-v4` 或 `text-embedding-v3`（均为 512 维，维度不可更改）；图片问答类固定使用 `multimodal-embedding-v1`（1024 维）；视觉理解场景自动切换为 `qwen3-vl-embedding`。 -- **本地 RAG 应用**：本地知识库方案默认调用百炼 embedding API 生成向量，也可替换为本地部署的 GTE 文本向量模型（如 `iic/nlp_gte_sentence-embedding_chinese-large`）。受 embedding API 限流影响，单文件不建议超过 100 MB。 -- **跨模态检索**：多模态向量模型（如 `qwen3-vl-embedding`、`multimodal-embedding-v1`）将文本、图像、视频映射到同一语义空间，支持以文搜图、以图搜视频等。支持「独立向量」（逐项生成）与「融合向量」（多输入合并为 1 个向量）两种模式。 -- **框架集成**：LlamaIndex 路线将知识库部署在百炼云端，使用官方向量模型与智能切分，不支持自定义嵌入模型；如需灵活选择嵌入模型，应改用本地知识库方案。 +### 向量与排序 API -## 模型与关键参数 +百炼提供通用文本向量和[多模态](multimodal.md)向量两类接口，用于将原始数据转为向量： -通用文本向量模型当前推荐 `text-embedding-v4`（Qwen3-Embedding 系列，支持 100+ 语种）。 +- **通用文本向量（同步）**：支持 `qwen3.7-text-embedding`、`text-embedding-v4/v3/v2/v1`，将字符串或文件实时转为向量。提供 [OpenAI 兼容接口](openai-compatible-interface.md)，可用 OpenAI SDK 直连。 +- **通用文本向量（批处理）**：面向大规模离线场景，仅支持异步模式，通过文件 URL 输入，支持 `text-embedding-async-v2/v1`。单文件最多 100,000 行、200MB。 +- **[多模态](multimodal.md)向量**：将文本、图片、视频编码到同一语义空间，支持跨模态检索。模型包括 `qwen3-vl-embedding`、`qwen2.5-vl-embedding`、`multimodal-embedding-v1` 等。 -| 模型 | 向量维度 | 最大行数 | 单行最大 Token | 语种 | -|------|---------|---------|---------------|------| -| text-embedding-v4 | 2048/1536/1024(默认)/768/512/256/128/64 | 10 | 8,192 | 100+ 语种 | -| text-embedding-v3 | 1024(默认)/768/512/256/128/64 | 10 | 8,192 | 50+ 语种 | -| text-embedding-v2 | 1,536 | 25 | 2,048 | 10 语种 | -| text-embedding-v1 | 1,536 | 25 | 2,048 | 6 语种 | +### 知识库索引与检索 -请求参数： +知识库在导入文档时自动调用向量模型对文本切片进行向量化，构建语义索引。检索阶段，用户查询同样经过向量化后与索引向量做相似度匹配，完成语义召回。 -- `model`（必选）：模型名称。 -- `input`（必选）：字符串、字符串列表或文件。 -- `dimensions`（可选）：指定向量维度，仅 v3/v4 支持。 -- `encoding_format`（可选）：当前仅支持 `float`。 +- **文档搜索**：支持 `text-embedding-v4` 和 `text-embedding-v3`（512 维）。 +- **视觉理解（富文本文档）**：自动切换为 `qwen3-vl-embedding`，对 PDF、图片等进行视觉级理解和索引，保留版面信息。 +- **图片问答**：使用 `multimodal-embedding-v1`（1024 维）。 -调用方式支持 OpenAI 兼容接口（base_url：`https://dashscope.aliyuncs.com/compatible-mode/v1`）和 DashScope SDK。 +### 知识检索与问答 API -## 批处理接口 +知识检索接口（`POST /api/v1/indices/knowledge/search`）跨多个知识库执行联合语义检索，底层依赖向量化的索引与查询。知识问答接口则在此基础上自动完成规划、检索、生成全流程。 -大规模文本向量化可使用异步批处理接口（`text-embedding-async-v1/v2`），单次最多 10 万行文本。需在 HTTP 请求头加 `X-DashScope-Async: enable` 启用异步模式，提交后通过 `task_id` 轮询结果。同时处理中任务不超过 50 个，并发运行上限 3 个，超出部分排队等待。 +## 关键参数与配置 -## 与 Rerank 的关系 +### 维度（dimensions） -向量嵌入负责「召回」，Rerank 模型负责「精排」。在知识库检索流程中，先由向量 + 关键词混合检索召回 TopK 切片，再由 `qwen3-rerank`、`qwen3-vl-rerank` 等排序模型对候选切片二次排序，最终按相似度阈值与最大召回数量返回。两者配合提升 RAG 命中准确率。 +仅部分模型支持自定义输出维度： -## 注意事项 +| 模型 | 支持维度 | 默认 | +| --- | --- | --- | +| `qwen3.7-text-embedding` | 2560/2048/1536/1024/768/512/256/128/64 | — | +| `text-embedding-v4` | 2048/1536/1024/768/512/256/128/64 | 1024 | +| `text-embedding-v3` | 1024/768/512/256/128/64 | 1024 | +| `text-embedding-v1/v2` | 固定 1536（不可调） | 1536 | -- 知识库向量模型与维度在创建时选定，**维度不可更改**；Meta 抽取、多轮对话改写等索引配置在创建后也无法追加，需重建知识库。 -- 嵌入与 Rerank 是两类不同模型，不要混用接口：qwen3-rerank 走 `/compatible-api/v1/reranks`，qwen3-vl-rerank / gte-rerank-v2 走 `/api/v1/services/rerank/text-rerank/text-rerank`。 -- `gte-rerank` 系列将于 2026 年 5 月 30 日下线，建议迁移到 `qwen3-rerank`。 +> 知识库场景下 `text-embedding-v3/v4` 统一使用 512 维。 + +### 输入上限 + +| 模型 | 单条最大 Token | 列表/文件最大条数 | +| --- | --- | --- | +| `qwen3.7-text-embedding` | 128,000 | 20 | +| `text-embedding-v3/v4` | 8,192 | 10 | +| `text-embedding-v1/v2` | 2,048 | 25 | + +### text_type（批处理） + +批处理接口的 `parameters.text_type` 可选 `document`（默认）或 `query`。检索类非对称任务建议区分 query 与 document，以提升检索效果。 + +### encoding_format + +同步接口可选 `encoding_format`，当前仅支持 `float`。 + +## 使用建议 + +- **检索场景**优先选 `text-embedding-v4` 或 `qwen3.7-text-embedding`，维度更高、效果更好。 +- **跨模态场景**（图文检索）使用 `qwen3-vl-embedding` 或 `multimodal-embedding-v1`。 +- **大规模离线**用批处理接口，注意请求头须带 `X-DashScope-Async: enable`，否则会报错。 +- **知识库**中向量模型由知识库类型决定，创建时选定后不可更改，务必提前规划。 +- 向量检索召回后通常配合 rerank 模型（如 `qwen3-rerank`）做二次精排，进一步提升相关性。 ## 关联主题页 - [vector and sort](../api/vector-and-sort.md) - [knowledge base](../guides/knowledge-base.md) - [knowledge](../api/knowledge.md) -- [frameworks](../api/frameworks.md) -- [application use cases](../guides/application-use-cases.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/evaluation.md b/skills/bailian-docs-llm-wiki/wiki/concepts/evaluation.md index a303b8d6..3aed4a8f 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/evaluation.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/evaluation.md @@ -1,77 +1,83 @@ -# 评测体系 +# 评测 -评测体系是百炼平台用于系统化衡量输出质量的一整套能力，覆盖**应用评测**（智能体/工作流应用）与**模型评测**（文本生成模型）两大场景，通过评测集、评测维度/评估器、评分方式与评测报告构成完整的评估闭环。 +评测是百炼平台对模型推理结果和应用输出质量进行系统化评估的能力，通过评测集、评分规则和报告三大要素，帮助开发者完成模型选型、调优效果验证和应用质量迭代。平台提供两套评测体系：**应用评测**（面向智能体应用与工作流应用）和**模型评测**（面向文本生成类模型），两者共用数据管理模块中的评测集基础设施。 -## 两类评测场景 +## 应用场景 -百炼的评测能力按被评测对象分为两条相对独立的链路： +### 应用评测 -- **应用评测**：评估已发布智能体应用、工作流应用的回答质量，支持自动评测（大模型基于知识库生成评测集并打分）和手动评测（人工标注打分）。当前存在新旧两套系统，新版以「评测任务 + 评估器 + 标签」组织。 -- **模型评测**：评估文本生成类模型的能力，支持自定义评测（自有数据集 + 自定义维度）和基线评测（C-Eval、MMLU、GSM8K、BBH 等公开数据集）。 +对已发布的智能体应用或工作流应用的输出质量进行评估，支持两种模式： -两者共享一套核心思路：准备评测数据 → 定义评分规则 → 创建评测任务 → 查看报告。 +- **自动评测**：基于应用关联的知识库自动生成评测集，由大模型（`qwen-max` / `qwen-plus`）自动评分并生成报告与调优建议。支持单应用深度评估和多应用横向对比（最多 8 个），适用于选型决策和版本迭代验证。 +- **手动评测**：人工构建评测集并逐条标注打分，适用于需要领域专家主观判断的场景。新版系统通过「评测任务 + 标签」组合替代旧版手动评测流程。 -## 评分方式 +前提条件：应用须已发布且配置知识库；须开通应用观测功能并将应用加入观测列表；多应用横向评测要求各应用至少关联一个相同知识库。 -无论应用还是模型评测，评分方式基本归为三类，选型取决于是否有标准答案与是否需要语义理解： +### 模型评测 -| 评分方式 | 原理 | 适用场景 | 费用 | -|----------|------|----------|------| -| 大模型评估（LLM/AI 自动评测） | 裁判模型按 Prompt 语义打分或分类 | 相关性、幻觉、内容安全等语义场景 | 产生 Token/裁判模型费用 | -| 规则评估（Code/自动化指标） | ROUGE、BLEU、Cosine、字符串匹配等算法 | 翻译、摘要、格式校验、精确匹配等确定性场景 | 无额外费用 | -| 人工评估 | 人工逐条标注 Pass/Fail 或打分 | 创意写作、专业领域主观判断 | 无裁判费用 | +对文本生成类模型的推理结果打分和对比，支持两种方式： -**选型路径**：有标准答案且格式固定 → 字符串匹配；有标准答案但表述多样 → 文本相似度；无标准答案需语义理解 → 大模型评估；需主观判断 → 人工评估。 +- **自定义评测**：使用自有数据集和自定义评测维度，评分方式包括大模型评估（裁判模型语义打分）、规则评估（ROUGE / BLEU / Cosine 等算法指标）和人工评估（逐条 Pass/Fail 标注）。 +- **基线评测**：使用公开标准数据集（C-Eval、MMLU、GSM8K、BBH 等）快速评估模型基础能力，无需自备数据。仅北京地域可用。 -## 评测集（数据基础） +模型评测常用于：调优后模型效果验证（对比基座模型与调优模型）、多模型选型对比、以及排行榜提交。 -评测集是评测任务的数据输入，需在数据管理模块或评测系统中准备： +### 在调优流程中的位置 -- **应用评测集**：旧版分对话分析（`.xls/.xlsx`）与知识问答（`.jsonl`）；新版分智能体、工作流、自定义三类，具备版本管理，创建后类型不可改。 -- **模型评测集**：单轮对话，Excel 格式，每行含 **Prompt**（用户输入）与 **Completion**（期望输出）。参评模型对每条 Prompt 推理，评分参考 Completion。 +自定义模型调优的标准流程为：模型调优 → 模型部署 → **模型评测**。评测不满意时可调整训练策略（数据质量、超参数、训练方式）重新迭代。已部署调优模型的评测不额外计费（推理费用包含在部署算力费用中）。 -数据量建议：小规模验证 50-100 条，正式评测 200-500 条，全面评估 500 条以上。 +## 核心组件 -## 评测维度与评估器（评分规则） +### 评测集 -评分规则是评测体系的核心组件，创建为模板后可被多个任务复用： +评测的数据基础，存储输入（Prompt/Query）与期望输出。 -- **模型评测的评测维度**提供五种评分器类型：大模型评估-数值型、大模型评估-分类型、规则评估-文本相似度、规则评估-字符串匹配、人工评估-分类型。维度类型创建后不可修改。 -- **应用评测的评估器**支持预置模板（通用质量、智能体、文本匹配、文本相似度、格式校验）、自定义（LLM 评估器 / Code 评估器）以及基于历史评测任务标注结果抽象生成。每个评测任务最多添加 10 个评估器，建议组合 3-5 个从不同维度评估。 +| 体系 | 类型 | 格式要求 | +|------|------|----------| +| 应用评测（旧版） | 对话分析 | `.xls` / `.xlsx`，含 Prompt、Completion、SessionId | +| 应用评测（旧版） | 知识问答 | `.jsonl`，含 query、referenceAnswer、关键词字段 | +| 应用评测（新版） | 智能体 / 工作流 / 自定义 | 按应用出入参定义表结构，支持版本管理 | +| 模型评测 | 文本生成评测集 | Excel，含 Prompt 和 Completion 两列 | -## 关键参数与配置 +评测集可通过手动上传、基于知识库自动生成、从应用观测导入、或日志回流（SLS 推理日志转结构化数据）四种方式创建。数据量建议：小规模验证 50-100 条，正式评测 200-500 条，全面评估 500 条以上。 -- **评分范围**（数值型）：整数打分区间，默认 0-5，建议不超过 10，范围过大会降低 LLM 评分一致性。 -- **通过阈值**：判定 Pass/Fail 的分界线，数值型步长 0.1，相似度型步长 0.01。 -- **评分器 Prompt**（大模型评估）：至少含一个变量（`${prompt}`、`${output}`、`${completion}`），长度不超过 50000 字符；用于指导裁判模型如何打分。 -- **System Prompt**：配置于评测任务，为被评测模型设定角色/行为规范，通常可留空，注意勿与评分器 Prompt 混淆。 -- **数据来源**：评测数据集（含 Prompt+Completion，产生推理费用）或推理结果集（已含 Output，不产生推理费用）。 -- **推理参数**：Temperature、TopP 等，按所选模型动态加载。 -- **标签类型**（应用评测）：分类、布尔值、数字、文本，用于标注评测与观测数据。 +### 评估器与评测维度 -## 计费说明 +定义评分规则的组件： -费用主要由**被评测模型推理费用**与**裁判模型评分费用**两部分构成： +- **应用评测评估器**（新版）：支持预置模板（通用质量、智能体、文本匹配、文本相似度、格式校验）和自定义两种。自定义分 LLM 评估器（语义理解，产生 Token 费用）和 Code 评估器（Python 规则判断，无额外费用）。每个评测任务最多 10 个评估器，建议组合 3-5 个。 +- **模型评测维度**：五种评分器类型——大模型评估-数值型（0-5 打分）、大模型评估-分类型（Pass/Fail）、规则评估-文本相似度、规则评估-字符串匹配、人工评估-分类型。维度类型创建后不可修改。 -- 使用推理结果集可免去推理费用。 -- 规则评估、人工评估无裁判模型费用。 -- 已部署的调优模型评测不额外计费（推理费用包含在部署算力费用中）。 +选型决策路径：有标准答案且格式固定 → 字符串匹配；有标准答案但表述多样 → 文本相似度；无标准答案需语义理解 → 大模型评估；需主观判断 → 人工评估。 -成本优化建议：先用 50-100 条小规模验证 → 保存推理结果集复用 → 确定性场景优先用规则评估。 +## 关键参数 -## 限制与注意事项 +| 参数 | 说明 | +|------|------| +| 评分范围（数值型） | 裁判模型打分区间，默认 0-5，建议不超过 10 | +| 通过阈值 | 判定 Pass/Fail 的分界线，步长 0.1（数值型）或 0.01（相似度型） | +| 评分器 Prompt | 指导裁判模型打分的提示词，须含 `${prompt}`、`${output}`、`${completion}` 中至少一个变量，≤50000 字符 | +| 数据来源 | 评测数据集（产生推理费用）或推理结果集（已有 Output，免推理费用） | +| 裁判模型 | 推荐千问-Max；评测集自动生成仅支持 `qwen-max` / `qwen-plus` | -- 模型评测当前仅支持文本生成类模型，且仅支持控制台操作，不提供公开 API/SDK。 -- 基线评测仅北京地域可用，不支持下载评测结果。 -- 数据管理与数据清洗/增强能力仅适用于华北2（北京）地域。 -- 任务提交后不可更换目标模型、维度类型创建后不可修改，选错需删除重建。 -- 应用自动评测仅面向已发布且已配置知识库的应用，须开通应用观测并具备相应权限。 -- LLM 评分器存在位置偏差与自我偏好偏差，1-3% 的分差通常为噪声，建议定期人工抽查校准。 +## 计费 + +费用由两部分构成：被评测模型推理费用 + 裁判模型评分费用。成本优化建议：先用 50-100 条小规模验证 → 保存推理结果集复用（免重复推理费用）→ 有确定性标准的场景优先用规则评估（无裁判模型费用）。 + +## 注意事项 + +- 当前仅支持文本生成类模型评测；模型评测仅支持控制台操作，无公开 API/SDK。 +- 评测任务提交后不可更换目标模型，需删除重建。 +- 1-3% 的分数差异通常为评测噪声，不建议仅凭微小分差做决策。 +- LLM 评分器存在位置偏差和自我偏好偏差，建议定期人工抽查校准。 +- 应用评测新旧两套系统并存，新版在评测任务管理、评估器和标签体系上功能更完整。 ## 关联主题页 - [application evaluation](../guides/application-evaluation.md) - [model evaluation introduction](../guides/model-evaluation-introduction.md) - [model data overview](../guides/model-data-overview.md) +- [fine tuning](../guides/fine-tuning.md) +- [use cases](../guides/use-cases.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/event-stream.md b/skills/bailian-docs-llm-wiki/wiki/concepts/event-stream.md new file mode 100644 index 00000000..deac152c --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/event-stream.md @@ -0,0 +1,85 @@ +# 事件流 + +事件流（Event Stream）是百炼平台中智能体与客户端之间传递原子消息的机制，涵盖会话状态变更、用户输入、工具调用回执、模型增量输出等。在 Managed Agents API 中通过 SSE（Server-Sent Events）单向推送，在 Realtime API 中通过 WebSocket 双向收发，两者共享"事件即会话内最小消息单元"的语义。 + +## 在百炼平台中的使用场景 + +### Managed Agents API 的 SSE 事件流 + +Managed Agents 会话（Session）内所有交互以 Event 记录存储。开发者通过 REST API 注入事件，并通过 SSE 长连接实时接收平台产生的事件： + +- **发送事件**：`POST /sessions/{session_id}/events` — 注入用户消息、工具审批结果、[函数调用](function-calling.md)返回值等。 +- **订阅事件流**：`GET /sessions/{session_id}/events/stream` — 建立 SSE 长连接，流式接收实时事件推送，包括状态机流转（`idle` → `running` → `idle` / `terminated`）、工具调用回执等。 +- **查询历史**：`GET /sessions/{session_id}/events` — 分页列出会话的事件历史。 + +Event 类型包括用户消息、工具调用回执、状态变更等，是会话生命周期内不可变的消息记录。 + +### Realtime API 的双向事件流 + +Realtime API 基于 WebSocket（以及 WebRTC、AOQ 协议），通过客户端事件与服务端事件的双向交互实现低延迟实时对话： + +**客户端事件**（Client → Server）： + +| 事件 | 用途 | +| --- | --- | +| `session.update` | 更新会话配置（模态、音色、VAD、工具等） | +| `input_audio_buffer.append` | 追加音频数据（Base64 编码） | +| `input_audio_buffer.commit` | 提交音频缓冲区（Manual 模式必需） | +| `input_audio_buffer.clear` | 清空音频缓冲区 | +| `input_image_buffer.append` | 追加图像数据 | +| `response.create` | 触发模型生成响应 | +| `response.cancel` | 取消正在进行的响应 | +| `conversation.item.create` | 回传工具调用结果 | + +**服务端事件**（Server → Client）： + +| 事件 | 含义 | +| --- | --- | +| `session.created` | 连接建立，返回默认配置 | +| `session.updated` | 会话配置更新成功 | +| `error` | 错误信息 | +| `input_audio_buffer.speech_started` | VAD 检测到语音开始 | +| `input_audio_buffer.speech_stopped` | VAD 检测到语音结束 | +| `input_audio_buffer.committed` | 音频缓冲区已提交 | +| `response.audio.delta` | 增量音频输出 | +| `response.audio_transcript.delta` | 增量文本转录 | +| `response.done` | 响应完成 | +| `response.function_call_arguments.done` | 工具调用参数完成 | +| `conversation.item.input_audio_transcription.delta` | 实时语音识别中间结果 | + +## 关键参数与配置 + +### Managed Agents API + +事件交互无额外配置，事件结构由发送时的 payload 决定。Session 状态机（`idle` → `running` → `idle` / `terminated`）驱动事件流转，归档后 Session 进入 `terminated` 终态不再产生新事件。删除 Session 会硬删除全部事件历史。 + +### Realtime API + +会话参数通过 `session.update` 客户端事件配置： + +| 参数 | 说明 | 默认值 | +| --- | --- | --- | +| `modalities` | 输出模态：`["text"]` 或 `["text","audio"]` | `["text","audio"]` | +| `turn_detection.type` | VAD 类型：`server_vad` / `semantic_vad` | `server_vad` | +| `turn_detection.threshold` | VAD 灵敏度，范围 [-1.0, 1.0] | 0.5 | +| `turn_detection.silence_duration_ms` | 静音触发时间（ms），范围 [200, 6000] | 800 | + +交互模式分两种：VAD 模式下服务端自动检测语音起止并触发响应，支持语音打断；Manual 模式下由客户端通过 `input_audio_buffer.commit` + `response.create` 手动控制对话节奏。 + +## 协议与鉴权 + +事件流依赖底层传输协议的建连鉴权： + +- **Managed Agents API**：REST 请求通过 `Authorization: Bearer ` 鉴权，SSE 流复用同一会话的鉴权上下文。 +- **Realtime API WebSocket**：握手时携带 API Key，建连成功后无需重复鉴权。 +- **Realtime API AOQ**：采用服务端代理鉴权，API Key 仅保留在 AppServer 侧，客户端使用网关返回的临时 Token 建连。 + +三种 Realtime 传输协议（WebSocket、WebRTC、AOQ）在弱网表现和端侧支持上差异显著：WebSocket 接入门槛最低适合服务端集成；WebRTC 浏览器原生支持内置回声消除；AOQ 基于 QUIC 抗弱网能力最强，提供移动端原生 SDK。 + +## 关联主题页 + +- [managed agents api](../api/managed-agents-api.md) +- [omni realtime api](../api/omni-realtime-api.md) +- [realtime api user guide](../guides/realtime-api-user-guide.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/function-calling.md b/skills/bailian-docs-llm-wiki/wiki/concepts/function-calling.md index 93ca5304..d3e10eb8 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/function-calling.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/function-calling.md @@ -1,41 +1,73 @@ -# 函数调用（Function Calling） +# 函数调用 -函数调用（Function Calling）是让大模型在推理过程中根据用户输入，自主判断并"调用"外部工具（自定义函数、内置能力或插件）以获取实时信息、执行精确计算或操作外部系统的能力。它是弥补大模型原生局限、构建 Agent 与复杂应用的核心机制。 +函数调用（Function Calling）是大模型根据用户输入和预定义的工具描述，自主决定是否调用外部函数并生成结构化调用参数的能力。它是百炼平台实现工具调用、智能体编排和插件集成的底层机制。 -## 在百炼平台的使用场景 +## 在百炼平台中的使用场景 -百炼在多个层面暴露了 Function Calling 能力： +### API 直接调用 -- **文本 / 视觉生成模型**：所有 Qwen3 及以上通用文本与视觉理解模型均支持自定义工具调用。以 `qwen3.7-plus` 为代表的旗舰模型工具调用完整、上下文长（1M），适合 AI 编程与 Agent 开发；效果确认后可切到 `qwen3.6-flash` 降本，功能与上下文保持一致。 -- **实时多模态（Omni-Realtime API）**：基于 WebSocket 的实时音视频对话同样支持工具调用。客户端通过 `session.update` 事件在会话中声明工具，模型触发调用后由 `response.function_call_arguments.done` 服务端事件返回调用参数，客户端执行后再用 `conversation.item.create` 事件回传工具结果。 -- **应用构建（智能体 / 工作流）**：新版智能体（Agent 2.0）将知识库、MCP、插件等能力统一抽象为"工具"，由智能体自主规划调用顺序，形成"规划-执行-反思"链路。工作流应用则把工具作为固定节点按编排顺序执行，不由模型主动规划。 -- **插件（Plug-in）与 Assistant API**：插件是工具集合，本质也是工具调用。智能体应用 / Assistant API 中，模型依据工具名称与描述判断是否调用；无需调用时直接生成结果。 +通过 Chat Completions 接口调用时，开发者在请求的 `tools` 参数中声明可用函数（名称、描述、参数 JSON Schema），模型根据对话上下文判断是否需要调用，并在响应中返回函数名和参数。开发者执行函数后将结果追加到消息列表，再次请求模型生成最终回复，支持多轮工具调用。 -## 内置工具与自定义工具 +各接口对函数调用的支持情况： -- **自定义工具（Function Calling）**：开发者自行定义工具名称、描述与参数结构，模型据此决定何时调用、如何填参，应用侧执行后将结果回填模型生成最终回复。 -- **内置工具**：联网搜索、代码解释器、网页抓取等由平台预置，无需复杂配置即可开启，是 Function Calling 的开箱即用形态。 +- **OpenAI 兼容 Chat Completions**：通过 `tools` 参数支持，兼容 OpenAI SDK 的 function calling 字段约定。 +- **Anthropic 兼容 Messages**：以 tool use 形式支持，字段遵循 Anthropic 协议。 +- **DashScope 原生接口**：参数支持最完整，适合需要精细控制调用行为的场景。 +- **OpenAI 兼容 Responses**：除自定义函数外，还内置联网搜索、代码解释器等平台工具，自动管理调用流程。 + +### 智能体应用与工作流 + +在百炼应用层面，函数调用是智能体（Agent）的核心运行机制： + +- **智能体应用（Agent 2.0）**：知识库、MCP 工具、插件、内置工具均统一为工具，由模型基于函数调用能力自主规划调用顺序。系统提示词中应明确提及工具名称及使用时机以引导调用。ReAct 最大轮次（1-50）限制单次会话中工具调用的最大次数。 +- **工作流应用**：插件作为工作流节点按预定义流程执行，不依赖模型主动规划，属于确定性调用。 +- **Assistant API**：通过工具 ID 调用插件，底层同样基于模型的函数调用判断。 + +### Managed Agents API + +在托管智能体运行时中，函数调用由平台沙箱执行。Session 内的 Event 记录包含工具调用回执，开发者可通过 `POST /sessions/{session_id}/events` 注入函数执行结果，平台驱动"规划-执行-反思"循环直至任务完成。 + +### 插件调用 + +百炼插件（官方、三方、自定义）本质上是函数的集合。模型根据用户输入、工具名称和工具描述判断是否调用；需要调用时选择合适工具，应用内部完成调用后将结果与用户内容合并再次输入模型生成最终回复。每个智能体应用最多添加 10 个工具。 + +## 支持函数调用的模型 + +- **文本生成**：Qwen3 及以上通用模型均支持 Function Calling，推荐 `qwen3.7-plus` 起步（完整 Function Calling 与内置工具支持）。 +- **视觉理解**：Qwen3.7/3.6/3.5 及 Qwen-VL 系列支持。 +- **语音实时对话**：`qwen-audio-3.0-realtime-plus` 支持 Function Calling。 +- **意图理解**：`tongyi-intent-detect-v3` 可同时输出意图与函数调用信息。 +- **GUI 交互**：`gui-plus` 通过 `computer_use` 工具定义实现界面操作调用。 + +插件场景目前支持 qwen-turbo、qwen-plus、qwen-max、qwen-vl-max、qwen-vl-plus。 ## 关键参数与配置 -- **模型选型**：推荐具备强工具调用能力的模型（如千问-Max / `qwen3.7-plus` 系列）。 -- **思考模式**：可通过 `enable_thinking` 开启（Responses API 用 `reasoning.effort` 控制），配合工具调用提升规划质量。 -- **ReAct 最大轮次**：智能体中取值 1-50，限制单次会话内工具调用的最大次数，防止无限循环。 -- **实时 API 工具配置**：通过 `session.update` 的 `tools` 字段声明可用工具；工具调用结果需经 `conversation.item.create` 回传。 -- **插件调用**：通过 Assistant API 调用时需正确传递工具 ID（如 `calculator`、`code_interpreter`），且每个智能体应用最多添加 10 个工具。 +| 参数 | 位置 | 说明 | +| --- | --- | --- | +| `tools` | 请求体 | 函数声明数组，每个元素包含函数名称、描述和参数 Schema | +| `tool_choice` | 请求体 | 控制调用策略（自动/强制/指定函数），具体取值以接口文档为准 | +| `tool_calls` | 响应体 | 模型返回的调用请求，含函数名和 JSON 格式参数 | +| `role: "tool"` | messages | 将函数执行结果回传模型的消息角色 | +| ReAct 最大轮次 | 智能体应用配置 | 限制单次会话工具调用次数（1-50） | +| 工具 ID | 插件/Assistant API | 调用插件工具时的唯一标识（如 `calculator`） | ## 开发建议 -- 优先用内置工具满足通用需求（搜索、计算、代码执行），减少自定义成本。 -- 自定义工具时，工具名称与描述要清晰准确，直接影响模型是否正确触发调用。 -- 需要精确、可控流程时用工作流把工具固化为节点；需要动态规划时用智能体让模型自主调用。 -- 旧版智能体的自定义插件有 5 秒超时限制，设计工具时注意执行时长。 +- 函数描述应清晰准确，模型依赖描述判断调用时机；参数 Schema 越严格，生成的调用参数越可靠。 +- 多步任务推荐选用工具调用能力强的模型（如千问-Max 系列），并在提示词中引导工具使用。 +- 开启思考模式（`enable_thinking`）有助于提升智能体的工具规划与反思效果。 +- 跨接口迁移时注意字段映射差异：DashScope 参数最全，OpenAI/Anthropic 兼容接口以对应生态约定为准。 ## 关联主题页 -- [omni realtime api](../api/omni-realtime-api.md) +- [qwen api reference](../api/qwen-api-reference.md) +- [more models](../api/more-models.md) +- [toolkits and frameworks](../api/toolkits-and-frameworks.md) +- [plug in](../guides/plug-in.md) - [model experience](../guides/model-experience.md) +- [managed agents api](../api/managed-agents-api.md) - [llm application](../guides/llm-application.md) -- [plug in](../guides/plug-in.md) + diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/knowledge-base.md b/skills/bailian-docs-llm-wiki/wiki/concepts/knowledge-base.md index 3062222e..7c5ae48a 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/knowledge-base.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/knowledge-base.md @@ -1,87 +1,94 @@ # 知识库 -知识库是阿里云百炼平台基于 RAG（[检索增强生成](rag.md)）技术构建的私有数据管理能力，用于为大模型补充私有数据与最新信息，使应用能够准确回答特定领域问题。知识库仅支持在中国站华北2（北京）地域开通和使用，提供标准版与旗舰版两种规格，并配套日志监控、API、效果优化与计费体系。 +知识库是百炼平台将企业私有数据、业务文档和最新信息接入大模型应用的核心能力。它通过解析、切片、向量化、检索和重排等流程，在生成回答前召回相关内容，实现 RAG（检索增强生成），提升回答准确性、可追溯性并降低幻觉。 -## 核心能力 +## 典型使用场景 -知识库对私有数据或文件进行语义检索，可找出语义相同或相近的内容，即使关键词匹配度极低甚至为零。检索结果可作为上下文喂给大模型生成回答，也可仅作为检索能力单独使用。支持挂载到[智能体应用](agent-application.md)、工作流应用，或通过阿里云百炼 SDK、LlamaIndex、Spring AI Alibaba 等框架集成到外部应用。 +### 私有知识问答 -知识库类型分为四类，单一知识库不支持同时选择多个类型： +在零代码场景中，开发者可以先创建智能体应用，再创建并挂载知识库，让应用回答企业内部文档、产品手册、FAQ、参数表等私域问题。知识库通常先从文件连接器、OSS 或数据连接器导入数据，完成解析和索引后，再在应用配置中添加到智能体或工作流。 -- 文档搜索：对非结构化文档做语义检索，细分为基础文档问答、图文并茂回复、视觉理解（富文本文档）、极速问答四种使用场景。 -- 数据查询：对结构化表格数据做检索。 -- 图片问答：基于 multimodal-embedding-v1（1024 维）对图片做检索。 -- 音视频搜索：对音频、视频内容按时间轴结构化检索。 +### 智能体应用中的工具能力 -## 在百炼平台中的使用场景 +在新版智能体应用中，知识库会作为工具由智能体自主规划调用。开发者可通过系统提示词说明知识库的使用边界，并结合标签过滤、元数据过滤等方式限定检索范围。适合面向客户服务、内部助手、产品问答、知识检索等需要动态调用私有知识的场景。 -### [智能体应用](agent-application.md) +### 工作流中的检索节点 -在 Agent 2.0 架构中，知识库作为工具由智能体自主规划调用，与 MCP 等外部工具统一调度；在旧版（Agent 1.0）中，则先检索知识库再决策是否调用其他工具。可在应用配置页文档知识库右侧点击「+ 添加知识库」接入，并设置相似度阈值与权重，还可通过标签限定查询范围以提升准确性。开启「展示回答来源」后，回答会以角标形式展示知识来源与源文件/源网页地址。 +在工作流应用中，知识库可作为节点接收用户 query，输出检索到的切片，再传给大模型节点生成答案。该方式适合需要固定流程控制的业务，例如先检索知识、再判断条件、再调用工具或生成结构化结果。 -### 工作流应用 +### 知识检索与知识问答服务 -将知识库节点拖入画布，配置输入变量（query）、选择固定知识库或动态引入、设置 TopK，用于在预定义流程中精确控制检索环节。 +百炼提供知识检索和知识问答两类应用网关接口：知识检索只返回排序后的切片，适合开发者自行拼接 [prompt](../guides/prompt.md) 并控制生成流程；知识问答会在检索基础上生成自然语言回答，并通过 SSE 流式返回规划、工具调用和生成阶段，适合快速接入端到端问答能力。 -### 外部应用集成 +### 开源框架集成 -- SDK：通过阿里云百炼 SDK 调用检索能力，子账号需获取 `AliyunBailianDataFullAccess` 策略并加入[业务空间](workspace.md)。 -- LlamaIndex（Python 3.9+）：将知识库部署在云端，使用默认智能文档切分与官方向量模型，不支持自定义切分与嵌入模型；如需灵活切分应改用本地知识库方案。 -- Spring AI Alibaba（Spring Boot 3.x，JDK 17+）：可调用百炼智能体/工作流应用并检索百炼知识库，应用集成推荐变量名 `DASHSCOPE_API_KEY`，知识库检索推荐 `AI_DASHSCOPE_API_KEY`。 -- 本地 RAG：检索环节在本地执行，生成环节调用通义千问 API，适合需要灵活切分与嵌入模型选择的场景。 +开发者也可以通过 LlamaIndex 或 Spring AI Alibaba 集成百炼知识库。LlamaIndex 方案可将本地文档上传并构建云端知识库，再基于检索结果构建 RAG 应用；Spring AI Alibaba 可在 Java 应用中检索百炼知识库，并与智能体或工作流应用结合使用。 -### API 编排 +## 数据来源与知识库类型 -通过百炼 OpenAPI（`bailian/2023-12-29`，ROA 风格）可程序化管理知识库全链路：申请上传租约 → 上传文件 → 添加文件到类目 → 轮询文件解析状态（INIT/PARSING/PARSE_SUCCESS）→ 初始化知识库 → 提交索引任务 → 轮询任务状态直至 COMPLETED。每[业务空间](workspace.md)最多创建 500 个类目，类目类型目前仅支持 `UNSTRUCTURED`。 +知识库可处理多种数据形态，包括文档、表格、图片、音视频和结构化数据。常见来源包括本地上传文件、OSS、文件连接器、表格连接器、数据库连接器以及语雀等外部系统。 + +常见类型包括： + +- 文档搜索：面向 PDF、DOCX、Markdown、Excel、HTML 等非结构化或半结构化文档。 +- 数据查询：面向表格和结构化数据。 +- 图片问答与[多模态](multimodal.md)检索：面向图片或富文本文档中的视觉信息。 +- 音视频搜索：面向课程回放、直播回放、客服质检、会议录音等音视频内容。 + +创建时通常需要选择规格、知识库类型、数据来源、解析方式、切片策略和索引配置。部分能力创建后不可直接变更，例如知识库类型、部分向量配置、创建时未开启的 Meta 信息抽取或多轮对话改写能力，实际规划时应提前确认。 ## 关键参数与配置 -### 规格与并发 +### 检索效果参数 + +- 相似度阈值：只召回分数高于阈值的切片。阈值过高可能漏召回，过低可能引入噪声。 +- 召回片段数 / TopK：控制向量检索、关键词检索或最终返回的候选数量。增加 TopK 可提升覆盖率，但会增加上下文长度和成本。 +- 权重：多个知识库联合检索时，用于控制同类型知识库在相似度接近时的优先级。 +- 排序模型 / Rerank：对初步召回结果重排，提升最终输入给模型的片段相关性。 +- 路由：在多知识库场景中选择更合适的知识库参与检索。 -| 规格 | 最高检索并发 | 平台存储空间 | 价格 | -| --- | --- | --- | --- | -| 标准版 | 1 QPS（固定，不可调） | ≤ 100 GB | 0.03 元/知识库/小时 | -| 旗舰版 | 50–10,000 QPS（可调，对应 1–200 RCU） | ≤ 9,999 GB | 0.2 元/RCU/小时 | +### 过滤与结构化控制 -RCU（Retrieval Compute Unit）是检索并发能力度量单位，1 RCU 约支撑最高 50 QPS，所需 RCU = 向上取整（检索峰值 QPS ÷ 50）。变配按发生时间分段计费，同一知识库 1 个自然日内最多变配 1 次。 +- 标签过滤：适合按业务线、产品、权限域、文档类别等粗粒度限定检索范围。 +- 元数据过滤：适合按文件名、产品名、日期、作者、关键词等结构化字段精确过滤。 +- SearchFilters：Retrieve 等检索接口可使用过滤语法对结构化字段做条件筛选。 -### 向量与切片 +合理使用过滤条件可以减少无关切片召回，降低模型输入 Token 消耗，并提升答案稳定性。 -- 向量模型：文档搜索、数据查询、音视频搜索类支持 text-embedding-v4、text-embedding-v3（均为 512 维）；图片问答类仅支持 multimodal-embedding-v1（1024 维）。向量维度不支持更改。 -- 文本切片长度上限：单个切片 6,000 [Token](token.md)；编辑切片（UpdateChunk）长度限制为 10–6,000 字符；删除切片（DeleteChunk）单次最多 10 个。 -- 召回文本切片数量：单次查询最多召回 20 个切片。 -- 切片方式：智能切分（保留语义完整性）或按长度切分。**知识库一旦创建，无法再配置 metadata 抽取，也无法更改文档切分 chunk**，需在创建时一次性规划好元数据与切片策略。 +### 模型与索引配置 -### 检索参数 +文档搜索、数据查询和音视频搜索通常使用 text-embedding-v4 或 text-embedding-v3 等文本向量模型，并可结合 qwen3-rerank 等排序模型；图片问答、视觉理解和[多模态](multimodal.md)检索则使用[多模态](multimodal.md)向量与排序模型。向量维度通常由平台模型决定，不支持随意修改。 -- 相似度阈值：仅语义相似度高于此阈值的文本切片才会被召回。阈值过高（如 0.60）可能导致无召回结果或丢弃相关切片。 -- 召回片段数（K 值）：取值范围 1–20，调大可提升完整性但增加 [Token](token.md) 消耗；拼装后总长度超出大模型输入限制会被截断，并非越大越好。 -- 初步向量检索 TopK / 初步关键词检索 TopK：默认 50，取值范围 10–100，影响送入排序模型的切片数量与成本。 -- 权重：仅在**同类型知识库之间生效**，用于干预多知识库召回顺序。 +如果使用百炼云端知识库，部分框架方案会采用平台默认的智能文档切分和官方向量模型，不支持自定义切分或自定义嵌入模型。若需要完全控制切分、向量模型和本地检索流程，可选择本地 RAG 方案，在本地完成检索后调用通义千问等生成模型。 -### 解析方式 +## 开发接入要点 -导入文件时可选择解析方式：电子文档解析（不支持插图与图表）、文档智能解析（提取插图文本与摘要）、大模型文档解析（支持对插图和图表提问，需配合选择模型）、Qwen VL 解析（仅图片，可传入 Prompt）、音视频解析（语音识别、视频帧提取、剧情解析按时间轴结构化对齐）。 +调用知识检索与问答接口时，需要使用 [API Key](api-key.md) Bearer 鉴权，并通过[业务空间](workspace.md) ID 拼接 Base URL。[API Key](api-key.md) 与[业务空间](workspace.md)需要与知识库所在地域和权限匹配；若知识库创建在子[业务空间](workspace.md)，还需配置对应 workspaceId。 -## 效果优化 +典型开发流程如下： -当出现召回不完整或内容不准确时，建议先建立评测基线（至少 100 组问题，覆盖事实型/比较型/教程型/分析型），再按 RAG 三阶段诊断改进： +1. 准备数据源并创建数据连接器或上传文件。 +2. 创建知识库，选择类型、规格、解析方式、切片策略和索引参数。 +3. 导入数据并等待解析、向量化和索引构建完成。 +4. 通过命中测试或在线调试验证召回效果。 +5. 将知识库挂载到智能体、工作流、知识检索服务或知识问答服务。 +6. 根据日志、命中测试、[评测](evaluation.md)集和用户反馈持续调整阈值、TopK、权重、过滤条件和数据质量。 -1. 建立索引：优化源文件排版（优先 Markdown、移除水印、避免复杂表格）、统一实体表述、启用多轮对话改写（创建时开启，后续无法补开）。 -2. 检索召回：为文件添加标签过滤、配置元数据做结构化搜索、采用智能切分保留语义完整性、调整相似度阈值与召回片段数。 -3. 生成答案：更换为能力更强的商业模型（如通义千问 Max/Plus/QwQ）、优化提示词模板（限定输出、少样本提示、内容分隔标记且 `${documents}` 只出现一次）。 +## 注意事项 -## 日志监控 +知识库效果高度依赖源文档质量、切片策略、检索参数和问题表达。上线前应使用真实问题做命中测试，关注是否召回正确切片、是否引入噪声、是否展示引用来源以及是否满足权限隔离要求。 -检索日志由日志服务（SLS）承载，首次使用需授权角色 `AliyunServiceRoleForSFMAccessSLS` 并创建 LogStore。每条日志 topic 为 `log_dispatch`，包含 request_id、pipeline_id、workspace_id、latency、response_status_code、response_code 等字段。建议搭建调用量趋势、TopN 知识库排名、业务错误率与 HTTP 5xx 错误率等监控。 +多知识库联合检索时，不建议只依赖默认参数，应根据业务重要性配置权重、路由和过滤条件。对延迟或成本敏感的场景，应控制 TopK、召回片段数和上下文拼装长度；对准确性要求高的场景，可增加重排、[评测](evaluation.md)集和日志监控闭环。 ## 关联主题页 -- [knowledge base](../guides/knowledge-base.md) - [frameworks](../api/frameworks.md) +- [more](../api/more.md) +- [start using](../guides/start-using.md) - [llm application](../guides/llm-application.md) -- [application use cases](../guides/application-use-cases.md) - [data connection overview](../guides/data-connection-overview.md) -- [application component api reference](../api/application-component-api-reference.md) +- [knowledge base](../guides/knowledge-base.md) +- [application use cases](../guides/application-use-cases.md) +- [knowledge](../api/knowledge.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/long-term-memory.md b/skills/bailian-docs-llm-wiki/wiki/concepts/long-term-memory.md new file mode 100644 index 00000000..fe1989a3 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/long-term-memory.md @@ -0,0 +1,112 @@ +# 长期记忆 + +长期记忆是百炼平台提供的跨会话上下文持久化能力。它自动从对话中提取关键信息并存储，在后续对话中通过语义检索召回相关记忆并注入 Prompt，使智能体能够持续理解用户偏好与历史信息，解决大模型上下文窗口有限、对话结束后信息丢失的问题。 + +## 核心能力 + +长期记忆持久化两类内容，二者可独立或组合使用： + +- **记忆片段**：从对话中自动提取的关键事件和信息（如"用户每天上午9点需要喝水提醒"），适用于大多数长期记忆场景。支持自动去重、动态更新，也可通过 `custom_content` 直接写入指定内容。 +- **用户画像**：基于自定义画像模板从对话中提取的结构化属性（如年龄、职业、偏好等），适用于需要固定属性持久化存储的场景。属性字段应清晰具体，避免"姓名/名称/名字"等同义字段并存，且不应期望一次对话就提取全部信息。 + +## 接入方式 + +### API 直连 + +通过 HTTPS 调用 `https://dashscope.aliyuncs.com/api/v2/apps/memory/*` 系列接口。需在请求 Header 中添加 `Authorization: Bearer $DASHSCOPE_API_KEY`，`Content-Type` 设为 `application/json`。典型流程为：对话结束调用 `AddMemory` 写入记忆 → 调用 `SearchMemory` 语义检索 → 将结果注入 Prompt。 + +### OpenClaw 记忆插件 + +OpenClaw Agent 可通过插件实现零侵入的跨会话记忆。插件在 Gateway 内通过 `before_agent_start`（自动召回）和 `agent_end`（自动捕获）两个生命周期钩子与长期记忆 API 交互，所有读写均由百炼服务端完成提炼、[向量化](embedding.md)和语义检索。 + +```bash +# 安装 +openclaw plugins install @modelstudio/modelstudio-memory-for-openclaw + +# 验证 +openclaw plugins info modelstudio-memory-for-openclaw +openclaw modelstudio-memory stats +openclaw gateway restart +``` + +> **注意**：OpenClaw 记忆插件为统一配置，所有 Agent 共享同一记忆，暂不支持按 Agent 独立配置；不支持阿里云百炼 Coding Plan 的 API Key。 + +## 关键接口 + +| 接口名称 | HTTP 方法 | 路径 | 说明 | +| --- | --- | --- | --- | +| AddMemory | POST | `/add` | 添加记忆片段 | +| SearchMemory | POST | `/memory_nodes/search` | 语义搜索记忆片段 | +| ListMemory | GET | `/memory_nodes` | 列出记忆片段 | +| DeleteMemory | DELETE | `/memory_nodes/{memory_node_id}` | 删除记忆片段 | +| UpdateMemory | PATCH | `/memory_nodes/{memory_node_id}` | 更新记忆片段 | +| CreateProfileSchema | POST | `/profile_schemas` | 创建画像模板 | +| ListProfileSchemas | GET | `/profile_schemas` | 获取画像模板列表 | +| GetProfileSchema | GET | `/profile_schemas/{profile_schema_id}` | 获取画像模板详情 | +| UpdateProfileSchema | PATCH | `/profile_schemas/{profile_schema_id}` | 更新画像模板 | +| DeleteProfileSchema | DELETE | `/profile_schemas/{profile_schema_id}` | 删除画像模板 | +| GetUserProfile | GET | `/profile_schemas/{profile_schema_id}/user_profile` | 获取用户画像 | + +## AddMemory 请求参数 + +| 参数 | 类型 | 必填 | 说明 | +| --- | --- | --- | --- | +| `user_id` | string | 是 | 记忆实体 ID，用于标识归属对象，最大 64 个字符 | +| `messages` | array | 与 `custom_content` 二选一 | 对话消息列表，每个消息含 `role`（user/assistant）和 `content`，最多 50 条 | +| `custom_content` | string | 与 `messages` 二选一 | 自定义内容，最大 512 个字符，传入后忽略 `messages` | +| `profile_schema` | string | 否 | 画像模板 ID | +| `memory_library_id` | string | 否 | 记忆库 ID，最大 32 个字符，不传则使用默认记忆库 | +| `project_id` | string | 否 | 记忆片段规则 ID，不传则使用指定记忆库的默认规则 | +| `meta_data` | object | 否 | 用户自定义信息 | + +返回结果中 `memory_nodes` 数组包含每项的 `memory_node_id`、`content`、`event`（`ADD`/`UPDATE`/`DELETE`）及 `old_content`（仅 `UPDATE` 时有效）。 + +## 使用示例 + +```bash +# 写入记忆（从对话自动提取） +curl -X POST https://dashscope.aliyuncs.com/api/v2/apps/memory/add \ + --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ + --header "Content-Type: application/json" \ + --data '{ + "messages": [ + {"role": "user", "content": "每天上午9点提醒我喝水"}, + {"role": "assistant", "content": "好的，已记录"} + ], + "user_id": "user_001" + }' + +# 语义检索记忆 +curl -X POST https://dashscope.aliyuncs.com/api/v2/apps/memory/memory_nodes/search \ + --header "Authorization: Bearer $DASHSCOPE_API_KEY" \ + --header "Content-Type: application/json" \ + --data '{ + "user_id": "user_001", + "messages": [{"role": "user", "content": "我需要做什么？"}], + "top_k": 5 + }' +``` + +## 使用限制 + +| 接口 | 限流（阿里云账号级别） | +| --- | --- | +| 全部接口 | 总计不超过 3000 QPM | +| 记忆片段 add 接口 | 120 QPM | +| 记忆片段 search 接口 | 300 QPM | + +## 记忆有效期 + +记忆有效期在不同入口存在差异：通过 API 直写且不指定 `project_id` 时使用默认规则，API 文档指出"生成的记忆片段与用户画像暂无失效日期"；而控制台记忆库的默认规则预置了"默认有效期 180 天"，并支持按规则配置 7/30/180 天或永不过期。以控制台记忆规则配置为准。 + +## 与应用调用的关系 + +长期记忆通常与应用调用配合使用。在通过 Responses API 或 DashScope API 调用智能体应用时，可在对话前通过 `SearchMemory` 检索相关记忆注入 Prompt，在对话后通过 `AddMemory` 将新信息写入记忆库，从而实现多轮对话间的上下文连续性。使用 OpenClaw 插件时，这一过程在 Gateway 内自动完成，无需在应用代码中显式调用记忆 API。 + +## 关联主题页 + +- [long term memory new](../api/long-term-memory-new.md) +- [memory library overview](../guides/memory-library-overview.md) +- [application call](../api/application-call.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/migration.md b/skills/bailian-docs-llm-wiki/wiki/concepts/migration.md new file mode 100644 index 00000000..31c46434 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/migration.md @@ -0,0 +1,93 @@ +# 迁移指南 + +迁移指南是将已有应用（主要基于 OpenAI 或 Anthropic 生态构建）迁移到阿里云百炼平台模型服务的操作总览。核心目标是尽可能复用现有代码与客户端库，仅替换凭证、服务地址和模型名称即可完成接入。 + +## 迁移三要素 + +无论使用哪种兼容协议，迁移的关键是配置以下三项： + +- **API Key**：替换为百炼 API Key。各地域的 API Key 不同，切换地域时需同步更换。建议通过环境变量（如 `DASHSCOPE_API_KEY`）注入，避免硬编码。 +- **Base URL**：根据所选兼容协议和地域，替换为百炼对应的服务地址。 +- **Model**：替换为百炼支持的模型名称（如 `qwen-plus`、`qwen3-max` 等）。 + +对于已有 OpenAI 应用的最简迁移路径，通常只需替换这三项即可，无需改动业务逻辑。 + +## 接口选择 + +百炼提供四类文本生成接口，迁移时应根据现有技术栈和功能需求选择： + +| 现有技术栈 | 推荐接口 | 迁移成本 | +| --- | --- | --- | +| OpenAI SDK（Chat Completions） | OpenAI 兼容 Chat Completions | 最低，直接复用客户端库 | +| OpenAI SDK（需要内置工具/自动上下文） | OpenAI 兼容 Responses | 中，需适配新接口字段 | +| Anthropic SDK | Anthropic 兼容 Messages | 最低，直接复用客户端库 | +| 无特定生态绑定，需最全功能 | DashScope 原生接口 | 较高，需学习原生 SDK | + +迁移时需注意：不同接口的参数集合和功能覆盖存在差异。DashScope 参数最全，OpenAI/Anthropic 兼容接口以对应生态的字段约定为准。跨接口迁移时需核对参数映射，确认目标接口是否支持原有功能（如流式输出、function call、多模态等）。 + +## Base URL 配置 + +### OpenAI 兼容协议 + +| 地域 | Base URL | +| --- | --- | +| 华北2（北京） | `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` | +| 新加坡 | `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` | +| 日本（东京） | `https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1` | +| 德国（法兰克福） | `https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/compatible-mode/v1` | +| 美国（弗吉尼亚） | `https://dashscope-us.aliyuncs.com/compatible-mode/v1` | + +其中 `{WorkspaceId}` 为业务空间 ID，可在百炼控制台的业务空间详情页查看。 + +### Anthropic 兼容协议 + +Anthropic 兼容协议的 Base URL 在上述地址基础上将 `/compatible-mode/v1` 替换为 `/apps/anthropic`。 + +## 计费方案匹配 + +百炼提供四种计费方案：Token Plan 个人版、Token Plan 团队版、Coding Plan、按量计费。迁移时需特别注意： + +- 四种方案的 API Key 互不通用，Base URL 也各不相同。 +- 配置时必须确保 API Key、Base URL 和计费方案三者匹配，否则会返回 401 认证错误。 +- 例如 Token Plan 个人版 OpenAI 兼容地址为 `https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1`，与按量计费地址不同。 + +## 域名与路径迁移 + +### 旧域名迁移 + +百炼为北京、新加坡地域推出了业务空间专属域名，性能与稳定性更佳，建议从旧域名迁移： + +- 北京：`https://dashscope.aliyuncs.com` → `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com` +- 新加坡：`https://dashscope-intl.aliyuncs.com` → `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com` + +现有旧域名仍可正常使用，但建议尽快迁移。 + +### 旧版路径迁移 + +Responses 与 Conversations 接口的旧版路径 `/api/v2/apps/protocols/compatible-mode/v1/...` 即将停止维护，请迁移至新版 `/compatible-mode/v1/...` 路径。 + +## 客户端工具迁移 + +百炼支持通过多种聊天客户端和开发工具接入模型服务（如 Claude Code、Codex、Cursor、Cline、Cherry Studio 等）。这类工具的迁移步骤通常为： + +1. 选择计费方案并获取对应 API Key。 +2. 根据工具支持的协议（OpenAI 兼容或 Anthropic 兼容）选择 Base URL。 +3. 在工具的配置文件或设置界面中填入 API Key、Base URL 和模型名称。 + +各工具的具体配置文件路径和字段不同，需参照对应文档。 + +## 注意事项 + +- **地域限制**：部分模型仅支持特定地域（如 `qwen-deep-research` 仅支持华北2-北京），迁移前需确认目标模型的地域可用性。 +- **协议差异**：Qwen-Audio 不支持 OpenAI 兼容协议，仅支持 DashScope 协议；`qwen-deep-research` 仅支持 Python DashScope SDK，暂不支持 OpenAI 兼容接口。 +- **三方模型**：三方直供模型（如 DeepSeek、Kimi、GLM 等）仅在中国站的中国内地地域可用，调用前需在控制台开通对应服务。 +- **功能差异**：若原应用依赖联网搜索、代码解释器等内置工具，需使用 Responses 接口而非普通的 Chat Completions。 + +## 关联主题页 + +- [qwen api reference](../api/qwen-api-reference.md) +- [toolkits and frameworks](../api/toolkits-and-frameworks.md) +- [more models](../api/more-models.md) +- [use chat client or development tool](../guides/use-chat-client-or-development-tool.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/model-lifecycle.md b/skills/bailian-docs-llm-wiki/wiki/concepts/model-lifecycle.md new file mode 100644 index 00000000..0fee1b1b --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/model-lifecycle.md @@ -0,0 +1,93 @@ +# 模型生命周期 + +模型生命周期指一个大模型在百炼平台中从数据准备、调优训练、压缩、部署上线到运行监控的完整链路。开发者可以通过控制台或 API 完成全流程操作，并可根据监控反馈持续迭代优化。 + +## 生命周期阶段 + +百炼平台的模型生产链路包含以下阶段，部分阶段可选： + +``` +数据准备 → 模型调优 → 模型压缩（可选）→ 模型部署 → 模型监控 → 迭代优化 +``` + +### 1. 数据准备 + +在调优前创建训练集和评测集，并可通过数据清洗与增强提升数据质量。训练集用于模型调优，评测集用于评估模型泛化能力。 + +- 训练集格式：SFT 用 ChatML 格式 `{"messages":[...]}`；CPT 用纯文本 `{"text":"..."}`；DPO 用 ChatML + `chosen`/`rejected` 字段。 +- 最低数据量要求：CPT 需 1000 万 Token，SFT 需上千条，DPO 需上百组。 +- 评测集为 Excel 格式，每行包含 Prompt 和 Completion。 + +### 2. 模型调优 + +通过 CPT、SFT、DPO 三种递进式方法定制模型，推荐顺序为 `CPT（可选）→ SFT → DPO（可选）`。训练模式分为全参训练和高效训练（LoRA）。 + +| 方法 | 目标 | 数据量要求 | +| --- | --- | --- | +| CPT（持续预训练） | 注入领域知识 | 1000 万+ Token | +| SFT（监督微调） | 学会遵循指令 | 1000+ 条 | +| DPO（直接偏好优化） | 对齐人类偏好 | 100+ 组 | + +文本生成模型中全参训练与高效训练费用相同，推荐优先选择全参训练。图像生成、视频生成和语音合成模型目前仅支持高效训练。 + +### 3. 模型压缩（可选） + +将全精度微调模型量化为低精度版本，降低部署规格和推理成本。压缩位于调优与部署之间，操作不可逆——压缩后的模型不支持继续微调或二次压缩。 + +- 量化模板中 MU 编号越大，部署规格越小、成本越低，但精度损失可能越大。 +- 建议在免费期内对同一微调模型尝试多个量化模板，分别部署后验证效果再正式上线。 + +### 4. 模型部署 + +将预置模型或调优后的模型部署为资源专享的推理服务。部署计费方式在服务创建后无法更改，切换需先下线已有部署。 + +| 计费方式 | 适用场景 | +| --- | --- | +| 预置吞吐（PTU） | 高负载生产环境，需稳定吞吐保障 | +| 模型单元 | 自定义性能指标，资源独占，支持 PD 分离 | +| 按 Token 用量 | 调优后模型效果验证，不使用不计费 | + +PTU 支持长输入（最高 256K token）和前缀缓存，通过阶梯容量系数和缓存折扣管理额度消耗。外部 LoRA 模型可从 OSS 导入后部署。 + +### 5. 模型监控 + +部署后通过用量统计和运行监控掌握模型运行状态，并设置告警发现静默失败。 + +- 用量统计：按业务空间维度统计 Token 消耗，延迟约 1 小时。 +- 运行监控：普通监控为小时级，高级监控（Prometheus）为分钟级。 +- 推理日志：记录每次调用的输入、输出及耗时，用于故障排查与内容审计。 +- 告警：支持 Token 消耗阈值告警，通知方式包括短信、邮件、电话、钉钉群机器人等。 + +### 6. 迭代优化 + +根据监控指标（如失败率、调用时长、Token 消耗）和评测结果，决定是否回到数据准备阶段收集更多数据进行重新调优。如果调优后评测结果不佳，最简单的改进方法是收集更多训练数据。 + +## API 全流程 + +通过 HTTP API 可完成模型生产全流程自动化： + +1. 上传数据集（`POST /api/v1/files`） +2. 创建调优任务（`POST /api/v1/fine-tunes`） +3. 查询任务状态（`GET /api/v1/fine-tunes/{id}`） +4. 部署模型（`POST /api/v1/deployments`） +5. 调用部署后的模型端点进行推理 + +API 部署支持三种 `plan` 参数：`ptu`（预置吞吐）、`mu`（模型单元）、`lora`（按 Token 计费）。 + +## 使用限制 + +- 模型调优、压缩、部署功能仅在华北2（北京）地域可用。 +- 模型压缩仅支持百炼平台微调产出的自定义模型，不支持基础模型或第三方模型。 +- OSS 导入仅支持 LoRA 模型，不支持全参微调模型。 +- 按 Token 计费的模型一个月内不使用将自动释放。 + +## 关联主题页 + +- [model data overview](../guides/model-data-overview.md) +- [fine tuning](../guides/fine-tuning.md) +- [model compression](../guides/model-compression.md) +- [model deployment 1](../guides/model-deployment-1.md) +- [model monitoring](../guides/model-monitoring.md) +- [model production](../api/model-production.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/multimodal.md b/skills/bailian-docs-llm-wiki/wiki/concepts/multimodal.md index d238b34c..0af78caf 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/multimodal.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/multimodal.md @@ -1,45 +1,59 @@ -# 多模态能力 +# 多模态 -多模态能力指模型同时理解或生成文本、图像、音频、视频、3D 等多种模态内容的能力。在百炼平台上，它既体现为「输入多模态」（如图文混合输入、音视频实时对话），也体现为「输出多模态」（如文生图、文生视频、语音合成、3D 资产生成）。 +多模态（Multimodal）指模型能够同时理解或生成文本、图像、音频、视频等多种信息模态的能力。在百炼平台上，多模态贯穿于实时交互、内容生成、向量检索与模型选型等多个产品线，是构建语音助手、视频创作、跨模态搜索等应用的基础能力。 -## 在百炼平台的主要场景 +## 在百炼平台的主要使用场景 -百炼按模态与任务把能力拆分到多个方向，开发者可按需组合： +### 实时多模态交互（Realtime / Omni） -- **实时音视频对话**：Qwen-Omni-Realtime 系列通过 WebSocket 提供低延迟的语音输入/输出、图像输入、语音活动检测（VAD）、工具调用与联网搜索，适用于智能客服、语音助手等实时交互场景。 -- **图像生成与编辑**：覆盖文生图、图生图、局部重绘、扩图、背景生成、虚拟模特、AI 试衣、创意海报等，涉及千问 Qwen-Image、万相 Wan/Wanx、Z-Image、可灵 Kling、Vidu 等模型家族。 -- **视频生成与编辑**：聚合万相 Wan、HappyHorse、PixVerse、Vidu、可灵 Kling 及人像驱动模型，支持文生视频、图生视频（首帧/首尾帧/续写）、参考生视频、视频编辑与数字人。 -- **3D 资产生成**：基于 Tripo 模型支持文生 3D、单图生 3D、多图生 3D，产出带 PBR 材质或无贴图的 GLB 模型。 -- **视觉理解与 OCR**：以 Qwen 旗舰多模态模型理解图像与长视频（最长约 2 小时），并提供专优的 OCR/文档提取能力。 -- **语音合成 / 识别 / 语音转语音**：TTS、ASR、声音复刻/设计、S2S 实时对话与同传翻译，以及音乐生成。 +- **Qwen-Omni-Realtime API**：基于 WebSocket 的实时多模态对话接口，支持音频、图像（视频帧）与文本的流式输入输出，输出模态通过 `modalities` 配置为 `["text"]` 或 `["text","audio"]`（默认）。适用于语音助手、音视频客服、实时同传等场景。 +- **Realtime API 接入方案**：提供 WebSocket、WebRTC、AOQ（AI over QUIC）三种传输协议。服务端集成选 WebSocket；浏览器端选 WebRTC；移动端原生、弱网环境选 AOQ（内置回声消除、降噪与弱网对抗）。 +- 代表模型：`qwen3.5-omni-plus-realtime`、`qwen3.5-omni-flash-realtime`、`qwen3.5-livetranslate-flash-realtime`，以及多模态开发套件 `multimodal-dialog`。 -## 输入与协议 +### 多模态内容生成（视频生成） -- **实时流式（WebSocket）**：延迟最低，适合实时交互。Omni-Realtime 通过 `input_audio_buffer.append`（PCM 音频，Base64）、`input_image_buffer.append`（JPG/JPEG，Base64）等事件送入多模态素材，`session.update` 的 `modalities` 控制输出模态（`["text"]` 或 `["text","audio"]`）。 -- **HTTP 同步**：新一代图像模型（如 `wan2.6-image`、`wan2.7-image`、`z-image-turbo`）支持一次请求返回结果，路径为 `.../aigc/multimodal-generation/generation`，请求体用 `messages` 结构，`content` 内混排 `text` 与 `image`。 -- **HTTP 异步**：图像、视频、3D 生成等耗时任务（约 1-5 分钟）统一采用「创建任务拿 `task_id` → 轮询查询」两步流程，创建时必须携带请求头 `X-DashScope-Async: enable`，否则报错 `current user api does not support synchronous calls`。`task_id` 有效期 24 小时，切勿重复创建，轮询即可。 +视频生成 API 统一采用异步任务模式（创建任务 → 轮询结果），多个模型家族支持多模态输入： -## 关键参数与配置 +- **万相 wan2.7**：图生视频支持文本/图像/音频/视频多模态输入；参考生视频（`wan2.7-r2v`）以图片/视频/音频为参考，保持角色形象与音色一致。 +- **数字人/对口型**：`wan2.2-s2v`（图片 + 音频生成说话视频）、`pixverse-lipsync`（音频或 TTS 文本驱动对口型）等，均属于跨模态驱动生成。 -不同模态的请求体大多由 `model`、`input`、`parameters` 三部分组成： +### 多模态向量（Multimodal Embedding） -- **实时对话**（`session.update`）：`modalities`、`voice`（音色，因模型而异）、`input_audio_format` / `output_audio_format`（仅 `pcm`，输入 16kHz、输出 24kHz）、`turn_detection.type`（`server_vad` / `semantic_vad`）及 `threshold`、`silence_duration_ms` 等。 -- **图像生成**：`input.prompt` / `negative_prompt`（或新协议 `messages`），图像编辑用 `images` / `image_url` / `mask_image_url`；`parameters` 含 `size`（如 `1024*1024`、`1K`/`2K`/`4K`）、`n`、`aspect_ratio`、`watermark`、`prompt_extend` 等。 -- **视频生成**：`input.prompt` 描述画面镜头，`input.media` 承载 `first_frame` / `last_frame` / `reference_image` / `video` 等素材；`parameters` 含 `resolution`（`480P`/`720P`/`1080P`）、`duration`、`ratio` 等。 -- **3D 生成**：`prompt`、`image`、`images` 三者互斥；`parameters` 含 `texture_quality`、`geometry_quality`、`pbr`、`texture`。 +将文本、图片、视频映射到同一语义空间，支持跨模态检索、相似度计算与聚类： -## 注意事项 +- 端点：`POST .../services/embeddings/multimodal-embedding/multimodal-embedding`，输入为 `input.contents` 数组，元素支持 `text`、`image`（URL 或 Base64）、`video`（仅 URL）。 +- 支持**独立向量**（每个输入各生成一个向量）与**融合向量**（`qwen3-vl-embedding` 通过 `parameters.enable_fusion=true` 开启；`qwen2.5-vl-embedding` 仅支持融合向量）。 +- 代表模型：`qwen3-vl-embedding`、`multimodal-embedding-v1`、`tongyi-embedding-vision-*` 系列。 -- **地域隔离**：模型、Endpoint URL 与 API Key 必须属于同一地域，华北2（北京）、新加坡、美国（弗吉尼亚）等地域各自独立、不可混用；部分能力（如 Tripo 3D、Fun-Music）仅在特定地域可用。推荐迁移到业务空间专属域名（如 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com`）以获得更好性能与稳定性。 -- **产物有效期**：图像结果 URL 有效期 24 小时，3D 模型下载链接仅 2 小时，务必及时下载。 -- **计费**：图像等仅对成功生成的输出计费，输入与失败任务不计费；具体计费、上下文窗口等实时参数以模型广场为准。 +### 多模态理解与文件输入 + +- 视觉理解首选 `qwen3.7-plus`：1M 上下文、单图最高 1600 万像素、视频最长 2 小时 / 2GB。图片 Token 消耗约 `h x w / (32 x 32) + 2`。 +- 需要文件输入的多模态理解、文档解析等能力，需先通过文件管理 API 上传文件获取文件标识，再在模型调用中引用（"先上传、再引用、后清理"）。 +- 全模态选型：首选 `qwen3.5-omni-plus`，低成本可选 `qwen3-omni-flash`。 + +## 关键参数与限制 + +| 参数/限制 | 说明 | +| --- | --- | +| `modalities` | Realtime 会话的输出模态，`["text"]` 或 `["text","audio"]` | +| 音频格式 | Omni Realtime 输入仅 16 kHz PCM，输出仅 24 kHz PCM | +| 图像输入 | Realtime 场景 JPG/JPEG，建议 480p/720p，Base64 后 ≤256KB，约 1 张/秒，且须在发送过音频后再发图像 | +| `enable_fusion` | 多模态向量是否生成融合向量（`qwen3-vl-embedding`） | +| `fps` / `max_video_frames` | 多模态向量视频抽帧控制 | +| 内容条数 | 如 `qwen3-vl-embedding` 单次内容元素 ≤20、图片 ≤10 张、视频 ≤1 个 | + +## 开发者建议 + +- **实时对话**选 Omni Realtime 系列并按端侧环境选协议；**离线理解**选视觉理解模型 + 文件管理 API；**跨模态检索**选多模态 Embedding。 +- 多模态调用的计费和[限流](rate-limit.md)通常按模态分别计量（如图片按像素折算 Token），集成前先确认目标模型支持的模态组合与输入规格。 ## 关联主题页 - [omni realtime api](../api/omni-realtime-api.md) -- [image generation](../api/image-generation.md) - [video generation api](../api/video-generation-api.md) -- [3d generation](../api/3d-generation.md) +- [vector and sort](../api/vector-and-sort.md) +- [file management api](../api/file-management-api.md) +- [realtime api user guide](../guides/realtime-api-user-guide.md) - [model experience](../guides/model-experience.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/openai-compatible-interface.md b/skills/bailian-docs-llm-wiki/wiki/concepts/openai-compatible-interface.md index a4fbcd2b..8ab9a39c 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/openai-compatible-interface.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/openai-compatible-interface.md @@ -1,41 +1,39 @@ # OpenAI 兼容接口 -OpenAI 兼容接口是百炼平台提供的一套遵循 OpenAI API 规范的服务入口，让已有 OpenAI 应用只需替换 `api_key`、`base_url` 和 `model` 三项即可迁移到百炼，无需改动业务逻辑，是接入成本最低的调用方式。 +OpenAI 兼容接口是阿里云百炼提供的一组与 OpenAI API 规范对齐的调用协议（含 Chat Completions、Responses、Vision、Embedding、Completions、Files、Batch、Conversations 等），开发者只需替换 `base_url`、`api_key` 和 `model` 三个参数，即可将基于 OpenAI SDK 或 LangChain 等框架的现有应用平滑迁移到百炼，无需改写业务代码。 -## 在百炼平台的使用场景 +## 接口形态 -OpenAI 兼容接口贯穿百炼的多类使用场景： +| 接口 | 定位 | +| --- | --- | +| Chat Completions | 与 OpenAI 客户端库直接兼容，迁移成本最低的标准对话接口 | +| Responses API | Chat 的演进版本：内置联网搜索、代码解释器、网页提取等工具，通过 `previous_response_id` 自动管理对话历史（响应 id 有效期 7 天） | +| Vision | 视觉理解（Qwen-VL、QVQ、Qwen-OCR），通过 `image_url` 传图片 URL 或 Base64 | +| Embedding | 文本向量（text-embedding-v1~v4），v3/v4 支持 `dimensions` 指定维度 | -- **文本生成模型调用**：作为四类接口（OpenAI 兼容 Chat Completions、OpenAI 兼容 Responses、Anthropic 兼容 Messages、DashScope 原生）中迁移成本最低的一类，适合已基于 OpenAI SDK 构建的应用平滑迁移。其中 Chat Completions 为最常用入口，支持非流式、流式与工具调用（function call）；Responses 为其演进版本，内置联网搜索、代码解释器、网页抓取等工具，并通过 `previous_response_id` 自动管理多轮上下文。 -- **接入第三方客户端与开发工具**：Cherry Studio、Chatbox、Cursor、Cline、Dify 等聊天客户端和编程工具，统一通过「Base URL + API Key + 模型 ID」以 OpenAI 兼容协议接入百炼网关。 -- **专用模型调用**：`tongyi-intent-detect-v3`（意图理解）、`qwen-mt-plus`（翻译）、`qwen3.5-ocr`（OCR）、`gui-plus`（界面交互）等专用模型多数支持 OpenAI 兼容接口调用（注意 `qwen-deep-research` 仅支持 Python DashScope SDK，Qwen-Audio 仅支持 DashScope 协议）。 -- **智能体与工作流应用调用**：应用可通过 OpenAI 兼容的 Responses API 调用，支持同步/异步、多轮对话与[流式输出](streaming-output.md)。 +与之并列的还有 DashScope 原生接口（功能与参数最全）和 Anthropic 兼容 Messages 接口；OpenAI 兼容接口的优势在于生态兼容与最低迁移成本。 -## 关键参数与配置 +## 在百炼平台的使用场景 -迁移与调用的核心是配置以下三要素： +- **模型调用**：Qwen 商业版/开源版、Qwen-VL、Qwen-Coder、Qwen-Omni、Qwen-Math 及 DeepSeek、Kimi、GLM 等第三方模型均可通过该接口调用；支持[流式输出](streaming.md)（`stream=True`，配合 `stream_options={"include_usage": True}` 返回 token 用量）与 function call（`tools` 参数）。 +- **专用模型**：意图理解（tongyi-intent-detect-v3）、翻译（qwen-mt-plus，通过 `extra_body` 传 `translation_options`）、OCR（qwen3.5-ocr）、GUI 交互（gui-plus）等大多支持 OpenAI 兼容调用。 +- **应用调用**：智能体/工作流应用可通过 OpenAI 兼容的 Responses API 调用（`.../apps/agent/{APP_ID}/compatible-mode/v1/responses`），支持多轮对话、[多模态](multimodal.md)输入与 `background=true` 异步执行。 +- **第三方工具接入**：Codex、Cursor、Cline、Qwen Code、Cherry Studio、Dify 等工具选择 "OpenAI Compatible" 提供方并填入百炼端点即可接入。 -- **`api_key`**：使用百炼 API Key。各地域、各计费方案的 API Key **相互独立、不能混用**，建议配置到环境变量 `DASHSCOPE_API_KEY`，避免硬编码泄露。若 Base URL 与 API Key 不配套会返回 401。 -- **`base_url`**：OpenAI SDK 调用统一以 `/compatible-mode/v1`（部分方案为 `/v1`）结尾；HTTP 调用需在其后追加具体资源路径（如 `/chat/completions`、`/responses`、`/embeddings`、`/files`）。各地域 SDK Base URL 示例： +## 关键参数与配置 + +- **base_url**：统一格式为 `https:///compatible-mode/v1`，按地域区分，例如： - 华北2（北京）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` - 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` - 日本（东京）：`https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1` - - 德国（法兰克福）：`https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/compatible-mode/v1` - 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1` - - 其中 `{WorkspaceId}` 为业务空间 ID，可在控制台业务空间详情页查看。北京、新加坡地域推荐使用业务空间专属域名以获得更好的性能与稳定性。 -- **`model`**：替换为百炼支持的模型名称（如 `qwen-plus`、`qwen3-max` 等）。 - -常用请求参数： - -- `messages`（array，必选）：对话消息列表，每条含 `role`（system/user/assistant）与 `content`。 -- `stream`（bool，可选）：是否[流式输出](streaming-output.md)；流式统计 Token 需配合 `stream_options={"include_usage": True}`。 -- 部分专用模型的特有参数（如 Qwen-MT 的 `translation_options`、Qwen-OCR 的 `min_pixels`/`max_pixels`）在 OpenAI SDK 中通过 `extra_body` 传入。 + - Token Plan：`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1`；Coding Plan：`https://coding.dashscope.aliyuncs.com/v1` +- **api_key**：建议写入环境变量 `DASHSCOPE_API_KEY`。各地域、各计费方案（按量付费 / Token Plan / Coding Plan）的 API Key 互不通用，必须与 base_url 配套，否则返回 401。 +- **model**：目标模型名，如 `qwen3.7-plus`；美国地域使用带 `-us` 后缀的模型名。 +- **扩展参数**：百炼专属参数（如 `translation_options`、`vl_high_resolution_images`、`enable_thinking`）通过 OpenAI SDK 的 `extra_body` 传入。 ## 调用示例 -以北京地域业务空间专属域名为例（Python）： - ```python import os from openai import OpenAI @@ -45,31 +43,24 @@ client = OpenAI( base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( - model="qwen-plus", - messages=[ - {"role": "system", "content": "You are a helpful assistant."}, - {"role": "user", "content": "你是谁？"}, - ], + model="qwen3.7-plus", + messages=[{"role": "user", "content": "你好"}], ) -print(completion.choices[0].message.content) ``` -OpenAI Python SDK 要求 Python ≥ 3.8。 - ## 注意事项 -- **协议路径区分**：OpenAI 协议 Base URL 以 `/compatible-mode/v1`（或 `/v1`）结尾，Anthropic 协议以 `/apps/anthropic` 结尾；部分工具还要求在 Anthropic 端点后追加 `/v1`，以各工具原文为准。 -- **接口能力差异**：DashScope 原生接口参数最全；若依赖联网搜索、代码解释器等内置工具，需使用 Responses 而非普通 Chat Completions。跨接口迁移时需核对参数映射。 -- **旧路径迁移**：Responses、Conversations 接口的旧版路径 `/api/v2/apps/protocols/compatible-mode/v1/...` 即将停止维护，请迁移至新版 `/compatible-mode/v1/...`。 -- **地域约束**：Base URL、API Key 和模型列表均不能跨地域混用；限流按主账号维度合并计算。 +- **并非全量兼容**：Qwen-Audio 仅支持 DashScope 协议；qwen-deep-research 仅支持 Python DashScope SDK；[多模态](multimodal.md) Embedding 模型（如 qwen3-vl-embedding）不支持 OpenAI 兼容接口。跨接口迁移时需核对参数映射，DashScope 原生接口参数最全。 +- **域名迁移**：北京/新加坡地域推荐使用[业务空间](workspace.md)专属域名，旧域名 `dashscope.aliyuncs.com` / `dashscope-intl.aliyuncs.com` 仍可用但建议迁移；Responses API 旧路径 `/api/v2/apps/protocols/compatible-mode/v1/responses` 即将停止维护，应迁移到 `/compatible-mode/v1/...` 新路径。 +- **地域差异**：各地域可用模型、价格与 API Key 均不同；DeepSeek 等第三方直供模型仅限中国站中国内地且需先在控制台开通。 ## 关联主题页 - [qwen api reference](../api/qwen-api-reference.md) - [toolkits and frameworks](../api/toolkits-and-frameworks.md) -- [use chat client or development tool](../guides/use-chat-client-or-development-tool.md) -- [get started with models](../guides/get-started-with-models.md) - [more models](../api/more-models.md) +- [get started with models](../guides/get-started-with-models.md) - [application call](../api/application-call.md) +- [use chat client or development tool](../guides/use-chat-client-or-development-tool.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/prompt-engineering.md b/skills/bailian-docs-llm-wiki/wiki/concepts/prompt-engineering.md index 6241b301..13bea8d8 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/prompt-engineering.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/prompt-engineering.md @@ -1,74 +1,77 @@ -# 提示词工程 +# Prompt工程 -提示词工程（Prompt Engineering）是通过设计、组织与优化 Prompt 来引导大模型生成符合预期结果的方法论。在百炼平台中，它贯穿智能体配置、[工作流](workflow.md)节点、模型直调与多模态生成等几乎所有 LLM 场景，并提供模板化管理、自动优化、样例库与反馈优化等成熟能力。 +Prompt工程是指在百炼平台中通过结构化、模板化、自动优化等手段系统性地设计和改进提示词（Prompt），以提升大模型输出的准确性、稳定性和可控性。百炼围绕 Prompt 的全生命周期提供了一整套工具链，涵盖模板管理、自动优化、反馈迭代、样例注入以及多模态结构化指南。 -## 在百炼中的使用方式 +## 在百炼平台中的使用场景 -### 1. 直接编写 System Prompt +### Prompt 模板管理 -最基础的形态是在智能体应用的「系统提示词」中定义角色、行为指令与能力边界，支持通过 `/` 引用自定义变量。[工作流](workflow.md)应用的大模型节点同样通过「提示词 + 用户提示词」驱动推理。Prompt 越清晰、具体、无歧义，模型表现越稳定。 +Prompt 模板将提示词的固定结构与动态变量分离，实现可复用的统一管理。模板分为**预置模板**和**自定义模板**两类： -### 2. Prompt 模板（结构与变量分离） +- **预置模板**：由平台提供，涵盖营销文案、摘要抽取、文案润色、商品评论等通用场景，已优化且效果稳定，不支持直接修改，但可复制后编辑。 +- **自定义模板**：支持控制台创建（"自定义创建"或"基于 Prompt 工程创建"）和 API 创建（`CreatePromptTemplate` 接口）。自定义模板支持文本生成和图片生成两种类型。 -将固定结构与动态变量分离，统一管理、复用，是团队协作与版本一致性保障的推荐做法。入口位于控制台「应用开发 > 组件管理 > 提示词」。 +模板使用方式包括：控制台中点击"创建应用"将模板内容自动填充到智能体应用的提示词编辑框；或通过 `GetPromptTemplate` 接口拉取模板内容，将业务数据填入变量后生成最终 Prompt 发送给目标模型。提示词最大支持 6144 个字符。 -- **预置模板**：平台提供，覆盖创意文案、办公助理等通用场景，效果稳定、不可修改。 -- **自定义模板**：用户自行设计，支持「自定义创建」（直接粘贴现成 Prompt，可选「优化 Prompt」润色）和「基于 Prompt 工程创建」（选择 ICIO / CRISPE / RASCEF 框架，按字段结构化填写）两种模式。 +> Prompt 模板功能目前仅适用于华北2（北京）地域。 -框架选型建议： +### Prompt 自动优化 -| 框架 | 适用场景 | -| --- | --- | -| ICIO | 简单、明确的任务执行，如数据分析、内容生成、文本摘要 | -| CRISPE | 需要 AI 扮演特定角色的交互，如智能客服、创意写作 | -| RASCEF | 涉及多步骤的复杂业务流程，如项目规划、战略分析 | +当手动编写高质量 Prompt 成本较高时，可使用自动优化功能。该功能利用大模型对原始 Prompt 进行分析和重写，优化策略包括结构重组、角色扮演引导、指令增强、安全与边界注入。操作路径为"应用开发 > 组件管理 > 提示词 > 自动优化"。该功能不额外计费，且提交的数据不会被存储或用于模型训练。 -> 注意：Prompt 模板相关功能仅适用于华北2（北京）地域，使用前请确认业务空间所在地域。 +### Prompt 反馈优化 -### 3. Prompt 自动优化 +反馈优化基于用户提供的输入输出样例进行多轮自动化评估和迭代，生成更贴合实际业务场景的 Prompt。流程为：选择推理模型（推荐千问-max）→ 输入初始 Prompt → 上传样例数据（建议 5-10 条）→ 上传评测数据（建议至少 20 条）→ 系统自动多轮评测与优化。优化后的 Prompt 包含原始 Prompt、few-shot 样例和自动生成的内容提示三部分。 -当缺乏经验或手动编写耗时，可在「提示词 > 自动优化」页面输入原始 Prompt，由大模型进行结构重组、角色扮演引导、指令增强、安全与边界注入等重写，生成结构更优的新版本。该功能不计费，提交数据不会被存储或用于训练。优化失败常见原因：输入超长超出 Token 限制、触发内容审核、网络或服务临时不可用。 +### Prompt 样例库 -### 4. Prompt 样例库（Few-shot 检索） +采用少样本学习（Few-shot learning）思路，从预定义的高质量问答对中检索相关样例注入模型上下文，引导模型生成更准确、风格更一致的回复。适用于智能客服、特定领域知识问答、格式化内容生成等场景。该功能已不再维护，推荐将数据迁移到 RAG 表格库。 -针对特定领域专业任务，从预定义的高质量问答对中检索相关样例注入上下文，引导模型生成更准确、风格更一致的回复。适用于智能客服、特定领域问答、格式化内容生成。注意该功能已不再维护，官方推荐迁移到 RAG 表格库。 +### 多模态 Prompt 指南 -### 5. Prompt 反馈优化 +百炼为不同模态提供结构化 Prompt 指南，核心思路一致：用框架化结构替代模糊描述，描述越完整精确，生成质量越高。 -基于输入输出样例与评测数据，多轮自动评估、反思、优化 Prompt，涉及推理调用。适合对输出质量有持续提升需求的闭环场景。 +- **文生文**：推荐"背景、目的、风格、语气、受众、输出"六要素框架。 +- **文生图**：基础公式（主体 + 场景 + 风格）和进阶公式（主体描述 + 场景描述 + 定义风格 + 镜头语言 + 氛围词 + 细节修饰），支持 `prompt_extend` 智能改写和 `negative_prompt` 反向提示词。 +- **文生视频**：在图像基础上增加"运动"维度，进阶公式为"主体描述 + 场景描述 + 运动描述 + 美学控制 + 风格化"，支持多镜头连贯叙事和参考生视频。 +- **Vidu 视频生成**：公式为"主体/场景 + 场景描述 + 环境描述 + 艺术风格/媒介"，通过关键词词典触发动态控制、运镜和特效。 -## 关键参数与配置 +### 模型评测中的 Prompt -### 智能体应用 +模型评测场景中涉及两类 Prompt：**评分器 Prompt**配置于评测维度，指导裁判模型如何对被评测模型的回答打分，长度不超过 50000 字符，至少包含一个变量（`${prompt}`、`${output}`、`${completion}`）；**System Prompt**配置于评测任务，为被评测模型设定角色定位或行为规范，通常可留空。 -- **系统提示词最大长度**：6144 字符（从模板「使用 [prompt](../guides/prompt.md) > 创建应用」时自动填充到此上限）。 -- **模型参数**：`temperature`、最长回复长度、`enable_thinking`（开启思考模式以提升反思效果，仅支持思考模式模型）。 -- **变量引用**：在系统提示词中通过 `/` 嵌入自定义变量，运行时由业务数据填充。 +## Prompt 工程框架 -### 多模态生成 +百炼平台内置三种 Prompt 工程框架，可在创建自定义文本生成模板时选用： -- **文生图**：`prompt`（正向）、`negative_prompt`（反向，描述不希望出现的内容）、`prompt_extend`（V2 专用，大模型智能改写，默认 `true`）。基础公式 `主体 + 场景 + 风格`，进阶公式追加镜头语言、氛围词与细节修饰。 -- **文生视频**：基础公式 `主体 + 场景 + 运动`，进阶公式 `主体描述 + 场景描述 + 运动描述 + 美学控制 + 风格化`；图生视频简化为 `运动 + 运镜`。wan2.7 起单/多镜头由提示词控制，不再使用 `shot_type`。 +| 框架 | 组成要素 | 适用场景 | +|------|----------|----------| +| **ICIO** | 指令、背景信息、补充数据、输出格式 | 简单明确的任务，如数据分析、内容生成、文本摘要 | +| **CRISPE** | 角色与能力、背景信息、任务、输出风格、输出范围 | 需要 AI 扮演特定角色的交互，如客服、创意写作 | +| **RASCEF** | 角色、行动、步骤、上下文、示例、格式 | 多步骤复杂业务流程，如项目规划、战略分析 | -### 模板调用 API +## 关键参数与限制 -- **创建模板**：`CreatePromptTemplate`，需先获取 Workspace ID。 -- **获取模板**：`GetPromptTemplate`，传入 `workspaceId` 与 `promptTemplateId`，返回 `content`、`variables` 等字段，在代码中填充变量后调用模型。相比字符串拼接，可实现逻辑与内容分离、集中管理与版本一致。 +| 参数/限制项 | 值 | +|-------------|-----| +| 提示词最大字符数 | 6144 | +| 评分器 Prompt 最大长度 | 50000 字符 | +| 单个样例库容量 | 300 条样例 | +| 单应用关联样例库数 | 5 个 | +| 单次召回片段数 | 最多 10 个 | +| 批量导入文件大小 | 20MB（Excel） | +| 单次导入条数 | 100 条 | +| Prompt 模板可用地域 | 仅华北2（北京） | -## 设计要点 +## API 接口 -1. **结构化优先**：复杂任务优先采用 ICIO/CRISPE/RASCEF 框架或「背景-目的-风格-语气-受众-输出」六要素，避免笼统指令。 -2. **变量分离**：把不可变的结构与可变的业务数据拆开，通过模板或 `/` 变量注入，便于复用与协作。 -3. **样例引导**：对风格或格式敏感的任务，用 Few-shot 样例或反馈优化建立闭环，持续校准输出。 -4. **正反向结合**：图片与视频生成场景同时使用正向与反向 Prompt 精确控制画面内容。 -5. **迭代验证**：结合在线调试面板与评测能力，对 Prompt 变更做回归验证后再发布。 +Prompt 模板的生命周期管理通过应用组件 OpenAPI（`bailian/2023-12-29`）实现，核心接口包括 `CreatePromptTemplate`、`GetPromptTemplate`、`UpdatePromptTemplate` 等。调用时需提供 `workspaceId`（业务空间 ID），RAM 子账号需授予百炼 API 权限并加入对应业务空间。接口采用 ROA 签名风格，官方已封装多语言 SDK。 ## 关联主题页 - [prompt](../guides/prompt.md) -- [start using](../guides/start-using.md) -- [llm application](../guides/llm-application.md) - [use cases](../guides/use-cases.md) +- [model evaluation introduction](../guides/model-evaluation-introduction.md) - [application component api reference](../api/application-component-api-reference.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/rag.md b/skills/bailian-docs-llm-wiki/wiki/concepts/rag.md index f5102324..b59dac57 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/rag.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/rag.md @@ -1,62 +1,71 @@ # 检索增强生成（RAG） -检索增强生成（Retrieval-Augmented Generation, RAG）是一种在大模型生成回答前，先从外部知识库检索语义相关内容、再将其与用户问题一起送入模型的技术。它为大模型补充私有数据和最新信息，从而显著提升在特定领域问题上的准确性、降低幻觉。 +检索增强生成（Retrieval-Augmented Generation，RAG）是一种在大模型生成回答前，先从知识库中检索相关内容切片并注入上下文的技术，用于把企业私有数据、业务文档和最新信息接入大模型应用，提升回答的准确性与可追溯性。在百炼平台上，RAG 以知识库为核心载体，贯穿智能体、工作流、知识检索/知识问答服务以及开源框架集成等多种使用形态。 + +## 工作原理 + +一次典型的 RAG 调用包含以下环节： + +1. **数据接入**：将 PDF、DOCX、Markdown、Excel、图片、音视频等数据解析、切片、向量化后存入知识库（向量模型如 `text-embedding-v4`、`multimodal-embedding-v1`）。 +2. **检索**：用户 query 经向量检索 + 关键词检索召回候选切片，再由排序模型（如 `qwen3-rerank`、`qwen3-vl-rerank`）重排。 +3. **过滤与拼装**：按相似度阈值、标签/元数据过滤、召回片段数筛选切片，拼入提示词。 +4. **生成**：大模型（千问 Max/Plus/Turbo、千问 VL、DeepSeek 等）基于检索内容生成回答。 ## 在百炼平台的使用场景 -百炼把 RAG 拆成「建立索引 → 检索召回 → 生成答案」三个阶段，并在不同层面提供了对应能力： +### 挂载到智能体应用 -- **云端知识库（控制台）**：进入知识库按「填写基础信息 → 配置数据来源 → 设置索引参数」三步建库，创建时选定类型（文档搜索 / 数据查询表格库 / 图片问答 / 音视频搜索，创建后不可更改），随后关联到智能体应用、工作流应用或外部应用。工作流应用中知识库节点须接在开始节点之后、大模型节点之前，并在大模型提示词中引用 `result` 变量。注意：知识库功能仅在中国站**华北2（北京）**地域可用。 -- **知识检索服务**：面向多知识库联合检索（最多 15 个），提供 Query 改写、混合检索（向量+关键词）、Rerank 排序的流水线。 -- **知识问答服务**：在检索基础上由大模型生成自然语言回答，提供**极速模式**（单轮检索+生成）与**多轮智能模式**（Agentic 多轮规划搜索），并支持文件预解析、拒答、防泄漏、多模态回复、引用来源等生成控制。 -- **应用场景接入**：围绕「RAG + 智能体应用」可将问答能力接入网站、企业微信、微信公众号、钉钉等渠道；也支持基于本地知识库构建 RAG 应用（检索在本地执行、生成调用通义千问 API），适合需要灵活切分与自定义嵌入模型的场景。 -- **框架集成**：LlamaIndex（Python）可构建云端知识库与 RAG 应用；Spring AI Alibaba（Java）可集成智能体/工作流应用并检索百炼知识库。 +在智能体应用中添加文档知识库，配置调用方式（如"必定调用"）、相似度阈值、权重、标签/元数据过滤和召回片段数。新版智能体（Agent 2.0）将知识库统一视为工具，由模型自主规划调用时机，并支持通过文件标签 + 提示词规则限定检索范围。注意检索文本会占用上下文窗口并增加输入 Token 消耗。 -## HTTP REST 接口 +### 挂载到工作流应用 -除控制台外，百炼提供 DashScope 应用网关体系的两个 REST 接口，用 API Key Bearer 鉴权，Base URL 形如 `https://{workspaceId}.cn-beijing.maas.aliyuncs.com`： +在工作流中添加"知识库"节点，把输入变量（通常是 `${sys.query}`）传给检索节点，再将检索结果交给大模型节点生成。适合需要组合检索、工具调用、条件分支的固定流程。 -| 接口 | 路径 | 说明 | -| --- | --- | --- | -| 知识检索 | `POST /api/v1/indices/knowledge/search` | 跨多个知识库联合语义检索，返回按相关性排序的切片，适合需自定义生成流程的场景 | -| 知识问答 | `POST /api/v2/apps/knowledge/chat` | 基于知识库的智能问答，通过 SSE 流式返回规划、工具调用、生成三个阶段 | +### 知识检索 / 知识问答服务 -默认用户维度 25 QPS。此外还有 `CreateIndex`、`Retrieve` 等 OpenAPI RPC 接口用于建库流程。 +- **知识检索**（`POST /api/v1/indices/knowledge/search`）：只返回排序后的切片，由开发者自行拼接 [prompt](../guides/prompt.md) 并调用大模型，适合需要自定义生成流程的场景。 +- **知识问答**（`POST /api/v2/apps/knowledge/chat`）：服务端自动完成规划、检索、生成，通过 SSE 流式返回，适合开箱即用的端到端问答。 -## 关键参数与配置 +两类服务均支持最多绑定 15 个知识库，可配置权重、路由、TopK、排序模型、阈值和过滤条件；API 使用[业务空间](workspace.md)专属域名 `https://{workspaceId}.cn-beijing.maas.aliyuncs.com`，Bearer [API Key 鉴权](api-key.md)，默认[限流](rate-limit.md) 25 QPS。 + +### 通过开源框架构建 -检索效果主要由以下参数决定，可在命中测试、检索服务与问答服务中反复调优： +- **LlamaIndex（Python）**：读取本地文件上传构建云端知识库，用 `as_retriever()` / `as_query_engine()` 构建检索与问答链路；云端方案使用官方智能切分与向量模型，不支持自定义切分或嵌入模型。 +- **Spring AI Alibaba（Java）**：集成百炼智能体/工作流应用并检索百炼知识库，通过 `DASHSCOPE_API_KEY` / `AI_DASHSCOPE_API_KEY` 等环境变量配置鉴权。 +- **本地知识库方案**：检索在本地执行、生成调用通义千问 API，适合需要自定义切分与嵌入模型（如本地 GTE 模型）的场景。 -- **相似度阈值（0.01~1.0）**：仅语义相似度高于阈值的切片会被召回。阈值过高会导致相关切片被全部丢弃（如调至 0.60 可能无召回）。 -- **初步向量检索 TopK / 初步关键词检索 TopK（1~100，默认各 50）**：控制初步召回数量，直接影响送入 Rerank 的 Token 量与成本。 -- **最大召回数量 / 召回片段数（1~20）**：最终提供给大模型的切片数，对总结、列举、比较类复杂问题应适当调大。 -- **权重**：多知识库联合召回时干预排序，仅在**同类型知识库之间**生效。 -- **排序模型（Rerank）**：纯文本可选 `qwen3-rerank` / `qwen3-rerank(hybrid)`，多模态可选 `qwen3-vl-rerank`；`gte-rerank` 将于 2026-05-30 下线，新项目直接选 `qwen3-rerank`。 -- **向量模型（Embedding）**：将文本/图片/视频编码到同一语义空间，供余弦相似度匹配。云端知识库使用官方向量模型；本地方案可改用自部署 GTE 模型。 -- **Meta 信息抽取与标签过滤**：在向量检索前做结构化筛选，精准定位目标文件；元数据只能在创建知识库时配置，创建后无法开启。 +### 接入业务渠道 -在框架（LlamaIndex）中对应的参数为：`similarity_top_k`（召回数）、`similarity_cutoff`（最低相似度阈值）、`top_n`（重排后返回数），可通过 `node_postprocessors`（`SimilarityPostprocessor`、`DashScopeRerank`）做后处理。 +基于 RAG 的智能体应用可通过 AppFlow 快速接入网站、企业微信、微信公众号、钉钉等渠道，为应用添加私有知识库后即可覆盖私域问题。 -## 效果优化建议 +## 关键参数与配置 -优化前建议用自动评测建立至少 100 组用例的基线，再针对失败用例（打分 < 4）诊断改进： +| 参数 | 说明 | +| --- | --- | +| 相似度阈值 | 仅召回分数高于阈值的切片；过高漏召回，过低引入噪声 | +| 召回片段数 / TopK | 复杂问题可增大，但会增加上下文长度与 Token 成本 | +| 权重 | 多知识库联合检索时，相似度接近的切片优先返回高权重库 | +| 标签过滤 / 元数据过滤 | 限定检索范围，减少无关切片召回 | +| 排序（rerank）模型 | 对召回结果重排，如 `qwen3-rerank`、`gte-rerank` | +| 向量模型 | `text-embedding-v4`/`v3`（512 维）、`multimodal-embedding-v1`（1024 维），创建后不可修改 | +| Meta 信息抽取 / 多轮对话改写 | 需在创建知识库时配置，后补通常需重建知识库 | -- **检索无效（没找到）**：补充知识、优化源文件排版（推荐转 Markdown、移除水印、避免复杂表格）、统一实体表述、启用多轮对话改写。 -- **召回不相关**：使用标签过滤或元数据做结构化搜索。 -- **切片不完整**：采用「智能切分」（基于语义自适应切分），并人工检查修正异常切片。 -- **重排不佳**：在漏召回与噪声之间平衡相似度阈值与召回片段数。 +LlamaIndex 侧对应参数为 `similarity_top_k`、`similarity_cutoff`、`top_n`（重排返回数），并可通过 `node_postprocessors` 做相似度过滤与重排后处理。 -## 监控 +## 效果优化与注意事项 -所有检索调用以日志形式投递到日志服务（SLS），topic 为 `log_dispatch`，含 `request_id`、`pipeline_id`（知识库 ID）、`workspace_id`、`latency`、`response_status_code` 等字段，可用于审计、用量统计与慢查询/错误率监控。SLS 存储与流量单独计费，关闭检索日志开关只停止新投递，历史日志仍保留计费。 +- **调优闭环**：通过评测集、命中测试、标签/元数据过滤、切片修正、阈值与召回片段数调整来定位并优化 RAG 效果。 +- **监控**：检索调用可投递到 SLS，按知识库、[业务空间](workspace.md)、API 路径、错误码、耗时等维度分析告警。 +- **地域限制**：知识库相关功能仅在中国站华北2（北京）地域开通和使用。 +- **成本**：检索切片计入模型输入 Token；合理设置阈值和召回片段数可在效果与成本间取得平衡。 ## 关联主题页 - [knowledge base](../guides/knowledge-base.md) -- [frameworks](../api/frameworks.md) - [application use cases](../guides/application-use-cases.md) -- [vector and sort](../api/vector-and-sort.md) -- [knowledge](../api/knowledge.md) +- [llm application](../guides/llm-application.md) +- [frameworks](../api/frameworks.md) - [use cases](../guides/use-cases.md) +- [knowledge](../api/knowledge.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/rate-limit.md b/skills/bailian-docs-llm-wiki/wiki/concepts/rate-limit.md new file mode 100644 index 00000000..1e3e5879 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/rate-limit.md @@ -0,0 +1,48 @@ +# 限流 + +限流是百炼平台对模型 API 调用施加的流量控制机制：按**主账号 + 模型**维度限制请求速率与 Token 吞吐，超限请求返回 429 错误。理解限流规则并做好应对，是保障线上业务稳定性的基础工程能力。 + +## 三类限流规则 + +| 类型 | 含义 | 超限错误码 | +| --- | --- | --- | +| RPM / TPM | 分钟级配额：每分钟请求数（Requests Per Minute）与 Token 数（Tokens Per Minute） | `Throttling.RateQuota`（RPM）/ `Throttling.AllocationQuota`（TPM） | +| RPS / TPS | 瞬时频率：每秒请求数与 Token 数 | `Throttling.RateQuota`（RPS）/ `Throttling.AllocationQuota`（TPS） | +| Traffic Burst | 增速限制：短时间内流量拉升过快 | `Throttling.BurstRate` | + +注意维度差异：TPM 按账号维度限流，而 TPS 限流按请求维度（TPS ≈ 1 ÷ 非首包时长均值），排查时不要混淆。 + +## 平台各场景中的限流表现 + +- **接入域名**：不同域名限流待遇不同。[业务空间](workspace.md)专属域名（`{WorkspaceId}.{region}.maas.aliyuncs.com`）面向生产环境，支持高并发与流量隔离（SLA 99.9%）；试用域名 RPM 固定为 1000，仅适合快速验证，无 SLA。 +- **免费额度**：开启「免费额度用完即停」后，额度用尽返回 403 错误 `AllocationQuota.FreeTierOnly`，属于配额停用而非速率限流，需注意区分。 +- **TPM 预留**：预付费锁定专属推理容量，不与其他用户共享、不受公共限流影响。溢出策略二选一：自动溢出至按量计费（默认，业务不中断），或仅使用预留容量（超出请求直接返回 429）。 +- **快速模式（Fast mode）**：采用特殊限流策略——超出 TPM 额度不立即拒绝，请求进入排队队列。 +- **模型监控**：监控详情页的「错误」类指标含限流错误次数（429），可按 API-KEY、时间范围筛选，用于定位限流发生的时段与来源。 + +## 限流应对三层方案 + +按改动成本从低到高： + +1. **平台配置**： + - 服务端排队等待：请求头 `X-DashScope-Wait-Timeout`（建议 3~120 秒），仅对 `Throttling.BurstRate` 生效；同时需上调客户端超时（非流式：原超时 + Wait-Timeout 值）。 + - 申请提升限流额度、购买 PTU 专属部署，或将离线任务迁移到 Batch API。 +2. **客户端流控**：从基础重试（指数退避）、令牌桶 / 并发信号量，到双重令牌桶 / 平滑限速器、自适应拥塞控制，按工程复杂度递进选择。 +3. **架构兜底**：模型降级（Fallback 到备用模型）与基于消息队列的削峰填谷。 + +## 容量类方案选型 + +- 流量可预估、不能接受限流 → **TPM 预留**（按 kTPM 预付费、按天计费，创建后用专属模型 code 替换 `model` 参数即可）。 +- 追求更高吞吐与性能确定性 → **PTU 专属部署**。 +- 流量波动大或短期使用 → 维持按量付费，叠加客户端流控与重试；可用资源包 / 节省计划优化费用。 + +> 提示：即使使用 TPM 预留，短时间内请求量快速拉升时系统仍需短暂预热，预热期间部分请求可能出现延迟波动，建议保留请求排队或重试机制。 + +## 关联主题页 + +- [get started with models](../guides/get-started-with-models.md) +- [model high speed inference](../guides/model-high-speed-inference.md) +- [use cases](../guides/use-cases.md) +- [model monitoring](../guides/model-monitoring.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/rate-limiting.md b/skills/bailian-docs-llm-wiki/wiki/concepts/rate-limiting.md index c43445dd..d713cb47 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/rate-limiting.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/rate-limiting.md @@ -1,71 +1,139 @@ # 限流与配额 -限流与配额是百炼平台对模型推理调用施加的吞吐量与调用频率约束，用以保障公共推理资源的公平使用和整体稳定性；当业务流量超过共享上限或需要刚性容量保障时，可通过专属域名、TPM 预留、PTU 部署或异步通知等手段规避限流影响。 +限流与配额是百炼平台对各 API 接口、模型调用吞吐量及资源使用量施加的访问控制机制，用于保障平台整体稳定性，同时为开发者提供可预期的服务边界。在百炼的模型推理、异步任务管理、文件上传、应用组件 API 等不同场景中，限流的维度、阈值和处理策略各不相同。 -## 限流的产生场景 +## 限流的类型与维度 -百炼模型调用主要在以下维度受到约束： +百炼平台的限流主要分为以下几类： -- **TPM（Tokens Per Minute）**：每分钟可消耗的 Token 数上限，按主账号 + 业务空间 + 模型维度计算，超出会返回 `429` 限流错误。 -- **RPM（Requests Per Minute）**：每分钟请求数，RPM 越大建议 TPM 同比增大。 -- **QPS 限制**：部分通用接口有独立的 QPS 限制。例如异步任务管理（查询、批量查询、取消）三个接口统一限制为 20 QPS，按主账号 + 子账号维度计算。 -- **公共池共享**：按量付费方案无专属容量，调用进入公共共享池，受公共限流波动影响，业务高峰期可能被限流。 +| 限流类型 | 维度 | 典型场景 | 超限处理 | +| --- | --- | --- | --- | +| QPS 限流 | 每秒请求数（主账号维度） | 异步任务管理、文件上传、应用组件 API | 返回 429，需退避重试 | +| TPM 限流 | 每分钟 Token 数 | 模型推理调用 | 公共资源限流或溢出至按量计费 | +| RPM 限流 | 每分钟请求数 | 模型单元部署 | 部署时自行设定上限 | +| 容量配额 | 预付费锁定的吞吐量 | PTU / TPM 预留 | 自动溢出或返回 429 | -监控页面的「错误」指标中专门提供**限流错误次数（429）**，可用于定位限流问题；「性能」指标中的 RPM、TPM 则帮助评估容量是否接近上限。 +## 模型推理场景 -## 不同方案下的限流行为 +### 公共 TPM 限流 -| 方案 | 容量保障 | 超额处理 | 接入改动 | -| --- | --- | --- | --- | -| 按量付费 | 无（共享公共池） | 自动服务，受公共限流 | 无需改动 | -| 资源包 / 节省计划 | 承诺用量折扣（非专属容量） | 超出转按量 | 无需改动 | -| TPM 预留 | 专属容量刚性兑付 | 超出自动降级公共池按量，不中断 | 替换 `model` 参数为专属模型 code | -| PTU（模型部署） | 专属部署实例 | 超出转按量 | 替换 `model` 参数 | +按量付费调用模型时，请求受公共资源池的 TPM（每分钟 Token 数）限制。不同模型和地域的 TPM 上限不同，高并发场景下可能触发限流。 + +### TPM 预留（容量保障） + +TPM 预留通过预付费方式锁定专属推理吞吐量，预留容量内的调用不与其他用户共享、不受公共资源限流影响。关键参数： + +| 参数 | 说明 | +| --- | --- | +| 输入 TPM | 预留输入吞吐量，单位 kTPM（1 kTPM = 1,000 Tokens/分钟） | +| 输出 TPM | 预留输出吞吐量，单位 kTPM | +| 付费周期 | 当前支持按天 | +| 购买时长 | 支持 1~30、60、90、120、365 天 | +| 溢出策略 | 自动溢出至按量计费（默认）/ 仅使用预留容量（返回 429） | + +溢出策略的两种选择： + +- **自动溢出至按量计费**：超出部分降级为按量计费，服务不中断，详情页可查看超额降级次数。 +- **仅使用预留容量**：超出返回 429，不产生额外费用。 + +创建 TPM 预留后，控制台生成专属 model code，将 API 请求中的 `model` 参数替换为该 code 即可使用预留容量。短时间请求量快速拉升时系统需短暂预热，预热期间可能出现延迟波动。 + +### PTU 预置吞吐 + +PTU（预置吞吐）是另一种容量保障方式，适用于高负载生产环境。PTU 的溢出策略与 TPM 预留一致：可选「自动溢出」（切换为按量付费，业务不中断）或「仅使用 PTU 容量」（超出返回 429）。输入超过模型上限（千问 128K / DeepSeek 64K）时，一律自动转为按量计费。 + +部分 PTU 部署模型支持长输入阶梯系数和前缀缓存折扣，影响额度消耗： + +- **长输入阶梯系数**：超过 32K token 的输入按更高系数折算 TPM 消耗。例如 glm-5.1 在 [32K, 200K] 区间输入系数为 1.33、输出系数为 1.17。 +- **前缀缓存折扣**：命中缓存的输入 token 按折扣系数消耗额度。glm-5.1 折扣为 0.2，deepseek-v4-pro 折扣为 0.08。 + +API 响应中与额度相关的字段： + +| 字段 | 说明 | +| --- | --- | +| `service_tier` | `ptu-standard` 表示使用 PTU 额度；`default` 或不返回表示按量计费 | +| `provisioned_tokens` | 折算后实际消耗的 PTU 额度 token 数（含阶梯系数和缓存折扣） | +| `cached_tokens` | 前缀缓存命中的 token 数 | -当业务流量可预估且不能接受限流时，优先选择 TPM 预留；对极致性能与隔离有更高要求时，可考虑 PTU 专属部署。 +### 模型单元 -## 接入域名对限流的影响 +模型单元部署方式支持自定义性能指标，部署时可设置 `rpm_limit` 和 `tpm_limit` 参数来限制每分钟请求数和每分钟 Token 数。 -接入域名直接影响并发上限与超时表现： +## Base URL 与并发能力 -- **业务空间专属域名** `https://{WorkspaceId}.{region}.maas.aliyuncs.com`：生产推荐，请求超时 3600 秒、SLA 99.9%，提供更高吞吐与时延隔离。 -- **Dashscope 中心化域名** `https://dashscope.aliyuncs.com`：存量业务兼容，可跨业务空间调用。 -- **试用域名** `https://trial.{region}.maas.aliyuncs.com`：仅限快速验证，限流小，不建议生产。 +不同类型的 Base URL 在并发能力和超时设置上有差异，直接影响可获得的限流上限： -各地域的 API Key、模型列表、接入域名不能跨地域混用；美国（弗吉尼亚）暂不支持业务空间专属域名，需使用 `dashscope-us.aliyuncs.com`。 +| 域名类型 | 特点 | 请求超时 | +| --- | --- | --- | +| 业务空间专属（推荐） | 更高并发、更低延迟、空间级隔离 | 3600 秒 | +| Dashscope 域名 | 存量业务兼容，建议迁移 | 600 秒 | +| 试用域名 | 快速验证，限流值小 | - | -## TPM 预留的关键参数 +生产环境应使用业务空间专属域名以获得更高的并发上限和 SLA 保障（99.9%）。 -创建 TPM 预留时需指定输入 TPM 与输出 TPM，单位为 kTPM（1 kTPM = 1,000 Tokens/分钟），起步和步长因模型而异。容量计算受以下参数影响： +## 异步任务管理 API -- **每分钟请求数（RPM）**：业务高峰期每分钟请求数。 -- **平均输入/输出长度（token）**：输入越长，阶梯系数越大，所需输入 TPM 越高。 -- **预估缓存命中率（%）**：命中率越高，输入容量消耗越慢，所需输入 TPM 越低；仅影响输入 TPM。 +图像生成、视频生成等耗时模型采用异步调用机制，相关管理 API 有明确的 QPS 限制： -部分模型支持长输入阶梯系数和缓存折扣（如 glm-5.1 在 [32K, 200K] 区间输入系数 1.33、输出 1.17；deepseek-v4-pro 缓存命中部分按 8% 折算），TPM 容量计算器会自动应用这些参数。 +| 接口 | 限流 | +| --- | --- | +| 查询单个任务 `GET /tasks/{task_id}` | 20 QPS（主账号维度） | +| 批量查询任务状态 `GET /tasks/` | 20 QPS（主账号维度） | +| 取消任务 `POST /tasks/{task_id}/cancel` | 20 QPS（主账号维度） | -## 规避异步任务轮询限流 +批量查询任务状态时，单次查询时间跨度不超过 24 小时。已完成任务通常保留 24 小时后自动清理。 -图像/视频生成等耗时模型采用异步机制，频繁轮询结果接口会浪费资源并可能触发 20 QPS 限流。百炼已接入事件总线 EventBridge，任务完成（无论成功或失败）后主动上报 `dashscope:System:AsyncTaskFinish` 事件，可推送到 HTTP 回调 URL 或 RocketMQ 消息队列。通知方案不限流、实时性高，适合高并发、大规模或对实时性要求高的任务。 +为避免频繁轮询触发限流，推荐使用 EventBridge 事件通知机制，在任务完成后主动推送通知（事件类型 `dashscope:System:AsyncTaskFinish`），收到通知后只需调用一次查询接口即可获取结果。 -## 免费额度与配额停用 +## 文件上传 API -免费额度页面提供「免费额度用完即停」开关：开启后免费额度用尽时服务自动停止，返回 `403 AllocationQuota.FreeTierOnly`，避免产生免费额度以外的费用。仅在账户内仍有未消耗的免费额度时才能开启；关闭需等免费额度完全消耗后进行。 +多模态、图像、视频、音频模型调用时需要传入文件 URL。百炼提供临时存储空间上传本地文件： -## 应对限流的实践建议 +| 限制项 | 值 | +| --- | --- | +| 上传凭证接口限流 | 100 QPS | +| 单文件大小上限 | 1 GB | +| 文件有效期 | 48 小时 | +| 上传与调用的 API Key | 必须属于同一阿里云主账号 | -1. 生产环境优先迁移到业务空间专属域名，获得更高吞吐与时延隔离。 -2. 流量可预估且不能接受限流的核心业务，使用 TPM 预留锁定专属容量。 -3. 通过模型监控的限流错误次数（429）与 RPM/TPM 指标评估容量是否接近上限。 -4. 异步任务改用 EventBridge 通知方案，避免轮询触发 20 QPS 限制。 -5. 需要严格控制成本的场景，可开启「免费额度用完即停」避免超额消费。 +临时 URL 不适用于生产环境，生产环境建议使用阿里云 OSS 等稳定存储方案。 + +## 应用组件 API + +应用组件 API（`bailian/2023-12-29`）的限流普遍为 5-10 次/秒： + +| 接口 | 限流 | +| --- | --- | +| `ApplyFileUploadLease`、`AddFile`、`DescribeFile`、`DeleteFile` | 10 次/秒 | +| `ListCategory`、`AddCategory`、`ListFile`、`AddFilesFromAuthorizedOss` | 5 次/秒 | + +触发限流后需实现退避重试机制。应用组件 API 还存在资源配额限制，如每个业务空间最多 500 个类目、每文件最多 100 个标签（总长 ≤ 700）、批量删除单次最多 20 个文件等。 + +## 临时 API Key + +临时 API Key 具有有限时效，过期自动失效，相关参数： + +| 参数 | 范围 | 默认值 | +| --- | --- | --- | +| `expire_in_seconds` | [1, 1800] 秒 | 60 秒 | + +临时 API Key 继承生成它的永久 API Key 的全部权限，无法手动删除，到期后自动失效。各地域的 API Key 不同，新加坡地域需将 Endpoint 中的 WorkspaceId 替换为实际值。 + +## 最佳实践 + +1. **生产环境使用业务空间专属域名**：获得更高并发上限和 SLA 保障。 +2. **高负载场景使用容量预留**：通过 TPM 预留或 PTU 锁定专属吞吐量，避免公共限流影响。 +3. **异步任务用事件通知替代轮询**：避免频繁轮询触发 QPS 限流。 +4. **实现退避重试机制**：所有有限流的接口都应实现指数退避重试。 +5. **合理规划配额**：使用控制台提供的 TPM/PTU 额度计算器，根据 RPM、平均输入/输出长度、缓存命中率预估所需容量。 +6. **溢出策略选择**：对延迟敏感的业务选自动溢出，对成本敏感的业务选仅使用预留容量。 ## 关联主题页 -- [get started with models](../guides/get-started-with-models.md) -- [model high speed inference](../guides/model-high-speed-inference.md) -- [model monitoring](../guides/model-monitoring.md) - [more about models](../api/more-about-models.md) -- [use cases](../guides/use-cases.md) +- [model high speed inference](../guides/model-high-speed-inference.md) +- [get started with models](../guides/get-started-with-models.md) +- [application component api reference](../api/application-component-api-reference.md) +- [model deployment 1](../guides/model-deployment-1.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/region.md b/skills/bailian-docs-llm-wiki/wiki/concepts/region.md index 3006794c..b255bb69 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/region.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/region.md @@ -1,80 +1,73 @@ -# 地域与可用区 +# 地域与接入点 -地域（Region）是百炼平台在物理地理上划分的部署区域，决定接入点位置、数据存储位置与推理执行位置；可用区是地域内相互隔离的物理设施单元。在百炼中，"地域"主要与**服务部署范围**和**接入域名**共同决定调用路径、数据驻留约束与服务保障等级。 +地域（Region）是阿里云百炼服务部署和数据存储的地理位置，接入点（Base URL / 域名）则是调用该地域服务的网络入口。两者共同决定了你使用哪个 API Key、能调用哪些模型、按什么价格计费，以及数据在哪个国家/地区境内处理。 -## 可用地域 +## 核心规则 -百炼当前支持的地域及其 Region ID： +- **各地域完全隔离**：API Key、可用模型列表、价格、[业务空间](workspace.md)在不同地域之间相互独立，**API Key 不能跨地域使用**。 +- **Base URL 与 API Key 必须配套**：Base URL 必须与同一地域、同一计费方案的 API Key 搭配使用，否则返回 401/403 鉴权失败，或意外走按量计费通道。 +- **部分能力仅限特定地域**：选型前务必确认目标能力在目标地域是否可用。 -| 地域 | Region ID | 数据驻留约束 | -| --- | --- | --- | -| 华北2（北京） | `cn-beijing` | 数据不出中国内地 | -| 新加坡 | `ap-southeast-1` | 数据不经过中国内地 | -| 美国（弗吉尼亚） | — | 数据不出美国 | -| 德国（法兰克福） | `eu-central-1` | 数据不出欧盟 | -| 日本（东京） | `ap-northeast-1` | 数据不出日本 | - -若对驻留无限制但追求更大推理资源池，可选择美国/德国/日本的全球部署范围。 - -## 三个关键概念 - -调用前需同时确认**地域**、**服务部署范围**、**接入域名**： +## 支持的地域 -- **地域**：决定接入点与数据存储位置。 -- **服务部署范围**：决定推理执行位置（数据是否跨域流转）。 -- **接入域名**：影响并发上限、超时、SLA 等服务保障。 - -三者在控制台或 Base URL 中体现，**各地域的接入域名、API Key、模型列表不能跨地域混用**。 +| 地域 | Region ID | 说明 | +| --- | --- | --- | +| 华北2（北京） | `cn-beijing` | 功能最全，DeepSeek 等部分第三方模型仅此地域提供 | +| 新加坡 | `ap-southeast-1` | 国际站主力地域 | +| 日本（东京） | `ap-northeast-1` | 需先创建[业务空间](workspace.md)并选择服务部署范围（全球/日本） | +| 德国（法兰克福） | `eu-central-1` | 需先创建[业务空间](workspace.md)并选择服务部署范围（全球/欧盟） | +| 美国（弗吉尼亚） | `us-east-1` | 暂不支持业务空间专属域名；用带 `-us` 后缀的模型名（如 `qwen-plus-us`）限定美国境内推理 | -## 接入域名类型 +## 三种接入域名（按量付费） -| 域名类型 | 格式 | 适用场景 | 关键指标 | +| 域名类型 | 格式 | 适用场景 | 请求超时 | | --- | --- | --- | --- | -| 专属域名 | `{WorkspaceId}.{region}.maas.aliyuncs.com` | 生产环境 | SLA 99.9%、超时 3600 秒、支持 HTTP/SSE/WebSocket/WebRTC | -| 共享域名 | `dashscope.aliyuncs.com` | 存量兼容 | — | -| 试用域名 | `trial.{region}.maas.aliyuncs.com` | 快速体验 | 不建议生产 | +| 业务空间专属域名（推荐） | `{WorkspaceId}.{region}.maas.aliyuncs.com` | 生产环境，高并发、流量隔离，SLA 99.9% | 3600 秒 | +| Dashscope 域名 | `dashscope.aliyuncs.com` 等 | 存量业务兼容，建议迁移 | 600 秒 | +| 试用域名 | `trial.{region}.maas.aliyuncs.com` | 快速验证，RPM 固定 1000，无 SLA | 600 秒 | + +接口规范与路径的对应关系（以专属域名为例）： -从共享域名迁移到专属域名只需替换 Base URL 中的域名部分，业务逻辑代码无需修改。 +- OpenAI 兼容：`https://{WorkspaceId}.{region}.maas.aliyuncs.com/compatible-mode/v1` +- DashScope 原生：路径换为 `/api/v1` +- Anthropic 兼容：路径换为 `/apps/anthropic` -## 各地域调用差异 +使用北京、新加坡、东京、法兰克福地域的专属域名时，需要在 Base URL 中填入业务空间 ID（WorkspaceId），可在控制台业务空间管理页面查看。 -- **华北2（北京）**：默认推荐地域，专属域名 `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1`。知识库、[模型部署](model-deployment.md)、数据管理、Prompt 模板、模型调优等大量能力仅在此地域开放。 -- **新加坡**：专属域名 `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`。 -- **美国（弗吉尼亚）**：使用带 `-us` 后缀的模型名（如 `qwen-plus-us`）可限定美国境内推理；共享接入域名 `https://dashscope-us.aliyuncs.com/compatible-mode/v1`。 -- **德国（法兰克福）/日本（东京）**：通过业务空间区分全球/欧盟或全球/日本部署范围，调用前需先在业务空间管理页面创建并选择对应业务空间。 +## 订阅制产品的专属接入点 -## 业务空间与 WorkspaceId +Token Plan / Coding Plan 与按量付费是三套完全隔离的体系，各有专属 Key 和域名： -使用北京、新加坡、日本、德国等地域时，需在 Base URL 中填入 WorkspaceId。WorkspaceId 可在控制台「业务空间管理」页面查看。子账号需加入对应业务空间并获得相应策略后方可操作该空间下的资源。 +- **Token Plan**：仅支持华北2（北京）地域，Key 以 `sk-sp-` 开头，Base URL 为 `https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1`（Anthropic 兼容为 `/apps/anthropic`）。 +- **Coding Plan**：专属 Key + `https://coding.dashscope.aliyuncs.com/v1`。 +- **按量付费**：通用 `sk-` Key + 上文各地域域名。 -## 地域相关能力约束 +三者不可混用，混用会导致鉴权失败或意外扣费。 -部分能力仅在特定地域可用： +## 各场景的地域限制速查 -| 能力 | 可用地域 | +| 场景 | 地域限制 | | --- | --- | -| 知识库（标准版/旗舰版） | 仅华北2（北京） | -| [模型部署](model-deployment.md)（资源专享推理） | 仅华北2（北京） | -| 数据管理（训练集/[评测](evaluation.md)集/数据流） | 仅华北2（北京） | -| Prompt 模板 | 仅华北2（北京） | -| `qwen-deep-research` | 仅华北2（北京），且仅支持 Python DashScope SDK | -| `qwen-mt-plus` / `qwen3.5-ocr` | 北京、新加坡、美国（弗吉尼亚） | -| `gui-plus` | 华北2（北京） | +| 模型微调（全部模态） | 仅华北2（北京），必须使用北京地域的 API Key | +| 数据管理 / 数据清洗与增强 | 仅华北2（北京） | +| 日志回流 | 华北2（北京）和新加坡 | +| 微调数据集 OSS 挂载 | Bucket 仅支持 `cn-beijing` 和 `ap-southeast-1` | +| Token Plan 订阅 | 仅华北2（北京），需在控制台左上角切换地域后购买 | +| DeepSeek 等部分第三方模型 | 仅华北2（北京） | -## 选择建议 +## 开发者实践建议 -1. 明确数据驻留要求（是否允许数据出境）。 -2. 确认所需模型/能力是否在目标地域开放。 -3. 生产环境统一使用专属域名，避免使用试用域名。 -4. 同一业务的所有调用、API Key、模型列表保持在同一地域，不要混用。 +1. **先定地域再拿 Key**：根据数据合规要求（如欧盟、日本、美国境内处理）和所需模型/功能选定地域，再在该地域创建 API Key 与业务空间。 +2. **生产用专属域名**：业务空间专属域名提供更高超时（3600 秒）、流量隔离和 SLA 保障；试用域名仅用于快速验证。 +3. **将地域信息写入配置**：Base URL、WorkspaceId、API Key 建议通过环境变量（如 `DASHSCOPE_API_KEY`）管理，避免硬编码和跨地域误用。 +4. **权限按地域规划**：业务空间不能跨地域存在，不同地域的默认业务空间也是不同空间；RAM 权限和空间级[限流](rate-limit.md)需按地域分别配置。 ## 关联主题页 - [get started with models](../guides/get-started-with-models.md) -- [knowledge base](../guides/knowledge-base.md) -- [model deployment 1](../guides/model-deployment-1.md) -- [prompt](../guides/prompt.md) +- [token plan guide](../guides/token-plan-guide.md) +- [fine tuning](../guides/fine-tuning.md) - [model data overview](../guides/model-data-overview.md) -- [more models](../api/more-models.md) +- [security and compliance](../guides/security-and-compliance.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/streaming.md b/skills/bailian-docs-llm-wiki/wiki/concepts/streaming.md index 82223a7f..2eeaa868 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/streaming.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/streaming.md @@ -1,47 +1,57 @@ # 流式输出 -流式输出（Streaming）是指服务端在生成内容的过程中，以增量分片（delta）的方式持续返回结果，而非等待全部内容生成完毕后一次性返回。它能显著降低首字延迟、改善实时交互体验，广泛用于对话、语音助手等场景。 +流式输出（Streaming）是指模型在生成过程中将结果以增量片段（chunk / delta / 事件）形式逐步返回给客户端，而不是等全部内容生成完毕后一次性返回。它能显著降低首字延迟（TTFT），是构建实时对话、语音交互等体验的基础能力。 -## 在百炼平台的使用场景 +## 在百炼平台不同场景中的使用 -百炼平台在多类接口中都支持流式输出，核心场景包括： +### 文本生成模型 API -- **应用调用（Application Call）**：无论是 OpenAI 兼容的 Responses API 还是 DashScope 原生 API，调用智能体或工作流应用时均支持流式输出，可用于需要边生成边展示的实时交互场景。 -- **文本生成模型 API**：OpenAI 兼容 Chat Completions / Responses、Anthropic 兼容 Messages 以及 DashScope 原生接口均可开启流式返回，适合聊天补全类应用逐字/逐段渲染输出。 -- **实时多模态交互（Omni-Realtime API）**：基于 WebSocket 协议，流式是其原生工作方式。服务端通过一系列增量事件持续推送音频与文本，天然适配低延迟的语音对话场景。 +百炼的四类文本生成接口（OpenAI 兼容 Chat Completions、OpenAI 兼容 Responses、Anthropic 兼容 Messages、DashScope 原生接口）均支持流式输出： -## 关键参数与配置 +- **[OpenAI 兼容接口](openai-compatible-interface.md)**：在请求中设置 `stream=true`，服务端以 SSE（Server-Sent Events）分块返回增量内容，可直接复用官方 OpenAI 客户端库的流式迭代方式。 +- **DashScope 原生接口**：功能与参数最完整，同样支持流式返回，适合需要平台全部能力的场景。 +- 跨接口迁移时注意流式响应的字段结构以各生态的约定为准，需核对参数与返回格式的映射。 + +### 智能体 / 工作流应用调用 + +通过 Responses API（OpenAI 兼容）或 DashScope API 调用百炼应用时，均支持流式输出： + +- **Responses API**：请求参数 `stream`（boolean，默认 `false`），设为 `true` 即可流式接收应用回复；耗时任务可另设 `background=true` 走异步模式。 +- **DashScope API**：调用 `POST /api/v1/apps/{APP_ID}/completion` 时开启流式，结合 `session_id` 维护多轮对话上下文。 -### HTTP / SDK 接口 +### 实时[多模态](multimodal.md)交互（Realtime API） -- **`stream`**：布尔值，控制是否开启流式输出，默认 `false`。设为 `true` 后，服务端以 SSE（Server-Sent Events）方式逐片返回结果。 - - 适用于 OpenAI 兼容 Responses API、DashScope API 等应用调用与模型调用接口。 -- 使用 SDK 时，开启 `stream=true` 后通过迭代响应对象逐步获取增量内容；部分接口可配合 `stream_options` 等参数控制是否返回用量统计等附加信息（以对应接口文档为准）。 +Qwen-Omni-Realtime 等实时接口基于 WebSocket / WebRTC / AOQ 协议，将流式思想扩展到音频与图像： -### Omni-Realtime（WebSocket） +- **流式输入**：客户端通过 `input_audio_buffer.append` 持续追加音频帧、`input_image_buffer.append` 追加视频帧，VAD 模式下服务端自动检测语音起止并提交。 +- **流式输出**：服务端以事件流下发增量结果，如 `response.audio.delta`（音频分片）、`response.function_call_arguments.delta`（工具调用参数增量）、`conversation.item.input_audio_transcription.delta`（实时转录预览）等，客户端边收边播实现低延迟对话与打断。 +- 流式 ASR（FunASR 系列）与流式 TTS（CosyVoice 系列）也基于 WebSocket 流式协议工作。 -实时接口不使用 `stream` 参数，而是以事件流的形式天然流式返回。关键的增量事件包括： +### Managed Agents（智能体托管） -| 事件 | 含义 | -| --- | --- | -| `response.audio.delta` | 增量音频输出 | -| `response.audio_transcript.delta` | 增量文本转录 | -| `conversation.item.input_audio_transcription.delta` | 实时语音识别中间结果 | -| `response.done` | 本轮响应流结束 | +托管智能体的会话事件通过 SSE 长连接流式推送：调用 `GET /sessions/{session_id}/events/stream` 订阅，即可实时接收用户消息回执、工具调用、状态变更等事件，无需轮询。 -此外，可通过 `session.update` 事件中的 `smooth_output`（部分模型支持）等参数调节流式输出的平滑度。 +## 关键参数与配置 + +| 场景 | 参数 / 机制 | 说明 | +| --- | --- | --- | +| 文本生成 / 应用调用 | `stream=true` | 开启 SSE 流式返回，默认 `false` | +| Realtime API | `*.delta` 事件 | 音频、文本、转录、工具参数均以 delta 事件增量下发 | +| Realtime API | `turn_detection` | `server_vad` 自动断句触发流式响应；置 `null` 则手动 commit | +| Managed Agents | `GET .../events/stream` | SSE 长连接订阅会话事件流 | ## 开发建议 -- **注意结束标志**：SSE 流需处理到结束标记（如 `[DONE]`）或 `response.done` 事件后再收尾，避免内容截断。 -- **增量拼接**：客户端需将各 delta 分片按序拼接，才能得到完整结果。 -- **错误处理**：流式过程中仍可能收到错误事件，需在读取流的循环中做好异常捕获与连接重试。 -- **跨接口差异**：不同接口的分片结构与字段命名不同（OpenAI/Anthropic 兼容接口以对应生态约定为准，DashScope 参数最全），跨接口迁移时需核对字段映射。 +- 交互式聊天、语音助手等对首响应时间敏感的场景优先开启流式；离线批处理或需要完整结果再处理的场景可用非流式或异步（`background=true`）。 +- 客户端需按增量拼接内容，并妥善处理流中断、错误事件与重连逻辑。 +- 流式与工具调用可以叠加：工具调用参数本身也会以增量事件下发（如 Realtime 的 `response.function_call_arguments.delta`），收齐后再执行工具并回传结果。 ## 关联主题页 -- [application call](../api/application-call.md) - [qwen api reference](../api/qwen-api-reference.md) +- [application call](../api/application-call.md) - [omni realtime api](../api/omni-realtime-api.md) +- [realtime api user guide](../guides/realtime-api-user-guide.md) +- [managed agents api](../api/managed-agents-api.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/token-billing.md b/skills/bailian-docs-llm-wiki/wiki/concepts/token-billing.md index 169b6401..e8c34aa9 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/token-billing.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/token-billing.md @@ -1,59 +1,56 @@ -# Token 与计费 +# Token 计费 -Token 是百炼平台衡量模型处理文本量的基本单位，也是绝大多数计费、限流和用量统计的计量基础；平台的费用则围绕 Token 消耗，通过按量付费、免费额度、节省计划与资源包等机制综合结算。理解 Token 如何被计量与抵扣，是控制大模型使用成本的前提。 +Token 计费是百炼平台的核心计量与计费机制：模型的推理、训练等用量以 Token（文本的最小计量单位）为口径分别统计输入与输出，并按各模型单价结算费用。开发者关心的"调用花多少钱、如何抵扣、如何查账"几乎都围绕 Token 计费展开。 -## Token 在计费中的角色 +## 在不同场景中的使用 -对大语言模型、全模态模型和向量模型，用量与费用均按 **Token** 计量；图像生成按「张」、视频生成按「秒」、语音模型按「秒/字符/Token」（视模型而定）。因此谈「计费」时，Token 主要针对文本类调用。 +### 模型推理（调用） -模型调用按**输入 Token** 和**输出 Token** 分别计费，单价以「每百万 Token」为单位。部分模型采用**阶梯计费**：按单次请求的输入 Token 总量分档定价，落入某一区间后该请求全部 Token 按对应档位结算（例如 `qwen3-max` 华北2·北京划分为 `0 资源包 > 其他模型节省计划 > AI 通用型节省计划 > 按量付费** +- **按 Token 用量部署**：仅部分 LoRA 调优后模型支持，`费用 = 输入 Token 数 × 输入单价 + 输出 Token 数 × 输出单价`，不使用不计费，一个月不用自动释放。 +- **预置吞吐（PTU）**：虽按时长计费，但额度以 TPM（每分钟 Token 数）折算消耗——长输入（超 32K）按更高系数折算，命中前缀缓存的输入按折扣系数（如 0.08~0.2）消耗；「自动溢出」策略下超额请求转按量 Token 计费（响应头 `x-dashscope-ptu-overflow:true`），「仅使用 PTU 容量」则返回 429。 -- **免费额度**：首次开通时各模型自动发放（通常每模型 100 万 Token），仅抵扣实时推理，不抵扣 Batch、调优、部署；不同模型（含同一模型不同快照版本）额度相互独立。开启「免费额度用完即停」后，额度耗尽会停止响应并返回 `AllocationQuota.FreeTierOnly`（或 403），避免意外扣费。 -- **折扣**：Batch 调用输入/输出单价按实时推理价的 50% 计费；支持上下文缓存的模型仅输入 Token 享折扣，两者不能同时生效。 -- **地域差异**：同一模型在不同地域单价不同，境外地域通常无免费额度。 +### 订阅类产品 -### 2. 模型训练（调优） +Token Plan / Coding Plan 以 Credits 统一计量而非直接按 Token 计价，其专属 API Key（`sk-sp-` 开头）**不消耗免费额度**；与通用按量付费 Key 混用会导致鉴权失败或意外走按量 Token 计费通道。 -按训练 Token 计费。文本模型公式为 `(训练数据 Token + 混合数据 Token) × 循环次数 × 训练单价`；图像/视频生成模型的训练 Token 总量由 `max_steps`、`max_pixels`、`n_epochs` 等超参决定。免费额度和节省计划**均不抵扣**训练费用。 +## 抵扣与优惠 -### 3. 模型部署 +费用抵扣优先级固定：**免费额度 > 资源包 > 其他模型节省计划 > AI 通用型节省计划 > 按量付费**，系统自动按序抵扣。 -免费额度和节省计划同样**不抵扣**部署费用。三种计费方式围绕 TPM（每分钟 Token 数）： +- **新人免费额度**：通常每模型 100 万 Token，有效期 90 天，仅抵扣实时推理，不抵扣 Batch、调优、部署；不同模型（含快照版本）额度相互独立。可开启"免费额度用完即停"，耗尽时返回 `AllocationQuota.FreeTierOnly`（403）。 +- **资源包**：预购特定模型的 Token 量，仅抵扣该模型超出免费额度后的实时推理用量。 +- **节省计划**：AI 通用型节省计划按承诺月消费换阶梯折扣（最高 5.3 折），可抵扣模型调用、上下文缓存、批量推理等按量 Token 费用，不抵扣调优/部署。 -- **预置吞吐（PTU）**：`费用 = 使用时长 × (输入 TPM 单价 × 输入 TPM + 输出 TPM 单价 × 输出 TPM)`。PTU 下超出购买吞吐或输入超上限时自动转按量付费（响应头 `x-dashscope-ptu-overflow:true`）。长输入按阶梯系数折算 TPM 消耗，命中前缀缓存的 Token 按折扣系数消耗额度。 -- **模型单元（MU）**：`费用 = 使用时长 × 模型单元数量 × 模型单元单价`。 -- **按 Token 使用量**：仅对 LoRA 微调后的自定义模型开放，用于调优效果验证。 +## 用量查询与监控 -### 4. 订阅套餐（Token Plan / Coding Plan） +- **模型用量页面**：按[业务空间](workspace.md)统计各模型 Token 消耗与费用，数据延迟约 1 小时，仅保留 30 天；文本/思考/视觉理解/向量模型按 Token 统计，图像按张、视频按秒。 +- **模型监控**：成本类指标含平均单次请求调用量与 Token 消耗；开通推理日志后可查看单次调用的 Token 用量、请求与响应内容（分钟级延迟，仅记录开通后的调用）。 +- **高级监控 / Prometheus**：可通过 `model_usage` 等指标接入 Grafana，按 `workspace_id`、`model`、`apikey_id` 等维度分析 Token 消耗。 +- **应用观测**：应用维度可查看 Token 总量、输入/输出 Token、平均单次请求 Token 量、平均首 Token 耗时等指标，并支持按 Token 量筛选 Span。 -- **Token Plan 团队版**：按 Token 消耗抵扣 Credits，面向团队协作。 -- **Coding Plan**：按模型调用次数计量，面向个人开发。 +## 关键参数与注意事项 -两者均使用 `sk-sp-` 前缀的**专属 API Key**，**不消耗新人免费额度**，且 Base URL 与按量付费端点完全隔离，混用会导致意外扣费或 401/403 鉴权失败。 +| 项目 | 说明 | +| --- | --- | +| `model` 参数 | 免费额度按模型（含快照版本）隔离，用完不自动切换，需手动更换 | +| `service_tier` / `provisioned_tokens` / `cached_tokens` | PTU 场景响应字段：是否走 PTU 额度、折算后额度消耗、缓存命中 Token 数 | +| `x-dashscope-ptu-overflow` | 响应头为 `true` 表示该请求已溢出到按量 Token 计费 | +| `AllocationQuota.FreeTierOnly` | 开启"用完即停"后免费额度耗尽的报错码 | -### 5. 监控与用量统计 - -模型监控将 Token 消耗归入「成本」类指标，并提供首 Token 延时、RPM/TPM 等性能指标。应用观测则可查看每次调用的输入/输出/平均 Token 量与平均首 Token 耗时。开通推理日志后可查看单次调用的 Token 消耗，用于排查与审计。 - -## 关键参数与配置 - -- **`max_tokens`**：限制单次生成的最大输出 Token 数，是控制输出成本和防止过度生成的首要手段。 -- **TPM / RPM 限流**：部署时可配置 `tpm_limit`、`rpm_limit`；触发限流后等待时间取决于限流值。 -- **PTU 计费识别字段**：`service_tier`（`ptu-standard` 走 PTU 额度，`default` 或缺失表示按量）、`provisioned_tokens`（折算后实际消耗额度）、`cached_tokens`（缓存命中数，Anthropic 兼容格式暂不返回）。这些字段在 OpenAI Chat、OpenAI Responses、Anthropic 兼容、DashScope 四种协议下 JSON 路径不同，需按协议读取。 - -## 成本优化与出账 - -- **节省 Token**：优化 Prompt 减少输入 Token、简单任务选用轻量级模型、非实时任务走批量推理、合理设置 `max_tokens`。 -- **预付费方案**：AI 通用型节省计划（承诺月消费换阶梯折扣，最高 5.3 折，月额度不可跨月累积）、其他模型节省计划、资源包（预购具体 Token/图片数量，仅抵扣单个模型超免费额度后的实时推理）。 -- **出账时效**：大模型推理分钟级出账（约 2～10 分钟）；批量推理、模型训练等小时级出账。用量统计数据约 1 小时延迟，且不支持查看 30 天前数据。 +常见误区：模型部署（PTU/模型单元）按时长计费，**部署为"运行中"即持续扣费，与是否有 Token 调用无关**，这是"没调用也产生费用"的最常见原因；不再使用务必删除部署任务。 ## 关联主题页 @@ -61,9 +58,6 @@ Token 是百炼平台衡量模型处理文本量的基本单位，也是绝大 - [token plan guide](../guides/token-plan-guide.md) - [model monitoring](../guides/model-monitoring.md) - [application monitoring](../guides/application-monitoring.md) -- [support](../guides/support.md) - [model deployment 1](../guides/model-deployment-1.md) - - diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/token.md b/skills/bailian-docs-llm-wiki/wiki/concepts/token.md index 0b4e6e3c..71dd7216 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/token.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/token.md @@ -1,65 +1,68 @@ -# Token 与计量计费 +# Token -Token 是大语言模型处理文本的最小计量单位，百炼平台以 Token 为核心，对模型的输入、输出、训练用量进行计量、计费与监控。理解 Token 的产生方式与计费规则，是控制模型调用成本、优化应用性能的基础。 +Token 是大语言模型处理文本的最小计量单位，模型将输入和输出文本切分为若干 Token 进行计算。在百炼平台中，Token 是模型调用计费、用量统计、容量规划和性能监控的核心度量单位。 -## 什么是 Token +## 在百炼平台中的使用场景 -Token 是模型在处理文本时切分出的基本片段（一个汉字、单词或子词可能对应一个或多个 Token）。在百炼平台中，Token 既是**用量计量单位**，也是**大部分计费的核算基准**： +### 计费 -- 文本生成模型按**输入 Token** 和**输出 Token** 分别计量，思考模式下的输出 Token 同时包含「思维链 + 回答」两部分。 -- 不同模型类型的计量单位不同：大语言模型 / 全模态模型 / 向量模型按 **Token** 计量，图像生成按**张**，视频生成按**秒**，语音模型按**秒 / 字符 / Token**（视模型而定）。 +Token 是模型推理和训练计费的基本单位： -## 在各场景中的使用 +- **按量计费**：模型调用按输入 Token 和输出 Token 分别计价。部分模型实行阶梯计费——单次请求的输入 Token 总量决定整笔请求适用的单价（例如 qwen3-max 分 0–32K、32K–128K、128K–256K 三档）。 +- **Batch 调用**：输入和输出 Token 均按实时推理价格的 50% 计费。 +- **上下文缓存**：缓存命中的输入 Token 按标准单价的 10% 计费，显式缓存创建按 125% 计费。 +- **模型训练**：费用按训练数据 Token 总数 × 循环次数 × 单价计算；万相图像模型的 Token 总量由 max_steps × max_token_length 决定。 +- **按 Token 用量部署**：LoRA 微调后的模型可部署为按输入/输出 Token 数计费的推理服务，最小计费单位为 1 Token。 -### 1. 按量付费与免费额度 +费用抵扣优先级为：免费额度 > 资源包 > 其他模型节省计划 > AI 通用型节省计划 > 按量付费。 -- 首次开通时，各模型会发放新人专属免费额度（通常各 100 万 Token），仅抵扣**实时推理**费用，且不同模型（含同一模型不同快照版本）额度相互独立、不可合并。 -- 免费额度耗尽后默认转为**按量付费**，按输入 / 输出 Token 计费。部分模型采用**阶梯计费**：按单次请求的输入 Token 总量分档（如 `qwen3-max` 分 0–32K / 32K–128K / 128K–256K 三档），落在哪一档，该请求全部 Token 均按该档单价结算。 -- 同一模型在不同地域（北京、弗吉尼亚、新加坡、法兰克福、东京）单价不同。 +### 免费额度与订阅 -### 2. 订阅制套餐 +- 新人免费额度通常为每模型 100 万 Token，有效期 90 天，仅抵扣华北2（北京）地域的实时推理费用。 +- 资源包以预购 Token 数量的方式抵扣单个特定模型的实时推理用量。 +- Token Plan 订阅服务以 Credits 为统一计量单位（底层仍按 Token 消耗折算），采用 5 小时 + 7 天双层窗口限额（个人版）或月度总额度制（团队版）。 -- **Token Plan 团队版**：按 Token 消耗抵扣 Credits，工具调用（联网搜索、代码解释器等内置工具）产生的 Token 同样从套餐 Credits 抵扣，不额外收费。 -- **Coding Plan**：按模型调用**次数**计费（而非 Token），面向个人开发场景。 +### 容量与吞吐（TPM） -### 3. 模型训练与部署 +TPM（Tokens Per Minute）是衡量推理吞吐能力的单位： -- **训练**按训练 Token 计费，文本模型公式为 `（训练数据 Token + 混合训练数据 Token）× 循环次数 × 训练单价`；图像 / 视频模型的训练 Token 由 `max_steps`、`max_pixels`、`n_epochs` 等超参数推算。 -- **部署**（预置吞吐 TPM）按输入 / 输出 TPM 单价与时长计费，与 Token 用量间接相关。训练与部署**不能**用免费额度或节省计划抵扣。 +- **预置吞吐（PTU）**：按输入/输出 kTPM 预留专属容量，超出最长输入 Token 或 TPM 容量时按溢出策略处理（自动转按量付费或返回 429）。长输入场景下超出 32K 的部分按阶梯系数（如 1.33）折算 TPM 消耗。 +- **TPM 预留**：独立的高速推理方案，锁定专属吞吐量以规避公共限流，按每 10,000 TPM 日预付费计价。 +- **PD 分离模式**：将首 Token 计算（Prefill）与后续 Token 计算（Decode）拆分到不同节点，降低首 Token 延迟、提高吞吐。 -### 4. 监控与观测 +### 监控与观测 -- 应用观测可查看每次调用的 Token 量，监控统计提供 Token 总量（全部 / 输入 / 输出）、平均单次请求 Token 量、平均首 Token 耗时等指标。 -- 模型监控将 Token 消耗归入「成本」类指标，首 Token 延时归入「性能」类指标，支持按分钟 / 小时 / 天聚合，并可配置告警。 -- 用量统计按业务空间维度归集，数据延迟约 1 小时。 +- **模型用量统计**：按[业务空间](workspace.md)维度汇总 Token 消耗，支持按 [API Key](api-key.md)、模型、时间范围筛选，数据延迟约 1 小时。 +- **模型监控**：提供平均单次请求 Token 量、首 Token 延时、非首 Token 延时等性能指标；支持设置 Token 消耗阈值告警。 +- **应用观测**：端到端追踪应用调用链路中的 Token 总量（全部/输入/输出），支持按 Span 筛选 Token 相关字段。 +- **推理日志**：记录每次调用的 Token 用量明细，需开通后方可查看。 -## 关键参数与配置 +## 关键参数与字段 -| 项 | 说明 | +| 参数/字段 | 说明 | | --- | --- | -| `max_tokens` | 限制单次生成的最大输出 Token 数，用于控制成本、防止过度生成，也是降低幻觉的手段之一 | -| 上下文缓存 | 命中缓存的**输入** Token 享折扣；缓存折扣与 Batch 折扣不可同时生效，价格表输入单价不含缓存单价 | -| Batch 调用 | 支持的模型输入 / 输出单价按实时推理价的 50% 计费 | -| 免费额度用完即停 | 开启后额度耗尽即停服（返回 403），避免意外扣费，但也会阻断节省计划继续抵扣 | +| `usage.input_tokens` / `usage.output_tokens` | API 响应中返回的本次调用输入/输出 Token 数 | +| `cached_tokens` | 前缀缓存命中的 Token 数，可用于确认缓存是否生效 | +| `provisioned_tokens` | PTU 部署下折算后实际消耗的额度 Token 数（含阶梯系数和缓存折扣） | +| `service_tier` | 值为 `ptu-standard` 表示使用 PTU 额度，`default` 表示按量计费 | +| `max_tokens` | 请求参数，限制模型单次生成的最大输出 Token 数 | ## 成本优化建议 -- **优化 Prompt**：简洁清晰的 Prompt 可减少不必要的输入 Token 消耗。 -- **控制输出长度**：合理设置 `max_tokens`，避免冗长输出。 -- **按任务选模型**：分类、摘要等简单任务优先使用轻量级模型。 -- **批量推理**：非实时大批量任务走 Batch 接口，Token 单价更低。 -- **善用抵扣顺序**：`免费额度 > 资源包 > 其他模型节省计划 > AI 通用型节省计划 > 按量付费`，据此规划预付费方案。 - -## 账单中的 Token - -大模型推理为分钟级出账（通常 2～10 分钟），批量推理与训练为小时级。账单「实例 ID」以英文分号分隔，包含 `ApiKeyID;业务空间ID;模型名称;输入/输出类型;调用渠道;免费额度用完即停标识`，可据此区分输入 / 输出 Token 的费用来源与调用渠道（`app` 代码调用、`bmp` 控制台体验、`assistant-api`）。 +- 利用上下文缓存降低重复前缀的输入 Token 费用（命中后仅需 10% 单价）。 +- 对延迟不敏感的任务使用 Batch 调用享受半价。 +- 通过模型监控设置 Token 消耗告警，及时发现异常用量。 +- 使用预置吞吐额度计算器，根据 RPM、平均输入/输出长度和缓存命中率估算所需 TPM，避免容量不足导致请求溢出为按量计费。 ## 关联主题页 -- [token plan guide](../guides/token-plan-guide.md) - [test 1](../guides/test-1.md) +- [token plan guide](../guides/token-plan-guide.md) - [application monitoring](../guides/application-monitoring.md) - [model monitoring](../guides/model-monitoring.md) -- [support](../guides/support.md) +- [fine tuning](../guides/fine-tuning.md) +- [model deployment 1](../guides/model-deployment-1.md) +- [model high speed inference](../guides/model-high-speed-inference.md) + diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/workflow.md b/skills/bailian-docs-llm-wiki/wiki/concepts/workflow.md index 0badc9bf..c053dfda 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/workflow.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/workflow.md @@ -1,75 +1,109 @@ # 工作流 -工作流（Workflow）是百炼平台三种核心应用构建模式之一，通过可视化节点编排将复杂任务拆解为有序步骤，逻辑确定、稳定可复现，适合流程固定、要求可复现的业务场景。 +工作流（Workflow）是阿里云百炼提供的一种低代码可视化应用形态，通过拖拽预定义节点并以有向图方式编排处理链路，将大模型调用、知识库检索、意图分类、API 调用等能力串联为输出确定、可复现的执行流程。与依赖模型自主规划的智能体不同，工作流由开发者显式控制每个步骤的执行顺序与数据流转。 -## 在百炼平台中的定位 +## 应用形态定位 -百炼提供智能体、工作流、高代码应用三种互补的应用构建模式。工作流对应"可视化节点编排（低代码）"路线，由预定义节点精确控制流程，适合 IT 运维、业务分析师构建报告生成、订单处理、审批流等场景。与由大模型自主规划的智能体不同，工作流强调开发者对流程的显式控制，输出可复现、可调试。 +百炼提供三种 LLM 应用形态：智能体、工作流、高代码应用。工作流处于中间位置： -## 核心节点 +| 维度 | 说明 | +| --- | --- | +| 开发方式 | 可视化节点编排（低代码） | +| 控制方式 | 预定义节点顺序确定 | +| 适合人群 | 业务分析师 / IT 实施 | +| 典型场景 | 报告生成、审批流、数据标注 | -工作流画布由以下节点类型组成： +工作流输出结果确定性高、可复现。需要模型自主决策调用顺序时选智能体，需要完全自定义代码时选高代码应用。工作流应用已被新版智能体编排应用替代部分场景，但仍适用于需要稳定、可审计处理流程的业务。 -- **开始 / 结束节点**：定义输入与输出参数。开始节点预置 `query`（用户输入）、`historyList`（对话历史）、`imageList`（图片）等变量，可在下游节点中引用。 -- **大模型节点**：执行 LLM 推理，配置模型、提示词、用户提示词、记忆等。 -- **意图分类节点**：根据输入分流到不同下游分支，支持多意图。 -- **变量处理节点**：用于文本输出或变量加工。 -- **智能体群组节点**：将任务分解给多个已发布的子智能体协同完成。 +## 节点类型与编排 -## 会话变量 +工作流由节点串联构成，支持以下节点类型： -会话变量作为全局变量在工作流全生命周期内记录参数，可在各节点中引用，在画布右上角配置。适合跨节点传递状态、累积上下文信息。 +- **开始 / 结束节点**：定义流程入口与出口。 +- **大模型节点**：调用 LLM 生成、理解、改写文本。 +- **意图分类节点**：根据用户输入分流到不同下游分支，实现多意图路由。 +- **智能体群组节点**：将已发布的[智能体应用](agent-application.md)作为子智能体调度。 +- **变量处理节点**：对中间结果进行格式转换、拼接、提取。 +- **API 节点**：调用外部 HTTP 接口。 +- **函数计算节点**：执行自定义计算逻辑。 +- **[多模态](multimodal.md)生成节点**：生成图像、视频、音频。 +- **[多模态](multimodal.md)数据节点**：解析文档、图片、视频、音频。 -## 典型案例 +典型组合模式：意图分类节点做分流 → 多个大模型节点承接不同意图 → 汇总到结束节点。智能体群组节点可把已发布的[智能体应用](agent-application.md)作为子智能体调度，实现多层编排。 -- **诈骗信息识别**：开始 → 大模型（提示词判定诈骗嫌疑）→ 结束。 -- **智能导购**：意图分类节点将输入分流到手机 / 电视 / 冰箱等大模型分支，未命中分支走变量处理节点。 -- **日程管理助手**：通过智能体群组节点串联"信息收集"与"数据整理"两个子智能体。 +## 预置变量与记忆 -## 创建与发布流程 +### 预置变量 -1. 控制台 → 应用管理 → 创建应用 → 工作流应用。 -2. 在画布上拖拽编排节点，配置各节点的模型、提示词、变量引用关系。 -3. 右侧对话框调试。 -4. 右上角"发布"——发布是后续 API 调用与集成的前提。 +| 变量 | 说明 | +| --- | --- | +| `query` | 用户输入文本 | +| `historyList` | 对话历史 | +| `imageList` | 用户上传图片 URL 列表 | + +`historyList` 和 `imageList` 需在支持"记忆"的节点中选择"自定义缓存"才会注入。 + +### 记忆模式 + +- **本节点缓存**：仅当前节点上下文，不跨节点共享。 +- **自定义缓存**：跨节点共享全局对话历史，适合多轮记忆场景。 + +### 会话变量 + +会话变量作为全局变量在工作流全生命周期内维护，可在任意节点引用，用于在节点间传递状态。 + +## 扩展能力 + +- **异步运行模式**：文本生成模式下后台执行并返回 Task ID，适合长耗时任务。 +- **Dify 工作流一键导入**：支持从 Dify 平台导入已有工作流配置。 +- **批量节点**：支持批量处理输入数据。 ## API 调用 -工作流应用与智能体应用共用同一套调用接口，区别仅在应用内部的编排逻辑。发布后通过 `APP_ID` 调用： +工作流应用发布后通过 API 调用，与[智能体应用](agent-application.md)调用接口完全一致： + +``` +POST https://dashscope.aliyuncs.com/api/v1/apps/{APP_ID}/completion +``` + +请求体包含 `input.prompt`（用户输入）、`parameters`（业务参数）、`debug` 字段，响应中业务侧主要消费 `output.text`。 + +### 多轮对话 + +工作流支持多轮对话，有两种实现方式： -- **DashScope 原生 API**：`POST https://dashscope.aliyuncs.com/api/v1/apps/{APP_ID}/completion`，请求体为 `{"input": {"prompt": "..."}, "parameters": {}, "debug": {}}`，响应中业务侧主要消费 `output.text`。 -- **OpenAI 兼容 Responses API**：`POST https://dashscope.aliyuncs.com/api/v2/apps/agent/{APP_ID}/compatible-mode/v1/responses`，可复用现有 OpenAI 生态代码库，支持同步 / 异步、流式、多模态。 -- **SDK**：Python 使用 `dashscope.Application.call`，Java 使用 `com.alibaba.dashscope.app.Application`，Node.js 可直接以 `axios` 发起 POST。 +- **使用 `session_id`**：系统自动从云端管理对话上下文。 +- **使用 `historyList`**：由调用方自行维护对话历史并传入。 -调用前需在控制台获取应用 ID 与 API Key（推荐写入 `DASHSCOPE_API_KEY` 环境变量）；若应用位于子业务空间，还需提供 Workspace ID。 +## 组件化复用 -## 多轮对话 +工作流应用可发布为模块化组件，供其他应用（智能体或工作流）复用。发布途径有三处：发布应用时勾选"发布应用组件"、在"发布渠道"的组件区域创建、或在控制台"组件管理"面板创建。 -工作流应用支持多轮对话，有两种实现方式： +组件预设系统参数 `query`（用户输入文本）和 `imageList`（图像 URL 列表），预设参数不可删除，不需要时可将"是否可见"设为"否"隐藏。参数传参方式分两种： -- **使用 `session_id`**：系统自动从云端加载历史对话，实现简单。`session_id` 有效期 1 小时，最多支持 50 轮。 -- **自行管理 `messages`（推荐）**：手动维护消息列表，灵活性更高，不受 session 有效期限制。 +- **业务透传**：由使用者或上游节点提供具体值。 +- **模型识别**：由大模型根据参数描述自动推断填充。 -## 评测 +> **注意**：即使参数传参方式设为模型识别，在工作流应用中也不会自动推断参数值，必须从上游节点明确提供输入值。模型识别仅在智能体应用中生效。 -工作流应用可作为评测任务的被测对象。在新版应用评测中，评测集支持智能体、工作流、自定义三种类型，按所选应用的出入参形式自动生成数据模板。评测任务可关联工作流应用，由 LLM 评估器或 Code 评估器自动评分，也可人工标注。 +接入工作流应用时，组件作为组件节点接入，需手动传入参数（如 `系统变量/query`），并将 `组件1/result` 传递到下游节点。避免嵌套调用（A 调 B、B 调 A）和过多级调用（A 调 B、B 调 C），会因最长运行时间限制导致超时报错。应用重新发布后组件会自动更新。 -## 发布与分享 +## 应用观测 -工作流应用支持多种发布渠道：UI 应用（通过 UI 设计器构建自定义界面）、钉钉机器人、微信公众号、组件化复用（将工作流作为其他智能体或工作流的子组件）、音视频实时互动（仅限图文对话类）。需要注意的是，官方网页版分享当前只支持智能体应用，不支持工作流应用。 +工作流应用（WorkflowApp）在百炼应用观测功能中以 `WorkflowApp` 为根 CHAIN 节点展示完整调用链路，支持查看延时、[Token](token.md) 量等关键指标，数据更新频率为分钟级，可保留最长 30 天的调用记录。观测数据支持按 Request ID / Trace ID / Span ID 检索，并支持导出为 JSONL 或 Excel 格式。 -## 关键限制与注意事项 +## 框架集成 -- 工作流应用若配置了文件类型的自定义参数，在 UI 设计器中需指定 `{{{file_name:files[0]}}}`（将 `file_name` 替换为实际变量名），否则应用无法正确读取用户上传的文件。 -- 旧版"智能体编排"应用已被工作流应用替代，新建应用请直接选择工作流。 -- 工作流与智能体调用接口一致，但可附加的扩展能力（如自定义参数传递）取决于应用内部编排逻辑。 +通过 Spring AI Alibaba（Java，Spring Boot 3.x，JDK 17+）可集成已创建的工作流应用，支持流式与非流式调用。需提前在百炼控制台创建工作流应用并获取应用 ID，通过 `APP_ID` 环境变量配置，[API Key](api-key.md) 推荐使用 `DASHSCOPE_API_KEY` 环境变量。若应用创建在子[业务空间](workspace.md)，还需配置 `WORKSPACE_ID`。 ## 关联主题页 +- [frameworks](../api/frameworks.md) - [llm application](../guides/llm-application.md) -- [application call](../api/application-call.md) - [bailian application calling](../guides/bailian-application-calling.md) -- [application evaluation](../guides/application-evaluation.md) +- [start using](../guides/start-using.md) +- [application monitoring](../guides/application-monitoring.md) - [application publishing and sharing](../guides/application-publishing-and-sharing.md) +- [application use cases](../guides/application-use-cases.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/concepts/workspace.md b/skills/bailian-docs-llm-wiki/wiki/concepts/workspace.md index 51a00e4c..cd45e1ec 100644 --- a/skills/bailian-docs-llm-wiki/wiki/concepts/workspace.md +++ b/skills/bailian-docs-llm-wiki/wiki/concepts/workspace.md @@ -1,61 +1,67 @@ -# 业务空间（Workspace） +# 业务空间 -业务空间（Workspace）是阿里云百炼平台进行精细化权限管理（模型、用户）和阿里云账单分账的**最小管理单元**。平台按地理区域划分资源与业务空间，单个业务空间不能跨地域存在，即便是各地域的默认业务空间也彼此独立。 +业务空间（Workspace）是阿里云百炼平台上进行资源隔离、精细化权限管理和账单分账的最小管理单元。每个业务空间归属于一个地域，**不能跨地域存在**——即使是各地域的默认业务空间，也是彼此独立的空间。 -## 在不同场景中的使用 - -### 1. 权限与身份管理 +## 核心特性 -业务空间是权限体系的核心边界，权限管理围绕三种角色展开： +- **地域隔离**：百炼按地理区域（北京、上海、新加坡、弗吉尼亚等）划分资源，业务空间从属于单一地域。 +- **默认空间与子空间**：每个地域有一个默认业务空间，其下可创建多个子业务空间。默认业务空间的 API Key 拥有所有模型的调用权限，且**无法设置**模型调用/训练/部署限制和空间级[限流](rate-limit.md)；需要精细化隔离时应创建独立的子业务空间。 +- **API Key 归属**：单个 API Key 只能归属一个地域内的一个业务空间和一个用户，且不能转移。API Key 可调用的功能和模型[限流](rate-limit.md)与归属业务空间的权限保持一致，不受用户控制台权限管理影响。 -- **超级管理员**：阿里云主账号，或拥有 `AliyunBailianFullAccess` 系统策略的 RAM 用户。可跨空间统一管理用户权限、空间可用模型、模型限流和 API Key。 -- **业务空间管理员**：拥有访问某个业务空间「权限管理」页面的 RAM 用户，仅管理该空间内的用户与资源，其权限包含该空间下所有页面的访问权限。 -- **普通用户**：仅能访问/使用被授权的空间、页面与资源，不能管理用户或模型授权。 +## 在不同场景中的使用 -在非默认业务空间中，可对模型进行三类精细化授权（**默认业务空间无法设置这些限制**，即所有模型均可调用、调优、部署且无法限流）： +### 权限管理 -| 权限项 | 控制范围 | -| --- | --- | -| 限制模型调用 | 是否可调用（控制台 & API）+ 请求数限流 + Token 限流 | -| 限制模型训练 | 是否可调优（控制台 & API）及调优后部署 | -| 限制模型部署 | 是否可直接部署 | -| 用户控制台权限 | 管理 RAM 用户能否使用该空间控制台及可用功能 | +权限体系围绕三种角色展开：**超级管理员**（主账号或拥有 `AliyunBailianFullAccess` 的 RAM 用户，可跨空间管理）、**业务空间管理员**（仅管理特定空间内的用户与资源）、**普通用户**（使用被授权的空间、页面、资源）。超级管理员可按空间维度控制： -> **注意**：OpenAPI 接口权限不通过业务空间角色授予，必须由阿里云主账号在 RAM 控制台为 RAM 用户单独添加系统策略（如 `AliyunBailianDataFullAccess` / `AliyunBailianDataReadOnlyAccess`）。 +- 特定模型是否允许在该空间调用（控制台与 API），并设置请求数[限流](rate-limit.md)和 Token 限流； +- 特定模型是否允许在该空间调优、调优后部署或直接部署； +- RAM 用户在该空间控制台可访问哪些页面。 -### 2. API Key 归属 +### 模型调用（子业务空间） -单个 API Key 只能归属**一个地域内的一个业务空间和一个用户**，且不能转移。API Key 可调用的功能与模型限流与其**归属业务空间**的权限保持一致，不受用户控制台权限管理的影响。将 RAM 账号移出业务空间会使其 API Key 失效（重新加入后恢复生效）。 +按业务线隔离权限或分账时，使用子业务空间的模型调用： -### 3. 应用组件 API 调用 +- 必须使用子空间自身的 API Key 调用； +- 调用标准模型（如 `qwen-plus`）前需先为该空间设置模型调用权限； +- 在百炼上调优并部署的模型无需额外授权，但仅能由其所在空间的 API Key 调用，且仅支持 DashScope 方式。 -调用百炼应用组件 API（`bailian/2023-12-29`，含数据连接、知识库、Prompt 模板、长期记忆等）时，所有接口均需传入 `WorkspaceId`（业务空间 ID）。RAM 子账号需先获取对应权限策略并加入业务空间后才能调用。类目等资源也以业务空间为界，例如每个业务空间最多可新建 500 个类目。 +### API 端点 -### 4. 应用观测 +部分 OpenAI 兼容端点将业务空间 ID 嵌入域名，调用时需替换为真实值，例如文本向量接口： -应用观测以业务空间为范围，端到端查看空间内应用（智能体应用、工作流应用、高代码应用）的处理流程与延时、Token 等指标。若应用观测列表中看不到已创建的应用，常见原因之一即为该应用不属于当前业务空间。 +``` +POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/embeddings +``` -### 5. 数据管理 +### 监控与观测 -数据管理功能统一管理业务空间下的大模型相关数据集（训练集、评测集）。该能力目前仅适用于华北2（北京）地域。 +- **模型用量**页面按业务空间维度统计调用量、Token 消耗与费用（不支持按阿里云账号维度统计）。 +- **模型监控**按"模型 + 业务空间"维度生成监控列表；默认业务空间成员可查看所有空间数据，子空间成员仅能查看当前空间。Prometheus 高级监控指标支持 `workspace_id` 作为过滤标签。 +- **应用观测**只能观测当前业务空间内已发布的应用；若列表中找不到应用，常见原因是应用不属于当前空间。 -## 关键参数与配置 +## 关键配置要点 -- **`WorkspaceId`**：业务空间 ID，是应用组件 OpenAPI 的必传参数，用于标识资源所属空间。 -- **地域隔离**：业务空间绑定单一地域，跨地域需在对应地域分别创建/使用空间。 -- **默认业务空间的限制**：默认空间无法设置模型调用/训练/部署授权与限流，如需精细化管控请使用非默认业务空间。 +| 配置项 | 说明 | +| --- | --- | +| 模型调用授权与限流 | 超级管理员按空间控制模型可用性，并设置 RPS / Token 限流（默认空间不可设置） | +| 模型训练/部署权限 | 按空间限制特定模型的调优与部署（默认空间不可设置） | +| 用户页面权限 | 控制 RAM 用户在该空间控制台可用的页面，不影响其 API Key 调用 | +| WorkspaceId | 子空间调用部分接口时需在 Endpoint 或请求中指定真实的业务空间 ID | -## 生产环境实践 +## 最佳实践 -- **空间规划**：推荐按环境（dev/test/prod）划分业务空间实现隔离，或按业务线划分以便权限与成本管理。 -- **限流策略**：将主账号总配额按比例分配给各业务空间并预留缓冲。例如总配额 1000 QPM，可分配 prod 600 / test 200 / dev 100，预留 100。 +- 生产环境建议按开发、测试、预发、生产划分独立业务空间，并为生产空间预留更高限流配额和更严格的用户权限。 +- 需要按业务线分账或限制模型使用范围时，创建子业务空间而非在默认空间内混用。 +- OpenAPI 接口权限不通过业务空间角色授予，需主账号在 RAM 控制台为 RAM 用户添加专用系统策略（如 `AliyunBailianDataFullAccess`）。 ## 关联主题页 -- [application component api reference](../api/application-component-api-reference.md) - [application permission management](../guides/application-permission-management.md) -- [application monitoring](../guides/application-monitoring.md) - [security and compliance](../guides/security-and-compliance.md) -- [model data overview](../guides/model-data-overview.md) +- [application monitoring](../guides/application-monitoring.md) +- [vector and sort](../api/vector-and-sort.md) +- [more about models](../api/more-about-models.md) +- [model monitoring](../guides/model-monitoring.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/application-evaluation.md b/skills/bailian-docs-llm-wiki/wiki/guides/application-evaluation.md index 508d7604..a508b0ad 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/application-evaluation.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/application-evaluation.md @@ -1,6 +1,6 @@ # application evaluation -阿里云百炼提供完整的应用[评测体系](../concepts/evaluation.md)，支持对[智能体应用](../concepts/agent-application.md)和工作流应用的输出质量进行系统化评估。平台同时提供自动评测与手动评测两种模式，并通过评测集、评估器和标签三大组件构建多维度的评测闭环。当前平台存在新旧两套评测系统，新版在评测任务管理、评估器和标签体系上做了较大升级。 +阿里云百炼提供完整的应用[评测体系](../concepts/evaluation.md)，支持对[智能体应用](../concepts/agent-application.md)和[工作流](../concepts/workflow.md)应用的输出质量进行系统化评估。平台同时提供自动[评测](../concepts/evaluation.md)与手动[评测](../concepts/evaluation.md)两种模式，并通过[评测](../concepts/evaluation.md)集、评估器和标签三大组件构建多维度的[评测](../concepts/evaluation.md)闭环。当前平台存在新旧两套[评测](../concepts/evaluation.md)系统，新版在[评测](../concepts/evaluation.md)任务管理、评估器和标签体系上做了较大升级。 ## 评测模式 @@ -8,17 +8,17 @@ ### 自动评测 -[自动评测](../../raw/application-user-guide/application-evaluation/application-auto-evaluation.md)利用大模型基于应用关联的知识库自动生成评测集，并对智能体的回答进行自动评分，生成评测报告与调优建议。支持两种子模式： +[自动评测](../../raw/application-user-guide/application-evaluation/application-auto-evaluation.md)利用大模型基于应用关联的[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)自动生成评测集，并对智能体的回答进行自动评分，生成评测报告与调优建议。支持两种子模式： - **单应用评测**：深度评估单个[智能体应用](../concepts/agent-application.md)的表现，生成包含评分、错误分析和优化建议的详细报告。 - **多应用横向评测**：在同一评测基准下对比最多 8 个应用（或同一应用的不同版本），用于选型决策或版本迭代效果验证。 前提条件： -1. 仅面向**已发布**的[智能体应用](../concepts/agent-application.md)，且应用须已配置知识库。 +1. 仅面向**已发布**的[智能体应用](../concepts/agent-application.md)，且应用须已配置[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)。 2. 须开通**应用观测**功能，并将待评测应用添加到观测列表。 3. 子账号需获取`管理员`或`应用评测-操作`权限。 -4. 多应用横向评测时，所有被选应用必须关联至少一个相同的知识库。 +4. 多应用横向评测时，所有被选应用必须关联至少一个相同的[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)。 自动评测流程分四步：创建评测任务 → 设置评测集 → 配置评测规则 → 执行评测。评测集生成和评估模型当前仅支持 `qwen-max` 和 `qwen-plus`。 @@ -51,8 +51,8 @@ | 类型 | 说明 | |------|------| -| 智能体 | 根据选中智能体应用的出入参形式定义评测集 | -| 工作流 | 根据选中工作流应用的出入参形式定义评测集 | +| 智能体 | 根据选中[智能体应用](../concepts/agent-application.md)的出入参形式定义评测集 | +| [工作流](../concepts/workflow.md) | 根据选中[工作流](../concepts/workflow.md)应用的出入参形式定义评测集 | | 自定义 | 任意定义评测集表结构，适用于特殊评测场景 | 新版评测集支持手动上传和从应用观测导入两种创建方式，并具备版本管理能力，每次发布生成新版本。创建后类型不可修改。 @@ -66,7 +66,7 @@ 百炼提供多种预置评估器模板，覆盖以下分类： - **通用质量**：评估回答的基本质量指标 -- **智能体**：专门用于评测智能体应用 +- **智能体**：专门用于评测[智能体应用](../concepts/agent-application.md) - **文本匹配**：精确规则文本匹配 - **文本相似度**：计算文本相似度得分 - **格式校验**：验证输出格式规范性 @@ -137,9 +137,9 @@ 识别 BadCase → 分析归因定位问题 → 实施针对性优化 → 发布新版本再次评测 → 对比结果确认改进。若效果未达预期则继续迭代。 -## 计费说明 +## [计费](../concepts/billing.md)说明 -评测任务调用大模型产生的 [Token](../concepts/token.md) 费用正常计费。自动评测的评测集生成和评估均会消耗 [Token](../concepts/token.md)，预估平均消耗仅为参考值，最终以实际账单为准。评估器模型当前限时免费。 +评测任务调用大模型产生的 [Token](../concepts/token.md) 费用正常[计费](../concepts/billing.md)。自动评测的评测集生成和评估均会消耗 [Token](../concepts/token.md)，预估平均消耗仅为参考值，最终以实际账单为准。评估器模型当前限时免费。 ## 常见问题 @@ -164,6 +164,19 @@ + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/application-monitoring.md b/skills/bailian-docs-llm-wiki/wiki/guides/application-monitoring.md index 9ae2c8d0..41ce28b6 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/application-monitoring.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/application-monitoring.md @@ -61,7 +61,7 @@ ### 添加到[评测](../concepts/evaluation.md)集 -应用观测支持将 Span 数据直接加入[评测](../concepts/evaluation.md)集，将真实线上调用作为评测样本。配置时需选择目标评测集、导入方式（追加数据或全量覆盖）并完成字段映射。每个评测集最多支持 50 个字段映射。 +应用观测支持将 Span 数据直接加入[评测](../concepts/evaluation.md)集，将真实线上调用作为[评测](../concepts/evaluation.md)样本。配置时需选择目标[评测](../concepts/evaluation.md)集、导入方式（追加数据或全量覆盖）并完成字段映射。每个[评测](../concepts/evaluation.md)集最多支持 50 个字段映射。 ## 节点类型 @@ -75,17 +75,17 @@ | AGENT | 对智能体的调用 | | RETRIEVER | 检索操作；KnowledgeRetriever 表示在[知识库](../concepts/knowledge-base.md)中检索。子节点名称含 TextRetriever（改进 BM25，默认返回 100 个切片）、VectorRetriever（向量检索，默认返回 100 个切片） | | REWRITER | 基于会话上下文调整原始 Prompt 以提升检索效果 | -| EMBEDDING | 将 Prompt 转为向量，[Token](../concepts/token.md) 量为本次向量化的 [Token](../concepts/token.md) 数 | +| EMBEDDING | 将 Prompt 转为向量，[Token](../concepts/token.md) 量为本次[向量化](../concepts/embedding.md)的 [Token](../concepts/token.md) 数 | | RERANKER | 计算文本切片相似度分数并降序排列 | | LLM | 大模型推理/文本生成，[Token](../concepts/token.md) 量 = 输入 + 输出；延时包含输出回复过程 | | TOOL | 插件调用（官方或自定义） | | GUARDRAIL | 阿里绿网调用；ManualIntervention 为用户干预规则，SystemIntervention 为系统干预规则 | -> 目前暂不支持观测长期记忆中的检索过程；TextRetriever 与 VectorRetriever 默认返回 100 个切片，暂不支持调整数量。 +> 目前暂不支持观测[长期记忆](../concepts/long-term-memory.md)中的检索过程；TextRetriever 与 VectorRetriever 默认返回 100 个切片，暂不支持调整数量。 ### [工作流](../concepts/workflow.md)应用节点 -除上述 CHAIN、RETRIEVER、REWRITER、EMBEDDING、RERANKER、LLM、GUARDRAIL 外，还包含工作流专属节点：START（开始）、END（结束）、API、CLASSIFIER（意图分类）、TEXT_CONVERTER（文本转换）、SCRIPT（脚本转换）、CONDITION（条件判断）、FUNCTION_COMPUTE（函数计算）、APP_FLOW。 +除上述 CHAIN、RETRIEVER、REWRITER、EMBEDDING、RERANKER、LLM、GUARDRAIL 外，还包含[工作流](../concepts/workflow.md)专属节点：START（开始）、END（结束）、API、CLASSIFIER（意图分类）、TEXT_CONVERTER（文本转换）、SCRIPT（脚本转换）、CONDITION（条件判断）、FUNCTION_COMPUTE（函数计算）、APP_FLOW。 ### 高代码应用节点 @@ -98,7 +98,7 @@ ## 关键指标说明 - **延时（调用时长）**：对 LLM 节点，包含输出回复的完整过程。 -- **[Token](../concepts/token.md) 量**：Embedding 节点为本次向量化 [Token](../concepts/token.md) 数；LLM 节点为输入 [Token](../concepts/token.md) + 输出 [Token](../concepts/token.md)。 +- **[Token](../concepts/token.md) 量**：Embedding 节点为本次[向量化](../concepts/embedding.md) [Token](../concepts/token.md) 数；LLM 节点为输入 [Token](../concepts/token.md) + 输出 [Token](../concepts/token.md)。 - **数据时效**：指标更新频率为分钟级，调用记录最长可查 30 天。 - **应用总量 / 平均延时**：用于评估应用运营效果与成本，详见 [应用观测](../../raw/application-user-guide/application-monitoring/application-observation.md)。 @@ -119,6 +119,19 @@ + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/application-permission-management.md b/skills/bailian-docs-llm-wiki/wiki/guides/application-permission-management.md index 32803f47..722261eb 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/application-permission-management.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/application-permission-management.md @@ -6,7 +6,7 @@ 百炼的身份管理基于以下三种角色，权限范围自上而下递减： -- **超级管理员**：可跨空间统一管理用户权限、空间可用模型、空间模型限流和 [API Key](../concepts/api-key.md)。包含两类账号：阿里云主账号，以及拥有 `AliyunBailianFullAccess`（百炼管理员）系统策略的 RAM 用户。超级管理员可通过百炼全局管理菜单（北京 / 新加坡 / 弗吉尼亚）为任意 RAM 用户授权任意地域、任意空间的几乎所有权限，仅 OpenAPI 接口权限需阿里云主账号添加。 +- **超级管理员**：可跨空间统一管理用户权限、空间可用模型、空间模型[限流](../concepts/rate-limit.md)和 [API Key](../concepts/api-key.md)。包含两类账号：阿里云主账号，以及拥有 `AliyunBailianFullAccess`（百炼管理员）系统策略的 RAM 用户。超级管理员可通过百炼全局管理菜单（北京 / 新加坡 / 弗吉尼亚）为任意 RAM 用户授权任意地域、任意空间的几乎所有权限，仅 OpenAPI 接口权限需阿里云主账号添加。 - **[业务空间](../concepts/workspace.md)管理员**：拥有访问某个[业务空间](../concepts/workspace.md)「权限管理」页面的 RAM 用户，只负责该特定[业务空间](../concepts/workspace.md)内的用户权限和资源管理。管理员权限包含可访问该[业务空间](../concepts/workspace.md)下所有页面的权限。 - **普通用户**：根据分配的权限使用资源，可访问/使用被授权的空间、页面、资源。 @@ -14,7 +14,7 @@ | [业务空间](../concepts/workspace.md)权限 | 超级管理员 | [业务空间](../concepts/workspace.md)管理员 | 普通用户 | | --- | --- | --- | --- | -| 允许特定模型调用 & 限流 | 支持 | 不支持 | 不支持 | +| 允许特定模型调用 & [限流](../concepts/rate-limit.md) | 支持 | 不支持 | 不支持 | | 允许特定[模型调优](../concepts/fine-tuning.md) | 支持 | 不支持 | 不支持 | | 允许特定[模型部署](../concepts/model-deployment.md) | 支持 | 不支持 | 不支持 | | 用户管理 | 支持 | 支持 | 不支持 | @@ -29,23 +29,23 @@ 百炼按地理区域划分资源和[业务空间](../concepts/workspace.md)，**单个[业务空间](../concepts/workspace.md)不能跨地域存在**，即使是各地域的默认[业务空间](../concepts/workspace.md)，也是不同的空间。[业务空间](../concepts/workspace.md)是精细化权限管理的最小单元，可管理以下维度（默认[业务空间](../concepts/workspace.md)无法设置这些限制）： -- **限制模型调用**：管理某个模型可否在该[业务空间](../concepts/workspace.md)调用（控制台 & API），并设置该模型的请求数限流和 [Token](../concepts/token.md) 限流。默认[业务空间](../concepts/workspace.md)所有模型均可调用且无法限流。 +- **限制模型调用**：管理某个模型可否在该[业务空间](../concepts/workspace.md)调用（控制台 & API），并设置该模型的请求数[限流](../concepts/rate-limit.md)和 [Token](../concepts/token.md) 限流。默认[业务空间](../concepts/workspace.md)所有模型均可调用且无法限流。 - **限制模型训练**：管理某个模型可否在该[业务空间](../concepts/workspace.md)进行调优和调优后部署。默认[业务空间](../concepts/workspace.md)所有支持调优的模型均可调优及部署。 - **限制[模型部署](../concepts/model-deployment.md)**：管理某个模型可否在该[业务空间](../concepts/workspace.md)直接部署。默认[业务空间](../concepts/workspace.md)所有支持部署的模型均可部署。 -- **用户控制台权限管理**：管理某个 RAM 用户是否能使用该业务空间控制台的功能及能使用哪些功能，但无法限制归属该用户的 [API Key](../concepts/api-key.md) 的调用。阿里云主账号无须设置，可访问所有业务空间的所有页面。 +- **用户控制台权限管理**：管理某个 RAM 用户是否能使用该[业务空间](../concepts/workspace.md)控制台的功能及能使用哪些功能，但无法限制归属该用户的 [API Key](../concepts/api-key.md) 的调用。阿里云主账号无须设置，可访问所有[业务空间](../concepts/workspace.md)的所有页面。 -关于业务空间的地域隔离与限流细节，可进一步参考 [权限管理](../../raw/application-user-guide/application-permission-management/application-permission-management-overview.md)。 +关于[业务空间](../concepts/workspace.md)的地域隔离与限流细节，可进一步参考 [权限管理](../../raw/application-user-guide/application-permission-management/application-permission-management-overview.md)。 ## API-Key 权限 -单个 [API Key](../concepts/api-key.md) 只能归属一个地域内的一个业务空间和一个用户，且不能转移。[API Key](../concepts/api-key.md) 可调用的功能和模型限流与**归属业务空间**的权限保持一致，不受用户控制台权限管理的影响，也无需为不同模型（如文生文、文生图、语音合成）创建不同的 [API Key](../concepts/api-key.md)。 +单个 [API Key](../concepts/api-key.md) 只能归属一个地域内的一个[业务空间](../concepts/workspace.md)和一个用户，且不能转移。[API Key](../concepts/api-key.md) 可调用的功能和模型限流与**归属[业务空间](../concepts/workspace.md)**的权限保持一致，不受用户控制台权限管理的影响，也无需为不同模型（如文生文、文生图、语音合成）创建不同的 [API Key](../concepts/api-key.md)。 [API Key](../concepts/api-key.md) 的状态随归属用户操作变化： | 触发操作 | 主账号的 [API Key](../concepts/api-key.md) | RAM 账号的 [API Key](../concepts/api-key.md) | | --- | --- | --- | | 主动删除 [API Key](../concepts/api-key.md) | 失效，不可恢复 | 失效，不可恢复 | -| 将账号移出业务空间 | — | 失效（重新加入后恢复生效） | +| 将账号移出[业务空间](../concepts/workspace.md) | — | 失效（重新加入后恢复生效） | | 在 RAM 控制台删除账号/角色 | — | 失效，不可恢复 | | 为 [API Key](../concepts/api-key.md) 设置 IP 访问白名单 | 华北2（北京）地域支持 | 华北2（北京）地域支持 | @@ -55,7 +55,7 @@ ## OpenAPI 接口权限 -RAM 用户默认无权调用百炼应用的数据、[知识库](../concepts/knowledge-base.md)、Prompt 工程及长期记忆等功能的 Open API。需由阿里云主账号在 RAM 控制台为 RAM 用户添加以下权限之一： +RAM 用户默认无权调用百炼应用的数据、[知识库](../concepts/knowledge-base.md)、Prompt 工程及[长期记忆](../concepts/long-term-memory.md)等功能的 Open API。需由阿里云主账号在 RAM 控制台为 RAM 用户添加以下权限之一： - `AliyunBailianDataFullAccess`：可调用百炼应用 API 目录下的所有 API。 - `AliyunBailianDataReadOnlyAccess`：可调用百炼应用 API 目录下的只读类 API，如 `DescribeFile`、`GetIndexJobStatus` 等。 @@ -73,43 +73,43 @@ RAM 用户默认无权查看阿里云账单和购买预付费产品，需在 RAM 需要阿里云主账号或具备 `AliyunRAMFullAccess` 系统策略的 RAM 用户操作。前往 RAM 控制台为 RAM 用户添加 `AliyunBailianFullAccess` 和 `AliyunBSSOrderAccess` 权限后，即可通过百炼全局管理菜单授权任意地域、空间的权限并购买预付费产品。 -### 设置业务空间管理员 +### 设置[业务空间](../concepts/workspace.md)管理员 -需超级管理员或业务空间管理员操作。在百炼控制台左侧导航栏「权限管理」页签内为 RAM 用户添加「管理员」权限。 +需超级管理员或[业务空间](../concepts/workspace.md)管理员操作。在百炼控制台左侧导航栏「权限管理」页签内为 RAM 用户添加「管理员」权限。 ### 设置模型调用权限 -1. 不使用默认业务空间时，需先由超级管理员为业务空间开通特定模型的模型调用权限。 -2. 通过控制台调用时，需由超级管理员或业务空间管理员为 RAM 用户添加：**模型体验-操作**（控制台调用模型）、**批量推理-操作**（支持批量推理）、**模型观测-操作**（查看 [Token](../concepts/token.md) 消耗量）。 -3. 通过 API 调用时，需为 RAM 用户在对应业务空间创建或分配 API Key。 +1. 不使用默认[业务空间](../concepts/workspace.md)时，需先由超级管理员为[业务空间](../concepts/workspace.md)开通特定模型的模型调用权限。 +2. 通过控制台调用时，需由超级管理员或[业务空间](../concepts/workspace.md)管理员为 RAM 用户添加：**模型体验-操作**（控制台调用模型）、**批量推理-操作**（支持批量推理）、**模型观测-操作**（查看 [Token](../concepts/token.md) 消耗量）。 +3. 通过 API 调用时，需为 RAM 用户在对应[业务空间](../concepts/workspace.md)创建或分配 [API Key](../concepts/api-key.md)。 ### 设置[模型调优](../concepts/fine-tuning.md)权限 -1. 不使用默认业务空间时，需先由超级管理员为业务空间开通特定模型的[模型调优](../concepts/fine-tuning.md)（训练）权限。 +1. 不使用默认[业务空间](../concepts/workspace.md)时，需先由超级管理员为[业务空间](../concepts/workspace.md)开通特定模型的[模型调优](../concepts/fine-tuning.md)（训练）权限。 2. 在「权限管理」页签内为 RAM 用户添加以下权限：**模型体验-操作**、**模型调优-操作**、**我的模型-操作**（管理调优后模型快照）、**[模型部署](../concepts/model-deployment.md)-操作**（部署调优后的模型）、**模型[评测](../concepts/evaluation.md)-操作**、**数据管理-操作**（管理调优数据集）、**模型观测-操作**。 -3. 通过 API 调优时，无需额外控制台权限，只需为 RAM 用户分配 API Key 即可。 +3. 通过 API 调优时，无需额外控制台权限，只需为 RAM 用户分配 [API Key](../concepts/api-key.md) 即可。 完整的权限配置流程与截图说明见 [权限管理](../../raw/application-user-guide/application-permission-management/application-permission-management-overview.md)。 ## 生产环境实践 - **空间规划策略**：推荐按环境划分（开发 `project-dev-workspace`、测试 `project-test-workspace`、预发与生产 `project-prod-workspace`），实现严格的环境隔离；也可按业务线划分（如 `marketing-team-workspace`、`customer-team-workspace`），便于权限和成本管理。 -- **限流策略**：将主账号总配额按比例分配给各业务空间并预留缓冲。例如账号总配额 1000 QPM，可分配生产 600 QPM（60%）、测试 200 QPM（20%）、开发 100 QPM（10%）、预留缓冲 100 QPM（10%），以应对突发流量。 +- **限流策略**：将主账号总配额按比例分配给各[业务空间](../concepts/workspace.md)并预留缓冲。例如账号总配额 1000 QPM，可分配生产 600 QPM（60%）、测试 200 QPM（20%）、开发 100 QPM（10%）、预留缓冲 100 QPM（10%），以应对突发流量。 ## 限制与注意事项 -- 业务空间不能跨地域存在；不同地域的默认业务空间也是不同空间。 -- 默认业务空间无法设置模型调用、调优、部署限制，所有模型均按默认策略可用且无法限流。 -- API Key 不可跨业务空间或跨用户转移；账号移出业务空间后其 API Key 失效，重新加入后恢复。 -- OpenAPI 接口权限、账单与预付费权限必须由阿里云主账号在 RAM 控制台授权，业务空间管理员无法授予。 +- [业务空间](../concepts/workspace.md)不能跨地域存在；不同地域的默认[业务空间](../concepts/workspace.md)也是不同空间。 +- 默认[业务空间](../concepts/workspace.md)无法设置模型调用、调优、部署限制，所有模型均按默认策略可用且无法限流。 +- [API Key](../concepts/api-key.md) 不可跨[业务空间](../concepts/workspace.md)或跨用户转移；账号移出[业务空间](../concepts/workspace.md)后其 [API Key](../concepts/api-key.md) 失效，重新加入后恢复。 +- OpenAPI 接口权限、账单与预付费权限必须由阿里云主账号在 RAM 控制台授权，[业务空间](../concepts/workspace.md)管理员无法授予。 - `AliyunBSSReadOnlyAccess` / `AliyunBSSOrderAccess` 为全产品级权限，授权范围远超百炼本身，需谨慎。 - 开通 AI 安全护栏、模型监控、应用观测等功能，建议使用阿里云主账号在控制台一次性授权开通。 ## 常见问题 -- **如何获取业务空间 ID**：参考应用开发的「获取 Workspace ID」文档。 -- **如何使用子业务空间调用模型**：无需特殊设置，使用子业务空间的 API Key 即可。 -- **如何使用特定业务空间的应用**：使用 API 管理和调用特定业务空间的应用时，需同时设置 APP ID 和 Workspace ID。 +- **如何获取[业务空间](../concepts/workspace.md) ID**：参考应用开发的「获取 Workspace ID」文档。 +- **如何使用子[业务空间](../concepts/workspace.md)调用模型**：无需特殊设置，使用子[业务空间](../concepts/workspace.md)的 [API Key](../concepts/api-key.md) 即可。 +- **如何使用特定[业务空间](../concepts/workspace.md)的应用**：使用 API 管理和调用特定[业务空间](../concepts/workspace.md)的应用时，需同时设置 APP ID 和 Workspace ID。 ## 来源文档 @@ -128,6 +128,19 @@ RAM 用户默认无权查看阿里云账单和购买预付费产品，需在 RAM + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/application-publishing-and-sharing.md b/skills/bailian-docs-llm-wiki/wiki/guides/application-publishing-and-sharing.md index 0dd5decf..6988f4bd 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/application-publishing-and-sharing.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/application-publishing-and-sharing.md @@ -8,24 +8,24 @@ > **注意**：**Agent 2.0** [智能体应用](../concepts/agent-application.md)仅支持通过 API 调用，**不支持**上述任何分享渠道。若需分享，请确认应用版本。 -前提条件是已有构建好且**已发布**的智能体应用。所有分享渠道通过百炼控制台 **应用管理 → 目标应用卡片 → 发布** 进入。详见 [分享智能体应用](../../raw/application-user-guide/application-publishing-and-sharing/share-an-application.md)。 +前提条件是已有构建好且**已发布**的[智能体应用](../concepts/agent-application.md)。所有分享渠道通过百炼控制台 **应用管理 → 目标应用卡片 → 发布** 进入。详见 [分享智能体应用](../../raw/application-user-guide/application-publishing-and-sharing/share-an-application.md)。 ## 分享渠道 -智能体应用（Agent 1.0）支持四种分享或发布方式，以及音视频实时互动： +[智能体应用](../concepts/agent-application.md)（Agent 1.0）支持四种分享或发布方式，以及音视频实时互动： - **UI 应用 / 魔笔分享渠道**：进入 UI 设计器编辑并发布界面，在 **环境部署** 中获取应用地址后分享。持有链接的阿里云用户均可访问，**单击「下线」可停止服务**。 - **钉钉**：在 **发布平台** 授权计算巢 AppFlow（SLR 关联 + API-KEY 加密传输），配置钉钉模板 ID、Client ID、Client Secret 后创建，最终得到 **回调地址** 用于配置钉钉机器人。钉钉机器人的 **消息接收模式必须选 HTTP 模式**，选 Stream 模式会导致无法返回消息；并需申请 `Card.Streaming.Write` 与 `Card.Instance.Write` 权限。 - **微信公众号**：若已在钉钉步骤授权过则无需再次授权。选择 API KEY 与微信凭据（需 AppID 授权）后创建，生成二维码供用户扫码体验。 -- **音视频实时互动**：仅支持图文对话类应用（含智能体与工作流）。支持 H5/APP 扫码与 SDK 集成（基于 AICallKit SDK，含 UI/不含 UI 两种方案）两种渠道。 +- **音视频实时互动**：仅支持图文对话类应用（含智能体与[工作流](../concepts/workflow.md)）。支持 H5/APP 扫码与 SDK 集成（基于 AICallKit SDK，含 UI/不含 UI 两种方案）两种渠道。 > **注意**：临时体验二维码（音视频互动）与从已有应用发布的 UI 体验链接，**有效期均为 24 小时**，过期需重新生成或重新发布。 -**权限与计费**：共享应用可被应用创建者（主账号）、RAM 用户及持有链接的其他阿里云用户访问；所有通过分享链接产生的费用由**应用创建者 UID 账号**承担。上述钉钉/微信配置细节见 [分享智能体应用](../../raw/application-user-guide/application-publishing-and-sharing/share-an-application.md)。 +**权限与[计费](../concepts/billing.md)**：共享应用可被应用创建者（主账号）、RAM 用户及持有链接的其他阿里云用户访问；所有通过分享链接产生的费用由**应用创建者 UID 账号**承担。上述钉钉/微信配置细节见 [分享智能体应用](../../raw/application-user-guide/application-publishing-and-sharing/share-an-application.md)。 ## 组件化发布与接入 -智能体或工作流应用可发布为模块化组件，供其他应用复用。发布路径有三处：发布应用时勾选 **发布应用组件**、在 **发布渠道** 的组件区域 **+ 创建**、或在控制台 **组件管理** 面板创建。详见 [使用智能体或工作流作为组件](../../raw/application-user-guide/application-publishing-and-sharing/use-agent-or-workflow-as-component.md)。 +智能体或[工作流](../concepts/workflow.md)应用可发布为模块化组件，供其他应用复用。发布路径有三处：发布应用时勾选 **发布应用组件**、在 **发布渠道** 的组件区域 **+ 创建**、或在控制台 **组件管理** 面板创建。详见 [使用智能体或工作流作为组件](../../raw/application-user-guide/application-publishing-and-sharing/use-agent-or-workflow-as-component.md)。 ### 关键参数 @@ -35,11 +35,11 @@ - **别名**：调用者只能看到别名，用于避免参数重名。 - **传参方式**： - - **业务透传**：智能体中由使用者提供，工作流中由上游节点提供。 + - **业务透传**：智能体中由使用者提供，[工作流](../concepts/workflow.md)中由上游节点提供。 - **模型识别**：智能体中由大模型根据参数描述自动推断填充。 - **组件描述**：接入智能体时，大模型据此自动判断是否调用；接入工作流时仅作说明，不影响运行。 -> **注意**：即使参数的传参方式设为**模型识别**，在**工作流应用**中应用也**不会**自动推断参数值，必须像业务透传一样从上游节点明确提供输入值。模型识别仅在智能体应用中生效。 +> **注意**：即使参数的传参方式设为**模型识别**，在**工作流应用**中应用也**不会**自动推断参数值，必须像业务透传一样从上游节点明确提供输入值。模型识别仅在[智能体应用](../concepts/agent-application.md)中生效。 ### 接入方式 @@ -56,9 +56,9 @@ UI 设计器集成阿里云多端低代码平台魔笔的能力，提供可视化拖放式界面构建，可将应用发布为网页 UI。详见 [UI设计器](../../raw/application-user-guide/application-publishing-and-sharing/ui-designer.md)。 -**前提**：百炼应用、API Key 和 UI 设计必须归属于**同一业务空间**，否则无法在 UI 创建时选择对应的 API Key 与应用。 +**前提**：百炼应用、[API Key](../concepts/api-key.md) 和 UI 设计必须归属于**同一[业务空间](../concepts/workspace.md)**，否则无法在 UI 创建时选择对应的 [API Key](../concepts/api-key.md) 与应用。 -发布方式有两种：从已有应用一键创建 UI（自动填充标题、API-KEY、智能体、预设问题等），或通过 UI 设计器从模板（空白 / 智能出行助手 / 智能体门户 / AI 基础对话 / 企业 AI 知识库 Lite）创建。核心流程为：创建 UI → 拖放组件编辑页面 → 发布与分享。 +发布方式有两种：从已有应用一键创建 UI（自动填充标题、API-KEY、智能体、预设问题等），或通过 UI 设计器从模板（空白 / 智能出行助手 / 智能体门户 / AI 基础对话 / 企业 AI [知识库](../concepts/[knowledge](../api/knowledge.md)-base.md) Lite）创建。核心流程为：创建 UI → 拖放组件编辑页面 → 发布与分享。 **环境对比**： @@ -71,7 +71,7 @@ UI 设计器集成阿里云多端低代码平台魔笔的能力，提供可视 **权限**：UI 应用发布后默认持有链接的阿里云用户可访问，也可开启 **允许匿名访问** 并通过权限组限制其只访问会话页。 -**计费**：UI 设计器功能本身不计费，但会产生模型调用费用、UI 应用数据（超出 1GB 免费文件存储与 0.3GB 免费数据库容量后按量计费）、以及生产环境发布所需的套餐订阅费用。 +**[计费](../concepts/billing.md)**：UI 设计器功能本身不[计费](../concepts/billing.md)，但会产生模型调用费用、UI 应用数据（超出 1GB 免费文件存储与 0.3GB 免费数据库容量后按量[计费](../concepts/billing.md)）、以及生产环境发布所需的套餐订阅费用。 对于工作流应用，若配置了文件类型的自定义参数，需在 UI 设计器中指定 `{{{file_name:files[0]}}}`（将 `file_name` 替换为实际变量名），才能正确读取用户上传的文件。 @@ -84,3 +84,16 @@ UI 设计器集成阿里云多端低代码平台魔笔的能力，提供可视 + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/application-support.md b/skills/bailian-docs-llm-wiki/wiki/guides/application-support.md index adc2411d..8a4a0cdb 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/application-support.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/application-support.md @@ -1,87 +1,106 @@ # application [support](support.md) -本页汇总阿里云百炼平台应用与[知识库](../concepts/knowledge-base.md)使用过程中的常见问题、关键限制以及相关协议入口，帮助开发者在接入应用、调用插件、管理数据与合规备案时快速定位答案。内容主要参考 [常见问题](../../raw/application-user-guide/application-support/application-faq.md) 与 [相关协议](../../raw/application-user-guide/application-support/application-related-agreements.md)。 +阿里云百炼平台为开发者提供多层次的售后支持体系，涵盖 7×24 基础咨询、标准工单、付费增值服务等渠道，同时明确了平台自身与第三方工具之间的支持边界。本页汇总应用开发中的常见问题解答、相关服务协议及售后服务范围，帮助开发者快速定位支持资源。 -## 应用中心 +## 应用中心常见问题 -### 插件能力 +以下问题摘自 [常见问题](../../raw/application-user-guide/application-support/application-faq.md)，覆盖插件、RAG 检索、API 调用等高频场景。 -百炼应用中心官方提供六款插件：Python 代码解释器、计算器、图片生成、夸克搜索、生成二维码、GitHub 搜索，其中部分插件需申请通过后方可使用。自定义插件服务本身暂不收费，但配置智能体 API 时若涉及 [prompt](prompt.md) 优化、应用调用及测试窗测试，则会产生费用。 +**官方插件类型** -- **插件理解机制**：自定义 API 插件遵循协议传给大模型理解；自定义函数则由大模型学习传入的参数信息并返回完整结果。 -- **header 透传**：百炼调用自定义插件时**不支持自定义 header**，仅支持 `authorization`。若业务场景需要透传 header，需在服务端侧另行处理。 +目前系统提供六款内置插件：Python 代码解释器、计算器、图片生成、夸克搜索、生成二维码、GitHub 搜索。部分插件需申请审批后方可使用。自定义插件服务暂不收费，但涉及 [prompt](prompt.md) 优化、应用调用及测试窗测试时会产生费用。 -### Agent 与 Assistant API 的区别 +**自定义插件与 Assistant API** -Agent 偏重于调整插件模型与基于上下文的理解，由用户自行开发；Assistant API 则提供各类能力以方便调优。两者面向不同定制粒度，可根据应用复杂度选择。 +- 自定义 API 插件遵循协议传给大模型理解，[函数调用](../concepts/function-calling.md)会返回完整结果。 +- 不支持透传自定义 header，仅支持 `authorization`。 +- Agent 与 Assistant API 的核心区别：Agent 允许开发者自行调整插件模型和上下文理解逻辑；Assistant API 提供多种类以方便调优。 -### 输出控制 +**RAG 检索增强** -- **流式与增量输出**：默认为全量回复，若需增量输出，可设置 `stream=True`（[流式输出](../concepts/streaming-output.md)）与 `incremental_output=True`（增量式[流式输出](../concepts/streaming-output.md)）。 -- **Markdown 加粗**：模型输出中的 `**xxxxx**` 是 Markdown 加粗标识，需在前端渲染时解析 md 语法即可正常显示。 +- 多[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)检索采用**并行**方式，各[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)按用户配置独立检索，最终按得分取 topN。 +- 模型回复不准确时，可点击回复下方的问题反馈按钮提交，或复制 RequestId 后提交阿里云工单。 -## 知识检索（RAG） +**流式/增量输出** -RAG（[检索增强生成](../concepts/rag.md)）在问答系统、对话系统、文本摘要、知识图谱构建与推理、教育与培训、客户服务、新闻与内容创作、智能搜索与推荐等多个领域均有应用。 +若需增量[流式输出](../concepts/streaming.md)，在 SDK 调用时设置： -- **检索顺序**：RAG 检索为**并行**方式，依据每个[知识库](../concepts/knowledge-base.md)的用户配置进行检索，再根据得分选取 topN 结果。 -- **回复不准确优化**：可点击模型回复下方的问题反馈按钮，勾选问题类型提交；也可复制 RequestId 通过阿里云工单反馈。详见 [常见问题](../../raw/application-user-guide/application-support/application-faq.md)。 +```python +stream=True # 开启流式输出 +incremental_output=True # 增量式流式输出 +``` -## 数据管理 +**Markdown 渲染** -### 文件上传 +AI 输出中的 `**text**` 为标准 Markdown 加粗语法，需在前端渲染时解析 Markdown 并做对应展示。 -- **格式限制**：上传 PDF 文件时后缀必须为小写 `pdf`，否则会触发错误码 `140010`（"上传文件仅支持 pdf/doc/docx 文件"）。 -- **MD5 参数**：上传文件接口的必填 MD5 参数用于校验上传文件的完整性。 -- **容量上限**：每个[业务空间](../concepts/workspace.md)最多上传 10 万个文档，超出需提交阿里云工单申请扩容。 +## 数据管理常见问题 -### 结构化数据导入 +**文件上传** -结构化数据导入后若出现条数缺失（如 100 条仅导入 20 条），通常是因为表格中存在空行。产品策略规定：遇到空行后续数据不再识别；若第一行为空行，则整表视为空文件。 +- PDF 文件后缀必须为小写 `pdf`，否则报错码 `140010`。 +- 每个[业务空间](../concepts/workspace.md)最多上传 10 万个文档；如需扩容，提交阿里云工单申请。 +- 上传接口中的 MD5 参数用于校验文件完整性。 -## 应用与小程序备案 +**结构化数据导入** -产品接入通义千问大模型后，若需上架应用市场或小程序平台，需完成备案并申请合作协议： +若结构化数据导入后条数少于预期，请检查表格中是否含空行——空行之后的数据不会被识别；若第一行为空行，文件将被视为空文件。 -1. 参考[应用合规备案](https://help.aliyun.com/zh/model-studio/compliance-and-launch-filing-guide-for-ai-apps-powered-by-the-tongyi-model)进行备案。 -2. [提交工单](https://smartservice.console.aliyun.com/service/create-ticket)申请通义千问系列模型的合作协议。 +## 应用/小程序备案 -## 相关协议 - -接入百炼前请阅读以下协议条款： - -- [阿里云百炼服务协议](https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html) -- [阿里云百炼服务特别说明](https://help.aliyun.com/zh/model-studio/bailian-service-notes) -- [开源模型协议条款说明](https://help.aliyun.com/zh/model-studio/open-source-model-terms) - -协议入口同时收录在 [相关协议](../../raw/application-user-guide/application-support/application-related-agreements.md) 中，建议在正式接入前逐项确认。 +接入通义千问大模型并准备上架应用市场或小程序平台时： -## 常见限制与注意事项 +1. 参考[应用合规备案](https://help.aliyun.com/zh/model-studio/compliance-and-launch-filing-guide-for-ai-apps-powered-by-the-tongyi-model)完成备案。 +2. 提交阿里云工单申请通义千问系列模型的合作协议。 -- 自定义 header 不被支持，仅 `authorization` 可透传。 -- PDF 后缀必须为小写，避免触发 `140010` 错误码。 -- 结构化数据表格中的空行会导致后续数据被截断识别。 -- [业务空间](../concepts/workspace.md)文档数上限为 10 万，超额需工单申请。 -- 自定义插件本身免费，但 [prompt](prompt.md) 优化、应用调用与测试窗测试会收费。 - -## 来源文档 +## 售后服务范围 -- [常见问题](../../raw/application-user-guide/application-support/application-faq.md) -- [相关协议](../../raw/application-user-guide/application-support/application-related-agreements.md) +详细说明见 [阿里云百炼平台售后服务范围说明](../../raw/application-user-guide/application-support/application-after-sales-service-scope.md)。 +### 基础支持（免费，服务期内） +通过官网、电话（95187、400）及阿里云 APP 提供 7×24 咨询，支持范围： +- 模型服务与产品功能、架构咨询 +- 使用与配置最佳实践 +- API 及官方 SDK 故障诊断 +- 管理控制台相关问题 +- 账号、财务、合同及计费咨询 +### 付费增值服务 +阿里云提供付费版售后增值服务（含支持计划），需在阿里云官网订购后生效。如项目需要更深度的技术支持（业务代码编写指导、定制化集成方案等），建议联系阿里云商务经理沟通定制化方案。 +### 第三方工具对接的支持边界 +**支持范围（方向性建议）**： +- 确认百炼 API 接口及服务端可用状态 +- 官方 API 调用示例及 SDK 使用说明参考 +- 协助核查服务端调用明细和计费记录 +- 基本连通性测试建议（如 curl 测试） +**不在支持范围内**： +- 第三方工具（Cursor、Windsurf、Cline 等）的安装、配置、升级及使用指导 +- 第三方工具内部功能、交互逻辑问题排查 +- 用户业务代码的编写、调试与实现 +- 用户本地环境（内网、代理、VPN、防火墙等）导致的连通性或兼容性问题 +- 第三方工具显示的 [Token](../concepts/token.md) 数/费用预估与阿里云计费数据之间的差异解释 +> **注意**：第三方工具不构成阿里云的代理或联合服务主体，阿里云不对外部第三方工具的任何陈述、承诺或行为承担责任。第三方工具的运行维护责任由用户及相应工具提供方承担。 +## 相关协议 +以下协议文本详见 [相关协议](../../raw/application-user-guide/application-support/application-related-agreements.md)： +- [阿里云百炼服务协议](https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20230728213935489/20230728213935489.html) +- [阿里云百炼体验功能特别说明](https://terms.alicdn.com/legal-agreement/terms/common_platform_service/20260716114753386/20260716114753386.html) +- [开源模型协议条款说明](https://help.aliyun.com/zh/model-studio/open-source-model-terms) +## 来源文档 +- [常见问题](../../raw/application-user-guide/application-support/application-faq.md) +- [相关协议](../../raw/application-user-guide/application-support/application-related-agreements.md) +- [阿里云百炼平台售后服务范围说明](../../raw/application-user-guide/application-support/application-after-sales-service-scope.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/application-use-cases.md b/skills/bailian-docs-llm-wiki/wiki/guides/application-use-cases.md index cbd307ef..e9deebfd 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/application-use-cases.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/application-use-cases.md @@ -38,7 +38,7 @@ 百炼应用侧： - Prompt：可设置角色人设引导回答，如“你叫小助，可以帮助用户解答产品选购、使用等方面的问题。” -- 知识库调用方式：可选**必定调用**；知识文档支持配置相似度阈值与权重。 +- [知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)调用方式：可选**必定调用**；知识文档支持配置相似度阈值与权重。 - 文件处理：可选全文引用、切片检索或自定义处理。 - 向量存储：标准版默认即可；如需集中管理多应用向量数据可选 ADB-PG。 @@ -47,7 +47,7 @@ - 模型参数：模型选择、温度（越高随机性越高）、最大回复长度、携带上下文轮数（设为 1 时不参考历史）。 - RAG 参数：召回片段数（越大参考越多但噪声可能增加）、相似度阈值（越大参考越少但噪声减少，为 0 不剔除）。 - 嵌入模型：默认用百炼 embedding API；可改用本地部署的 GTE 文本向量模型（如 `iic/nlp_gte_sentence-embedding_chinese-large`）。 -- 文件限制：受 embedding API 限流，不建议传入超过 100 MB 的文件。 +- 文件限制：受 embedding API [限流](../concepts/rate-limit.md)，不建议传入超过 100 MB 的文件。 AppFlow 连接流： @@ -61,7 +61,7 @@ AppFlow 连接流： 企业微信 / 公众号 / 钉钉：发布连接流后，在目标平台配置 API 接收消息（URL 填 WebhookUrl，[Token](../concepts/token.md) / EncodingAESKey 填 AppFlow 凭证生成的值），配置可信 IP，即在聊天中 @机器人 或直接对话使用。 -本地 RAG：解压 `local_rag.zip`，Python 3.8–3.12 环境安装依赖，配置百炼 [API Key](../concepts/api-key.md) 环境变量，运行 `uvicorn main:app --port 7866`，访问 `http://127.0.0.1:7866`。支持临时性文件上传（对话框直接传 pdf/docx/txt/xlsx/csv，刷新后失效）与长期知识库创建（上传到 File/Unstructured 或 File/Structured 后在界面创建知识库存于 VectorStore）。通过 Gradio 界面下方的“通过 API 使用”可获取 API 文档集成到业务。 +本地 RAG：解压 `local_rag.zip`，Python 3.8–3.12 环境安装依赖，配置百炼 [API Key](../concepts/api-key.md) 环境变量，运行 `uvicorn main:app --port 7866`，访问 `http://127.0.0.1:7866`。支持临时性文件上传（对话框直接传 pdf/docx/txt/xlsx/csv，刷新后失效）与长期[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)创建（上传到 File/Unstructured 或 File/Structured 后在界面创建[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)存于 VectorStore）。通过 Gradio 界面下方的“通过 API 使用”可获取 API 文档集成到业务。 ## 日志与扩展 @@ -98,6 +98,19 @@ AppFlow 连接流可在百炼步骤后添加 SLS 日志云服务节点，将对 + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/bailian-application-calling.md b/skills/bailian-docs-llm-wiki/wiki/guides/bailian-application-calling.md index 5d694cf8..18d76f1f 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/bailian-application-calling.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/bailian-application-calling.md @@ -13,7 +13,7 @@ ## 基本调用方式 -[智能体应用](../concepts/agent-application.md)与工作流应用的调用接口完全相同，详见[调用智能体应用](../../raw/application-user-guide/bailian-application-calling/call-single-agent-application.md)与[调用工作流应用](../../raw/application-user-guide/bailian-application-calling/invoke-workflow-application.md)。核心请求结构如下： +[智能体应用](../concepts/agent-application.md)与[工作流](../concepts/workflow.md)应用的调用接口完全相同，详见[调用智能体应用](../../raw/application-user-guide/bailian-application-calling/call-single-agent-application.md)与[调用工作流应用](../../raw/application-user-guide/bailian-application-calling/invoke-workflow-application.md)。核心请求结构如下： ``` POST https://dashscope.aliyuncs.com/api/v1/apps/{APP_ID}/completion @@ -84,14 +84,14 @@ curl -X POST https://dashscope.aliyuncs.com/api/v1/apps/YOUR_APP_ID/completion \ ## 多轮对话 -工作流应用支持多轮对话，详见[调用工作流应用](../../raw/application-user-guide/bailian-application-calling/invoke-workflow-application.md)。两种实现方式： +[工作流](../concepts/workflow.md)应用支持多轮对话，详见[调用工作流应用](../../raw/application-user-guide/bailian-application-calling/invoke-workflow-application.md)。两种实现方式： - **使用 `session_id`**：系统自动从云端加载历史对话，实现简单。`session_id` 有效期 1 小时，最多支持 50 轮对话。 - **自行管理 `messages`（推荐）**：手动维护 `messages` 数组传递每轮历史，无需传 `prompt`，控制更灵活。 > **注意**：若请求中同时包含 `session_id` 和 `messages`，系统将优先使用 `messages`。 -使用 `messages` 时，需先在工作流的大模型节点中配置提示词变量 `historyList` 并发布应用，再发起调用。 +使用 `messages` 时，需先在[工作流](../concepts/workflow.md)的大模型节点中配置提示词变量 `historyList` 并发布应用，再发起调用。 ## 自定义参数传递 @@ -186,6 +186,19 @@ Java SDK 通过 `JsonUtils.parse(...)` 将 JSON 字符串转为对象传入 `App + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/data-connection-overview.md b/skills/bailian-docs-llm-wiki/wiki/guides/data-connection-overview.md index c0b98b7e..348f1972 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/data-connection-overview.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/data-connection-overview.md @@ -11,7 +11,7 @@ - **表格**：导入并查询结构化表格数据（CSV、Excel 等）。 - **流处理**：数据保留在原数据源，实时访问。 - **MySQL / PostgreSQL / PolarDB-X 2.0**：连接对应数据库，支持执行 SQL 查询（仅 DMS 导入方式支持）。 - - **语雀**：访问语雀文档和知识库（仅公网版本）。 + - **语雀**：访问语雀文档和[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)（仅公网版本）。 - **OSS**：访问对象存储中的文件。 各类型的适用场景与存储方式详见 [数据连接](../../raw/application-user-guide/data-connection-overview/data-connection.md) 的连接器类型表。 @@ -24,7 +24,7 @@ - MySQL：已有 MySQL 实例（RDS 或自建），网络可达（公网或私网）。 - PostgreSQL：账号具备高权限（Superuser 或 REPLICATION），且 `wal_level` 设置为 `logical`；自建实例还需配置 `listen_addresses` 允许 `100.64.0.0/16` 网段访问。 - PolarDB-X 2.0：已有阿里云 PolarDB-X 2.0 实例且所在地域支持私网访问；DMS 导入方式需先在 DMS 录入实例。 - - 语雀：已有公网版语雀知识库并获取访问 Token。 + - 语雀：已有公网版语雀[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)并获取访问 [Token](../concepts/token.md)。 - OSS：已创建 Bucket 并开通向量检索服务。 ## 数据库连接器关键差异 @@ -63,7 +63,7 @@ - 数据库连接器执行 SQL 的限制见上文注意框（仅 DMS 导入方式支持）。 - **OSS 连接器**：使用需开通[向量检索服务](https://help.aliyun.com/zh/oss/user-guide/vector-retrieval/)，否则无法使用 `searchOSSFile` / `searchOSSFileByFileName` 工具；不支持归档/冷归档/深度冷归档类型的 Bucket；支持内容加密与私有 Bucket；开启 Referer 防盗链时需将 `*.console.aliyun.com` 加入白名单。 -- **文件导入**：文件作为独立副本存储在平台免费空间（当前无容量限制），仅支持查看最近 **90** 天内导入的文件（超期不可查看但不删除），且仅供当前业务空间使用。请求高峰期解析可能耗时数小时甚至偶现超时，需耐心等待或重试。 +- **文件导入**：文件作为独立副本存储在平台免费空间（当前无容量限制），仅支持查看最近 **90** 天内导入的文件（超期不可查看但不删除），且仅供当前[业务空间](../concepts/workspace.md)使用。请求高峰期解析可能耗时数小时甚至偶现超时，需耐心等待或重试。 - **语雀连接器**：仅支持公网版本语雀，需提供有效的 Tenant access token。 以上流程、字段和限制的完整细节，请以原文 [数据连接](../../raw/application-user-guide/data-connection-overview/data-connection.md) 为准。 @@ -78,3 +78,16 @@ + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/fine-tuning.md b/skills/bailian-docs-llm-wiki/wiki/guides/fine-tuning.md index 3781b4b1..13c610b5 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/fine-tuning.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/fine-tuning.md @@ -1,77 +1,94 @@ # fine tuning -模型微调（Fine-tuning）是阿里云百炼在 Prompt 工程、插件调用等手段仍无法满足效果时提供的深度定制手段。它覆盖文本生成、视觉理解（Qwen-VL）、图像/视频生成（万相）以及语音合成（CosyVoice）等多种模态，通过 SFT、CPT、DPO 等训练方式，把领域知识、任务能力、人类偏好或特定音色/风格直接写入模型参数。 +模型微调（fine tuning）是阿里云百炼平台在 Prompt 工程等手段无法满足需求时的核心模型优化方式，覆盖文本生成、视觉理解、图像/视频生成、语音合成等多种模态。平台提供 CPT（继续预训练）、SFT（监督微调，含 LoRA 高效训练）、DPO（直接偏好优化）以及 RL（强化学习）等训练方式，统一走"上传数据集 → 创建训练任务 → 部署 → 调用"的流程，支持控制台零代码操作和 DashScope API/命令行两种入口。 -> **注意**：以下所有微调、部署与调用能力均**仅在华北2（北京）地域可用**，且必须使用该地域的 API Key；子账号（RAM 用户）需预先被授予调用、训练和部署权限。 +> **注意**：本主题下所有微调能力均仅支持**华北2（北京）**地域，必须使用该地域的 API Key；RAM 子账号需先授予模型调用、训练和部署权限。 -## 支持的模型与训练方式 +## 调优方式与选型 -按模态划分，不同模型支持的训练方式差异明显： +根据 [模型调优简介](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md)，各调优方式是递进而非互斥的，推荐顺序为 `CPT（可选）→ SFT → DPO（可选）→ RL（可选）`： -- **文本生成（千问系列）**：支持 CPT、SFT（全参 `sft` / 高效 `efficient_sft`）、DPO（全参 `dpo_full` / 高效 `dpo_lora`）。是否支持某种方式因模型而异，例如 Qwen3-32B、Qwen3-4B/1.7B/0.6B、Qwen2.5 系列支持全部 5 种；而 Qwen3.5-Plus/Flash、Qwen3.6/3.7 等新模型往往仅支持 `sft`。详见 [模型调优简介](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md)。 -- **视觉理解（千问 VL）**：Qwen3-VL、Qwen2.5-VL 系列支持 SFT 全参与高效训练，不支持 CPT/DPO。 -- **图像生成（万相）**：`wan2.7-image-pro`、`wan2.7-image`，仅支持 SFT-LoRA 高效微调，见 [微调图像生成模型](../../raw/model-user-guide/fine-tuning/wan-image-generation-finetune-guide.md)。 -- **视频生成（万相）**：图生视频-基于首帧 `wan2.7-i2v`/`wan2.5-i2v-preview`/`wan2.2-i2v-flash`，基于首尾帧 `wan2.2-kf2v-flash`，同样仅支持 SFT-LoRA，见 [微调视频生成模型](../../raw/model-user-guide/fine-tuning/wan-video-generation-finetune-guide.md)。 -- **语音合成（CosyVoice）**：`cosyvoice-v3-flash`，仅支持 `efficient_sft`，且**当前只能通过 API 发起，控制台暂不支持**，见 [CosyVoice模型调优](../../raw/model-user-guide/fine-tuning/fine-tune-speech-synthesis-model/fine-tune-speech-synthesis-model-by-api.md)。 +| 方式 | 一句话定位 | 数据要求 | 典型场景 | +| --- | --- | --- | --- | +| CPT 持续预训练 | 补知识（领域适应） | 1000万+ Token 无标签领域文本，格式 `{"text":"..."}` | 金融术语、医疗病理、法律条文 | +| SFT 监督微调 | 学做事（遵循指令） | 1000+ 条高质量 ChatML 问答对 | 客服流程、代码助手、Agent 工具调用 | +| DPO 直接偏好优化 | 做得更好（对齐偏好） | 100+ 组 chosen/rejected 回答对 | 抑制幻觉、安全对齐、bad case 修复 | +| RL 强化学习 | 学推理（Reward 驱动） | 数十至数百条起步（含 `rollout_extra` 参考答案） | 数学推理、代码生成、Agent 调用稳定性 | -## 三种调优方式（文本生成） +训练模式上分**全参训练**与**高效训练（LoRA）**：两者费用相同，若模型支持全参训练官方建议优先全参（效果更好）；LoRA 训练更快、适合快速验证与小数据集场景。LoRA 关键参数 `lora_rank` 需为 2 的幂（如 16/32/64），秩越大拟合能力越强但更易过拟合。 -推荐按递进顺序组合使用：`CPT（可选）→ SFT → DPO（可选）`。 +## 支持的模型 -| 方式 | 目标 | 数据量 | 数据形态 | -| --- | --- | --- | --- | -| CPT（持续预训练） | 补领域知识 | 1000 万+ Token | 无标签领域文本 `{"text":"..."}` | -| SFT（监督微调） | 学会遵循指令 | 1000+ 条 | ChatML「问-答」对 | -| DPO（直接偏好优化） | 对齐人类偏好 | 100+ 组 | 同指令下「更好/更差」回答对（`chosen`/`rejected`） | +- **文本生成**：Qwen3.7/3.6/3.5 商业版（仅 SFT 全参）、Qwen3 系列（qwen3-32b、qwen3-14b、qwen3-8b 等，多数支持 CPT/SFT/efficient_sft/DPO）、Qwen2.5 系列等，详细支持矩阵与训练单价见 [模型调优简介](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md)。 +- **视觉理解（千问 VL）**：Qwen3-VL-8B/4B、Qwen2.5-VL-7B/32B/72B 等，支持 SFT 全参与高效训练。 +- **图像生成**：wan2.7-image-pro、wan2.7-image（文生图 t2i / 图生图 i2i），仅支持 SFT-LoRA。 +- **视频生成**：wan2.7-i2v、wan2.5-i2v-preview、wan2.2-i2v-flash（首帧）、wan2.2-kf2v-flash（首尾帧），仅支持 SFT-LoRA。 +- **语音合成**：cosyvoice-v3-flash，仅支持 `efficient_sft`，且仅能通过 API 发起。 +- **RL 训练**：qwen3.5-9b、qwen3.6-flash、qwen3.5-flash，需联系商务经理开通。 + +## 使用方式 -训练模式分**全参训练**与**高效训练（LoRA）**：两者费用相同，官方建议在模型支持全参训练时优先选择全参（效果更好、性价比更高）；LoRA 适合对训练时间/成本敏感或数据集较小的场景。 +### 控制台（零代码） -## 关键超参数 +在控制台"模型调优"页面创建训练任务，选择训练方法（高效/全参）、模型、数据集与超参即可，详见 [在控制台进行模型调优](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-on-console.md)。关键超参建议： -文本生成调优的常用超参及默认值（以控制台实际显示为准）： +- `learning_rate`：高效训练 1e-4 量级，全参/CPT 1e-5 量级 +- `n_epochs`：数据量 < 10,000 时循环 3~5 次，> 10,000 时 1~2 次 +- `batch_size`：一般建议 16/32，具体范围以控制台显示为准 +- 其他默认值：`lora_rank=8`、`lora_alpha=16`、`max_length=8192`、`warmup_ratio=0.05` 等 -- `learning_rate`：高效训练建议 `1e-4` 量级，全参/CPT 建议 `1e-5` 量级。 -- `n_epochs`：默认 `3`，范围 `[1, 200]`；数据量 <10000 建议循环 3~5 次，>10000 建议 1~2 次。 -- `batch_size`：一般 16/32。 -- `max_length`：建议设为模型支持的最大值；SFT 会**丢弃**超长数据，DPO 则**截断**后仍训练。 -- `lora_rank` / `lora_alpha` / `lora_dropout`：LoRA 专用，秩越大效果略好但更慢、更易过拟合。 -- 通过 API 创建任务时，`n_epochs`、`batch_size`、`max_length` 因影响计费而**必填**。 +训练中可通过 Training Loss / Validation Loss 曲线判断拟合状态：双双仍在下降为欠拟合（加 `n_epochs` 或 `lora_rank`），Validation Loss 回升为过拟合（减小两者）。一个端到端的零代码实践（含裁判模型自动评测与超参对照实验）见 [0 代码强化大模型安全合规能力](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/enhance-the-security-compliance-of-large-models.md)。 -> **注意**：默认学习率各文档取值不一致。控制台参数面板列出的 `learning_rate` 默认值为 `3e-4`（对应高效训练默认场景），而 API 示例中 SFT 全参使用的是 `1.6e-5`。请以实际训练方式对应的量级为准，切勿照搬。 +### API / 命令行 -万相图像/视频与 CosyVoice 使用各自独立的超参集，例如万相有 `max_steps`/`generation_type`/`val_img_size`，CosyVoice 分 `lm_*`（影响韵律）与 `fm_*`（影响音色）两组网络的 `*_max_epoch`/`*_step`/`*_num`/`*_batch_size`（8 个子字段全部必填）。 +统一走 DashScope 三个接口，详见 [使用 API 或命令行进行模型调优](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/fine-tuning-api-guide.md)： -## 使用方式 +1. **上传数据集**：`POST /api/v1/files`，`purpose=fine-tune`，返回 `file_id`；也支持 `oss_mount` 挂载未压缩的数据集目录（OSS Bucket 仅支持 cn-beijing 和 ap-southeast-1）。 +2. **创建微调任务**：`POST /api/v1/fine-tunes`，指定 `model`、`training_type`（sft / efficient_sft / cpt / dpo_full / dpo_lora）与 `hyper_parameters`，返回 `job_id` 与 `finetuned_output`（新模型名）；轮询 `GET /api/v1/fine-tunes/{job_id}` 直到 `status=SUCCEEDED`。 +3. **部署与调用**：`POST /api/v1/deployments`（LoRA 模型用 `"plan": "lora"`），轮询部署状态至 `RUNNING` 后即可调用。 + +> **注意**：通过 API 创建的训练任务仅支持按 [Token 计费](../concepts/token-billing.md)，不支持模型训练单元（MTU）；而 RL 训练恰好相反，**仅支持** MTU 计费。需要训练单元时请走控制台创建任务。 -**控制台（推荐入门）**：在[模型调优](https://bailian.console.aliyun.com/?tab=model#/efm/model_manager)页面创建训练任务 → 选训练方式与模型 → 配置训练集/验证集（可自动切分）→ 配置 Checkpoint 保存 → 开始训练 → 部署 → 评测。零代码场景可参考 [0 代码强化大模型安全合规能力](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/enhance-the-security-compliance-of-large-models.md)，其中给出了以 Qwen3-8B 为例的完整安全对齐 SFT 流程与超参实验对照（全参 `n_epochs=3`/`lr=1e-5` 或 LoRA `n_epochs=3`/`lr=3e-4` 效果较好）。 +### 数据格式要点 -**API / 命令行**：统一四步流程，详见 [使用 API 或命令行进行模型调优](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/fine-tuning-api-guide.md)： +- **SFT ChatML**：`{"messages":[{"role":"system"...},{"role":"user"...},{"role":"assistant"...}]}`，所有 assistant 输出都会被训练；不支持 OpenAI 的 `name`/`weight` 参数；`loss_weight` 为邀测参数。 +- **SFT 思考模型**：只训练最后一个 assistant 输出，思考内容用 `\n...\n\n\n` 包裹且换行必须保留；若训练成不输出 ``，训练后不建议再开思考模式调用。 +- **SFT 视觉理解**：system 消息的 `content` 必须为数组格式；ZIP 包最大 2 GB，`data.jsonl` 必须位于压缩包根目录，图片单张宽高不超过 1024px 且文件名全局唯一。 +- **DPO**：在 `messages` 之外提供 `chosen` 与 `rejected` 两个 assistant 回答。 +- **CPT**：纯文本 `{"text":"文本内容"}`。 -1. 上传数据集到 `POST /api/v1/files`（`purpose=fine-tune`），获取 `file_id`。 -2. `POST /api/v1/fine-tunes` 创建任务，关注返回的 `job_id`、`finetuned_output`、`status`。 -3. 轮询 `GET /api/v1/fine-tunes/` 直到 `status` 变为 `SUCCEEDED`。 -4. `POST /api/v1/deployments` 部署（`plan=lora`），轮询直到 `status` 为 `RUNNING`，再用 `deployed_model` 调用。 +## [多模态](../concepts/multimodal.md)微调 -> **注意**：通过 API 创建的训练任务**仅支持按 Token 计费**，不支持模型训练单元（预付费/后付费）；如需使用训练单元，必须通过控制台创建。 +### 图像/视频生成（万相 LoRA） -数据集除 `file_id` 外还可用 OSS 挂载（`data_source_type=oss_mount`），OSS Bucket 地域支持 `cn-beijing` 与 `ap-southeast-1`，挂载时只需指定 `data.jsonl` 路径。 +用于定制特定人物、IP 形象、风格或特效，仅当 Prompt 优化无法满足时使用。流程与文本模型一致（上传 zip 数据集 → fine-tunes → 部署 → [异步调用](../concepts/async-invocation.md)），细节见 [微调图像生成模型](../../raw/model-user-guide/fine-tuning/wan-image-generation-finetune-guide.md) 与 [微调视频生成模型](../../raw/model-user-guide/fine-tuning/wan-video-generation-finetune-guide.md)。要点： -## 数据格式要点 +- 图像模型超参以 `max_steps`（建议 ≥ 500）+ `max_token_length` 控制训练量；视频模型以 `n_epochs × ⌈数据集/batch_size⌉` 控制，建议总步数 ≥ 800。 +- 视频部署时可通过 `aigc_config`（`prompt` 模板 + `lora_prompt_default`）实现"无提示词自动触发特效"。 +- 部署后的图像模型仅支持**[异步调用](../concepts/async-invocation.md)**（`X-DashScope-Async: enable`）。 -- **SFT**：ChatML `{"messages":[...]}`，支持多轮；不支持 OpenAI 的 `name`/`weight`，所有 assistant 行都会被训练；思考模型（thinking）只训练**最后**一个 assistant 输出且须保留 `` 标签前后的换行。 -- **DPO**：在 `messages` 基础上追加 `chosen`/`rejected`。 -- **CPT**：纯文本 `{"text":"..."}`。 -- **视觉理解**：`content` 用数组，`image`/`video` 声明文件名（不含路径），打包为 ZIP（≤2GB），`data.jsonl` 必须在根目录，文件名全局唯一且仅含 ASCII。 -- **CosyVoice**：`data.jsonl` 每行 `{"wav_fn":"train/xxx.wav","text":"..."}`，`wav_fn` 必须以 `train/` 前缀；`text` 须为纯文本，禁止 SSML/LaTeX/情感标注。 +> **注意**：图像模型创建任务用 `training_datasets` 字段（对象数组，支持 file_id / oss_mount），而视频模型示例用的是 `training_file_ids` 字符串数组，两类接口入参形态不同，编写代码时请以对应文档为准。 -## 限制与注意事项 +### 语音合成（CosyVoice） -- **成本与耗时高**：文本模型微调需构建大规模数据集，且调优后模型**必须部署才能使用**，部署费用较高；官方明确将模型调优定位为「最后的手段」，建议先充分尝试 Prompt 工程与插件调用。 -- **训练耗时差异大**：万相文生图约数十分钟，视频微调可达数小时；文本 LoRA 通常 15~30 分钟；部署一般需 3~10 分钟。 -- **计费方式各异**：文本/VL 按训练 Token 计费（`训练Token × 循环次数 × 单价`）；CosyVoice 按 `(lm_max_epoch+fm_max_epoch)×25×音频总秒数` 估算 Token，单价 0.2 元/千 Token，另加部署时长费用。 -- **能力边界不可突破**：CosyVoice 调优产物为单音色模型（`voice` 锁定 `default`），无法新增基础模型不支持的语种、也不支持 `instruction` 指令控制。 -- **过拟合/欠拟合判断**：观察 Training/Validation Loss 曲线，欠拟合可增大 `n_epochs`/`lora_rank`，过拟合则反向调整。 -- 万相 LoRA 调用需在提示词中包含**触发词**以激活风格；图像模型部署后当前仅支持异步调用。 +面向**同一发音人**多条录音的高还原度音色定制，产出独立部署的单音色模型（调用时 `voice` 固定为 `default`），仅支持 API 发起，见 [CosyVoice模型调优](../../raw/model-user-guide/fine-tuning/fine-tune-speech-synthesis-model/fine-tune-speech-synthesis-model-by-api.md)。关键限制： + +- 仅有单条音频、或只想调整朗读风格时，应优先使用声音复刻 / 声音设计 / SSML，而非调优。 +- 调优无法扩展基础模型的语种范围；调优产物不支持 `instruction` 指令控制。 +- 数据建议：`.wav`、≥16kHz、单条 2~30 秒、总时长 1~10 小时、不少于 150 条；`text` 必须为纯文本（禁止 SSML/LaTeX 标签）。 +- 超参分 LM（韵律）与 FM（音色）两组，推荐 `lm_max_epoch=60`、`fm_max_epoch=100`；轮次越高基础能力"遗忘"越严重。 +- 训练费 0.2 元/千 Token，估算公式：`(lm_max_epoch + fm_max_epoch) × 25 × 训练集总秒数`；平台同一时刻仅运行一个训练任务，超出的排队（QUEUING）。 + +### 强化学习（RL） + +通过"Rollout 采样 → Reward 评分 → 策略更新"循环提升推理/Agent 能力，无需标注标准答案，见 [强化学习训练概述](../../raw/model-user-guide/fine-tuning/rl-training-overview.md)。使用离线 DashScope SDK（Python ≥ 3.10）提交，需自定义 Rollout / Reward 函数并部署到函数计算（FC），首次使用需在控制台授权 OpenTelemetry、FC、SLS 三项服务。推荐算法 GSPO，核心超参含 `batch_size`、`learning_rate=2e-6`、`n_rollouts` 等；训练中重点观察 `critic/rewards/mean` 指标。 + +## 计费与限制 + +- **按 [Token 计费](../concepts/token-billing.md)**：训练费用 =（训练数据 Token + 混合训练数据 Token）× 循环次数 × 训练单价；如 qwen3-8b 为 ¥0.006/千 Token，qwen2.5-72b-instruct 为 ¥0.15/千 Token，完整价目见 [模型调优简介](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/model-training-overview.md)。 +- **训练单元（MTU）**：预付费包月（IV 型 19,914 元/月/实例）或后付费（41 元/小时/实例）；RL 训练仅支持此方式。 +- **文件配额**：单文件最大 300MB（zip 数据集最大 2GB），总空间 100GB，总数 10000 个，无存储时限。 +- **部署**：LoRA 产物需以 `plan=lora` 部署为在线服务后才能调用，部署约 5~10 分钟；调优后模型的部署与调用会产生独立的部署费用。 ## 来源文档 @@ -82,6 +99,6 @@ - [使用 API 或命令行进行模型调优](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/fine-tuning-api-guide.md) - [0 代码强化大模型安全合规能力](../../raw/model-user-guide/fine-tuning/fine-tune-text-generation-model/enhance-the-security-compliance-of-large-models.md) - [CosyVoice模型调优](../../raw/model-user-guide/fine-tuning/fine-tune-speech-synthesis-model/fine-tune-speech-synthesis-model-by-api.md) - +- [强化学习训练概述](../../raw/model-user-guide/fine-tuning/rl-training-overview.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/get-started-with-models.md b/skills/bailian-docs-llm-wiki/wiki/guides/get-started-with-models.md index 56974632..3609b9ab 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/get-started-with-models.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/get-started-with-models.md @@ -1,35 +1,62 @@ # get started with models -阿里云百炼是一站式大模型开发与应用平台，集成千问（Qwen）全系列及 DeepSeek、Kimi、GLM 等主流第三方模型，并提供兼容 OpenAI 的 API。本页面帮助开发者快速完成从选择模型、获取 API Key、配置 Base URL 到发起首次调用的全流程，并梳理地域、接入域名与限流等关键约束。 +阿里云百炼是一站式大模型开发与应用平台，提供千问（Qwen）全系列及 DeepSeek、Kimi、GLM 等第三方模型的开箱即用服务，并兼容 OpenAI / Anthropic / DashScope 三种接口规范。本页汇总从开通账号、选择模型、获取 API Key，到选择地域与 Base URL、理解[限流](../concepts/rate-limit.md)规则的完整入门路径，帮助开发者快速发起第一次模型调用。 -## 支持的模型与能力 +## 前置准备：账号与 API Key -百炼提供开箱即用的模型服务，无需自行部署或运维即可调用。文本生成方面，千问旗舰模型按能力与成本分层选择（参见 [选择模型](../../raw/model-user-guide/get-started-with-models/models.md)）： +按照[首次调用千问API](../../raw/model-user-guide/get-started-with-models/first-api-call-to-qwen.md)的引导完成以下步骤： -- **千问 Max**（如 `qwen3.7-max`）：Qwen 系列效果最好的模型，适合复杂、多步骤任务。 -- **千问 Plus**（如 `qwen3.7-plus`）：效果、速度和成本均衡，多数场景的**推荐选择**。 -- **千问 Flash**（如 `qwen3.6-flash`）：高性价比、低延迟，适合需要快速响应的简单任务。 +1. **注册阿里云账号**并完成实名认证。 +2. **开通百炼**：使用阿里云主账号访问百炼控制台，同意服务协议后自动开通（开通免费，调用按量计费）。 +3. **创建 API Key**：在控制台 API Key 页面创建。各地域的 API Key 相互独立，**不能跨地域使用**。 +4. **获取[业务空间](../concepts/workspace.md) ID（WorkspaceId）**：使用华北2（北京）、新加坡、日本（东京）或德国（法兰克福）地域时，Base URL 中需填入 WorkspaceId，可在[业务空间](../concepts/workspace.md)管理页面查看。 +5. **配置环境变量**：建议将 API Key 写入 `DASHSCOPE_API_KEY` 环境变量（Linux/macOS 用 `export`，Windows 用系统属性或 `setx`），避免硬编码泄露。 -此外还覆盖视觉理解、图像生成、视频生成、语音识别与合成、嵌入向量等多模态能力，以及长文本、翻译、法律等细分领域模型。平台同时支持模型调优（SFT / CPT / DPO）、模型部署与模型评测，详见 [什么是阿里云百炼](../../raw/model-user-guide/get-started-with-models/what-is-model-studio.md)。 +## 选择模型 -## 关键概念与参数 +根据[什么是阿里云百炼](../../raw/model-user-guide/get-started-with-models/what-is-model-studio.md)与[选择模型](../../raw/model-user-guide/get-started-with-models/models.md)，千问旗舰文本模型按"能力—成本"梯度分为三档： -调用前需要先确定四个维度（详见 [选择地域、服务部署范围和接入域名](../../raw/model-user-guide/get-started-with-models/regions.md)）： +| 系列 | 代表模型 ID | 定位 | +| --- | --- | --- | +| 千问 Max | `qwen3.7-max` | 效果最强，适合复杂多步骤任务 | +| 千问 Plus | `qwen3.7-plus` | 效果、速度、成本均衡，多数场景的推荐选择 | +| 千问 Flash | `qwen3.7-flash` | 高性价比、低延迟，适合简单快速任务 | -- **地域（Region）**：决定接入点和数据存储位置。目前提供华北2（北京，`cn-beijing`）、新加坡（`ap-southeast-1`）、日本（东京，`ap-northeast-1`）、德国（法兰克福，`eu-central-1`）、美国（弗吉尼亚，`us-east-1`）。就近选择可降低延迟。 -- **服务部署范围**：决定推理执行位置。有数据合规需求时选择特定地理边界（如中国内地、欧盟、美国），无合规需求可选全球部署（推理资源池更大）。德国、日本地域通过[业务空间（Workspace）](../concepts/workspace.md)区分部署范围；美国地域可用带 `-us` 后缀的模型名（如 `qwen-plus-us`）限定境内推理。 -- **接入域名**：影响并发上限、超时等服务保障，推荐使用**业务空间专属域名**（`{WorkspaceId}.{region}.maas.aliyuncs.com`，SLA 99.9%、请求超时 3600 秒、支持 HTTP/SSE/WebSocket/WebRTC），另有 Dashscope 域名（现有，超时 600 秒）和试用域名（限流小，不建议生产）。 -- **API Key**：各地域相互独立、不能跨地域混用；Base URL 也必须与同一计费方案的 API Key 配套使用，否则报 401。 +此外平台覆盖视觉理解、图像/视频生成、语音识别与合成、嵌入向量等[多模态](../concepts/multimodal.md)能力，以及长文本、翻译、法律等细分领域模型。 -> **注意**：各地域接入点（Base URL）、API Key 和模型列表均不能跨地域混用；使用北京、新加坡、日本、德国地域时，业务空间专属域名中的 `{WorkspaceId}` 需替换为真实业务空间 ID（可在业务空间管理页查看）。 +> **注意**：`qwen3.8-max-preview` 目前仅面向 Token Plan 订阅用户开放，按量付费用户不可直接调用。 -## 使用方式 +> **注意**：DeepSeek 等部分第三方模型仅在华北2（北京）地域提供，选型前请核对目标地域的可用模型列表。 -**1. 账号与凭证准备**（参见 [首次调用千问API](../../raw/model-user-guide/get-started-with-models/first-api-call-to-qwen.md)）：注册并开通百炼 → 在 API Key 页面创建 Key → 获取业务空间 ID。建议将 Key 配置到环境变量 `DASHSCOPE_API_KEY`，避免硬编码泄露。 +## 地域与 Base URL -**2. 选择 Base URL**（详见 [Base URL总览](../../raw/model-user-guide/get-started-with-models/base-url.md)）：百炼提供 OpenAI 兼容（`/compatible-mode/v1`）、Anthropic 兼容（`/apps/anthropic`）、DashScope（`/api/v1`）三类接口。以北京地域业务空间专属域名的 [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)为例：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1`。 +地域决定接入点和数据存储位置，各地域的 Base URL、API Key、可用模型与价格均不同。详见[地域及接入域名](../../raw/model-user-guide/get-started-with-models/regions.md)与[Base URL总览](../../raw/model-user-guide/get-started-with-models/base-url.md)。 -**3. 发起调用**：兼容 OpenAI 接口规范，迁移现有代码只需调整 API Key、base_url 和模型名称。Python 示例： +### 三种接入域名（按量付费） + +| 域名类型 | 格式 | 适用场景 | 请求超时 | +| --- | --- | --- | --- | +| [业务空间](../concepts/workspace.md)专属域名（**推荐**） | `{WorkspaceId}.{region}.maas.aliyuncs.com` | 生产环境，高并发、流量隔离，SLA 99.9% | 3600 秒 | +| Dashscope 域名 | `dashscope.aliyuncs.com`（北京）等 | 存量业务兼容，建议迁移 | 600 秒 | +| 试用域名 | `trial.{region}.maas.aliyuncs.com` | 快速验证，RPM 固定 1000，无 SLA | 600 秒 | + +### 各地域 OpenAI 兼容 Base URL + +- 华北2（北京，`cn-beijing`）：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` +- 新加坡（`ap-southeast-1`）：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` +- 日本（东京，`ap-northeast-1`）：`https://{WorkspaceId}.ap-northeast-1.maas.aliyuncs.com/compatible-mode/v1` +- 德国（法兰克福，`eu-central-1`）：`https://{WorkspaceId}.eu-central-1.maas.aliyuncs.com/compatible-mode/v1` +- 美国（弗吉尼亚，`us-east-1`）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1`（暂不支持业务空间专属域名） + +DashScope 原生接口将路径 `/compatible-mode/v1` 换为 `/api/v1`，Anthropic 兼容接口换为 `/apps/anthropic`。 + +> **注意**：Base URL 必须与同一计费方案的 API Key 配套使用，否则返回 401。Token Plan / Coding Plan 有独立的专属域名和 API Key，仅限 Claude Code、Codex 等 AI 工具交互式使用，不能用于后端服务。 + +> **注意**：德国（法兰克福）和日本（东京）地域需先创建业务空间并选择服务部署范围（全球/欧盟、全球/日本）；美国（弗吉尼亚）地域用带 `-us` 后缀的模型名（如 `qwen-plus-us`）限定美国境内推理。 + +## 发起第一次调用 + +以 OpenAI Python SDK 为例（`pip install -U openai`，Python ≥ 3.8）： ```python import os @@ -37,36 +64,42 @@ from openai import OpenAI client = OpenAI( api_key=os.getenv("DASHSCOPE_API_KEY"), + # 北京地域示例，{WorkspaceId} 替换为真实业务空间 ID base_url="https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1", ) completion = client.chat.completions.create( - model="qwen-plus", - messages=[ - {'role': 'system', 'content': 'You are a helpful assistant.'}, - {'role': 'user', 'content': '你是谁？'} - ] + model="qwen3.7-plus", + messages=[{"role": "user", "content": "你是谁？"}], ) print(completion.choices[0].message.content) ``` -也可通过 DashScope SDK（`pip install -U dashscope`）、curl，或 Chatbox、Claude Code 等客户端/开发工具调用。OpenAI Python SDK 要求 Python ≥ 3.8。 +DashScope SDK（`pip install -U dashscope`）则设置 `dashscope.base_http_api_url = 'https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1'` 后用 `Generation.call()` 调用。已有 OpenAI 代码只需替换 `api_key`、`base_url` 和模型名即可迁移。 -## 限制与注意事项 +## [限流](../concepts/rate-limit.md)规则 + +根据[限流](../../raw/model-user-guide/get-started-with-models/rate-limit.md)： -- **限流按主账号维度合并计算**：账号下所有 RAM 子账号、业务空间和 API Key 的调用量合并统计，不同模型限流额度相互独立。超限请求被拒绝，通常一分钟内自动恢复。详见 [限流](../../raw/model-user-guide/get-started-with-models/rate-limit.md)。 -- **RPM 与 TPM 双重约束**：`Requests rate limit exceeded` 表示触发每分钟请求数（RPM）限流；`Allocated quota exceeded` 表示触发每分钟 Token 数（TPM）限流；`Request rate increased too quickly` 表示请求瞬时激增触发稳定性保护。限流可能按秒级 RPS（RPM/60）、TPS（TPM/60）执行。 -- **规避限流**：优先选用高限流额度模型（稳定版比日期快照版更宽松）、平滑请求速率（匀速/指数退避/队列）、配置备选模型自动切换、拆分任务、无需实时响应时改用 Batch API（不受实时限流约束）。北京与新加坡地域支持在控制台申请临时 TPM 提额（生效 30 天）。 -- **计费独立**：模型推理按 Token 用量计费，知识库（RAG）按规格时长与调用独立计费，两者互不相通。限流只约束速率、不限制累计用量；如需控费可设置费用告警、开启"免费额度用完即停"或订阅 Coding Plan（固定月费）。 -- **各地域功能差异**：批量推理、模型调优、应用开发等能力目前主要在华北2（北京）支持，海外地域功能覆盖较少，选型前请核对目标地域的功能与模型列表。 +- **账号级别[限流](../concepts/rate-limit.md)**：按主账号维度计算，所有 RAM 子账号、业务空间、API Key 的调用量合并统计；不同模型的额度相互独立。 +- **双重限制**：RPM（每分钟请求数）与 TPM（每分钟 Token 数，含输入和输出）超出任一即触发；系统还可能按秒级 RPS（RPM/60）、TPS（TPM/60）执行，瞬时爆发也会被限。 +- **典型额度（北京地域）**：`qwen3.7-max` / `qwen3.7-plus` / `qwen3.7-flash` 均为 30,000 RPM；带日期的快照版本额度显著更低（多为 60–600 RPM）。 +- **常见错误**：`Requests rate limit exceeded`（RPM 超限）、`Allocated quota exceeded`（TPM 超限）、`Request rate increased too quickly`（请求激增触发保护）。限流通常一分钟内自动恢复。 + +规避建议：优先选稳定版高额度模型、匀速调度 + 指数退避、配置备选模型自动切换、大任务改用批量推理（Batch API 不受实时限流约束），或在控制台"限流提额"页面申请临时 TPM 提额（立即生效，有效期 30 天，仅支持北京和新加坡地域）。 + +## 限制与注意事项 -> **注意**：Token Plan 与 Coding Plan 的 Base URL 及专属 API Key 仅限 Claude Code、Codex 等 AI 工具交互式使用，不能用于后端服务；使用非专属 Base URL 调用将按量付费。 +- **地域间不通用**：API Key、Base URL、模型列表、价格均按地域隔离；批量推理仅北京和新加坡支持，模型调优与应用开发仅北京支持。 +- **计费**：开通免费，调用按 Token 量计费，按分钟出账。新用户有北京地域专属免费额度，可开启"免费额度用完即停"避免意外扣费。 +- **调用统计延迟**：模型监控数据在调用完成约一小时后可见，高峰期可能有小时级延迟。 +- **数据安全**：阿里云不会将用户数据用于模型训练，传输全程加密。 ## 来源文档 - [什么是阿里云百炼](../../raw/model-user-guide/get-started-with-models/what-is-model-studio.md) - [选择模型](../../raw/model-user-guide/get-started-with-models/models.md) - [首次调用千问API](../../raw/model-user-guide/get-started-with-models/first-api-call-to-qwen.md) -- [选择地域、服务部署范围和接入域名](../../raw/model-user-guide/get-started-with-models/regions.md) +- [地域及接入域名](../../raw/model-user-guide/get-started-with-models/regions.md) - [限流](../../raw/model-user-guide/get-started-with-models/rate-limit.md) - [Base URL总览](../../raw/model-user-guide/get-started-with-models/base-url.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/knowledge-base.md b/skills/bailian-docs-llm-wiki/wiki/guides/knowledge-base.md index 8f2e139a..10743970 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/knowledge-base.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/knowledge-base.md @@ -1,81 +1,173 @@ # [knowledge](../api/knowledge.md) base -知识库（Knowledge Base）是阿里云百炼平台基于 RAG（检索增强生成）技术为大模型补充私有数据和最新信息的能力。大模型在生成回答前先从知识库中检索语义相关的内容，从而显著提升在特定领域问题上的准确性。围绕知识库，平台还提供了知识检索、知识问答、API 集成、日志监控与计费等一整套配套能力。 +阿里云百炼知识库用于把企业私有数据、业务文档和最新信息接入大模型应用，通过 RAG（检索增强生成）在回答前检索相关切片，提升回答准确性与可追溯性。开发者可以在控制台创建知识库并挂载到智能体、工作流、知识检索或知识问答服务，也可以通过 API 将文档搜索类知识库集成到外部系统。 -> **注意**：知识库功能仅能在中国站 **华北2（北京）** 地域开通和使用，新加坡、德国（法兰克福）等其他地域均不支持。 +## 适用范围与核心能力 + +知识库主要覆盖以下能力： + +- **私有知识增强**：将 PDF、DOCX、Markdown、Excel、图片、音视频等数据解析、切片、向量化后用于检索，适合企业内部文档、产品手册、FAQ、参数表等场景。 +- **多形态知识库**：包括文档搜索、数据查询、图片问答、音视频搜索等类型；文档搜索还可选择基础文档问答、视觉理解（富文本文档）或极速问答等使用场景。 +- **多知识库联合检索**：知识检索和知识问答服务均支持最多绑定 15 个知识库，并可配置权重、路由、TopK、排序模型、阈值和过滤条件，详见 [知识检索](../../raw/application-user-guide/knowledge-base/rag-knowledge-retrieval.md) 与 [知识问答](../../raw/application-user-guide/knowledge-base/rag-knowledge-qa.md)。 +- **效果调优闭环**：可通过[评测](../concepts/evaluation.md)集、命中测试、标签/元数据过滤、切片修正、阈值和召回片段数调整来定位并优化 RAG 效果，参考 [RAG效果优化](../../raw/application-user-guide/knowledge-base/rag-optimization.md)。 +- **日志审计与监控**：所有检索调用可投递到 SLS，便于按知识库、[业务空间](../concepts/workspace.md)、API 路径、错误码、耗时等维度做分析和告警，参考 [知识库日志与监控](../../raw/application-user-guide/knowledge-base/rag-knowledge-base-log-monitoring.md)。 + +> **注意**：知识库相关功能仅能在中国站华北2（北京）地域开通和使用。API 指南明确仅适用于文档搜索类知识库，不应将其中的创建/上传示例直接套用于所有知识库类型。 ## 支持的模型与知识库类型 -预置模型（千问-QwQ/Long/Max/Plus/Turbo/Coder、千问VL 系列、Qwen 开源版，以及 DeepSeek-R1/V3.1、Llama3.1 等第三方文本生成模型）和部分调优后的自定义模型均可挂载知识库。具体可选模型以[应用管理](https://bailian.console.aliyun.com/?tab=app#/app-center)页面创建应用时实际可选项为准。 +### 生成模型 + +可在百炼应用中使用知识库的模型包括： + +- 千问系列预置模型：QwQ、Long、Max、Plus、Turbo、Coder、Deep-Research 等。 +- 千问视觉模型：VL-Max、VL-Plus、VL-Flash、OCR 等。 +- 千问开源版：Qwen3、Qwen2.5、Qwen2 等。 +- 第三方文本生成模型：DeepSeek、Llama、Yi、abab 等。 +- 在百炼调优后的部分自定义模型，如千问 Plus/Turbo、千问 VL、千问开源版等。 + +实际可用模型会随平台更新变化，创建或配置应用时应以控制台可选项为准，详见 [知识库](../../raw/application-user-guide/knowledge-base/rag-knowledge-base.md)。 + +### 向量与排序模型 + +不同知识库类型使用的模型有所差异： + +| 场景 | 向量模型 | 排序模型 | +| --- | --- | --- | +| 文档搜索、数据查询、音视频搜索 | `text-embedding-v4`、`text-embedding-v3` | `qwen3-rerank` 等 | +| 图片问答、视觉理解/[多模态](../concepts/multimodal.md)检索 | `multimodal-embedding-v1` 或 `qwen3-vl-embedding` | `qwen3-vl-rerank` | -创建知识库时按场景选择类型（创建后不可更改）： +向量维度不支持修改：`text-embedding-v4` 和 `text-embedding-v3` 为 512 维，`multimodal-embedding-v1` 为 1024 维。视觉理解场景创建后会自动切换为 qwen3 [多模态](../concepts/multimodal.md)向量模型，且不可更改。 -- **文档搜索**：企业内部文档、产品手册等非结构化数据检索。可细分为基础文档问答、图文并茂回复、视觉理解（富文本文档）、极速问答四种使用场景。选择视觉理解后向量模型自动切换为 qwen3 多模态向量（qwen3-vl-embedding），不可更改。 -- **数据查询（表格库）**：结构化 Excel/CSV，单库仅支持 1 篇。 -- **图片问答类**：仅支持 multimodal-embedding-v1 向量模型。 -- **音视频搜索类**：支持语音识别、视频帧提取与剧情解析。 +## 创建与数据处理流程 -不同解析方式（电子文档解析、文档智能解析、大模型文档解析、Qwen VL 解析、音视频解析）在速度与图表理解能力上有明显差异，详见[知识库](../../raw/application-user-guide/knowledge-base/rag-knowledge-base.md)。 +创建知识库通常包括三步：填写基础信息、配置数据来源、设置索引参数。 -## 关键参数 +1. **选择规格与类型**：标准版适合 PoC、小规模场景；旗舰版适合高并发生产场景。知识库类型创建后不可更改。 +2. **导入数据**：支持本地上传或从 OSS/数据连接器导入；单次控制台导入最多 50 个文件，API 批量导入不受该限制但建议单次不超过 500 个。 +3. **解析与切片**：根据文档类型选择电子文档解析、文档智能解析、大模型文档解析、Qwen VL 解析、音视频解析等方式。 +4. **索引配置**：设置切片方式、向量模型、向量存储、Meta 信息抽取、多轮对话改写等参数。 +5. **命中测试与发布**：导入完成后建议通过命中测试验证召回结果，再挂载到应用或服务。 -知识库的检索效果主要由以下参数决定，在命中测试、检索服务和问答服务中可反复调优： +Meta 信息抽取适合需要按文件名、产品名、日期、作者、关键词等结构化字段精确过滤的场景。标签过滤适合按业务类别、产品线、权限域等粗粒度划分检索范围。两者都可以减少无关切片召回，降低模型输入 [Token](../concepts/token.md) 消耗。 -- **相似度阈值（0.01~1.0）**：仅语义相似度高于阈值的切片会被召回。阈值过高会导致相关切片被全部丢弃（例如调至 0.60 可能返回无召回结果）。 -- **初步向量检索 TopK / 初步关键词检索 TopK（1~100，默认各 50）**：控制初步召回的切片数量，直接影响送入 Rerank 模型的 Token 量与成本。 -- **最大召回数量 / 召回片段数（1~20）**：即多路召回的 K 值，最终提供给大模型的切片数。对总结、列举、比较类复杂问题应适当调大。 -- **权重**：多知识库联合召回时用于干预排序，但**仅在同类型知识库之间生效**。 -- **排序模型（Rerank）**：纯文本可选 qwen3-rerank、qwen3-rerank(hybrid)；多模态可选 qwen3-vl-rerank。支持问答模式与相似模式。 -- **Meta 信息抽取与标签过滤**：通过元数据（常量/变量/大模型/正则/关键词方式提取）和标签在向量检索前做结构化筛选，精准定位目标文件。注意元数据只能在创建时配置，创建后无法再开启。 +> **注意**：Meta 信息抽取和多轮对话改写都需要在创建知识库时配置；若创建时未开启，后续通常需要重新创建知识库才能补齐该能力。 ## 使用方式 -**控制台快速构建**：进入[知识库](https://bailian.console.aliyun.com/?tab=app#/knowledge-base)选择规格后，按"填写基础信息 → 配置数据来源 → 设置索引参数"三步完成创建，随后可关联到智能体应用、工作流应用或外部应用。工作流应用需将知识库节点接在开始节点之后、大模型节点之前，并在大模型提示词中引用 `result` 变量。 +### 挂载到智能体应用 -**知识检索服务**：面向多知识库联合检索（最多 15 个），提供 Query 改写、混合检索（向量+关键词）、Rerank 排序的流水线，支持知识库路由、混排模型模式等全局与单库独立参数配置，详见[知识检索](../../raw/application-user-guide/knowledge-base/rag-knowledge-retrieval.md)。 +在应用管理中选择目标智能体应用，为应用选择模型后，在文档知识库区域添加知识库。常用参数包括： -**知识问答服务**：在检索基础上由大模型（如 qwen3.6-plus）生成自然语言回答，提供**极速模式**（单轮检索+生成）和**多轮智能模式**（Agentic 多轮规划搜索），并支持文件预解析、拒答、防泄漏、多模态回复、引用来源等生成控制。 +- **相似度阈值**：仅召回排序后分数高于阈值的切片。阈值过高可能导致漏召回，过低可能引入噪声。 +- **权重**：多知识库同类型召回时，权重更高的知识库在相似度接近时优先返回。 +- **标签过滤/元数据过滤**：用于限定检索范围。 +- **召回片段数**：复杂问题可适当增加，但需注意模型上下文长度和成本。 -**API/SDK 集成**：通过[阿里云百炼 SDK](https://api.aliyun.com/api-tools/sdk/bailian?version=2023-12-29)调用，典型创建流程为：申请上传租约（ApplyFileUploadLease）→ 上传文件 → AddFile → CreateIndex → SubmitIndexJob → 轮询 GetIndexJobStatus。子账号需先获取 AliyunBailianDataFullAccess 策略并加入业务空间，详见[知识库API指南](../../raw/application-user-guide/knowledge-base/rag-knowledge-base-api-guide.md)。 +### 挂载到工作流应用 -## 效果优化 +工作流中可添加“知识库”节点，将输入变量（通常是用户 query）传给知识库检索，再把检索结果传给大模型节点。适合需要在流程中组合知识检索、工具调用、条件分支和生成逻辑的业务。 -RAG 效果由建立索引、检索召回、生成答案三个阶段决定。优化前建议用[自动评测](https://help.aliyun.com/zh/model-studio/application-auto-evaluation)建立至少 100 组用例的评估基线，再针对失败用例（打分 < 4）诊断改进： +### 使用知识检索服务 -- **检索无效（没找到）**：补充知识、优化源文件排版（推荐转 Markdown、移除水印、避免复杂表格）、统一实体表述、启用多轮对话改写。 -- **召回不相关**：使用标签过滤或元数据做结构化搜索。 -- **切片不完整**：采用"智能切分"（基于语义自适应切分），并人工检查修正异常切片。 -- **重排不佳**：调整相似度阈值与召回片段数，在漏召回与噪声之间平衡。 +知识检索服务只返回检索切片，不负责生成最终答案，适合已有应用自行控制提示词、上下文拼装和生成模型的场景。全局配置包括知识库绑定、知识库路由、混排模型、混排模型模式、最大召回数量；每个知识库还可独立配置向量 TopK、关键词 TopK、排序模型、阈值、标签过滤和结构化字段过滤。 -## 日志与监控 +### 使用知识问答服务 + +知识问答服务在检索基础上进一步调用大模型生成自然语言回答，适合快速构建端到端问答能力。它支持极速模式和多轮智能模式： + +- **极速模式**：单轮检索后直接生成，低延迟，适合明确问题。 +- **多轮智能模式**：通过 Agentic 多轮规划、Query 改写和知识库路由处理复杂或模糊问题。 + +生成控制参数包括系统提示词、拒答、防泄漏、[多模态](../concepts/multimodal.md)回复、引用来源展示和文件预解析。开启引用后，回答会展示来源，便于用户核验。 + +### 通过 API 集成 + +外部系统可通过阿里云百炼 SDK 或签名 HTTP 请求调用知识库相关 API。典型流程包括：配置 AccessKey 和 `WORKSPACE_ID`、申请文件上传租约、上传文件、添加文件、创建索引、提交索引任务、轮询索引状态、执行检索。详细示例见 [知识库API指南](../../raw/application-user-guide/knowledge-base/rag-knowledge-base-api-guide.md)。 + +## 关键参数速查 + +| 参数 | 常见位置 | 作用 | 调优建议 | +| --- | --- | --- | --- | +| 初步向量检索 TopK | 检索服务、问答服务、命中测试 | 控制语义召回候选数量 | 默认常见为 50；降低可省排序成本，提高可能增加召回但也增加费用 | +| 初步关键词检索 TopK | 检索服务、问答服务、命中测试 | 控制关键词召回候选数量 | 适合包含精确术语、编号、字段名的问题 | +| 排序模型 | 知识库/服务检索参数 | 对初步召回切片精排 | 精度优先时开启；成本敏感且可接受排序下降时可关闭 | +| 排序模型模式 | 检索/问答服务 | 问答模式、相似模式或高级模式 | 问答任务选问答模式；相似文档搜索选相似模式 | +| 相似度阈值 | 应用调试、检索/问答服务 | 过滤低分切片 | 通过命中测试反复验证；过高会漏召回 | +| 最大召回数量 | 检索/问答服务 | 控制最终返回切片数 | 总结、列举、比较类问题可适当增大，最高通常为 20 | +| 标签过滤 | 应用/API/服务参数 | 按文件标签限定范围 | 多类别文件混合时优先使用 | +| 结构化字段/元数据过滤 | 应用/API/服务参数 | 按元数据精准定位文件或切片 | 多文件结构相似、实体易混淆时优先使用 | -所有检索调用都会以日志形式投递到日志服务（SLS），topic 为 `log_dispatch`，包含 `request_id`、`pipeline_id`（知识库 ID）、`workspace_id`、`latency`、`response_status_code`、`response_code`、`request_body`、`response_body` 等索引字段，可用于调用审计、用量统计、慢查询与错误率监控。首次使用需在知识库列表页的**监控配置**中授权 SLS 角色、开通日志服务并创建 LogStore。SLS 存储与流量单独计费，关闭检索日志开关只停止新投递，历史日志仍保留计费，需彻底停止请到 SLS 删除对应 LogStore。 +## 配额、限制与文件要求 -## 限制与配额 +关键限制如下，完整清单见 [知识库配额与限制](../../raw/application-user-guide/knowledge-base/rag-knowledge-base-specifications.md)： -| 类别 | 上限 | -| --- | --- | -| 知识库数量 | RDS 数据源 100，其它数据源无限制 | -| 存储容量 | 旗舰版 9,999 GB / 标准版 100 GB | -| 单个文档搜索类知识库文件数 | 无硬性上限（数据查询类仅 1 篇） | -| 单次控制台导入文件数 | 50（API 批量建议单次 ≤ 10,000） | -| 单文件标签数 | 32 | -| 文本切片长度 | 6,000 Token | -| 召回文本切片数量 | 20 | -| 检索并发 | 旗舰版 50-10,000 QPS（1-200 RCU）/ 标准版 1 QPS 固定 | +- **知识库数量**：RDS 数据源每个主账号最多 100 个；其他数据源无限制。 +- **平台存储**：标准版 100 GB，旗舰版 9,999 GB；使用自购 ADB-PG 时按 ADB-PG 另行计费和限制。 +- **[业务空间](../concepts/workspace.md)资源**：类目最多 500 个，文件最多 100,000 个，数据表最多 1,000 个。 +- **单知识库文件数量**：文档搜索类无硬性上限；数据查询类单库 1 篇 Excel 或 CSV。 +- **检索并发**：标准版固定 1 QPS；旗舰版 50–10,000 QPS，可按 1–200 RCU 调整。 +- **单次最终召回切片数**：通常最多 20。 +- **切片长度**：单个文本切片最多 6,000 [Token](../concepts/token.md)；编辑切片内容限制为 10–6000 字符。 +- **切片操作**：文档搜索、数据查询、图片问答支持编辑、新增、删除；音视频搜索支持编辑和删除，不支持新增。 -文件格式限制：pdf/docx/ppt 等最大 150MB 且页数 ≤ 1,000；txt/markdown/html 最大 10MB；图片最大 20MB；音视频最大 512MB。向量模型仅支持 text-embedding-v3/v4（512 维）与 multimodal-embedding-v1（1024 维），维度不可更改。完整清单见[知识库配额与限制](../../raw/application-user-guide/knowledge-base/rag-knowledge-base-specifications.md)。 +常见文件限制： -## 计费注意事项 +| 类型 | 支持格式 | 主要限制 | +| --- | --- | --- | +| 文档搜索 | pdf、docx、doc、wps、pptx、ppt | 最大 150 MB，页数不超过 1,000 | +| 文档搜索 | txt、markdown、html | 最大 10 MB | +| 文档搜索 | xlsx、xls | 最大 10 MB，10 万行以内 | +| 文档搜索图片 | png、jpg/jpeg、bmp、gif | 最大 20 MB，需满足尺寸比例限制 | +| 数据查询/图片问答 | xlsx、xls | 最大 100,000 行，列数不超过 100 | +| 音视频搜索 | aac、mp3、wav、mp4、mkv 等 | 最大 512 MB | -知识库服务自 **2026 年 1 月 4 日**起正式计费，总费用由**规格费用**（运行时长）和**模型调用费用**（向量化 + Rerank）两部分构成，扣费顺序为免费额度 > 资源包 > 按量付费。 +上传 Excel 时不要使用合并单元格表头，不要把备注写在首行；修改表格格式后需重新上传并重新发布应用。 -- 规格费用：标准版 0.03 元/知识库/小时；旗舰版 0.2 元/RCU/小时（1 RCU ≈ 50 QPS）。 -- 模型调用费用独立计费，公式为 `(输入 Token 总数 / 1000) × 模型单价`。**Rerank 费用取决于初步召回的总切片数，而非最终返回数**，因此降低初步 TopK 或关闭排序可显著省钱。 -- 应用挂载多个知识库时，模型 Token 消耗按知识库数量倍增（N 个库则 × N）。 -- 平台提供一次性 720 小时免费额度（仅抵扣标准版规格费用，不含模型调用费）。删除知识库以停止计费，但删除会**永久清除数据且无法恢复**。 +## 计费与成本控制 -> **注意**：免费额度有效期存在新老用户差异——老用户统一截至 2026 年 2 月 3 日 23:59，新用户自开通起 30 天内有效，逾期作废。此外《知识库计费说明》以 2026 年计费规则描述，而《知识库》文档仍按旧的即时计费口径（0.03/0.2 元/小时）介绍创建流程，接入时以[知识库计费说明](../../raw/application-user-guide/knowledge-base/billing-for-knowledge-base.md)为准。 +知识库费用包括规格费用和模型调用费用，详见 [知识库计费说明](../../raw/application-user-guide/knowledge-base/billing-for-knowledge-base.md)。 + +- **规格费用**：标准版按知识库运行小时计费；旗舰版按 RCU 小时计费。删除知识库可停止规格费用，但会永久清除数据。 +- **模型调用费用**：创建/更新时的向量化、检索时的 Query 向量化和 Rerank、知识库路由、问答生成、文件预解析等均可能按 [Token](../concepts/token.md) 独立计费。 +- **多知识库成本**：应用挂载 N 个知识库并全部检索时，Query 向量化和排序相关 [Token](../concepts/token.md) 消耗会按知识库数量放大。 +- **排序成本**：Rerank 费用取决于初步召回总切片数和平均切片 [Token](../concepts/token.md) 数，而不是最终返回给用户的切片数量。 + +成本优化建议： + +1. 对搜索精度要求不高时关闭排序模型。 +2. 降低初步向量检索 TopK 和初步关键词检索 TopK，减少进入排序模型的切片数。 +3. 使用标签、元数据、结构化字段过滤减少无关召回。 +4. 对多知识库场景开启路由，但需同时评估路由模型费用。 +5. 删除不再使用的知识库，避免持续产生规格费用。 +6. 监控命中测试、调试和线上调用的模型消耗，避免调试阶段产生非预期费用。 + +> **注意**：计费文档包含 2026 年 1 月 4 日起正式计费、免费额度有效期和历史数据保留期限等时间相关规则。落地前应以控制台账单页和最新计费文档为准。 + +## 效果优化方法 + +建议按“评估基线 → 诊断原因 → 定向调整 → 再评估”的方式优化： + +1. **建立[评测](../concepts/evaluation.md)集**：至少覆盖事实型、比较型、教程型、分析型等真实问题，记录初始召回内容和回答质量。 +2. **检索不到知识**：补充源文件，优化排版，统一实体表述，确保源文件语言与用户问题一致，必要时开启多轮对话改写。 +3. **召回不相关**：通过标签过滤和元数据过滤限定范围，尤其适合多类别文件或同结构文件混合的知识库。 +4. **切片不完整**：优先使用智能切分；导入后人工检查切片，必要时编辑、新增或删除切片。 +5. **重排不佳**：通过命中测试调整相似度阈值、召回片段数、排序模型模式和 TopK。 +6. **模型理解有误**：更换更适合长文本、推理或指令遵循的生成模型，并优化提示词中对引用材料的使用规则。 + +## 日志与监控 + +开通知识库检索日志后，所有检索调用会以 `log_dispatch` topic 投递到 SLS。常用字段包括： + +- `request_id`：定位单次请求。 +- `pipeline_id`：知识库 ID，用于按知识库聚合。 +- `workspace_id`：[业务空间](../concepts/workspace.md) ID。 +- `path`：API 路径。 +- `latency`：请求耗时。 +- `response_status_code`、`response_code`、`response_message`：定位 HTTP 或业务错误。 +- `request_body`、`response_body`：排查请求参数和召回结果，`response_body.data.nodes[]` 中包含切片文本、分数和元数据。 + +建议在 SLS 中建立调用量趋势、TopN 知识库、错误率、5xx、慢查询等仪表盘和告警。关闭检索日志开关只会停止新日志投递，历史日志仍按 SLS 配置保留和计费；如需彻底停止相关 SLS 费用，需要删除对应 LogStore。 ## 来源文档 @@ -89,3 +181,7 @@ RAG 效果由建立索引、检索召回、生成答案三个阶段决定。优 - [知识问答](../../raw/application-user-guide/knowledge-base/rag-knowledge-qa.md) + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/llm-application.md b/skills/bailian-docs-llm-wiki/wiki/guides/llm-application.md index cf82ffab..6df52af6 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/llm-application.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/llm-application.md @@ -1,124 +1,105 @@ # llm application -阿里云百炼平台提供三种核心应用构建模式：智能体（Agent）、工作流（Workflow）和高代码应用，用于突破大模型在私有知识访问、实时信息获取和复杂任务规划方面的原生局限。开发者可根据开发门槛、控制粒度和业务场景选择合适的应用类型，并通过集成知识库、MCP 工具、插件等能力构建完整的 AI 应用。 +阿里云百炼提供三种核心应用构建模式——智能体（Agent）、工作流（Workflow）和高代码应用，用于突破大模型在私有知识、实时信息、固定流程与复杂任务规划上的原生局限。三类应用通过知识库检索增强（RAG）、MCP/插件工具调用、记忆等能力扩展模型边界，并统一支持发布后经 API/SDK 集成到业务系统。 -## 应用类型与选型 +## 应用类型选型 -| 对比维度 | 智能体（Agent） | 工作流（Workflow） | 高代码应用 | -|---------|---------------|-------------------|-----------| -| 开发方式 | 自然语言配置（零代码） | 可视化节点编排（低代码） | Python 编码 | -| 核心特点 | AI 自主决策、动态规划 | 预定义流程精确控制 | 完全由代码控制 | -| 适合人群 | 业务人员、产品经理 | IT 运维、业务分析师 | AI 工程师、开发者 | +根据 [应用类型介绍](../../raw/application-user-guide/llm-application/application-introduction.md)，三种应用的定位如下： + +| 维度 | 智能体（Agent） | 工作流（Workflow） | 高代码应用 | +| --- | --- | --- | --- | +| 开发方式 | 自然语言配置（零代码） | 可视化节点编排（低代码） | Python 编码（专业代码） | +| 核心特点 | 大模型按提示词自主规划、动态决策 | 预定义节点精确控制，逻辑确定可复现 | 逻辑与执行路径完全由代码定义 | +| 适合场景 | 智能客服、知识问答、任务助理等开放式对话 | 报告生成、订单处理、审批流等固定流程自动化 | 私有算法部署、复杂系统集成、专业 AI 后端 | | 开发门槛 | 低 | 中 | 高 | -详细的类型介绍参见 [应用类型介绍](../../raw/application-user-guide/llm-application/application-introduction.md)。 +选型原则：希望模型根据可用工具自主决策时选智能体；需要稳定可控、可复现的多步骤链路时选工作流；需要深度定制并部署为云上后端服务时选高代码应用。 -## [智能体应用](../concepts/agent-application.md) +## 智能体应用（Agent 1.0 与 Agent 2.0） -### 新版智能体（Agent 2.0） +智能体由提示词驱动：接收请求后模型识别意图、规划任务、自主调用外部工具并整合结果，详见 [智能体应用](../../raw/application-user-guide/llm-application/single-agent-application.md)。 -新版智能体将知识库、MCP 等能力统一为工具，由智能体自主规划调用顺序，支持完整的"规划-执行-反思"链路展示。推荐在无旧版依赖时使用新版。 +### 版本差异 -核心能力配置： +[新版智能体应用](../../raw/application-user-guide/llm-application/new-single-agent-application.md)（Agent 2.0）与旧版（Agent 1.0）的关键区别： -- **模型选择**：推荐具备强工具调用能力的模型（如千问-Max 系列）。支持配置最长回复长度、temperature、enable_thinking（思考模式）等参数。 -- **提示词**：定义角色、行为指令与能力边界，支持自定义变量嵌入。 -- **内置工具**：沙箱环境中的 bash、write、read、edit、glob、grep、download_file 等工具，默认关闭需按需开启。 -- **知识库**：作为工具由智能体自主调用，支持标签过滤限定查询范围。 -- **MCP**：外部工具以 MCP 协议接入，支持动态非固定顺序调用。 -- **记忆**：短期记忆支持 0-30 轮上下文；长期记忆暂未支持。 -- **ReAct 最大轮次**：取值 1-50，限制单次会话中工具调用最大次数。 +- **规划与调度**：旧版先检索知识库、再决策是否调用其他工具；新版将知识库、MCP 统一为工具，由智能体自主规划调用时机与顺序。 +- **过程透明度**：旧版只展示最终结果；新版以卡片流完整展示每轮"规划-执行-反思"链路（含思考步骤与工具调用入参/结果）。 +- **外部工具形态**：旧版使用插件；新版全部以 MCP 协议接入，插件支持一键转换为 MCP 服务。 -详细配置方法参见 [新版智能体应用（Agent 2.0）](../../raw/application-user-guide/llm-application/new-single-agent-application.md)。 +> **注意**：旧版智能体与新版基于不同技术架构，互不兼容，不支持直接升级/降级；如需使用 Agent 2.0 能力，必须在控制台重新创建新版应用。 -### 旧版智能体（Agent 1.0） +### 关键配置与参数 -旧版智能体通过知识库（RAG）和插件扩展能力，适合意图单一、流程固定的简单任务。知识库检索后再决策是否调用其他工具。 +- **模型**：多步规划推荐千问-Max 系列等强工具调用能力模型。可配置最长回复长度、温度系数、`enable_thinking`（思考模式，仅支持思考的模型可配）。 +- **系统提示词**：定义角色、输出格式、约束条件与工具使用引导；支持嵌入自定义变量（输入 `/` 引用）。 +- **知识库（RAG）**：作为工具由智能体自主调用；新版支持通过文件标签 + 提示词规则限定检索范围。检索文本会占用上下文窗口并增加输入 Token。 +- **内置工具**（新版，沙箱运行，默认关闭）：`bash`、`write`、`read`、`edit`、`glob`、`grep`、`download_file`。 +- **记忆**：短期记忆支持 0-30 轮上下文；长期记忆在新版中尚未支持（计划迭代）。 +- **ReAct 最大轮次**：取值 1-50，限制单次会话工具调用次数，超限后强制生成最终回复。 +- **预解析文件**：开启时用预置解析器提取文本给模型；关闭时仅传文件 URL，由智能体决策是否调用工具处理。千问-VL 系列即使关闭预解析也能直接理解图片/视频。 -> **注意**：新版智能体与旧版智能体基于不同技术架构，不支持直接升级或版本切换。需要迁移时必须重新创建新版应用。 +### 限制与注意事项 -旧版智能体的自定义插件有 5 秒超时限制。详情参见 [智能体应用](../../raw/application-user-guide/llm-application/single-agent-application.md)。 +- 应用必须先**发布**才能通过 API 调用或集成；RAM 账号发布需 `ram:CreateServiceLinkedRole` 权限。 +- 自定义插件超时限制为 5 秒。 +- 上下文缓存仅支持隐式缓存（自动生效，命中部分输入按 20% 计费），不支持显式缓存。 +- Assistant API 可创建类似应用但不支持控制台管理。 ## 工作流应用 -工作流通过可视化节点编排将多步骤任务串联为稳定可控的执行链路，适合固定流程自动化场景。 - -主要节点类型： - -- **开始/结束节点**：定义输入输出参数结构，预置 query、historyList、imageList 变量 -- **大模型节点**：配置模型、提示词和用户提示词 -- **意图分类节点**：根据用户输入分发到不同处理分支 -- **智能体群组节点**：将子智能体作为工具组合调用 -- **变量处理节点**：文本输出或变量转换 +[工作流应用](../../raw/application-user-guide/llm-application/workflow-application.md)通过可视化节点（开始/结束、大模型、意图分类、变量处理、智能体群组等）编排多步骤任务： -工作流支持会话变量作为全局参数在节点间传递，支持记忆功能（本节点缓存或自定义缓存）。 - -创建和配置详情参见 [工作流应用](../../raw/application-user-guide/llm-application/workflow-application.md)。 +- **预置变量**：开始节点提供 `query`（用户输入）、`historyList`（对话历史）、`imageList`（图片列表）；节点中用 `${sys.query}` 引用。 +- **记忆**：大模型、意图分类等节点支持"本节点缓存"（仅记当前节点）或"自定义缓存"（记全局对话）。 +- **会话变量**：全局变量，在工作流全生命周期内记录参数并可被各节点引用。 +- **智能体群组节点**：可挂载多个已发布的子智能体（需先创建并发布），依据各子智能体的描述信息由决策模型分派任务。 +- 典型案例：诈骗信息识别（开始→大模型→结束）、意图分类导购、多智能体日程管理。 ## 高代码应用 -面向专业开发者，支持基于 Python 项目部署 AI 后端服务。 - -关键特性： - -- **部署方式**：Serverless Function（无状态快速拉起）和 K8s（高性能有状态长程任务） -- **MCP 工具接入**：控制台直接关联知识库、工作流、插件等 MCP 服务 -- **前端体验**：支持直接体验、自定义交互卡片、基于 Spark Design 的自定义 WebUI -- **企业级能力**：自动化运维、可观测、日志服务、API 网关 -- **代码提交**：支持控制台模板创建或命令行上传 .whl 代码包 - -生产环境建议开启网关功能，通过自定义域名访问。时延敏感业务建议最小实例数大于等于 1。 - -详细开发和部署流程参见 [高代码应用](../../raw/application-user-guide/llm-application/rich-code-application.md)。 - -## 文件问答 +[高代码应用](../../raw/application-user-guide/llm-application/rich-code-application.md)面向专业开发者，将 Python 项目部署为公网可访问的云上后端 API 服务： -[智能体应用](../concepts/agent-application.md)支持上传文件进行智能问答，提供三种处理模式： +- **部署方式**：Serverless Function（默认，低负载、无状态、快速拉起）或 K8s（高性能、有状态、长程任务，需开通 ACK 并授权）。 +- **代码提交**：控制台选模板（基础对话 / 工具调用 / 深度研究 Agent）或上传 `.whl` 代码包（AgentScope-AI 命令行）。 +- **资源配置**：vCPU/内存/磁盘规格、最小实例数、单实例并发度、部署地域；时延敏感业务建议最小实例数 ≥ 1。 +- **管理能力**：部署（重新部署/环境变量/触发器/日志/停止服务）、工具（MCP 接入知识库、工作流、插件）、网关（云原生 API 网关 + Token 鉴权，生产环境建议禁止测试域名公网访问；网关须与部署同地域）、前端（直接体验 / 自定义交互卡片 / 基于 Spark Design 的自定义 WebUI）。 +- 创建时需授权函数计算 FC 与 API 网关角色权限；部署后即开始计费。 -| 模式 | 适用场景 | 特点 | -|------|---------|------| -| 全文引用 | 文档总结、全文翻译 | 简单直接，受上下文长度限制 | -| 切片检索（RAG） | 长文档问答、知识库检索 | 能处理超长文件，效果依赖检索策略 | -| 自定义处理 | 图片转换、视频分析等需工具介入的任务 | 功能灵活，依赖配置的工具 | +## 文件问答（智能体内能力） -文件限制：单会话最多 10 个文件，单文件不超过 10MB。超过 10MB 需使用文件上传 API。 +[文件问答](../../raw/application-user-guide/llm-application/file-q-a.md)支持在智能体应用中对文档、图片、音视频进行问答，提供三种处理模式： -支持格式：文档（doc/docx/pdf/md/txt 等）、图片（png/jpg/bmp/gif）、视频（mp4/mkv/avi 等）、音频（mp3/wav/flac 等）。 +| 模式 | 机制 | 适用 | +| --- | --- | --- | +| 全文引用 | 文件全文（截断后）作为整体输入 | 总结、翻译、润色；Token 消耗大 | +| 切片检索（RAG） | 切片后检索相关片段拼接输入 | 长文档问答、精确定位来源 | +| 自定义处理 | 传文件 URL/内容，模型自主调用工具 | 图片风格转换、视频分析等 | -详细使用方式参见 [文件问答](../../raw/application-user-guide/llm-application/file-q-a.md)。 +关键限制： -## 发布与调用 +- 仅适用于中国大陆版（北京地域）。 +- 单会话最多 10 个文件、单文件不超过 10MB；聊天窗口上传的文件仅当前会话有效。 +- 更大文件走文件上传 API（`session_file_id`，有效期约 24 小时，生产推荐）；也可用 `image_list` / `file_list` 传 URL。 +- API 调用无法动态切换处理模式，以控制台保存的配置为准；应用级[限流](../concepts/rate-limit.md)默认 100 次/分钟，所有 API 请求共享配额。 +- 关键参数：全文引用的"单文件最大解析长度 / 最大拼装长度"（超出从末尾截断），切片检索的"召回片段数 / 最大拼装长度"（超出按相关性从低到高丢弃）。 -所有应用类型均需先发布才能通过 API 集成： +## 计费要点 -1. 在应用配置页点击"发布"，确认变更后完成发布 -2. 在"发布渠道"页签查看 API 调用方式 -3. [智能体应用](../concepts/agent-application.md)还支持发布到钉钉、微信公众号等第三方平台 +- 仅创建应用不收费；调用时按所选模型的输入/输出 [Token 计费](../concepts/token-billing.md)。 +- 知识库按量付费，召回切片会增加输入 Token；文件上传本身免费。 +- 部分官方 MCP 按模型调用计费（文生图、语音合成等）；第三方 MCP 的 API 费用由第三方收取。 +- 高代码应用的函数计算、网关、存储、模型调用均单独计费。 -RAM 账号发布前需确认拥有 `ram:CreateServiceLinkedRole` 权限。 - -## 计费说明 - -- **模型调用**：按模型类型和 [Token](../concepts/token.md) 用量计费 -- **知识库**：按量付费，召回的文本切片会增加输入 [Token](../concepts/token.md) -- **MCP/插件**：部分官方 MCP 按调用计费，第三方 MCP 费用由第三方收取 -- **高代码应用**：部署后函数计算、API 网关、存储均按量计费 -- **文件上传**：上传本身不收费，问答消耗按所选模型标准计费 +> **注意**：旧版智能体文档提到"长期记忆数据存储不收费、记忆体 Token 暂不计费"，而新版智能体尚不支持长期记忆，该计费说明仅适用于 Agent 1.0。 ## 来源文档 - [应用类型介绍](../../raw/application-user-guide/llm-application/application-introduction.md) -- [新版智能体应用（Agent 2.0）](../../raw/application-user-guide/llm-application/new-single-agent-application.md) +- [新版智能体应用](../../raw/application-user-guide/llm-application/new-single-agent-application.md) - [智能体应用](../../raw/application-user-guide/llm-application/single-agent-application.md) - [高代码应用](../../raw/application-user-guide/llm-application/rich-code-application.md) - [工作流应用](../../raw/application-user-guide/llm-application/workflow-application.md) - [文件问答](../../raw/application-user-guide/llm-application/file-q-a.md) - - - - - - - diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/managed-agents.md b/skills/bailian-docs-llm-wiki/wiki/guides/managed-agents.md index 6ad7d7d1..d4bbee6e 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/managed-agents.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/managed-agents.md @@ -1,62 +1,63 @@ # managed agents -Managed Agents 是百炼提供的智能体托管运行时，面向多步工具调用、代码执行、文件处理等长时运行任务。与无状态的[智能体应用](../concepts/agent-application.md)不同，它由平台在服务端托管会话状态、沙箱环境与工具执行，智能体在独立云端容器中自主执行命令、读写文件、安装依赖，事件历史在服务端持久化并支持中断与续接。 +Managed Agents 是百炼提供的智能体托管运行时，面向多步工具调用、代码执行、文件处理等长时运行任务。平台托管会话状态、沙箱环境和工具执行，智能体在独立的云端容器沙箱中自主运行命令、读写文件、安装依赖并处理数据，事件历史在服务端持久化。开发者只需专注 Agent 逻辑，无需自建代理循环、沙箱编排或工具执行基础设施。 -## 与[智能体应用](../concepts/agent-application.md)的区别 +## 与智能体应用的区别 -| 维度 | [智能体应用](../concepts/agent-application.md) | Managed Agents | +| 维度 | 智能体应用 | Managed Agents | | --- | --- | --- | | 运行模式 | 无状态调用，应用侧维护上下文 | 服务端维护会话状态，支持中断与续接 | | 执行环境 | 共享运行时 | 独立沙箱，云端容器 | | 事件模型 | 响应级[流式输出](../concepts/streaming.md) | 会话级 SSE 事件流，事件历史持久化 | -| 典型场景 | 问答、对话、轻量任务 | 多步工具调用、代码执行、文件处理等长时任务 | +| 典型场景 | 问答、对话、轻量任务 | 多步工具调用、代码执行、文件处理等长时运行任务 | ## 核心概念 -四个核心对象构成完整的运行链路，详见 [概述](../../raw/application-user-guide/managed-agents/managed-agents-introduction.md)： +详见[概述](../../raw/application-user-guide/managed-agents/managed-agents-introduction.md)： - **智能体（Agent）**：模型、系统提示词、工具、MCP 服务和 Skill 的组合配置。创建后通过 ID 引用，可在多个会话中复用。 - **运行环境（Environment）**：会话运行的沙箱配置，由百炼托管的云端容器，独立于智能体管理，可被多个会话复用。 -- **会话（Session）**：智能体在指定环境中的一次运行实例，承载任务执行与输出。 +- **会话（Session）**：智能体在指定环境中的一次运行实例，执行任务并生成输出。 - **事件（Event）**：应用与智能体之间交换的消息，包括用户消息、工具调用结果和状态变更。 -## 支持的工具 +## 支持的模型与工具 -智能体通过以下工具与运行环境交互： +- **模型**：创建智能体时从下拉列表选择或在 API 中指定，如 `qwen3-max`。 +- **内置工具**：共 7 个，默认全选——`bash`（命令执行）、`read`、`write`、`edit`、`glob`、`grep`（文件操作与搜索）、`download_file`（从 URL 下载）。可按需取消勾选。 +- **MCP 服务**：接入外部工具服务。 +- **Skill（技能）**：挂载预置的工具组合，封装端到端任务流程。 -- **命令执行**：在沙箱中运行 shell 命令（`bash`）。 -- **文件操作**：`read`、`write`、`edit`、`glob`、`grep`，以及从 URL 下载的 `download_file`；也可上传本地文件挂载到沙箱。 -- **MCP 服务**：接入外部工具服务，详见 [构建 Agent](../../raw/application-user-guide/managed-agents/managed-agents-agent.md)。 -- **Skill**：挂载预置的工具组合，封装端到端任务流程。 +> **注意**：[快速开始](../../raw/application-user-guide/managed-agents/managed-agents-quick-start.md)控制台示例中的模型名 `qwen3.7-plus` 与其 API 示例中的 `qwen3-max` 不一致，前者疑似笔误，实际以控制台下拉列表可选模型为准。 -快速开始阶段默认全选 7 个内置工具（`bash`、`read`、`write`、`edit`、`glob`、`grep`、`download_file`），可按需取消勾选。 +## 使用方式（4 步工作流程） -## 使用方式 +按[快速开始](../../raw/application-user-guide/managed-agents/managed-agents-quick-start.md)，可通过控制台向导或 API 完成： -典型流程分为四步（控制台向导或 API 均可完成），参见 [快速开始](../../raw/application-user-guide/managed-agents/managed-agents-quick-start.md)： +1. **创建智能体**：`POST /api/v1/agentstudio/agents`，指定 `name`、`model`、`system`（系统提示词）和 `tools`（`builtin_toolkit` 配置各工具的 `enabled`）。返回 `agent_xxx` ID 与版本号。 +2. **配置运行环境**：`POST /api/v1/agentstudio/environments`，`config.type` 为 `cloud`（云端托管），可通过 `packages.apt` / `packages.pip` 预装依赖（如 `ffmpeg`、`pandas`），`networking.type` 可设为 `unrestricted`。 +3. **创建会话**：`POST /api/v1/agentstudio/sessions`，用 `agent` 和 `environment_id` 绑定智能体与环境，可选 `title`。 +4. **发送事件并接收响应**：向 `sessions/{id}/events` 写入 `role: user` 的 message 事件触发处理；通过 `sessions/{id}/events/stream` 以 SSE（`Accept: text/event-stream`）实时接收工具调用过程与输出，直到会话状态变为 `idle` 或 `terminated`。 -1. **配置智能体**：指定名称、模型（如 `qwen3-max`）、系统提示词与工具。API 为 `POST /api/v1/agentstudio/agents`。 -2. **配置运行环境**：默认云端托管沙箱，可通过 `config.packages` 预装 apt / pip 依赖并设置网络策略。API 为 `POST /api/v1/agentstudio/environments`。 -3. **发起会话**：绑定智能体 ID 与环境 ID 创建会话实例。API 为 `POST /api/v1/agentstudio/sessions`。 -4. **发送事件并接收响应**：向会话写入用户消息触发处理（`POST /sessions/{id}/events`），通过 SSE 事件流实时接收工具调用过程与输出（`GET /sessions/{id}/events/stream`）。 +所有接口以 `https://{workspace_id}.cn-beijing.maas.aliyuncs.com` 为 endpoint，使用 `Authorization: Bearer $DASHSCOPE_API_KEY` 认证；同时提供 Python / Java SDK。执行过程中可发送新事件引导方向，或中断当前任务；会话操作还支持状态机管理与审批工具调用（详见[委派任务给 Agent](../../raw/application-user-guide/managed-agents/managed-agents-session.md)）。 -控制台的**预览调试**标签页可直接对话并按事件类型（User、Agent、Tool、Tool_output、Error、Model、System）筛选查看执行过程。 +## 上下文与文件挂载 -### 上下文与资源挂载 +根据 [Agent 上下文管理](../../raw/application-user-guide/managed-agents/managed-agents-context.md)： -上下文中的挂载资源独立于会话管理，可被多个会话复用，详见 [Agent 上下文管理](../../raw/application-user-guide/managed-agents/managed-agents-context.md)： +- 挂载资源是上下文中唯一独立于会话管理的部分，通过控制台或 API 创建后可挂载到一个或多个会话，生命周期与会话解耦。 +- **挂载时机**：创建会话时在 `resources` 字段指定资源列表和挂载路径。 +- **路径约定**：统一位于 `/mnt/session/uploads` 前缀下，可在系统提示词中直接引用完整路径。 +- **会话隔离**：挂载时平台做内部拷贝放入会话沙箱；会话内修改不影响原始资源，也不影响挂载同一资源的其他会话；卸载后仅清理会话内副本。 -- **挂载时机**：创建会话时在 `resources` 字段指定，或运行时通过 `POST /sessions/{session_id}/resources` 追加，实时生效且无需重启会话。 -- **路径约定**：挂载资源统一放在 `/mnt/session/uploads` 前缀下，可在系统提示词中直接引用完整路径。 -- **会话隔离**：平台为挂载资源做内部拷贝放入沙箱，会话内的修改不影响原始资源，也不影响挂载同一资源的其他会话；卸载后副本被清理。 +环境层面还可管理外部服务的**凭证**（详见[配置 Agent 环境](../../raw/application-user-guide/managed-agents/managed-agents-environment.md)）。 ## 限制与注意事项 -- 单个上传文件不超过 **10 MB**。 -- 沙箱内文件路径遵循 `/mnt/session/uploads` 约定，代码中引用文件应使用该完整路径。 -- 会话状态、中断续接与工具审批由会话状态机管理，详见 [委派任务给 Agent](../../raw/application-user-guide/managed-agents/managed-agents-session.md)。 - -> **注意**：文档中出现的模型名称不一致——控制台向导示例填写 `qwen3.7-plus`，而 API 代码示例使用 `qwen3-max`。请以控制台模型下拉列表中实际可选的模型 ID 为准。 +- 单个挂载文件不超过 **10 MB**。 +- 需已开通阿里云百炼，且当前账号在目标工作空间内具备 Managed Agents 操作权限。 +- 会话有状态：跨多轮交互保持上下文和文件系统状态，适合持续数分钟到数小时的任务；轻量问答场景应使用普通智能体应用。 +- 控制台"预览调试"页可按事件类型筛选（User、Agent、Tool、Tool_output、Error、Model、System），便于排查工具调用过程。 +- 智能体的完整配置项（模型、提示词、内置工具、MCP、技能）入口见[构建 Agent](../../raw/application-user-guide/managed-agents/managed-agents-agent.md)。 ## 来源文档 @@ -68,8 +69,3 @@ Managed Agents 是百炼提供的智能体托管运行时，面向多步工具 - [委派任务给 Agent](../../raw/application-user-guide/managed-agents/managed-agents-session.md) - - - - - diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/memory-library-overview.md b/skills/bailian-docs-llm-wiki/wiki/guides/memory-library-overview.md index 3ee13beb..e9041ea3 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/memory-library-overview.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/memory-library-overview.md @@ -1,12 +1,12 @@ # memory library overview -百炼记忆库（Memory Library）通过长期记忆 API 解决大模型跨会话上下文丢失的问题：自动从对话中提取关键信息并持久化存储，再在后续对话中基于语义检索召回相关记忆注入 Prompt，使智能体能够持续理解用户偏好与历史信息。该能力既可在百炼控制台可视化管理，也提供开放的 HTTP API 接入任意应用，并支持通过 OpenClaw 插件以"自动捕获 / 自动召回"的方式零侵入接入 Agent。详见 [记忆库](../../raw/application-user-guide/memory-library-overview/memory-library.md)、[长期记忆 API](../../raw/application-user-guide/memory-library-overview/long-term-memory-2-0.md) 与 [为 OpenClaw 配置长期记忆插件](../../raw/application-user-guide/memory-library-overview/modelstudio-memory-for-openclaw.md)。 +百炼记忆库（Memory Library）通过[长期记忆](../concepts/long-term-memory.md) API 解决大模型跨会话上下文丢失的问题：自动从对话中提取关键信息并持久化存储，再在后续对话中基于语义检索召回相关记忆注入 Prompt，使智能体能够持续理解用户偏好与历史信息。该能力既可在百炼控制台可视化管理，也提供开放的 HTTP API 接入任意应用，并支持通过 OpenClaw 插件以"自动捕获 / 自动召回"的方式零侵入接入 Agent。详见 [记忆库](../../raw/application-user-guide/memory-library-overview/memory-library.md)、[长期记忆 API](../../raw/application-user-guide/memory-library-overview/long-term-memory-2-0.md) 与 [为 OpenClaw 配置长期记忆插件](../../raw/application-user-guide/memory-library-overview/modelstudio-memory-for-openclaw.md)。 ## 核心能力 记忆库提供两类持久化记忆内容，二者可独立或组合使用： -- **记忆片段**：从对话中自动提取的关键事件和信息（如"用户每天上午9点需要喝水提醒"），适用于大多数长期记忆场景。支持自动去重、动态更新，也可通过 `custom_content` 直接写入指定内容。 +- **记忆片段**：从对话中自动提取的关键事件和信息（如"用户每天上午9点需要喝水提醒"），适用于大多数[长期记忆](../concepts/long-term-memory.md)场景。支持自动去重、动态更新，也可通过 `custom_content` 直接写入指定内容。 - **用户画像**：基于自定义画像模板从对话中提取的结构化属性（如年龄、职业、偏好等），适用于需要固定属性持久化存储的场景。属性字段及描述应清晰具体，避免"姓名/名称/名字"等同义字段并存，且不应期望一次对话就提取全部信息。 > **注意**：记忆有效期在不同入口存在差异。[长期记忆 API](../../raw/application-user-guide/memory-library-overview/long-term-memory-2-0.md) 文档指出"生成的记忆片段与用户画像暂无失效日期"，而 [记忆库](../../raw/application-user-guide/memory-library-overview/memory-library.md) 控制台的默认记忆片段规则预置了"默认有效期 180 天"，并支持按规则配置 7/30/180 天或永不过期。以控制台记忆规则配置为准；通过 API 直写且不指定 `project_id` 时使用默认规则。 @@ -45,7 +45,7 @@ Python 用户可安装 `agentscope-runtime`，使用 `AddMemory`、`SearchMemory ### 方式二：OpenClaw 记忆插件 -OpenClaw Agent 可通过插件实现零侵入的[跨会话记忆](../concepts/cross-session-memory.md)。插件在 Gateway 内通过 `before_agent_start`（自动召回）和 `agent_end`（自动捕获）两个生命周期钩子与长期记忆 API 交互，所有读写均由百炼服务端完成提炼、向量化和语义检索。 +OpenClaw Agent 可通过插件实现零侵入的[跨会话记忆](../concepts/cross-session-memory.md)。插件在 Gateway 内通过 `before_agent_start`（自动召回）和 `agent_end`（自动捕获）两个生命周期钩子与[长期记忆](../concepts/long-term-memory.md) API 交互，所有读写均由百炼服务端完成提炼、[向量化](../concepts/embedding.md)和语义检索。 ```bash # 安装 @@ -157,6 +157,19 @@ CLI 等效：`openclaw modelstudio-memory search|list|stats`。 + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/model-compression.md b/skills/bailian-docs-llm-wiki/wiki/guides/model-compression.md index 7b9ed189..6a2d2304 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/model-compression.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/model-compression.md @@ -1,77 +1,76 @@ # model compression -模型压缩是百炼平台提供的量化功能，用于将全精度微调模型转换为低精度版本，在保持模型能力的前提下降低部署所需的 MU 规格，从而减少推理成本。该功能属于模型生产链路中的可选环节，位于模型调优与模型部署之间。需要注意的是，百炼平台的模型压缩特指量化，不涉及结构剪枝或知识蒸馏。 +模型压缩是百炼平台模型生产链路中的可选环节，通过量化技术将全精度微调模型转为低精度版本，在保持模型能力的前提下降低部署所需的 MU 规格与推理成本。百炼的模型压缩特指量化，不涉及结构剪枝或知识蒸馏。完整链路为：模型调优 → 模型压缩 → 模型部署，详见[模型压缩](../../raw/model-user-guide/model-compression/model-compression-introduction.md)。 -## 核心概念 +## 支持的模型与使用限制 -模型压缩通过量化技术降低模型参数精度来实现部署成本优化。完整的模型生产链路为：模型调优 → 模型压缩（可选）→ 模型部署。以 qwen3.5-flash-2026-02-23 微调模型为例，压缩前部署规格为 MU1*2（108 元/小时），压缩后降至 MU8*1（47 元/小时），部署成本节省约 56%。详细说明参见[模型压缩](../../raw/model-user-guide/model-compression/model-compression-introduction.md)。 +- **支持范围**：仅支持通过百炼平台微调产出的自定义模型，不支持基础模型或第三方模型。当前支持的模型系列以控制台展示为准，示例：Qwen 系列的 qwen3.5-flash-2026-02-23。 +- **地域限制**：仅华北2（北京）地域可用。 +- **不可逆性**：压缩是单向操作。压缩后的模型不支持继续微调，也不支持二次压缩；如需调整，须从上游全精度微调模型重新压缩。 -> **注意**：压缩操作不可逆。压缩后的模型不支持继续微调，也不支持二次压缩。如需调整，必须从上游全精度微调模型重新压缩。 +以 qwen3.5-flash-2026-02-23 微调模型为例，压缩前部署规格为 MU1\*2（¥108/小时），压缩后降至 MU8\*1（¥47/小时），部署成本节省约 56%。 -## 支持的模型 +## 关键参数 -当前支持压缩的模型以控制台展示为准。已知支持的模型系列及规格如下： - -| 模型系列 | 基础模型 | 压缩前部署规格 | 压缩后部署规格 | -|---------|---------|--------------|--------------| -| Qwen | qwen3.5-flash-2026-02-23 | MU1*2（108 元/小时） | MU8*1（47 元/小时） | - -仅支持通过百炼平台微调产出的自定义模型，不支持基础模型或第三方模型。 - -## 使用限制 - -- 仅华北2（北京）地域可用。 -- 仅支持百炼平台微调产出的自定义模型。 -- 压缩后的模型不支持继续微调或二次压缩。 -- 压缩任务创建后不可修改配置。 - -## 创建压缩任务 - -在使用模型压缩前，需确保工作空间中已有微调训练完成的自定义模型。操作路径：控制台 → 模型 → 模型训练 → 模型压缩 → 创建压缩任务。 - -创建任务时需配置以下参数（详见[模型压缩](../../raw/model-user-guide/model-compression/model-compression-introduction.md)中的操作步骤）： +创建压缩任务时需配置以下参数（详见[模型压缩](../../raw/model-user-guide/model-compression/model-compression-introduction.md)的操作步骤）： | 参数 | 必填 | 说明 | -|-----|------|-----| +| --- | --- | --- | | 任务名称 | 是 | 最长 50 字符，建议包含模型简称、量化方式和版本号 | | 任务描述 | 否 | 最长 200 字符 | -| 选择源模型 | 是 | 仅展示可压缩的微调模型，切换源模型会清空已选量化模板 | -| 量化产出模型名后缀 | 是 | 仅支持小写字母和数字，最长 8 位 | -| 量化模板 | 是 | 须先选择源模型，MU 编号越大表示部署规格越小、成本越低 | -| 校准数据 | 条件选填 | 仅当量化模板包含校准输入参数时显示，最多选择 5 个数据集 | +| 选择源模型 | 是 | 仅展示可压缩的微调模型；切换源模型会自动清空已选量化模板 | +| 量化产出模型名后缀 | 是 | 仅小写字母和数字，最长 8 位，拼接到源模型名后作为压缩后模型标识 | +| 量化模板 | 是 | 须先选源模型；模板名中 MU 编号越大，部署规格越小、成本越低，但精度损失可能越大 | +| 校准数据 | 条件选填 | 仅当模板含校准输入参数时显示；最多 5 个数据集，不支持 OSS 挂载数据集，需在数据管理中创建并发布 | -## 量化模板与校准数据选择 +**参数选择建议**： -**量化模板**决定了压缩后模型的部署规格。模板名称中 MU 编号越大，部署规格越小、成本越低，但精度损失可能越大。建议根据业务对成本和精度的权衡来选择。 +- 量化模板：根据成本与精度权衡选择；不确定时可在免费期内尝试多个模板，部署后用业务测试集验证。 +- 校准数据：选择与目标推理场景语义相近的数据集（如客服场景选客服对话样本），可提升量化精度。 -**校准数据**用于提升量化精度，建议选择与目标推理场景语义相近的数据集。例如用于客服问答场景时，应选择包含客服对话样本的数据集。 +## 使用方式 -## 任务状态与管理 +### 创建压缩任务 -压缩任务有 7 种状态：PENDING（待开始）→ QUEUING（排队中）→ RUNNING（运行中）→ SUCCEEDED（成功）/ FAILED（失败）/ CANCELING（停止中）→ CANCELED（已取消）。 +1. 确保当前工作空间已有微调训练完成的自定义模型（前提条件）。 +2. 登录百炼控制台，选择**模型** > **模型训练** > **模型压缩**，单击**创建压缩任务**。 +3. 按上表配置参数后单击**开始压缩**，创建成功后自动跳转任务列表页。 -任务管理操作： -- **停止任务**：仅 PENDING 和 RUNNING 状态可停止，停止后不可恢复。 -- **删除任务**：仅终态（SUCCEEDED、FAILED、CANCELED）可删除，删除任务记录不影响已产出的压缩模型。 +> **注意**：压缩任务创建后不可修改配置，请在开始压缩前仔细确认量化模板选择。 -任务失败时，可在详情页查看错误信息，或切换到日志页签搜索 ERROR 级别日志进行排查。如仍无法解决，可提交工单并附上任务 ID 和日志文件。更多管理细节参见[模型压缩](../../raw/model-user-guide/model-compression/model-compression-introduction.md)。 +### 查看进度与日志 -## 计费说明 +在任务列表单击任务名称进入详情页： -- 压缩任务本身限时免费，截止时间以控制台公告为准。 -- 压缩后的模型在部署阶段按 MU 规格计费。 -- 建议在免费期内对同一微调模型尝试多个量化模板，分别部署后用业务测试集验证推理效果，选择最优方案后再正式上线。 +- **详情页签**：任务 ID、运行状态、创建时间、任务耗时（实时更新）、压缩配置；失败时显示错误信息。 +- **日志页签**：可选显示最近 100/500/2000/5000 行，支持 30 秒自动刷新、日志级别着色（INFO/DEBUG/ERROR/WARN）、下载全量日志、暗色/亮色切换。 -## 来源文档 +失败排查顺序：详情页看错误信息 → 日志页搜 ERROR → 下载全量日志分析 → 提交工单（附任务 ID 和日志）。 -- [模型压缩](../../raw/model-user-guide/model-compression/model-compression-introduction.md) +### 管理任务 +任务共 7 种状态：PENDING → QUEUING → RUNNING → SUCCEEDED / FAILED，以及 CANCELING → CANCELED。 +- **停止**：仅 PENDING 和 RUNNING 状态可停止，需二次确认，停止后不可恢复。 +- **删除**：仅终态（SUCCEEDED、FAILED、CANCELED）可删除，QUEUING 不可删除；删除任务记录不影响已产出的压缩后模型。 +列表支持按量化模板、任务状态、时间范围和任务名称/ID 筛选，每页 10 条。也可通过 API 管理压缩任务，参见原文中的模型压缩 API 参考链接。 + +## 计费说明 +- 压缩任务本身限时免费，截止时间以控制台公告为准；免费期结束后的计费方式同样以公告为准。 +- 压缩后模型在部署阶段按 MU 规格计费，不受免费期影响。 +建议在免费期内对同一微调模型尝试多个量化模板，分别部署并用业务测试集对比精度与成本，确定最优方案后再正式上线，具体做法见[模型压缩](../../raw/model-user-guide/model-compression/model-compression-introduction.md)的常见问题部分。 +## 注意事项 +- 量化可能带来一定精度损失。部署后应使用业务测试集验证推理效果；若精度不满足需求，可更换量化模板重新压缩，或添加校准数据提升量化质量。 +- 量化、剪枝、蒸馏是三种不同的模型优化技术：量化降低参数精度，剪枝移除冗余结构，蒸馏用大模型训练小模型。百炼当前仅支持量化。 +- 创建任务时若无可选源模型，说明当前工作空间没有符合条件的微调模型，需先完成模型微调训练且状态为成功。 +## 来源文档 + +- [模型压缩](../../raw/model-user-guide/model-compression/model-compression-introduction.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/model-context-protocol.md b/skills/bailian-docs-llm-wiki/wiki/guides/model-context-protocol.md index d021cb14..53fd856a 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/model-context-protocol.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/model-context-protocol.md @@ -1,6 +1,6 @@ # model context protocol -模型上下文协议（Model Context Protocol, MCP）是 Anthropic 提出的开源标准协议，用于在大模型与外部工具之间搭建统一的信息传递通道。阿里云百炼基于 MCP 提供全周期服务：开发者无需为每个外部工具编写专用接口，即可让智能体、工作流应用接入海量第三方工具，也能通过外部调用集成到第三方应用或个人项目中。 +模型上下文协议（Model Context Protocol, MCP）是 Anthropic 提出的开源标准协议，用于在大模型与外部工具之间搭建统一的信息传递通道。阿里云百炼基于 MCP 提供全周期服务：开发者无需为每个外部工具编写专用接口，即可让智能体、[工作流](../concepts/workflow.md)应用接入海量第三方工具，也能通过外部调用集成到第三方应用或个人项目中。 ## 服务类型 @@ -16,10 +16,10 @@ MCP 服务既可在平台内部集成，也可通过外部调用集成到第三方应用。 -### 平台内部：智能体与工作流 +### 平台内部：智能体与[工作流](../concepts/workflow.md) - **[智能体应用](../concepts/agent-application.md)**：大模型根据对话内容自动判断是否调用 MCP 服务，单个智能体最多可同时添加 **5 个** MCP 服务。适合路径规划、逻辑推理、多工具组合（如天气查询 + 图表绘制）等场景。 -- **工作流应用**：每个 MCP 节点只能使用一个工具，需手动指定输入参数并将输出传递到下一节点。通常需先用大模型节点把自然语言解析为 MCP 工具所需的输入参数（在 System Prompt 中描述工具的名称、功能、输入输出格式），再接入 MCP 节点。 +- **[工作流](../concepts/workflow.md)应用**：每个 MCP 节点只能使用一个工具，需手动指定输入参数并将输出传递到下一节点。通常需先用大模型节点把自然语言解析为 MCP 工具所需的输入参数（在 System Prompt 中描述工具的名称、功能、输入输出格式），再接入 MCP 节点。 > **注意**：在工作流中仅使用单一工具（如 Amap Maps 的 `maps_weather`）时，工作流只能回答与该工具相关的问题。 @@ -51,11 +51,11 @@ MCP 服务既可在平台内部集成，也可通过外部调用集成到第三 > **注意**：百炼 MCP 服务已从旧版 SSE 协议升级为新版 **Streamable HTTP** 协议。已开通用户需在 MCP 广场执行"取消开通 → 立即开通"完成协议升级；SDK 调用请使用 `streamablehttp_client` 连接。 -## 计费 +## [计费](../concepts/billing.md) -- **云部署 MCP 服务**：限时免部署费用；部分服务涉及第三方 API 调用，费用由第三方收取。联网搜索 MCP 服务免费额度 2000 次，用尽后按 29 元/千次计费，限流 15 QPS（主账号与 RAM 子账号共享）。 +- **云部署 MCP 服务**：限时免部署费用；部分服务涉及第三方 API 调用，费用由第三方收取。联网搜索 MCP 服务免费额度 2000 次，用尽后按 29 元/千次[计费](../concepts/billing.md)，[限流](../concepts/rate-limit.md) 15 QPS（主账号与 RAM 子账号共享）。 - **自定义部署 MCP 服务**： - - **基础模式**：无部署费用，按调用时长计费（0.000156 元/秒），首次调用有冷启动延迟，适合偶尔调用。 + - **基础模式**：无部署费用，按调用时长[计费](../concepts/billing.md)（0.000156 元/秒），首次调用有冷启动延迟，适合偶尔调用。 - **极速模式**：有部署费用（0.000036 元/秒）+ 调用费用（0.000156 元/秒），适合长时间在线、调用频繁的场景。 ## 限制与注意事项 @@ -66,8 +66,8 @@ MCP 服务既可在平台内部集成，也可通过外部调用集成到第三 - **无法访问本地资源**：自定义 MCP 服务托管在函数计算 FC，暂不支持访问用户本地数据库、文件、硬件等资源；需要访问本地资源的 MCP Server 建议在本地部署。 - **访问远程资源需配置网络**：FC 无固定出口公网 IP，访问云数据库等远程资源需配置 FC 的 IP 白名单或打通 VPC 网络。 - **仓库与版本限制**：私有 npm 仓库暂不支持，需发布到公共仓库或改用 SSE；通过 npx/uvx 部署的服务在源版本更新后不会自动更新，需手动重新部署。 -- **调用增加 Token 消耗**：MCP 返回的内容会作为上下文传入模型，增加输入 Token，并可能间接增加输出 Token。 -- **调用失败排查**：优先确认已开通/升级服务、API Key 有效、额度未用尽；若模型无报错但不调用工具，应在提示词中明确工具名称与能力，必要时更换更强的推理模型（如千问 3 系列）。自定义服务的连接、超时、鉴权、协议等错误可对照 `11200044`~`11200060` 系列错误码逐项排查。 +- **调用增加 [Token](../concepts/token.md) 消耗**：MCP 返回的内容会作为上下文传入模型，增加输入 [Token](../concepts/token.md)，并可能间接增加输出 [Token](../concepts/token.md)。 +- **调用失败排查**：优先确认已开通/升级服务、[API Key](../concepts/api-key.md) 有效、额度未用尽；若模型无报错但不调用工具，应在提示词中明确工具名称与能力，必要时更换更强的推理模型（如千问 3 系列）。自定义服务的连接、超时、鉴权、协议等错误可对照 `11200044`~`11200060` 系列错误码逐项排查。 ## 来源文档 @@ -80,3 +80,16 @@ MCP 服务既可在平台内部集成，也可通过外部调用集成到第三 + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/model-data-overview.md b/skills/bailian-docs-llm-wiki/wiki/guides/model-data-overview.md index a2d4f8ac..49831466 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/model-data-overview.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/model-data-overview.md @@ -1,173 +1,80 @@ # model data overview -百炼平台的数据管理功能用于在[模型调优](../concepts/fine-tuning.md)和[评测](../concepts/evaluation.md)前创建、清洗、增强训练集与[评测](../concepts/evaluation.md)集。它统一管理[业务空间](../concepts/workspace.md)下的大模型相关数据集，分为训练集（用于[模型调优](../concepts/fine-tuning.md)）和[评测](../concepts/evaluation.md)集（用于模型评测）两类，并支持基于数据流的可视化数据处理能力。本文汇总训练集与评测集的格式规范、关键参数以及数据清洗与增强的使用方式。 +阿里云百炼的模型数据管理围绕"数据集的构建、加工与积累"三个环节展开：通过数据管理功能统一维护用于模型调优的**训练集**和用于模型评测的**评测集**，通过数据流实现训练数据的**清洗与增强**，并可借助**日志回流**将线上推理日志转化为可复用的结构化数据集。三者共同构成模型调优/评测前的数据准备链路。 -> **注意**：本文涉及的数据管理与数据处理能力**仅适用于华北2（北京）地域**。此外，阿里云百炼目前暂未提供可用的数据处理 API，所有数据处理操作需在控制台完成。 +> **注意**：数据管理与数据处理功能仅适用于华北2（北京）地域；日志回流在华北2（北京）和新加坡两个 Region 可用，其他 Region 不显示入口。 -## 支持的数据集类型 +## 数据集类型与格式 -数据集分为训练集和评测集两类，详见 [训练集与评测集](../../raw/model-user-guide/model-data-overview/training-set-and-evaluation-set.md)。 +根据[训练集与评测集](../../raw/model-user-guide/model-data-overview/training-set-and-evaluation-set.md)，数据管理支持两大类数据集： -| 类型 | 用途 | 支持的子类型 | -| --- | --- | --- | -| 训练集 | 用于[模型调优](../concepts/fine-tuning.md)，通过在特定任务上进行有监督训练提升模型表现 | 文本生成、[多模态](../concepts/multimodal.md)理解、图生视频（首帧）、图生视频（首尾帧） | -| 评测集 | 用于评估模型在未见过数据上的泛化能力 | 文本生成 | +- **训练集**：用于模型调优，支持文本生成、[多模态](../concepts/multimodal.md)理解、图生视频（首帧）、图生视频（首尾帧）场景。 +- **评测集**：用于评估调优后模型的泛化能力，目前仅支持文本生成场景（Excel 格式，Prompt/Completion 两列）。 -## 训练集格式 +### 文本生成与[多模态](../concepts/multimodal.md)理解训练集 -### SFT 训练集（文本生成） - -采用 ChatML 格式，支持多轮对话和多种角色设置。一行训练数据为一个 JSON 对象，结构如下： - -```json -{"messages": [ - {"role": "system", "content": "系统输入1"}, - {"role": "user", "content": "用户输入1"}, - {"role": "assistant", "content": "期望的模型输出1"}, - {"role": "user", "content": "用户输入2"}, - {"role": "assistant", "content": "期望的模型输出2"} -]} -``` - -不支持 OpenAI 的 `name`、`weight` 参数，所有的 assistant 输出都会被训练。单条训练数据的所有 assistant 行支持 `loss_weight` 参数（设置范围 `0.0~1.0`，数值越大重要性越高），该参数属于邀测参数，如需使用请联系商务经理。 - -### SFT 思考模型（thinking） - -只能针对**最后**的 assistant 输出进行训练，思考内容必须用 `导读` / `` 标签包裹，且思考标签前后的若干个 `\n` 必须保留。中间的 assistant 输出不应添加思考标签。也可以在训练样本中设置模型不输出 `导读` 标签，但训练完成后不建议再开启思考模式调用。 - -### SFT 视觉理解（千问 VL） - -支持图片、视频文件路径和图片帧列表三种输入。如需传入 `system` 消息，对应 `content` 必须使用数组格式 `[{"text":"..."}]`，不能使用字符串格式。关键参数： - -| 字段 | 类型 | 必填 | 说明 | -| --- | --- | --- | --- | -| `image` | str | 是 | 图片文件路径 | -| `resized_width` / `resized_height` | int | 否 | 图片目标缩放尺寸（像素） | -| `video`（路径模式） | str | 是 | 视频文件路径，仅 qwen3.5 及以后 VL 模型支持 | -| `video`（帧列表模式） | `List[str]` | 是 | 图片帧列表，需配合 `sample_fps` | -| `fps` / `sample_fps` | float | 否 | 训练输入频率 / 帧率 | -| `video_start` / `video_end` | float | 否 | 视频截取起止时间（秒） | - -关于物体定位坐标：Qwen2.5-VL 使用相对缩放后图像左上角的绝对像素坐标；Qwen3-VL 使用相对坐标，坐标值会缩放到 `[0, 999]` 范围。 - -### DPO 数据集 - -DPO ChatML 格式将 `messages` 内所有内容作为输入，通过 `chosen` 与 `rejected` 字段训练模型对最后一条用户输入的正负反馈。针对深度思考内容需使用 `导读` 标签包裹。`chosen` 模块支持 `loss_weight` 参数（邀测参数）。 +| 类型 | 格式要点 | +| --- | --- | +| SFT | ChatML 格式 JSONL，支持多轮对话；不支持 OpenAI 的 `name`、`weight` 参数，所有 assistant 输出都会被训练 | +| SFT 思考模型（thinking） | 只训练**最后一个** assistant 输出，思考内容用 `\n...\n\n\n` 包裹（换行必须保留）；中间轮次的 assistant 不加 `` 标签 | +| SFT 视觉理解（千问VL） | `content` 为数组格式（含 `text` / `image` / `video` 元素）；system 消息的 content 必须用数组格式而非字符串 | +| DPO | messages + `chosen` / `rejected` 结构，训练模型对最后一条用户输入的正负偏好 | +| CPT | 纯文本 JSONL：`{"text":"文本内容"}` | -### CPT 训练集 +关键参数与限制： -纯文本格式，一行训练数据结构为 `{"text":"文本内容"}`。 +- `loss_weight`（0.0~1.0）：可对 assistant 行（SFT）、最后一条 assistant（thinking）或 `chosen` 模块（DPO）设置训练相对重要性。**邀测参数**，需联系商务经理。 +- 视觉理解压缩包：ZIP 格式，最大 2 GB；`data.jsonl` 必须位于根目录；图片单张不超过 1024px、10MB；文件名全局唯一且仅支持 ASCII 字母/数字/下划线/连字符。 +- 视频文件路径模式与图片帧列表模式仅 qwen3.5 及以后的 VL 模型支持。 +- 物体定位坐标：Qwen2.5-VL 使用缩放后图像的绝对像素坐标；Qwen3-VL 使用缩放到 `[0, 999]` 的相对坐标。 +- 规模建议：CPT 至少一千万 Token 预训练数据；SFT 至少上千条微调数据；DPO 一般需要上百条偏好数据。 ### 图生视频训练集 -分为基于首帧和基于首尾帧两种。训练集必须提供，验证集可选（无需提供视频，训练任务会在评估节点自动调用模型生成预览视频）。 - -- 标注文件固定命名为 `data.jsonl`，最大 20MB，每行为一个 JSON 对象。 -- 字段包括 `prompt`、`first_frame_path`、`last_frame_path`（仅首尾帧）、`video_path`（仅训练集）。 -- 图像最大分辨率 4096*4096，支持 BMP、JPEG、PNG、WEBP；视频支持 MP4、MOV。 - -## 文本生成评测集 - -单轮对话评测数据，使用 Excel 格式，每行包含 Prompt（用户输入）和 Completion（模型期望输出）。参评模型基于评测集中每条 Prompt 进行推理，评分员或自动化评分系统参考 Completion 对推理结果评分。 - -## 数据集构建规模要求 - -不同调优方式对训练集规模有不同最低要求： - -| 调优方式 | 最低规模 | -| --- | --- | -| CPT | 一千万 [Token](../concepts/token.md) 优质预训练数据 | -| SFT | 上千条优质微调数据 | -| DPO | 上百条人类偏好数据 | - -> 如果调优后评测结果不佳，最简单的改进方法是收集更多数据进行训练。 +- **首帧模式**：训练集含首帧图像 + 训练视频 + `data.jsonl`（字段 `prompt` / `first_frame_path` / `video_path`）。 +- **首尾帧模式**：额外提供尾帧图像（字段增加 `last_frame_path`）。 +- 验证集可选，仅需图像和 [prompt](prompt.md)，训练任务在评估节点自动生成预览视频。 +- 限制：`data.jsonl` 最大 20MB；图像/视频最大分辨率 4096*4096；图像支持 BMP/JPEG/PNG/WEBP，视频支持 MP4/MOV。 ## 数据清洗与增强 -在模型调优前，可使用数据处理功能对训练集进行数据清洗和数据增强，从而获得更高质量的训练集，详见 [数据清洗或增强](../../raw/model-user-guide/model-data-overview/data-processing.md)。 +参见[数据清洗或增强](../../raw/model-user-guide/model-data-overview/data-processing.md)，通过控制台可视化数据流对训练集做加工（**暂无 API**）： -| 处理方式 | 适用场景 | -| --- | --- | -| 数据清洗 | 修正训练数据中的规范性、合规性、一致性及重复等问题（如特殊内容移除、敏感信息打码等十种清洗操作） | -| 数据增强 | 增加训练数据的多样性和均衡性，或扩展数据规模 | - -> **注意**：如果训练集数据不适合清洗与增强（如法律文件、医学记录、文学作品、方言汇总、用户评论、技术手册等），建议直接跳过数据处理。此外，数据处理目前仅支持 SFT-文本生成训练集，不支持图片理解训练集和 DPO-文本生成训练集。 - -### 创建数据流任务 +- **数据清洗**：修正规范性、合规性、一致性和重复问题，提供特殊内容移除、敏感信息打码、敏感词过滤、毒性消除等十种算子；算子按编排顺序执行，建议每个清洗节点只开启一个算子便于排错。 +- **数据增强**：基于 Few-Shot 策略调用千问-Max（不可选其他模型）生成新样本，支持通用、文本分类、文本抽取、文本创作四种场景；单次任务最多生成 2000 条样本。 -数据处理通过在控制台搭建自定义数据流完成。典型流程为「先清洗后增强」——确保增强操作在干净、高质量的数据集上进行，保证模型调优数据源的准确性。两步操作： +使用方式：在数据管理页面创建数据流（开始节点 → 清洗/增强/条件判断节点 → 结束节点）并发布，再基于数据流创建任务处理训练集。处理结果自动生成训练集**新版本**，不覆盖原数据。 -1. **创建数据流**：在数据管理页面创建空白数据流，将数据清洗节点（开启所需算子）和数据增强节点拖入画布并依次连接，发布后即可使用。也可直接使用预置的数据流模板。 -2. **创建数据流任务**：选择已发布的数据流，输入任务名称，选择模型数据作为数据来源并指定训练集，任务自动执行。执行期间不支持手动终止。 +限制与注意事项： -任务处理状态包括：**处理中**（执行中，高峰时段需排队）、**已完成**（成功完成，可查看处理结果）、**处理失败**（建议提交工单咨询）。 +- 仅支持 SFT-文本生成训练集（ChatML 格式），不支持图片理解和 DPO 训练集。 +- 任务执行期间不支持手动终止；任务删除后执行过程不可找回。 +- 法律文件、医学记录、文学作品等不适合清洗/增强的数据建议跳过此环节。 -> 训练集在清洗或增强后会自动生成一个新版本，新版本独立保存，不会覆盖原训练集。建议检查清洗后的训练集，确保数据完整性和真实性未被破坏。 +## 日志回流 -### 节点说明 +参见[日志回流](../../raw/model-user-guide/model-data-overview/model-log-backflow.md)，可将 SLS 推理日志转化为结构化 JSONL 数据集： -数据流由若干节点组成，每个数据流必须包含一个开始节点和一个结束节点： +- **支持范围**：训练集（文本生成，SFT/DPO/CPT）和评测集（文本生成）。 +- **前置条件**：需按序开启审计日志和推理日志（先审计后推理，关闭时反序），并授权 `AliyunServiceRoleForSFMAccessSLS` 等服务关联角色；OSS 挂载模式还需额外授权两个角色。 +- **入口**：模型监控列表页、模型监控详情页、数据管理页三处均可创建。 +- **存储方式**：平台存储（默认，导入完成自动发布版本）或 OSS 挂载（可直接在 Bucket 管理文件，但不支持评测集和"新增版本"操作）。 -| 节点 | 作用 | -| --- | --- | -| 开始/结束 | 开始节点接收待处理训练集（`对话文本`参数无法更改）；结束节点输出处理结果，自动生成新版本 | -| 条件判断 | 设置条件分支，支持且/或配置，多条件自上而下顺序执行 | -| 数据清洗 | 选择清洗算子（如文章相似度去重、敏感词过滤、毒性消除等），按编排顺序执行，输出清洗后训练集及 `dataSetCount` 变量 | -| 数据增强 | 增加数据多样性和规模，分为通用、文本分类、文本抽取、文本创作四种场景 | - -### 数据增强节点关键参数 +限制和最佳实践： -数据增强节点本质上是基于 `千问-Max` 大模型的 Few-Shot 生成器，暂不支持选择其他模型。每次最多生成 2000 条样本。 +- 单次回流上限 10 万条，但可多次回流到同一数据集的不同版本积累数据，总量不受此限制。 +- 存储方式、数据类型、训练方式创建后不可更改；时间范围仅支持最近 30 天。 +- 推荐分批次按时段/模型回流，优先选择表现良好的模型和业务高峰时段的数据。 +- 推理日志开启后 SLS 持续产生费用，不再需要时应及时关闭。 -| 参数 | 说明 | -| --- | --- | -| 生成样本数 | 需要生成的数据量。原训练集 N 条 + 生成 M 条 = 增强后 N+M 条 | -| 指令生成依赖样本数 | 从原训练集中选出的种子数量，拼入 Prompt 提供给大模型。若种子+Prompt 总长度超过千问-Max 最大输入 [Token](../concepts/token.md)，系统会自动调整 | -| 过滤相似度阈值 | 控制生成数据的相似度过滤 | -| Prompt 配置 | 定义增强任务输入输出要求，支持 `few_shot_examples` 参数。提供默认模板 | +## 三个功能的协作关系 -Few-Shot 策略示例：训练集 1000 条、指令生成依赖样本数 5、生成样本数 200，则每次从 1000 条中抽样 5 条拼入 `few_shot_examples`，请求千问-Max 生成 1 条，重复 200 次。 - -> 输出数据中 `foreignKey` 为系统后添加的标识字段，增强后的训练集可直接用于模型调优，无需删除该字段。 - -### 数据增强建议 - -- **任务相关性**：确保生成数据与目标任务高度相关，避免引入不相关变体。 -- **多样化策略**：使用同义词替换、随机抽样、翻译变换等多种策略提升数据多样性。 -- **平衡增强**：生成数据应在类别、难度和结构上相对平衡，避免过度接触特定类型数据导致过拟合。 - -## 限制与注意事项 - -- 数据管理与数据处理能力仅适用于华北2（北京）地域。 -- 数据处理暂未提供可用 API，需在控制台完成；仅支持 SFT-文本生成训练集（ChatML 格式），不支持图片理解和 DPO 训练集。 -- 压缩包格式为 ZIP，最大 2GB；VL 训练集图片单张尺寸宽高均不超过 1024px、单张不超过 10MB；图生视频训练集图像/视频最大分辨率 4096*4096。 -- `loss_weight`、思考模式相关参数等部分能力属于邀测参数，如需使用请联系商务经理。 -- 数据流任务执行期间暂不支持手动终止，处理失败建议提交工单咨询。 - -> **注意**：构建有效的 SFT 训练集通常需要 1000+ 样本；数据增强节点暂不支持选择模型，固定使用千问-Max。更多信息参考 [训练集与评测集](../../raw/model-user-guide/model-data-overview/training-set-and-evaluation-set.md) 与 [数据清洗或增强](../../raw/model-user-guide/model-data-overview/data-processing.md)。 +典型工作流：线上推理日志经**日志回流**沉淀为训练集 → 通过**数据清洗/增强**提升质量与规模 → 在**数据管理**中以版本形式维护，最终用于模型调优或评测。回流产出的数据同样支持后续清洗，形成"生产数据 → 高质量训练集"的闭环。 ## 来源文档 - [训练集与评测集](../../raw/model-user-guide/model-data-overview/training-set-and-evaluation-set.md) - [数据清洗或增强](../../raw/model-user-guide/model-data-overview/data-processing.md) - - - - - - - - - - - - - - - - - - - +- [日志回流](../../raw/model-user-guide/model-data-overview/model-log-backflow.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/model-deployment-1.md b/skills/bailian-docs-llm-wiki/wiki/guides/model-deployment-1.md index 0491c527..99340401 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/model-deployment-1.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/model-deployment-1.md @@ -1,93 +1,72 @@ # model deployment 1 -模型部署让你为平台预置模型或调优后的自定义模型获得独立、资源专享的推理服务，以满足高并发、低延迟等生产需求。本页汇总三种计费方式的选型、PTU 长输入与前缀缓存机制、LoRA 模型导入约束，以及通过控制台或 API 完成部署的完整流程，面向需要落地专属推理服务的开发者。 +模型部署用于将百炼平台的预置模型或调优/导入后的自定义模型发布为独立、资源专享的推理服务，满足高并发、低延迟场景的性能需求。本主题覆盖三种计费部署方式（预置吞吐 PTU、模型单元、按 Token 用量）的选型与计费规则、PTU 长输入与前缀缓存的额度消耗机制、从 OSS 导入 LoRA 模型的完整流程，以及通过 API/命令行完成部署、查询、推理和下线的操作步骤。 -## 三种计费方式与选型 +## 三种部署计费方式选型 -百炼提供三种互斥的部署计费方式，计费方式在服务创建后无法更改，如需切换必须先下线已部署的模型再重新部署（详见 [模型部署](../../raw/model-user-guide/model-deployment-1/model-deployment-introduction.md)）： +根据 [模型部署](../../raw/model-user-guide/model-deployment-1/model-deployment-introduction.md)，平台提供三种部署方式，**计费方式在服务创建后无法更改**，切换需先下线再重新部署： -- **预置吞吐（PTU，Provisioned Throughput Unit）**：平台预留资源保障特定 TPM 吞吐能力，额度内不限速。相比按 Token 计费，TPS 通常提升约 1.5～2.0 倍，适合流量可预估的高负载生产环境（智能客服、实时内容审核）。支持预付费（按天）与后付费（按小时），可自助增减吞吐量并设置自动续费。 -- **模型单元（MU）**：按使用时长 × 模型单元数量计费，资源独占，延迟/吞吐等性能指标可自定义。支持部分预置模型与所有调优后模型，可自助增减模型单元数量，支持 PD 分离计算模式（拆分 Prefill 与 Decode 阶段以降低首 Token 延迟、提高吞吐）。 -- **按 Token 使用量**：以每次调用的输入/输出 Token 计量，不使用不计费。仅支持对基础模型完成 SFT 高效训练后的自定义模型，主要用于调优后模型的效果验证；扩缩容需在控制台提交申请等待人工审核。 +| 方式 | 定义 | 支持模型 | 计费 | 典型场景 | +| --- | --- | --- | --- | --- | +| 预置吞吐（PTU） | 平台预留资源，保障特定 TPM 吞吐，额度内不限速 | 部分预置模型 | 按使用时长和预置吞吐；随用随付、包天 | 高负载生产环境（智能客服、内容审核），TPS 比按量计费提升约 1.5~2.0 倍 | +| 模型单元（MU） | 按模型单元数量配置算力、资源独占，性能指标可自定义 | 部分预置模型与所有调优后模型 | 按使用时长 × 模型单元数量；随用随付、包月 | 私有微调模型、长时独占计算任务；支持 PD 分离模式 | +| Token 用量 | 按输入/输出 Token 计量，不使用不计费 | 部分 LoRA 调优后模型 | 按 Token 用量随用随付 | 调优后模型效果验证 | 关键计费公式： -- 预置吞吐（按时长）：`费用 = 使用时长 × (输入 TPM 单价 × 输入 TPM + 输出 TPM 单价 × 输出 TPM)` -- 模型单元（按时长）：`费用 = 使用时长（小时）× 模型单元数量 × 模型单元单价`；预付费按月时改为 `包月数 × 模型单元数量 × 月单价` -- 按 Token：`费用 = 输入 Token 数 × 输入单价 + 输出 Token 数 × 输出单价` +- PTU：`费用 = 使用时长 × (输入 TPM 单价 × 输入 TPM + 输出 TPM 单价 × 输出 TPM)`，后付费按小时、预付费按天。 +- 模型单元：`费用 = 使用时长（小时）× 模型单元数量 × 模型单元单价`；预付费为 `包月数 × 模型单元数量 × 月单价`。 +- Token 用量：`费用 = 输入 Token 数 × 输入单价 + 输出 Token 数 × 输出单价`（最小计费单位 1 token），仅 SFT 高效训练（LoRA）后的自定义模型支持。 -## PTU 长输入与前缀缓存 +限制与注意事项： -PTU 部署支持长输入请求（部分模型最高 200K token）和前缀缓存，通过阶梯容量系数和缓存折扣管理额度消耗，详见 [预置吞吐长输入与缓存](../../raw/model-user-guide/model-deployment-1/ptu-long-input-and-cache.md)： +- PTU 预付费按天计费、无法提前退费；预付费订单到期后延迟 2 小时停服，资源再保留 14 小时后释放。 +- 模型单元预付费首月内提前退订，日单价（≈ 月单价 / 30）按 1.2 倍计费；后付费算力资源先买到先得，购买不成功全额退款。 +- Token 用量方式的部署一个月内不使用将自动释放；扩缩容需在控制台提交申请人工审核。 +- 后付费欠费后资源继续保留并计费 24 小时，超时后停止计费、底层资源删除，补足欠费后可恢复。 +- PD 分离模式将 Prefill 与 Decode 拆到不同计算节点，降低首 Token 延迟、提高吞吐，仅模型单元部署支持。 -- **长输入阶梯系数**：超过 32K token 的输入按更高阶梯系数折算 TPM。例如 glm-5.1 在 `[32K, 200K]` 区间输入系数为 1.33、输出为 1.17；deepseek-v4-pro 与 qwen3.7-plus-2026-05-26 无阶梯（1.0）。 -- **前缀缓存折扣**：命中缓存的输入 token 按折扣系数消耗额度（glm-5.1 为 0.2，deepseek-v4-pro 为 0.08，qwen3.7-plus 为 0.2），可显著降低多轮对话和重复前缀场景的额度消耗。 -- **自动转按量计费**：超出 PTU 额度或输入超过模型上限时，请求自动转为按量计费，无需修改调用代码，业务不中断。 +## PTU 长输入、前缀缓存与溢出策略 -API 响应关键字段：`service_tier`（值为 `ptu-standard` 表示使用 PTU 额度，`default` 或不返回表示按量计费）、`provisioned_tokens`（折算后实际消耗的额度 token 数）、`cached_tokens`（前缀缓存命中数）。不同 API 格式（OpenAI Chat / Responses、Anthropic、DashScope）下这些字段的 JSON 路径不同，需按对应格式取值。 +[预置吞吐长输入与缓存](../../raw/model-user-guide/model-deployment-1/ptu-long-input-and-cache.md) 描述了 PTU 的额度消耗细则： -> **注意**：模型输入上限存在两处口径。[模型部署](../../raw/model-user-guide/model-deployment-1/model-deployment-introduction.md) 的价格表中千问系列多为 128K、部分新模型达 256K，而 PTU 文档明确将「千问 128K / DeepSeek 64K」作为触发自动转按量计费的上限。请以控制台实际展示与所选具体模型为准。 +- **长输入阶梯系数**：部分模型支持超 32K token 输入（如 glm-5.1 最高 200K），超出 32K 部分按更高系数折算 TPM（glm-5.1：[32K, 200K] 区间输入 1.33 / 输出 1.17）。deepseek-v4-pro、qwen3.7-plus-2026-05-26 无阶梯（1.0）。 +- **前缀缓存折扣**：命中缓存的输入 token 按折扣系数消耗额度（glm-5.1 为 0.2，deepseek-v4-pro 为 0.08），适合多轮对话与重复前缀场景。 +- **溢出策略**（创建 PTU 时选择）：「自动溢出」超额请求转为按量计费，业务不中断，响应头带 `x-dashscope-ptu-overflow:true`；「仅使用 PTU 容量」超额返回 429，不产生额外费用。 +- **API 响应字段**：`service_tier`（`ptu-standard` 表示走 PTU 额度）、`provisioned_tokens`（折算后额度消耗）、`cached_tokens`（缓存命中数）。不同 API 格式（OpenAI Chat / Responses、Anthropic 兼容、DashScope）的 JSON 路径不同；Anthropic 兼容格式暂不返回 `cached_tokens`。 +- 监控中 PTU 利用率超过 100% 属正常现象（长输入阶梯系数导致折算消耗高于原始 token 数）。 -> **注意**：长输入场景下 PTU 利用率可能超过 100%，这是阶梯系数导致折算消耗高于原始 token 数的正常现象，超出部分自动转按量计费，不影响服务可用性。 +> **注意**：关于单次输入超出模型上限的行为，该文档 FAQ 称"千问系列输入上限 128K、DeepSeek 系列 64K"，但同文档额度规则表及 [模型部署](../../raw/model-user-guide/model-deployment-1/model-deployment-introduction.md) 价格表中已出现 256K 输入的模型（qwen3.7-max、deepseek-v4 系列）。各模型实际上限请以控制台展示为准。 -## 模型导入（LoRA） +创建或扩容前建议使用控制台的**预置吞吐额度计算器**，输入 RPM、平均输入/输出长度、缓存命中率，估算建议购买的输入/输出 KTPM。 -通过**我的模型**页面可将本地训练的 LoRA 模型从 OSS 导入百炼平台，详见 [模型导入](../../raw/model-user-guide/model-deployment-1/model-import.md)。当前版本**仅支持 LoRA 模型，不支持全参微调模型**。 +## 从 OSS 导入 LoRA 模型 -导入前提与约束： +按 [模型导入](../../raw/model-user-guide/model-deployment-1/model-import.md)，可将本地训练的 LoRA 模型从阿里云 OSS 导入百炼后再部署。要点： -- **OSS Bucket**：需为目标 Bucket 添加 `bailian-datahub-access` 标签（标签值 `read`）；不支持归档/冷归档类存储；不支持访问 Bucket 根目录文件，需放入子目录。首次导入需先完成 OSS 服务关联角色授权（子账号还需主账号授予 `ram:CreateServiceLinkedRole` 权限）。 -- **必需文件**：`adapter_model.safetensors`（权重）与 `adapter_config.json`（含 rank、alpha 等配置）。 -- **rank 限制**：必须为 8、16、32、64 之一，且同一模型所有 LoRA 层使用相同 rank。 -- **词汇表与对话模板**：不得修改原始 vocab 或 chat_template，必须与开源基础模型默认配置一致，否则无法导入。 -- **VL 模型**：必须冻结 VIT，若 adapter 中包含 `visual` 开头的权重参数则无法导入。 +1. **首次授权**：一键创建服务关联角色 `AliyunServiceRoleForSFMDataHubOSSImport`（服务名 `datahub.sfm.aliyuncs.com`）；子账号需先由主账号授予 `ram:CreateServiceLinkedRole` 权限。目标 Bucket 必须添加标签 `bailian-datahub-access=read`，否则下拉不可选。 +2. **文件要求**：子目录（不支持 Bucket 根目录）中须包含 `adapter_model.safetensors`、`adapter_config.json`、`config.json`。 +3. **训练约束**：rank 必须为 8/16/32/64 且所有 LoRA 层一致；词汇表与 chat_template 不可修改；视觉语言模型必须冻结 VIT（adapter 中不能含 `visual` 开头的参数键）。 +4. **限制**：仅支持 LoRA 模型（全参微调不可导入）；导入来源仅 OSS；导入的模型不支持增量训练；删除仅移除百炼侧记录、不影响 OSS 源文件；OSS 源文件变更后模型状态变为"已失效"，需重新导入。 +5. **推理效果对齐**：百炼推理引擎默认参数与 vLLM/SGLang 可能不同，如需对齐 vLLM 默认值可设置 `temperature=1.0`、`top_p=1.0`、`presence_penalty=0`、`repetition_penalty=1.0` 等。 -支持导入的基础模型涵盖千问3、千问3-VL、千问2.5、千问2.5-VL 系列的指定版本。导入后模型状态包括创建中、创建成功（可部署）、创建失败、已失效。 +常见错误：`AvailableModelFileNotFound`（文件校验未通过）、`10041495`（主账号未开通 OSS）。 -> **注意**：导入模型若与本地 vLLM/SGLang 推理效果不一致，通常是推理引擎参数默认值差异所致。可将 `temperature`、`top_p`、`repetition_penalty` 设为 1.0、`presence_penalty` 设为 0 以对齐 vLLM 默认行为。 +## 通过 API/命令行部署 -## 使用 API/命令行部署 +[使用 API或命令行进行模型部署](../../raw/model-user-guide/model-deployment-1/model-deployment-quick-start.md) 给出了完整调用流程（**仅适用于华北2（北京）地域**，需先配置 `DASHSCOPE_API_KEY`）： -除控制台外，可通过 DashScope HTTP API 完成部署，**仅适用于华北2（北京）地域**，需先获取并配置 API Key，详见 [使用 API或命令行进行模型部署](../../raw/model-user-guide/model-deployment-1/model-deployment-quick-start.md)。核心接口为 `POST/GET/DELETE https://dashscope.aliyuncs.com/api/v1/deployments`，通过 `plan` 字段区分计费方式： +1. **创建部署**：`POST https://dashscope.aliyuncs.com/api/v1/deployments`，核心参数： + - `model_name`：模型代码或调优后的模型 ID； + - `plan`：`ptu`（预置吞吐，配 `ptu_capacity.input_tpm/output_tpm`）、`mu`（模型单元，配 `deploy_spec`、`capacity`、`enable_thinking`、`max_context_length`、`rpm_limit`、`tpm_limit`）或 `lora`（按 [Token 计费](../concepts/token-billing.md)，`capacity` 必填但无效）。 + - 返回的 `deployed_model` 为专属服务唯一 ID。 +2. **查询状态**：`GET /api/v1/deployments/`，`status` 为 `RUNNING` 即部署完成。 +3. **推理调用**：通过 DashScope SDK 调用，注意 API Key 所在[业务空间](../concepts/workspace.md)必须与部署所在[业务空间](../concepts/workspace.md)一致。 +4. **删除服务**：`DELETE /api/v1/deployments/`，删除后立即下线、停止计费、不可恢复。 -- **PTU**：`plan: "ptu"`，配合 `ptu_capacity.input_tpm` / `output_tpm`。 -- **模型单元**：`plan: "mu"`，配合 `deploy_spec`（如 `MU1`）、`capacity`（副本数）、`enable_thinking`、`max_context_length`、`rpm_limit`、`tpm_limit`。 -- **按 Token（LoRA 自定义模型）**：`plan: "lora"`，`capacity` 必填但设置无效，扩缩容需在控制台申请。 +> **注意**：部署成功后即开始计费，即使尚未发起任何调用；执行部署命令前请先确认计费规则。 -典型部署命令（模型单元）： - -```bash -curl "https://dashscope.aliyuncs.com/api/v1/deployments" \ ---header "Authorization: Bearer $DASHSCOPE_API_KEY" \ ---header 'Content-Type: application/json' \ ---data '{ - "name": "my_qwen_plus", - "model_name": "qwen-plus-2025-12-01", - "plan": "mu", - "deploy_spec": "MU1", - "enable_thinking": true, - "capacity": 4, - "max_context_length": 10000, - "rpm_limit": 500, - "tpm_limit": 1000 -}' -``` - -部署流程：创建部署 → 返回 `deployed_model`（专属服务唯一 ID）→ 轮询 `GET /deployments/{id}` 直到 `status` 为 `RUNNING` → 通过 DashScope SDK 或兼容 API 发起推理 → 不再使用时 `DELETE /deployments/{id}` 下线并停止计费。 - -## 部署配置与列表管理 - -在控制台部署时可配置：服务名称、选择模型、模型单元类型（部署规格）、部署副本数、部署模板（如「单机部署」，仅模型单元模式可用）、推理模式（Instruct 非思考 / Thinking 思考）、最长上下文、服务限流（RPM/TPM）。 - -部署列表页展示服务名称、模型名称、**模型 Code**（API 调用时指定模型的唯一标识）、部署状态（待部署、部署中、运行中、部署失败、下线中、已停止、变配中等）、计费方式、部署详情与限流详情。 - -## 限制与注意事项 - -- **计费不可逆变更**：计费方式创建后不可改；预付费按天/按月无法提前退费，首月内提前退订按日单价 1.2 倍计费。 -- **部署即计费**：模型部署成功后即产生费用，即便尚未发起任何调用；后付费欠费后资源保留并继续计费 24 小时，超时后停止计费并删除底层资源（部署任务保留）。 -- **按 Token 模式约束**：仅支持 LoRA 调优后模型，一个月内不使用将自动释放。 -- **权限**：API 部署报错 `Workspace ... does not have deployment privilege` 或 `Workspace access denied` 时，需检查 API Key 归属业务空间的模型部署授权与账号操作权限。 -- **删除不可恢复**：执行 DELETE 后服务立即下线且不可恢复。 +权限报错排查：`Workspace xxx does not have deployment privilege for model xxxx` 需在[业务空间](../concepts/workspace.md)管理中对目标模型授权"模型部署"；`Workspace access denied` 需确认 API Key 归属账号在该业务空间的权限管理用户列表中。 ## 来源文档 @@ -97,4 +76,3 @@ curl "https://dashscope.aliyuncs.com/api/v1/deployments" \ - [使用 API或命令行进行模型部署](../../raw/model-user-guide/model-deployment-1/model-deployment-quick-start.md) - diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/model-evaluation-introduction.md b/skills/bailian-docs-llm-wiki/wiki/guides/model-evaluation-introduction.md index 674be24b..0fb1ed8d 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/model-evaluation-introduction.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/model-evaluation-introduction.md @@ -1,97 +1,80 @@ # model evaluation introduction -模型评测是百炼平台提供的模型能力评估功能，支持自定义评测和基线评测两种方式，通过评测维度对模型推理结果进行打分和对比，帮助开发者选择最优模型或验证调优效果。当前仅支持文本生成类模型评测。 - -## 核心概念 - -模型评测涉及两个容易混淆的核心概念： - -- **评分器 Prompt**：配置于评测维度，指导裁判模型如何给被评测模型的回答打分。 -- **System Prompt**：配置于评测任务，为被评测模型设定角色定位或行为规范，通常可留空。 - -两者作用对象和费用归属不同，详见[模型评测](../../raw/model-user-guide/model-evaluation-introduction/model-evaluation-overview.md)。 +模型评测是百炼平台提供的模型能力评估功能，支持**自定义评测**和**基线评测**两种方式，通过评测维度对模型推理结果打分和对比，生成评测报告与排行榜。可用于模型选型对比、调优效果验证、能力量化评估和持续质量监控，当前仅支持文本生成类模型评测，且仅支持控制台操作（不提供公开 API/SDK）。 ## 评测方式 -### 自定义评测 - -使用自有数据集和自定义评测维度，支持三种评分方式： - -| 评分方式 | 说明 | 适用场景 | -|---------|------|---------| -| 大模型评估（AI 自动评测） | 由裁判模型（推荐千问-Max）对回答评分 | 问答质量、内容安全等语义理解场景 | -| 规则评估（自动化指标） | ROUGE、BLEU、Cosine 等算法直接计算 | 翻译、摘要、Function Calling 等确定性场景 | -| 人工评估（人工标注） | 人工逐条标注 Pass/Fail | 创意性写作、专业领域判断 | - -### 基线评测 - -使用公开标准数据集（C-Eval、MMLU、GSM8K、BBH 等）快速评测模型基础能力，无需自行准备数据集或配置维度。 - -> **注意**：基线评测仅北京地域可用，不支持下载评测结果，不支持推理结果集数据来源。 +| 对比项 | 自定义评测 | 基线评测 | +| --- | --- | --- | +| 数据来源 | 用户数据集或推理结果集 | 公开标准数据集（平台预设） | +| 评测维度 | 自定义创建 | 系统预设（不可配置） | +| 评分方式 | AI 评测 / 规则评估 / 人工标注 | 系统自动评分 | +| 结果下载 | 支持 | 不支持 | +| 可用范围 | 全地域 | 仅北京地域 | -## 评测维度类型 +基线评测的公开数据集分为学科类（C-Eval、MMLU、ARC）、数学类（GSM8K）和推理类（BBH、HellaSwag）。基线评测任务在任务列表中不显示综合得分和维度详情列，也不可进入人工标注页面。详见[模型评测](../../raw/model-user-guide/model-evaluation-introduction/model-evaluation-overview.md)。 -评测维度定义模型的评分规则，创建为模板后可被多个评测任务复用。百炼提供五种评分器类型，详细说明参见[评测维度](../../raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md)。 +## 评测维度（评分器类型） -| 维度类型 | 评分方式 | 需要参考答案 | 裁判模型费用 | -|---------|---------|------------|------------| -| 大模型评估-数值型 | 裁判模型打分（整数，如 0-5） | 否 | 有 | -| 大模型评估-分类型 | 裁判模型标签（Pass/Fail） | 否 | 有 | -| 规则评估-文本相似度 | ROUGE/BLEU/Cosine 等算法 | 是 | 无 | -| 规则评估-字符串匹配 | 相等/不相等/包含 | 是 | 无 | -| 人工评估-分类型 | 人工标注 Pass/Fail | 否 | 无 | +评测维度定义评分规则，创建为模板后可被多个评测任务复用。共 5 种类型： -**选型决策路径**：有标准答案且格式固定用字符串匹配；有标准答案但表述多样用文本相似度；无标准答案需语义理解用大模型评估；需主观判断用人工评估。 +| 维度类型 | 评估方式 | 适用场景 | 需要参考答案 | 裁判模型费用 | +| --- | --- | --- | --- | --- | +| 大模型评估-数值型 | 裁判模型按区间打分（整数，默认 0~5） | 问答质量、内容生成 | 否 | 有 | +| 大模型评估-分类型 | 裁判模型输出 Pass/Fail 标签 | 内容安全、正确性判断 | 否 | 有 | +| 规则评估-文本相似度 | ROUGE-1/2/L、BLEU、Cosine、Fuzzy Match、Accuracy 共 7 种算法 | 翻译、摘要、改写 | 是 | 无 | +| 规则评估-字符串匹配 | 相等 / 不相等 / 包含 | Function Calling、NL2SQL | 是 | 无 | +| 人工评估-分类型 | 人工逐条标注 Pass/Fail | 创意写作、专业审核 | 否 | 无 | -## 关键参数 +类型选择决策路径：有标准答案且格式固定 → 字符串匹配；表述多样 → 文本相似度；需语义理解 → 大模型评估；需主观判断 → 人工评估。完整参数说明见[评测维度](../../raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md)。 -### 评测维度参数 +### 关键参数 -- **评分范围**（数值型）：裁判模型打分区间，整数，默认 0-5。建议不超过 10，范围过大会降低 LLM 评分一致性。 -- **通过阈值**（数值型/相似度型）：判定 Pass/Fail 的分界线，步长 0.1（数值型）或 0.01（相似度型）。 -- **评分器 Prompt**（大模型评估类型）：至少包含一个变量（`${prompt}`、`${output}`、`${completion}`），长度不超过 50000 字符。 -- **Pass/Fail 标签**（分类型/人工评估）：两组标签不可重复。 +- **维度名称** ≤ 20 字符，描述 ≤ 100 字符；**类型创建后不可更改**，选错只能删除重建。 +- **评分器 Prompt**（大模型评估专属）：支持 `${prompt}`（用户问题）、`${output}`（模型回答）、`${completion}`（参考答案）三个变量，须至少包含一个变量，长度 ≤ 50000 字符。切换模板会覆盖当前 Prompt，编辑内容需先备份。 +- **评分范围**（数值型）：整数，最小值 ≥ 0、最大值 ≥ 1，默认 0~5，建议不超过 10（范围过大会降低 LLM 评分一致性）。 +- **通过阈值**：数值型步长 0.1（默认 3.0）；文本相似度型范围 0~1、步长 0.01。评分范围决定"能得几分"，通过阈值决定"几分算及格"，两者独立配置。 +- **Pass/Fail 标签**（分类型）：每个标签 ≤ 20 字符，Pass 与 Fail 标签不可重复，应互斥且穷尽。 +- **裁判模型**：推荐千问-Max（推理能力强）。 -> **注意**：维度类型创建后不可修改，选错需删除重建。 - -### 评测任务参数 +## 使用流程 -- **数据来源**：评测数据集（含 Prompt + Completion，会产生推理费用）或推理结果集（已含 Output，不产生推理费用）。 -- **推理参数**：Temperature、TopP、System Prompt 等，按所选模型动态加载。 -- **排行参与**：开启后须绑定排行榜，结果加入排名对比。 +端到端流程四步：**准备数据集 → 创建评测维度 → 创建评测任务 → 查看结果**。 -## 使用流程 +1. **准备数据**：在数据管理模块上传评测集（EvaluationSet）类型数据（含 Prompt 和 Completion 列）并发布版本。训练集不可用于评测。 +2. **创建评测维度**：控制台**模型评测 > 评测维度** Tab 创建。 +3. **创建评测任务**：**评测任务** Tab 选择被评测模型（支持预置模型和调优后模型）、数据来源、关联维度。任务名称 ≤ 50 字符。数据来源二选一： + - **评测数据集**：系统调用被评测模型推理后评分，产生推理费用； + - **推理结果集**：已含模型 Output，直接评分不调推理，不产生推理费用。 +4. **查看结果**：状态流转为 待执行 → 进行中 → 评测完成/失败/终止。详情页**指标统计** Tab 看综合得分、通过率、分数分布；**数据明细** Tab 看逐条评分。结果可下载（待执行状态和基线评测除外）。 -端到端流程分四步： +数据量建议：小规模验证 50~100 条，正式评测 200~500 条，全面评估 500 条以上。 -1. **准备数据集**：在数据管理模块上传评测集类型数据（含 Prompt 和 Completion 列）。 -2. **创建评测维度**：定义评分标准和方式，选择评分器类型并配置参数。 -3. **创建评测任务**：选择被评测模型、关联数据集和维度，提交评测。 -4. **查看结果**：在指标统计 Tab 查看综合得分和通过率，在数据明细 Tab 查看逐条评分。 +### 排行榜与人工标注 -数据量建议：小规模验证 50-100 条，正式评测 200-500 条，全面评估 500 条以上。 +- **排行榜**：得分范围 0-100（与维度评分范围不同），绑定维度创建后不可修改；建议榜内所有任务使用相同评测数据集以保证公平。 +- **人工标注**：使用人工评估维度的任务须全部数据标注完成后才变为评测完成状态。 -## 计费说明 +## 计费 -费用由两部分构成：被评测模型推理费用 + 裁判模型评分费用。 +费用 = 被评测模型推理费（按 Token）+ 裁判模型评分费（仅大模型评估维度，按 Token）。要点： -- 使用推理结果集可免去推理费用。 -- 规则评估和人工评估无裁判模型费用。 -- 已部署的调优模型评测不额外计费（推理费用包含在部署算力费用中）。 +- 使用推理结果集不产生被评测模型推理费用；已部署的调优模型评测不额外计费。 +- 规则评估和人工评估无裁判模型费用；评测前无法预估总费用，以实际发生为准。 +- 成本优化：先小规模验证 → 保存推理结果集复用 → 有确定性标准的场景优先规则评估。 -**成本优化**：先用 50-100 条小规模验证 → 保存推理结果集复用 → 有确定性标准的场景优先用规则评估。 +详细计费构成见[模型评测](../../raw/model-user-guide/model-evaluation-introduction/model-evaluation-overview.md)的计费说明章节。 -## 限制与注意事项 +## 限制和注意事项 -- 当前仅支持文本生成类模型评测。 -- 基线评测仅北京地域可用。 -- 任务提交后不可更换目标模型，需删除重建。 -- 评测维度类型创建后不可修改。 -- 模型评测当前仅支持控制台操作，不提供公开 API/SDK，如需编程化评测可参考 PAI Judge Model API。 -- 1-3% 的分数差异通常为评测噪声，不建议仅凭微小分差做决策。 -- LLM 评分器存在位置偏差和自我偏好偏差，建议定期人工抽查校准。 +- 仅支持文本生成类模型评测；仅控制台操作，无公开 API/SDK（可参考 PAI Judge Model API 作为替代）。 +- 基线评测仅北京地域可用，其他地域不显示该选项，属预期行为。 +- 不可逆操作：终止任务、删除任务、删除排行榜、删除评测维度；任务提交后不可更换目标模型，已消耗 Token 费用不可追回。 +- 维度被评测任务引用时无法删除，须先移除关联任务；被排行榜绑定的维度删除后会阻止在该排行榜下创建新任务。 +- **评分器 Prompt**（配置于维度，指导裁判模型打分）与 **System Prompt**（配置于任务，为被评测模型设定角色，通常可留空）作用对象和费用归属不同，勿混淆。 +- LLM 评分器存在位置偏差和自我偏好偏差，建议定期人工抽查校准；1-3% 的分差通常为评测噪声，不建议据此做决策；关注分数分布而非仅看综合得分。 -更多操作细节及常见问题排查请参见[模型评测](../../raw/model-user-guide/model-evaluation-introduction/model-evaluation-overview.md)和[评测维度](../../raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md)。 +> **注意**：关于删除维度对已有任务的影响，两篇文档表述不同——[模型评测](../../raw/model-user-guide/model-evaluation-introduction/model-evaluation-overview.md)称"选错类型需删除维度重新创建，已关联该维度的评测任务不受影响"，而[评测维度](../../raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md)称"维度已被评测任务引用时无法删除，需先移除关联的评测任务"。实际操作时以控制台行为为准，建议删除维度前先检查任务和排行榜的引用关系。 ## 来源文档 @@ -99,9 +82,3 @@ - [评测维度](../../raw/model-user-guide/model-evaluation-introduction/evaluation-metrics.md) - - - - - - diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/model-experience.md b/skills/bailian-docs-llm-wiki/wiki/guides/model-experience.md index 5cedd681..45d2197e 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/model-experience.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/model-experience.md @@ -1,83 +1,91 @@ # model experience -百炼平台按模态和场景组织了一整套可直接调用的模型能力：文本生成、视觉理解、图片/视频/3D 生成、语音合成与识别、语音转语音、全模态、音乐生成，以及向量与重排序。本页汇总各场景的推荐模型、关键参数和选型要点，帮助开发者快速定位到合适的模型；模型的实时上下文窗口、计费等详细参数请以模型广场为准。 +本页是百炼平台各模态模型选型的总览，覆盖文本生成、视觉理解、图片/视频生成与编辑、3D 生成、语音合成/识别/转语音、音乐生成、向量与重排序、全模态共 11 类能力。面向开发者提供"按场景选模型"的决策路径、关键参数与限制，并给出从 GPT/Claude/Gemini/Sora/ElevenLabs 等闭源服务迁移到百炼的对位建议。 + +## 能力总览与首选模型 + +| 能力方向 | 首选模型 | 备选/低成本 | +| --- | --- | --- | +| 文本生成 / Agent 开发 | `qwen3.7-plus` | `qwen3.7-flash`、`qwen3.7-max`（最强推理） | +| 视觉理解（图像/视频/OCR） | `qwen3.7-plus` | `qwen3.7-flash`、`qwen3.5-ocr`（文档提取） | +| 视频生成与编辑 | `happyhorse-1.1-t2v` / `happyhorse-1.1-i2v` | `wan2.7-*` 系列（自定义音频、首尾帧） | +| 图片生成与编辑 | `wan2.7-image-pro` | `z-image-turbo`（快 10 倍、约 1/5 价格）、`qwen-image-2.0-pro` | +| 3D 生成 | `Tripo/Tripo-P1.0`（快速验证） | `Tripo/Tripo-H3.1`（高精度，最高 200 万面） | +| 语音合成（TTS） | `qwen-audio-3.0-tts-plus` | `cosyvoice-v3.5-plus`（声音设计）、`MiniMax/speech-2.8-hd` | +| 语音转语音（S2S） | `qwen-audio-3.0-realtime-plus` | `qwen-audio-3.0-realtime-flash`、`qwen3.5-livetranslate-flash-realtime`（同传） | +| 语音识别（ASR） | `fun-asr-realtime` / `fun-asr` | `qwen3.5-omni-plus`（Prompt 上下文、多语种） | +| 音乐生成 | `fun-music-v1` | `fun-music-preview` | +| 向量与重排序 | `text-embedding-v4`、`qwen3-rerank` | `qwen3-vl-embedding`（[多模态](../concepts/multimodal.md)）、`gte-rerank-v2` | +| 全模态 | `qwen3.5-omni-plus` | `qwen3-omni-flash`（低成本、支持思考模式） | ## 文本生成 -通用文本场景（聊天机器人、内容生成、摘要、文档处理、办公任务）推荐从 `qwen3.7-plus` 起步——能力与成本均衡，具备 1M 上下文、Function Calling 和内置工具；效果确认后可切到 `qwen3.6-flash` 降本，功能与上下文一致；需要最强推理时用 `qwen3.7-max`。超长文档（多合同审阅、大规模文献）可用 `qwen-long`（10M 上下文）。AI 编程 / Agent 开发推荐 `qwen3.7-plus`（工具调用完整、1M 上下文适合大代码库）。 +详见[文本生成](../../raw/model-user-guide/model-experience/text-generation-model.md)。 -关键能力： +- **通用场景**（聊天机器人、内容生成、办公文档处理）：`qwen3.7-plus`，能力与成本均衡，1M 上下文（约 70 万汉字），支持思考模式、Function Calling、内置工具和结构化输出。效果确认后可换 `qwen3.7-flash` 降低成本。 +- **AI 编程 / Agent 开发**：推荐 `qwen3.7-plus`；最强推理选 `qwen3.8-max-preview`（仅 Token Plan 可用）或 `qwen3.7-max`。 +- **超长文档**：`qwen-long`，上下文达 10M Token，但不支持思考模式和 Function Calling。 +- **关键参数**：`enable_thinking` 开启思考模式（Responses API 用 `reasoning.effort`），所有 Qwen3 及以上模型支持；结构化输出可获得有效 JSON；批量推理可降低大量低时效请求的成本。 +- 三方模型（`deepseek-v4-pro`、`glm-5.2`、`MiniMax-M3`、`kimi-k2.7-code` 等）均支持 Function Calling，但普遍不支持内置工具。 -- **思考模式**：通过 `enable_thinking` 参数开启（Responses API 用 `reasoning.effort` 控制开关与深度），所有 Qwen3 及以上模型均支持，多为混合模式可按请求切换。 -- **Function Calling**：所有通用模型均支持自定义工具调用；**内置工具**（联网搜索、代码解释器、网页抓取）免复杂配置。 -- **结构化输出**：可强制返回有效 JSON，适合信息抽取。 -- **批量推理**：适合大量、低时延要求的请求以降本。 +> **注意**：`qwen3.7-max` 在文本生成文档的推荐模型表中标注"结构化输出不支持"，选型时如需 JSON 输出请优先使用 `qwen3.7-plus` / `qwen3.7-flash`。 -详细的模型对位（从 GPT / Claude / Gemini 迁移）与完整模型清单见 [文本生成](../../raw/model-user-guide/model-experience/text-generation-model.md)。 +## 视觉理解 -## 视觉理解与 OCR +详见[视觉理解](../../raw/model-user-guide/model-experience/vision-model.md)。 -图像 / 视频理解推荐 `qwen3.7-plus`（1M 上下文、最长 2 小时视频、Function Calling + 内置工具），稳定后可降本到 `qwen3.6-flash`。要点见 [视觉理解](../../raw/model-user-guide/model-experience/vision-model.md)： +- 首选 `qwen3.7-plus`：1M 上下文、单图最高 1600 万像素、视频最长 2 小时 / 2GB、最多 2048 张图片。 +- Token 消耗：每张图片约 `h x w / (32 x 32) + 2` 个 Token。 +- OCR 与文档提取：`qwen3.5-ocr`（文档、表格、试卷、手写内容专用）。 +- 内置工具（联网搜索、代码执行）仅限 `qwen3.7-max-2026-06-08`、`qwen3.7/3.6/3.5` 的 plus 与 flash 版本；结构化输出仅在**非思考模式**下支持。 -- **图像分辨率**：多数模型支持每张最高 1600 万像素，Token 数按 `h x w / (32 x 32) + 2` 计算。 -- **视频支持**：`qwen3.7-plus` / `qwen3.6-plus` / `qwen3.6-flash` / `qwen3.5-plus` / `qwen3.5-flash` 最长 2 小时 / 2GB；`qwen3-vl-plus` / `qwen3-vl-flash` 最长 1 小时。 -- **OCR / 文档提取**：`qwen3.5-ocr` 针对文档、表格、试卷、手写内容优化；通用图片文字提取也可用旗舰模型。 +## 图片与视频生成 -## 图片生成与编辑 +- **文生图/图片编辑**：`wan2.7-image-pro` 集成文字渲染、品牌色控制、角色一致性多图生成与编辑，文生图最高 4096x4096、编辑最高 2048x2048。仅需生成且追求速度/成本选 `z-image-turbo`；需要负向提示词或单次最多 6 张变体选 Qwen Image 系列（`qwen-image-3.0-pro` 邀测中）。 +- **文生视频**：`happyhorse-1.1-t2v`（有声、1080P、3-15 秒）；需传入自定义音频用 `wan2.7-t2v-2026-06-12`。 +- **图生视频**：首帧用 `happyhorse-1.1-i2v`；首尾帧串联长视频用 `wan2.7-i2v-2026-04-25`。 +- **参考生视频**：`happyhorse-1.1-r2v`；需视频参考主体或自定义音色用 `wan2.7-r2v-2026-06-12`。 +- **视频编辑与角色动画**：`happyhorse-1.0-video-edit` / `wan2.7-videoedit`（特效/运镜复刻）；动作迁移 `wan2.2-animate-move`，人物替换 `wan2.2-animate-mix`（均有 wan-std / wan-pro 两种模式）。 -见 [图片生成与编辑](../../raw/model-user-guide/model-experience/image-model.md)。推荐 `wan2.7-image-pro`，集成文字渲染、品牌色控制、角色一致性多图生成和图片编辑（文生图最高 4096x4096，编辑最高 2048x2048，支持最多 9 张输入图参考）。仅需生图且追求速度/成本时用 `z-image-turbo`（约快 10 倍、价格约 1/5，写实人像与产品照）；需要负向提示词或单次最多 6 张变体时用 `qwen-image-2.0-pro`（生成和编辑同一模型 ID）。 +## 3D 与音乐生成（仅北京地域） -## 视频生成与编辑 +- **Tripo 3D**：文生/单图生/多图生 3D，异步任务接口（创建任务 → 轮询 `task_id`，建议间隔 15 秒，24 小时有效）。`input` 中 `prompt` / `image` / `images` 三字段互斥。贴图质量用 `texture_quality`（standard/detailed）控制；几何精度 `geometry_quality`（ultra 最高 200 万面）仅 `Tripo/Tripo-H3.1` 支持。详见 [Tripo 3D模型生成](../../raw/model-user-guide/model-experience/tripo-3d-generation-guide.md)。 +- **Fun-Music**：`prompt`（描述风格）或 `lyrics`（自定义歌词）二选一或组合，`is_instrumental=true` 生成纯音乐（此时 lyrics/gender 被忽略），`gender` 仅 `fun-music-v1` 支持，输出 MP3/WAV。**邀测中**，需在模型广场申请开通。 -见 [视频生成与编辑](../../raw/model-user-guide/model-experience/video-generate-edit-model.md)，按子场景选型： +> **注意**：Tripo 与 Fun-Music 均仅在华北2（北京）地域可用，且必须使用该地域的 API Key。 -- **文生视频**：`happyhorse-1.1-t2v`（1080P、单片段最长 15 秒、有声）；需传入自定义音频文件用 `wan2.7-t2v-2026-06-12`。 -- **图生视频**：首帧生视频用 `happyhorse-1.1-i2v`；首尾帧串联长视频用 `wan2.7-i2v-2026-04-25`。 -- **参考生视频**：`happyhorse-1.1-r2v`（保持角色一致性）；需自定义音色或视频参考主体用 `wan2.7-r2v-2026-06-12`。 -- **视频编辑 / 角色动画**：编辑用 `happyhorse-1.0-video-edit`，特效/运镜复刻用 `wan2.7-videoedit`；动作迁移用 `wan2.2-animate-move`，人物替换用 `wan2.2-animate-mix`（均支持 wan-std / wan-pro 两种模式）。 +## 语音能力 -## Tripo 3D 模型生成 +### 语音合成（TTS） -支持文生 3D、单图生 3D、多图生 3D 三种模式，通过 `input` 中互斥的 `prompt` / `image` / `images` 字段区分。`Tripo/Tripo-H3.1` 面向高精度（最高 200 万面，较慢），`Tripo/Tripo-P1.0` 面向快速预览（最高 2 万面，更快）。贴图质量用 `parameters.texture_quality`（`standard` / `detailed`）控制，几何精度用 `parameters.geometry_quality`（仅 H3.1 支持，`standard` 最高 150 万面 / `ultra` 最高 200 万面）。调用为异步任务，需轮询任务状态（`PENDING` → `RUNNING` → `SUCCEEDED` / `FAILED`），建议间隔 15 秒。 +先决定内置音色（标准合成，`qwen-audio-3.0-tts-plus`、`MiniMax/speech-2.8-hd`）还是自定义音色；自定义再分**声音复刻**（有音频样本，`qwen-audio-3.0-tts-plus/flash`）与**声音设计**（文字描述音色，`cosyvoice-v3.5-plus/flash`），音色统一通过 `voice-enrollment` 服务管理。实时交互用 WebSocket（流式输入输出、延迟最低），内容制作用 HTTP。Qwen-Audio-TTS / CosyVoice / Qwen3-TTS-Instruct 系列支持指令控制（自然语言控制语速、情绪、风格）。 -> **注意**：Tripo 3D 生成仅适用于**华北2（北京）**地域，且必须使用该地域的 API Key。 +### 语音识别（ASR） -## 语音合成（TTS） +按四个维度选型：实时（`fun-asr-realtime`）还是非实时（`fun-asr`，最长 12 小时 / 2GB）；专业术语用热词（Fun-ASR）还是 Prompt 上下文注入（`qwen3.5-omni-plus`）；说话人分离仅 Fun-ASR 非实时模型（`fun-asr`、`fun-asr-mtl`）支持；情感识别用 Qwen-ASR（`qwen3-asr-flash-realtime` / `qwen3-asr-flash-filetrans`）。Paraformer 为较早一代模型，建议迁移。 -见 [语音合成](../../raw/model-user-guide/model-experience/tts-model.md)。先确定内置音色还是自定义音色：标准合成推荐 `qwen-audio-3.0-tts-plus` / `MiniMax/speech-2.8-hd`；自定义音色分**声音复刻**（提供音频样本，用 `qwen-audio-3.0-tts-flash` / `MiniMax/speech-2.8-hd`）和**声音设计**（用文字描述音色，用 `cosyvoice-v3.5-plus` / `cosyvoice-v3.5-flash`），音色统一由 `voice-enrollment` 服务注册管理。接入方式上，WebSocket 双向流式延迟最低（实时交互），HTTP 适合有声阅读等；Qwen 系列以 `-realtime` 后缀区分 WebSocket / HTTP。指令控制可用自然语言动态调节语速、情绪和风格。 +### 语音转语音（S2S） -## 语音识别（ASR） +两条路线：**S2S 单模型**（低延迟、端到端感知语调情绪）与 **Pipeline（ASR + LLM + TTS）**（可自定义音色、逐组件选优）。S2S 路线中：对话选 `qwen-audio-3.0-realtime-plus`（支持语义 VAD 与 Function Calling），同传选 `qwen3.5-livetranslate-flash-realtime`（60 种语言），文件翻译选 `qwen3-livetranslate-flash`。文件（HTTP）模式支持 Function Calling、联网搜索、思考模式等附带能力；实时（WebSocket）模式下 Omni 系列不支持 Function Calling，思考模式下不生成语音。详见[语音转语音](../../raw/model-user-guide/model-experience/s2s-model.md)。 -见 [语音识别](../../raw/model-user-guide/model-experience/asr-model.md)，按维度选型： - -- **实时 vs 非实时**：实时（WebSocket）用 `fun-asr-realtime` 或 `qwen3.5-omni-plus-realtime`；非实时文件转写（HTTP）用 `fun-asr` 或 `qwen3.5-omni-plus`。 -- **专业术语**：Prompt 上下文注入（Qwen3.5-Omni，无需预配置）或热词表（Fun-ASR，适合稳定术语列表）。 -- **说话人分离**：仅 Fun-ASR 非实时模型（`fun-asr`、`fun-asr-mtl`）支持。 -- **情感识别**：Qwen-ASR 与 Qwen3.5-Omni 系列支持，推荐 `qwen3-asr-flash-realtime` / `qwen3-asr-flash-filetrans`。 - -Paraformer 为较早一代模型，新业务建议迁移到 Fun-ASR 或 Qwen-ASR。 - -## 语音转语音（S2S）与全模态 - -构建语音应用可选 **S2S 单模型**（延迟低、端到端感知语调情绪）或 **Pipeline（ASR + LLM + TTS）**（可自定义音色、各阶段独立选优）。S2S 路线推荐：语音助手/客服用 `qwen3.5-omni-plus-realtime`，成本敏感用 `qwen3.5-omni-flash-realtime`，同传/直播翻译用 `qwen3.5-livetranslate-flash-realtime`，视频配音/播客翻译用 `qwen3-livetranslate-flash`。全模态（同时理解文本/音频/图片/视频）三大系列为 Qwen3.5-Omni（旗舰）、Qwen3-Omni-Flash（轻量、支持思考模式）、Qwen3.5-Livetranslate（专业翻译，开箱即用，60 种语言）。 - -> **注意**：Function Calling 与联网搜索能力在不同接入模式下差异较大——例如 `qwen3-omni-flash` 在 HTTP 模式支持 Function Calling 和思考模式，但其 WebSocket（`-realtime`）版本均不支持；联网搜索仅 Qwen3.5-Omni（HTTP / WebSocket）支持，且联网搜索与 Function Calling 不可同时开启；思考模式下不输出语音。选型前务必核对目标模型的具体接入模式。 - -## 音乐生成 +## 向量与重排序 -Fun-Music 是端到端音乐生成模型，通过 `prompt` 描述风格/场景/情绪自动作词谱曲，或通过 `lyrics` 提供自定义歌词，用 `gender` 选男女声（仅 `fun-music-v1`），`is_instrumental=true` 生成纯音乐（此时 `lyrics` / `gender` 被忽略），`format` 指定 mp3 / wav 输出。 +- **文本 Embedding**：`text-embedding-v4`，维度 64~2048（默认 1024）；大规模检索省存储选 256/512，高精度选 1536/2048；已有 v3 索引可继续用 `text-embedding-v3`。 +- **[多模态](../concepts/multimodal.md) Embedding**：图文混合检索（融合向量）用 `qwen3-vl-embedding`；跨模态搜索（独立向量）用 `tongyi-embedding-vision-plus`。纯文本数据仍建议 `text-embedding-v4`。 +- **重排序**：RAG 的 Top-N 精排用 `qwen3-rerank`（100+ 语言，最多 500 文档）；[多模态](../concepts/multimodal.md)排序用 `qwen3-vl-rerank`。 -> **注意**：Fun-Music 处于邀测阶段，需在模型广场申请开通，且服务仅在**华北2（北京）**地域可用。 +## 全模态 -## 向量与重排序 +`qwen3.5-omni-plus`（旗舰）可同时理解文本、音频、图片、视频并输出文本和语音：音频最长 3 小时、视频最长 1 小时，支持 Function Calling、联网搜索（二者不可同时开启）、声音复刻。`qwen3-omni-flash` 成本更低、支持思考模式（仅 HTTP、仅文本输出、单次输入限 150 秒）。翻译场景：快速搭建选 Qwen3.5-Livetranslate（60 种语言、约 3 秒延迟），最高质量选 Qwen3.5-Omni（29 种输出语言），成本敏感选 Qwen3-Omni-Flash（11 种输出语言）。 -见 [向量与重排序](../../raw/model-user-guide/model-experience/embedding-rerank-model.md)。纯文本搜索 / RAG / 聚类推荐 `text-embedding-v4`（维度 64~2048，默认 1024，最大 8192 Token；迁移旧索引可用 `text-embedding-v3`）；跨模态检索用 `qwen3-vl-embedding`（融合向量）或 `tongyi-embedding-vision-plus`（独立向量）。重排序用于 Embedding 检索后对 Top-N 结果精排：纯文本用 `qwen3-rerank`（100+ 语言、最多 500 文档），多模态用 `qwen3-vl-rerank`（文本/图片/视频混排）。 +> **注意**：关于粤语翻译，[语音转语音](../../raw/model-user-guide/model-experience/s2s-model.md)文档标注 Qwen3.5-Livetranslate 对粤语为"仅文本"，而全模态文档同一表格写作"支持 仅文本"，实际应以"仅输出文本、不输出语音"理解。 -## 选型与使用注意事项 +## 通用限制与注意事项 -- 旧版模型（如旧版 Qwen、Paraformer、`qwen-omni-turbo`、`qwen-tts` 等）不再作为首选，新项目建议使用各系列最新版本。 -- 上下文窗口、计费、地域可用性等以模型广场实时信息为准；本页面数据可能滞后。 -- 图片/视频/3D/音乐等生成类模型多为异步任务，需按文档轮询或配置回调。 +- 模型的上下文、计费等详细参数以模型广场（北京/新加坡/美国/法兰克福各地域）实时数据为准。 +- 快照版本（如 `qwen3.7-plus-2026-05-26`）适合版本回归和稳定性需求；主模型 ID 会随升级更新行为。 +- 各文档中的"旧版模型"（Qwen3、Qwen2.5-Omni、Wan2.1、Paraformer、按 [Token 计费](../concepts/token-billing.md)的旧版 Qwen-TTS 等）不再作为首选推荐，新项目应使用文中的推荐系列。 +- `qwen3.8-max-preview` 等部分模型仅对 Token Plan 用户开放。 ## 来源文档 diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/model-high-speed-inference.md b/skills/bailian-docs-llm-wiki/wiki/guides/model-high-speed-inference.md index 712b5a20..f3c05517 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/model-high-speed-inference.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/model-high-speed-inference.md @@ -1,34 +1,40 @@ # model high speed inference -百炼平台针对高吞吐、高速率的推理场景提供两类能力：**TPM 预留**用于为指定模型锁定专属推理容量，避免业务高峰期受公共限流影响；**快速模式（Fast mode）**则针对输出速度敏感的场景提升 TPS。两者都通过替换或指定 `model` 参数接入，无需大改代码。 +百炼平台为高吞吐、低[限流](../concepts/rate-limit.md)影响的推理场景提供两类高速推理能力：**TPM 预留**通过预付费锁定专属推理容量，保障业务高峰期不受公共[限流](../concepts/rate-limit.md)影响；**快速模式（Fast mode）**则以更高的输出 TPS（80~100 TPS，约为标准 API 的 1.5~2 倍）服务对输出速度敏感的场景。两者定位不同：前者解决"容量确定性"，后者解决"输出速度"。 -## TPM 预留：锁定专属容量 +## 能力概览与选型 -TPM（Tokens Per Minute）预留为指定模型锁定专属的推理吞吐量，预留容量内的调用不受公共资源限流影响，容量为业务专属、不与其他用户共享。详见 [TPM 预留](../../raw/model-user-guide/model-high-speed-inference/tpm-reservation.md)。 +| 能力 | 解决的问题 | 计费方式 | 接入改动 | 状态 | +| --- | --- | --- | --- | --- | +| TPM 预留 | 锁定专属吞吐容量，不受公共[限流](../concepts/rate-limit.md)影响 | 按 kTPM 预付费（按天） | 将 `model` 参数替换为专属模型 code | 正式 | +| 快速模式 | 提升输出速度（TPS 1.5~2 倍） | 按 token 计费，与标准 API 一致 | 使用专属域名 + fast 模型 ID | preview | -核心特性： +结合 [TPM 预留](../../raw/model-user-guide/model-high-speed-inference/tpm-reservation.md) 中的方案对比：流量可预估、不能接受限流选 TPM 预留；追求更高吞吐/性能可考虑 PTU 专属部署；流量波动大或短期使用则维持按量付费（可叠加资源包/节省计划做费用优化）。 -- **专属模型 code**：创建预留后系统自动生成专属模型 code，需将 API 请求中的 `model` 参数替换为该 code 才能使用预留容量。 -- **超额不中断**：超出预留容量的请求自动降级为按量计费处理，服务不中断，无需修改代码。可在详情页的**超额降级统计**查看降级次数。 -- **计费单位**：按 kTPM 预付费（1 kTPM = 1,000 Tokens/分钟），一次性支付，从购买成功起连续生效。 +## TPM 预留 -### 方案选型 +### 核心机制 -[TPM 预留](../../raw/model-user-guide/model-high-speed-inference/tpm-reservation.md) 文档给出了多种容量方案的对比，便于按业务诉求选型： +- **专属容量**：预留的 TPM（Tokens Per Minute）容量业务专属，不与其他用户共享，刚性兑付。 +- **专属模型 code**：创建后系统自动生成专属模型 code，API 请求中把 `model` 参数替换为该 code 即可使用预留容量，无需其他代码改动。 +- **溢出策略**（创建时选择）： + - 自动溢出至按量计费（默认）：超出部分降级为标准按量计费，业务不中断，可在详情页"超额降级统计"查看降级次数； + - 仅使用预留容量：超出请求返回 429，不产生额外费用。 -| 方案 | 计费单位 | 容量保障 | 适用场景 | 超额处理 | 代码改动 | -| --- | --- | --- | --- | --- | --- | -| 按量付费 | 按 token | 无（共享公共池） | 流量波动大/短期 | 自动服务，受公共限流 | 无需改动 | -| 资源包/节省计划 | 预付费额度 | 承诺用量折扣（非专属） | 费用优化 | 超出转按量 | 无需改动 | -| TPM 预留 | 按 kTPM 预付费 | 专属容量刚性兑付 | 流量可预估、不能接受限流 | 超出自动降级公共池按量，不中断 | 替换 model 参数 | -| PTU 专属部署 | 按 kTPM 预付费 | 专属部署实例 | 高吞吐高性能 | 超出转按量 | 替换 model 参数 | +### 关键参数（创建时） -### 创建与接入 +| 参数 | 说明 | +| --- | --- | +| 选择模型 | 仅支持已开放 TPM 预留的模型（如千问3.7-Max/Plus、千问3.6-Flash、GLM-5.2/5.1、DeepSeek-v4-Flash/Pro、Kimi-K2.6，以控制台为准） | +| 输入/输出 TPM | 单位 kTPM（1 kTPM = 1,000 Tokens/分钟），起步和步长因模型而异 | +| 付费周期 | 按天 | +| 购买时长 | 1~30、60、90、120、365 天 | +| 到期自动续费 | 默认开启，到期前一天 08:00 自动扣款 | +| 溢出策略 | 自动溢出（默认）/ 仅使用预留容量 | -1. 登录百炼控制台创建 TPM 预留，填写预留名称、选择模型、付费周期（按天）、输入/输出 TPM（单位 kTPM）、购买时长（支持 1~30、60、90、120、365 天）等参数。建议先用 **TPM 容量计算器**（根据 RPM、平均输入/输出长度、缓存命中率估算）确认所需额度。 -2. 确认费用后完成支付。 -3. 在详情页**概览** Tab 复制**专属模型 code**。 -4. 将 API 请求的 `model` 参数替换为该 code 即可： +创建页右侧提供 **TPM 容量计算器**，按 RPM、平均输入/输出长度、缓存命中率估算所需容量；部分模型有长输入阶梯系数（如 glm-5.1 在 [32K, 200K] 区间输入按 1.33 折算）和缓存折扣（如 deepseek-v4-pro 缓存命中按 8% 折算容量），计算器会自动应用，详见 [TPM 预留](../../raw/model-user-guide/model-high-speed-inference/tpm-reservation.md)。 + +### 接入方式 ```python import dashscope @@ -41,53 +47,49 @@ response = dashscope.Generation.call( print(response.output.text) ``` -> **注意**：短时间内请求量快速拉升时，系统需短暂预热以匹配算力，预热期间部分请求可能出现延迟波动，请做好请求排队或重试机制。 - -### 容量换算参数 +> **注意**：短时间内请求量快速拉升时系统需要短暂预热，预热期间部分请求可能出现延迟波动，请做好请求排队或重试机制。 -部分模型支持长输入阶梯系数和缓存折扣，容量计算器会自动应用。例如 glm-5.2 输入长度上限 1M、缓存折扣 0.25、无阶梯；glm-5.1 缓存折扣 0.2，且在 [32K, 200K] 区间输入系数 1.33 / 输出 1.17；deepseek-v4-pro 缓存折扣低至 0.08。Qwen3.6-flash-2026-04-16 不支持缓存。 +### 计费与生命周期要点 -### 管理与生命周期 +- 部署成功即开始计费，预留容量内调用不额外收费；价格按地域和模型不同（如北京地域千问3.7-Max 输入 ¥121.00 / Per 10,000 TPM / 天），**以控制台为准**。 +- 缩容/退订退费公式：`退款 = 降量部分预付费 - (降量部分预付费 × 已用时长/购买时长 × 1.5)`，即已使用部分按 1.5 倍系数结算。 +- 支持扩缩容（不中断服务）；TPM 可归 0 以保留专属模型 code 且不产生容量费用，但归 0 属减配、按上述违约金结算。 +- 到期后 2 小时内仍可调用、可续费；2~14 小时实例停止、可续费恢复；14 小时后删除、不可恢复。退订后专属 code 失效，请求回退公共资源按量计费。 -- **扩缩容**：在详情页调整输入/输出 TPM，变配期间服务不中断。利用率持续接近 100% 或频繁降级时建议扩容。 -- **续费**：可手动续订，或开启**到期自动续费**（到期前一天 08:00 自动扣款）。 -- **退订**：跳转费用中心完成，退订后专属模型 code 失效、请求回退公共资源，不可恢复。缩容/退订退费按已使用部分 1.5 倍系数结算：`退款 = 降量部分预付费 - (降量部分预付费 × 已用时长/购买时长 × 1.5)`。 +## 快速模式（Fast mode） -预留实例状态：服务到期后 2 小时内仍为**运行中**（可调用、可续费）；2~14 小时转为**已停止**（不可调用、仍可续费）；到期 14 小时后**已过期/删除**，不可恢复。其余状态包括待生效、变配中、已取消。 +### 核心特性 -## 快速模式（Fast mode）：提升输出速度 +- **高速输出**：TPS 达 80~100，为标准 API 的 1.5~2 倍，适用于 AI 编程助手、Agent 多步推理、实时对话等场景。 +- **按 token 计费**：与标准 API 一致，按输入/输出 token 计费。 +- **特殊限流**：超出 TPM 额度不立即限流，请求进入排队队列。 -[快速模式](../../raw/model-user-guide/model-high-speed-inference/fast-mode.md) 面向对输出速度敏感的场景（AI 编程助手、Agent 多步推理、实时对话等），当前处于 **preview 阶段**，能力与规格可能随版本调整。 +> **注意**：快速模式当前为 preview 阶段，能力与规格可能随版本调整，详见 [快速模式](../../raw/model-user-guide/model-high-speed-inference/fast-mode.md)。 -关键特性： +### 使用方式 -- **高速输出**：TPS 提升至标准 API 的 1.5~2 倍，达 80~100 TPS。 -- **按 token 计费**：计费逻辑与标准 API 一致，按输入/输出 token 计费。 -- **特殊限流**：超出 TPM 额度不会立即限流，请求进入排队队列。 +将 `model` 指定为 fast 模型 ID（当前支持 `glm-5.2-fast-preview`，北京与新加坡地域）即可开启，无需额外参数。接入域名为专属格式： -### 接入方式 +``` +https://{workspace_id}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1 +``` -将 `model` 参数指定为支持的模型 ID（如 `glm-5.2-fast-preview`）即可开启，无需额外参数。接入域名格式为 `https://{workspace_id}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1`，其中 `{workspace_id}` 可在业务空间管理页面切换到对应地域后查看。 +`{workspace_id}` 在控制台[业务空间](../concepts/workspace.md)管理页面切换到对应地域后查看。 ```bash curl -X POST https://{workspace_id}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions \ -H "Authorization: Bearer $API_KEY" \ -H "Content-Type: application/json" \ - -d '{ - "model": "glm-5.2-fast-preview", - "messages": [{"role": "user", "content": "你是谁"}], - "stream": false -}' + -d '{"model": "glm-5.2-fast-preview", "messages": [{"role": "user", "content": "你是谁"}], "stream": false}' ``` -glm-5.2 默认返回 `reasoning_content` 思考字段；[流式输出](../concepts/streaming-output.md)时思考内容与回答内容分别通过 `delta.reasoning_content` 与 `delta.content` 推送。 +glm-5.2 默认返回 `reasoning_content` 思考字段；[流式输出](../concepts/streaming.md)时思考内容与回答内容分别通过 `delta.reasoning_content` 与 `delta.content` 推送，客户端需分别处理。 ## 限制与注意事项 -- **两者定位不同**：TPM 预留解决"容量保障/不受限流"，快速模式解决"输出速率提升"。快速模式仍为 preview，生产环境需评估稳定性。 -- **接入差异**：TPM 预留通过替换为专属模型 code 接入标准 dashscope 域名；快速模式使用带 `{workspace_id}` 的 maas 域名并指定 fast-preview 模型 ID。 -- **限流行为差异**：TPM 预留超额自动降级按量、不中断；快速模式超额不立即限流而是排队。 -- **计费口径**：具体价格、容量换算与费用以百炼控制台为准，[TPM 预留](../../raw/model-user-guide/model-high-speed-inference/tpm-reservation.md) 与快速模式的价格表可能随时间调整，请以控制台实时展示为准。 +- **TPM 预留**：仅支持控制台展示的已开放模型；创建需一次性预付费，建议先用容量计算器估算；退订不可恢复且专属 code 立即失效；频繁降级或 429 时应扩容。 +- **快速模式**：preview 阶段规格可能变化；当前支持模型有限（glm-5.2-fast-preview）；使用专属 workspace 域名而非标准 dashscope 域名。 +- **两者接入方式不同**：TPM 预留走标准域名 + 专属模型 code；快速模式走专属 workspace 域名 + fast 模型 ID，切勿混用，具体分别参见 [TPM 预留](../../raw/model-user-guide/model-high-speed-inference/tpm-reservation.md) 与 [快速模式](../../raw/model-user-guide/model-high-speed-inference/fast-mode.md)。 ## 来源文档 diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/model-monitoring.md b/skills/bailian-docs-llm-wiki/wiki/guides/model-monitoring.md index 6aa8dc00..7e3acd7b 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/model-monitoring.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/model-monitoring.md @@ -1,75 +1,89 @@ # model monitoring -阿里云百炼提供两套互补的用量与监控能力：**模型用量**（控制台聚合视图，用于查看调用量、Token 消耗和费用）与**模型监控**（面向指标、告警、日志的可观测体系）。前者侧重成本与免费额度管理，后者侧重性能、错误、安全指标的采集、告警与对话审计，二者共同覆盖从成本控制到线上运维的完整链路。 +阿里云百炼提供两套互补的用量与运行观测能力：**模型用量**页面按[业务空间](../concepts/workspace.md)统计各模型的调用量、Token 消耗与费用概览；**模型监控**则聚焦运行时指标（延时、失败率、RPM/TPM 等）、调用日志与主动告警，并支持通过 Prometheus HTTP API 接入 Grafana 或自建应用。两者结合可覆盖成本管理、性能观测与故障排查的完整链路。 -## 支持的模型与功能范围 +## 支持范围 -- **用量查看**：模型列表中的所有模型均支持查看用量，包括基于它们调优后的自定义模型。详见 [模型用量](../../raw/model-user-guide/model-monitoring/model-usage-statistics.md)。 -- **模型监控**： - - **普通监控**支持所有模型（含调优后的自定义模型），延迟通常为小时级。 - - **高级监控**支持北京、新加坡、弗吉尼亚地域下的所有模型，可提供分钟级数据洞察。 - - **告警功能**支持北京、新加坡地域下的所有模型。 -- 监控可查看调用记录、指标监控与告警（Token、延时、调用时长、RPM、TPM、失败率）、以及 Token 消耗统计。详见 [模型监控](../../raw/model-user-guide/model-monitoring/model-telemetry.md)。 +- **用量统计**：模型列表中的所有模型均支持查看用量，包括基于它们调优后的自定义模型，详见[模型用量](../../raw/model-user-guide/model-monitoring/model-usage-statistics.md)。 +- **普通监控**：支持所有模型（含调优后的自定义模型），数据延迟为**小时级**。 +- **高级监控**：支持北京、上海、新加坡、弗吉尼亚地域下的所有模型，提供**分钟级**数据洞察（收费功能）。 +- **告警功能**：支持北京、新加坡、弗吉尼亚地域。 +- **推理日志（请求/响应记录）**：仅适用于北京、新加坡、弗吉尼亚地域的**部分模型**（如 qwen3-max、qwen-plus 系列、qwen-flash、qwen-turbo、qwen3-coder 系列、部分 Qwen3 开源模型及 deepseek-v3.1/v3.2 等）；不支持的模型界面会提示"当前模型暂不支持日志"。 -## 用量与费用查看 +## 查看模型用量与免费额度 -数据按[业务空间](https://help.aliyun.com/zh/model-studio/use-workspace)维度统计，不支持按阿里云账号维度统计。 +在控制台**模型用量**页面按[业务空间](../concepts/workspace.md)维度查看（不支持按阿里云账号维度统计；如需账号级 Token 总量，需在账单详情页导出账单查看）。关键规则： -- **数据延迟约 1 小时**；不支持查看 30 天以前的统计数据，更早数据需前往「费用与成本」页面查询。 -- **时间精度**：支持分钟 / 小时 / 天三种精度。时间跨度超过 1 天时分钟精度不可选，超过 7 天时仅支持按天查看。 -- **筛选维度**：仅「大语言模型」页签支持按推理类型（实时推理 / 批量推理）筛选；支持按 API-KEY、模型名称（如 `qwen-plus`）筛选。 -- **费用概览**：可查看当前账期总消费、订阅费用、账单趋势（按月/按天，可按产品分类、API Key ID、模型筛选），并可设置**费用告警**。 +- 数据延迟约 **1 小时**；不支持查看 **30 天以前**的数据，更早数据到费用与成本页面查询。 +- 时间精度支持分钟/小时/天：跨度超 1 天不可选分钟，超 7 天仅支持按天。 +- 仅「大语言模型」页签支持按推理类型（实时推理/批量推理）筛选；支持按 API-KEY 筛选。 +- 统计单位随模型类型不同：文本/深度思考/视觉理解/全模态/向量模型按 **Token**，图像生成按**张**，视频生成按**秒**，语音模型按秒、字符或 Token（视模型而定）。 -不同模型的用量统计口径不同：大语言模型 / 全模态 / 向量模型按 **Token**，图像生成按**张**，视频生成按**秒**，语音模型按**秒、字符或 Token**（视模型而定）。完整口径见 [模型用量](../../raw/model-user-guide/model-monitoring/model-usage-statistics.md)。 +免费额度页面可查看各模型免费额度剩余量与过期时间，并可单个或批量切换**免费额度用完即停**开关（开启后额度用尽时返回 403 错误 `AllocationQuota.FreeTierOnly`）。 -## 免费额度管理 +> **注意**：「免费额度用完即停」只能在仍有未消耗免费额度时开启；开启后若要关闭，需等免费额度完全消耗后才能操作。 -「免费额度」页面提供使用概览（按模型总数、额度充沛、使用超 50%/80%、无免费额度等维度汇总）及「即将用尽 Top 3」列表。 +## 监控模型运行 -- **免费额度用完即停**：开启后免费额度用尽时服务自动停止（返回 `403 AllocationQuota.FreeTierOnly`），避免产生额度外费用。 -- 支持批量开启/关闭、一键开启/关闭所有模型；账号未绑定有效支付方式时批量操作会失败。 +系统自动采集主账号下所有[业务空间](../concepts/workspace.md)的模型调用数据，按"模型 + 业务空间"维度生成[模型监控](../../raw/model-user-guide/model-monitoring/model-telemetry.md)列表。默认业务空间成员可查看所有业务空间数据；子业务空间成员仅能查看当前空间。 -> **注意**：「免费额度用完即停」只能在账户仍有未消耗免费额度时开启；一旦开启，需在免费额度完全消耗后才能关闭。控制台免费额度数据为分钟级更新，账单记录按分钟汇总，请以控制台显示数值为准。 +点击操作列**监控**可查看 4 类指标： -## 监控指标与告警 +| 类别 | 示例指标 | +| --- | --- | +| 安全 | 内容安全错误次数（被内容安全服务拦截） | +| 成本 | 平均单次请求调用量、Token 消耗 | +| 性能 | 调用时长、首 Token 延时、非首 Token 延时、RPM、TPM | +| 错误 | 失败次数、失败率、[限流](../concepts/rate-limit.md)错误次数（429） | -在模型监控列表中点击目标模型操作列的**监控**，可查询 4 类指标： +调用统计支持按 API-KEY、推理类型、时间范围与时间精度（分钟/小时）筛选，失败次数图表可点击**失败详情**定位失败原因。 -- **安全**：如 `内容安全错误次数`（输入/输出被内容安全服务拦截）。 -- **成本**：如 `平均单次请求调用量`。 -- **性能**：`调用时长`、`首 Token 延时`、RPM、TPM、非首 Token 延时等。 -- **错误**：`失败次数`、`失败率`，其中**限流错误次数**指因 [429 状态码](https://help.aliyun.com/zh/model-studio/error-code)导致的失败。 +## 查看 Token 消耗与历史对话（日志） -**创建告警**（仅限新加坡、华北2（北京）地域）：需先开启高级监控（性能和用量指标监控），再在模型告警页面创建规则。 +- **历史 Token 消耗**：监控详情页调用量区域可查最近 30 天；更早数据到费用与成本页面查询。 +- **单次调用 Token 消耗 / 请求与响应内容**：需先在**模型监控配置**中开通审计日志和推理日志，随后在**日志**页签查看每次实时推理调用的 Request ID、调用时长、状态码、错误码、用量、请求和响应。 +- 推理日志从调用发生到可查询存在**分钟级**延迟；普通监控用量汇总为**小时级**延迟，高峰期可能达 1-2 小时。 +- **仅记录开启推理日志后**的调用，开通前的历史调用无法补录追溯。 -- **通知方式**：短信、电子邮件、电话、钉钉群机器人、企业微信机器人、Webhook。 -- **告警等级**：紧急（电话/短信/邮件）、错误（短信/邮件）、警告（短信/邮件）、普通（邮件），不支持自定义。 +## 建立主动告警 -## Token 消耗与历史对话 +针对静默失败（超时、Token 消耗突增等）可设置主动告警，步骤： -- **历史 Token 消耗**：最近 30 天可在监控页调用统计的「调用量」区域查看；更早数据前往「费用与成本」页面。 -- **单次调用 Token 消耗 / 历史对话（模型日志）**：需在「模型监控配置」中依次开通审计日志和推理日志，之后在**日志**页签查看每次调用的输入、输出与用量。开通后从调用到记录存在分钟级延迟。 +1. 在模型监控配置中开启**高级监控**（性能和用量指标监控）。 +2. 在模型告警页面创建告警规则，选择模型和监控模板。 -> **注意**：查看某次调用的 Token 消耗及历史对话（模型日志）功能**目前仅适用于华北2（北京）地域的部分模型**，且仅覆盖特定模型/快照版本（如 qwen3-max、qwen-plus、qwen3-coder 系列、部分开源与三方模型）。详见 [模型监控](../../raw/model-user-guide/model-monitoring/model-telemetry.md)。 +通知方式支持短信、邮件、电话、钉钉群机器人、企业微信机器人及 Webhook。告警等级固定为四级（不支持自定义）：紧急（电话/短信/邮件）、错误（短信/邮件）、警告（短信/邮件）、普通（邮件）。 ## 接入 Grafana 与自建应用 -高级监控的指标数据存储在私有 Prometheus 实例中，支持标准 Prometheus HTTP API，可接入 Grafana 或自建应用做可视化分析。 +高级监控数据存储在私有 Prometheus 实例中，支持标准 Prometheus HTTP API。使用方式： -1. 确保已开启高级监控，在模型监控配置中查看 Prometheus 实例详情，按网络环境（公网/VPC）复制 HTTP API 地址。 -2. 通过 `GET {HTTP API}/api/v1/query_range?query=<指标名>&start=...&end=...&step=60s` 查询，`Authorization` 需用 `Basic base64Encode(AccessKey:AccessKeySecret)`。 +1. 开启高级监控后，在模型监控配置中获取 Prometheus 实例的 HTTP API 地址（区分公网/VPC）。 +2. 以 `Authorization: Basic base64(AccessKey:AccessKeySecret)` 鉴权调用 `query_range` 等接口，AccessKey 必须与 Prometheus 实例归属同一阿里云账号。 -常用指标名包括 `model_call_count`（调用次数）、`model_call_duration`（调用时长均值）、`model_usage`（用量总和）等；可用 `{workspace_id="...",model="qwen-plus"}` 形式追加过滤条件（支持 `user_id`、`apikey_id`、`workspace_id`、`model`、`protocol`、`status_code`、`usage_type` 等 LabelKey）。 +``` +GET {HTTP API}/api/v1/query_range?query=model_usage{workspace_id="llm-xxxx",model="qwen-plus"}&start=2025-11-20T00:00:00Z&end=2025-11-20T23:59:59Z&step=60s +``` -> **注意**：`status_code`、`error_code` 仅 `model_call_count` 支持；`usage_type` 仅 `model_usage` 支持。AccessKey 必须与 Prometheus 实例归属同一阿里云账号。 +可用指标包括 `model_call_count`、`model_call_duration`（含 p50/p99）、`model_first_token_duration`、`model_generation_duration_per_token`、`model_usage`、`model_tps_per_request`（仅高级监控）等；过滤条件（LabelKey）支持 `user_id`、`apikey_id`（-1 表示控制台调用）、`workspace_id`、`model`、`protocol`（HTTP/SSE/WS）、`sub_protocol`（DEFAULT/ASYNC）。 -## 生产环境实践建议 +> **注意**：TPS（`model_tps_per_request`）与非首包时长呈倒数关系（TPS ≈ 1 ÷ 非首包时长均值），且 TPS [限流](../concepts/rate-limit.md)按请求维度，区别于 TPM 的按账号维度[限流](../concepts/rate-limit.md)；排查响应慢时应结合首 Token 延时、非首 Token 延时与输入 Token 量综合分析。 -- **控制输出长度**：合理设置 `max_tokens` 与限制思考长度以控制费用。 -- **按任务选模型**：分类、摘要等简单任务优先用轻量级模型。 -- **监控与告警**：通过模型监控跟踪用量趋势并配置告警。 -- **优化 Prompt**：简洁清晰的 Prompt 可减少输入 Token 消耗。 -- **使用批量推理**：非实时大批量任务用批量推理更具成本优势。 +## 成本控制建议 + +按[模型用量](../../raw/model-user-guide/model-monitoring/model-usage-statistics.md)的生产环境建议： + +- 合理设置 `max_tokens` 并限制思考长度，控制单次输出成本。 +- 简单任务（分类、摘要）优先选轻量级模型。 +- 通过[模型监控](../../raw/model-user-guide/model-monitoring/model-telemetry.md)配置用量告警，及时发现异常消耗。 +- 优化 Prompt 减少输入 Token；非实时大批量任务改用批量推理。 + +## 限制和注意事项 + +- 用量数据按业务空间统计，账号级汇总需导出账单。 +- 用量统计数据延迟约 1 小时，且仅保留 30 天。 +- 高级监控为收费功能，且有地域限制；推理日志仅覆盖部分模型。 +- 告警等级与通知渠道映射固定，不支持自定义新增或修改。 ## 来源文档 diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/plug-in.md b/skills/bailian-docs-llm-wiki/wiki/guides/plug-in.md index 5a84d01d..28fe7f97 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/plug-in.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/plug-in.md @@ -56,7 +56,7 @@ Python 代码解释器可用依赖包括：matplotlib、pandas、scipy、seaborn ### 调用插件 - **方式一（插件页面）**：在插件页面将工具添加至[智能体应用](../concepts/agent-application.md)。官方插件只能与位于相同[业务空间](../concepts/workspace.md)里的[智能体应用](../concepts/agent-application.md)关联。每个[智能体应用](../concepts/agent-application.md)最多支持添加 10 个工具，应用会根据输入选择调用一个或多个工具。 -- **方式二（应用管理页面）**：在指定智能体或工作流应用内添加插件，测试效果并发布应用。 +- **方式二（应用管理页面）**：在指定智能体或[工作流](../concepts/workflow.md)应用内添加插件，测试效果并发布应用。 - **方式三（Assistant API）**：通过 Assistant API 调用工具，需正确传递工具 ID。 子[业务空间](../concepts/workspace.md)调用官方插件前，需先在插件详情页为子[业务空间](../concepts/workspace.md)授权；默认[业务空间](../concepts/workspace.md)无需此步骤。 @@ -69,7 +69,7 @@ Python 代码解释器可用依赖包括：matplotlib、pandas、scipy、seaborn 当官方和三方插件无法满足业务需求时，可创建自定义插件。详见[自定义插件](../../raw/application-user-guide/plug-in/custom-plug-ins.md)。 -### 工作流程 +### [工作流](../concepts/workflow.md)程 1. **创建/导入插件**：定义插件基础信息，或直接从云市场导入。 2. **添加工具**（导入插件无需此步）：配置 API 路径、请求参数和返回数据。 @@ -149,6 +149,19 @@ Python 代码解释器可用依赖包括：matplotlib、pandas、scipy、seaborn + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/prompt.md b/skills/bailian-docs-llm-wiki/wiki/guides/prompt.md index 1a567cf3..7805b53d 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/prompt.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/prompt.md @@ -36,17 +36,17 @@ Prompt 模板将提示词的固定结构与动态变量分离，实现可复用 - **指令增强**：将模糊指令具体化、步骤化 - **安全与边界注入**：增加输出格式、内容限制等边界条件 -操作路径：**应用开发 > 组件管理 > 提示词 > 自动优化**。优化结果可直接复制使用或保存为模板。该功能**不计费**，且提交的数据不会被存储或用于模型训练。详见 [Prompt自动优化](../../raw/application-user-guide/prompt/optimize-prompt.md)。 +操作路径：**应用开发 > 组件管理 > 提示词 > 自动优化**。优化结果可直接复制使用或保存为模板。该功能**不[计费](../concepts/billing.md)**，且提交的数据不会被存储或用于模型训练。详见 [Prompt自动优化](../../raw/application-user-guide/prompt/optimize-prompt.md)。 ## Prompt 反馈优化 -相比普通自动优化，Prompt 反馈优化基于用户提供的**输入输出样例**进行多轮自动化评估和迭代，生成更贴合实际业务场景的 Prompt。其工作流程为： +相比普通自动优化，Prompt 反馈优化基于用户提供的**输入输出样例**进行多轮自动化评估和迭代，生成更贴合实际业务场景的 Prompt。其[工作流](../concepts/workflow.md)程为： 1. 选择推理模型（推荐千问-max） 2. 输入初始 Prompt（描述任务目标） 3. 上传样例数据（建议 5-10 条，每种场景至少 1 条） -4. 上传评测数据（建议至少 20 条，数据越多效果越好） -5. 系统自动进行多轮评测与优化 +4. 上传[评测](../concepts/evaluation.md)数据（建议至少 20 条，数据越多效果越好） +5. 系统自动进行多轮[评测](../concepts/evaluation.md)与优化 优化后的 Prompt 包含三部分：原始 Prompt、添加的样例（few-shot）、以及自动生成的内容提示（对分类边界等的补充说明）。优化结果可保存为模板或直接创建[智能体应用](../concepts/agent-application.md)。详见 [基于大模型输入输出样例的Prompt自动优化](../../raw/application-user-guide/prompt/prompt-feedback-optimization.md)。 @@ -66,7 +66,7 @@ Prompt 样例库采用少样本学习（Few-shot learning）思路，从预定 | 批量导入文件大小 | 20MB（Excel） | | 单次导入条数 | 100 条 | -### 计费说明 +### [计费](../concepts/billing.md)说明 样例库功能本身不收费，但启用后会增加大模型调用的 [Token](../concepts/token.md) 消耗。总输入 [Token](../concepts/token.md) 约等于：用户查询 [Token](../concepts/token.md) + 所有召回样例的总 [Token](../concepts/token.md) + 系统指令 [Token](../concepts/token.md)。 @@ -106,6 +106,19 @@ Prompt 样例库采用少样本学习（Few-shot learning）思路，从预定 + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/realtime-api-user-guide.md b/skills/bailian-docs-llm-wiki/wiki/guides/realtime-api-user-guide.md new file mode 100644 index 00000000..9abeaee9 --- /dev/null +++ b/skills/bailian-docs-llm-wiki/wiki/guides/realtime-api-user-guide.md @@ -0,0 +1,185 @@ +# realtime api user guide + +百炼 Realtime API 是针对实时[多模态](../concepts/multimodal.md)交互场景提供的一套接入方案，支持 WebSocket、WebRTC 和 AOQ（AI over QUIC）三种传输协议，开发者可根据业务场景在服务端集成、浏览器端和移动端原生应用之间灵活选择。核心能力包括实时语音/视频对话、流式 ASR/TTS、[多模态](../concepts/multimodal.md)输入，以及内置回声消除、降噪和弱网对抗。 + +## 协议选择 + +参见 [Realtime API简介](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-overview.md) 的完整对比表。简要说明： + +| 协议 | 适用场景 | 端侧平台 | 弱网对抗 | 回声消除/降噪 | +|------|----------|----------|----------|--------------| +| WebSocket | 服务端集成、快速原型 | 全平台 | 差 | 无，需自行处理 | +| WebRTC | 浏览器端、已有 RTC 基础设施 | 浏览器、移动端 | 良好 | 内置 | +| AOQ | 移动端原生、弱网、[多模态](../concepts/multimodal.md) | Android / iOS / HarmonyOS | 极致 | 内置 | + +## 支持的模型与应用 + +| 模型/应用类型 | 模型 | AOQ | WebRTC | WebSocket | +|-------------|------|-----|--------|-----------| +| 实时全模态 | qwen3.5-omni-plus-realtime | ✓ | ✓ | ✓ | +| 实时全模态 | qwen3.5-omni-flash-realtime | ✓ | ✓ | ✓ | +| 实时全模态 | qwen3.5-livetranslate-flash-realtime | ✓ | ✓ | ✓ | +| [多模态](../concepts/multimodal.md)开发套件 | multimodal-dialog | ✗ | ✓ | ✓ | +| 实时语音识别 | FunASR 系列 | ✗ | ✗ | ✓ | +| 实时语音合成 | CosyVoice 系列 | ✗ | ✗ | ✓ | +| 实时语音对话 | qwen-audio-3.0-realtime-plus/flash | ✗ | ✗ | ✓ | + +## 鉴权机制 + +三种协议均使用 [API Key](../concepts/api-key.md)，通过 HTTP 请求头 `Authorization: Bearer ` 完成建连鉴权，连接建立后无需重复鉴权。 + +**AOQ 特殊机制**：[API Key](../concepts/api-key.md) 仅在业务 AppServer 侧使用，客户端从 AppServer 换取临时 [Token](../concepts/token.md)。AppServer 向百炼网关发起如下请求： + +```bash +curl -X POST \ + "https://{endpoint}/api/v1/webrtc/realtime?model=qwen3.5-omni-plus-realtime" \ + -H "Content-Type: application/json" \ + -H "Authorization: Bearer ${DASHSCOPE_API_KEY}" \ + -H "x-dashscope-rtc-transport: moq" \ + -d '{"clientIp": "<客户端真实公网IP>"}' +``` + +响应包含 `aoqTokenForClient`、`sid`、`clientRelayEndpoints`、`clientRelayCertFingerprint` 等字段，客户端使用这些凭证调用 SDK `connect`。详见 [Token鉴权](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-token-authentication.md)。 + +> **注意**：[API Key](../concepts/api-key.md) 不要硬编码到客户端代码或代码仓库。AOQ 方案本身的设计目标就是避免 [API Key](../concepts/api-key.md) 下发到端侧，务必通过 AppServer 代理换取 [Token](../concepts/token.md)。 + +## SDK 下载与集成 + +WebSocket 协议使用 [DashScope SDK](../concepts/dashscope-sdk.md)，参见官方安装文档。 + +AOQ Client SDK 当前最新版本 v1.0.1，支持 Android（`.aar` + `libPluginOpus.so`）、iOS（`.framework` + `PluginOpus.framework`）和 HarmonyOS（`.har` + `libPluginOpus.so`）。如需 Opus 编解码，须同时下载音频插件。详见 [SDK下载](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-sdk-download.md)。 + +Android 集成要点： +- `minSdk 21`，ABI 过滤 `armeabi-v7a` 和 `arm64-v8a` +- Manifest 需声明 `INTERNET`、`RECORD_AUDIO`、`CAMERA` 等权限，其中 `RECORD_AUDIO` 和 `CAMERA` 为运行时权限 + +iOS 集成要点： +- framework 选 **Embed & Sign** +- Info.plist 声明 `NSMicrophoneUsageDescription` 和 `NSCameraUsageDescription` + +## 接入流程 + +### AOQ 接入 + +参见 [实现接通模型/应用](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-connect-model.md) 的 AOQ 章节。核心步骤： + +1. 调用 `AoqClientEngine.createEngine` 创建引擎，注册 `onConnectionStatusChange` / `onDataMsg` / `onError` 回调 +2. 调用 `startAudioCapture` / `startAudioPlayer`（可选 `startVideoCapture`）启动媒体设备 +3. AppServer 向百炼换取 [Token](../concepts/token.md)，下发给客户端 +4. **建连前**调用 `enableSendMediaStream(.audio, enable: false)` 关闭媒体发送 +5. 调用 `connect(config)` 建立连接 +6. 收到 `session.updated` 后调用 `enableSendMediaStream(.audio, enable: true)` 开启媒体发送 +7. 通话结束后调用 `disconnect()` → `AoqClientEngine.destroy()` + +> **注意**：AOQ SDK 建连成功后默认立即开始发送媒体数据。若未在 `connect` 前关闭媒体发送，模型侧可能在会话配置完成前收到数据。务必遵循"先禁用、收到 `session.updated` 后再开启"的模式。 + +### WebRTC 接入（浏览器） + +1. `new RTCPeerConnection({ iceServers: [] })` +2. `getUserMedia` 获取音频（必须）和视频（可选）媒体流 +3. 添加媒体轨道；建连前将 sender track 替换为 `null` 实现媒体门控 +4. 创建名为 `oai-events` 的 DataChannel +5. `createOffer` → `setLocalDescription`，等待 `iceGatheringState === "complete"` +6. HTTP POST Offer SDP 到 `https://{endpoint}/api/v1/webrtc/realtime?model=<模型名>`，Header 携带 [API Key](../concepts/api-key.md) +7. 将返回的 Answer SDP 设置为 `setRemoteDescription`（注意行尾规范化为 `\r\n`） +8. 收到 `session.created` 后恢复媒体发送 + +> **注意**：浏览器端受 CORS 限制，无法直接向百炼服务端发起 SDP 交换请求。Demo 中通过 curl 命令手动执行；正式产品应由业务 AppServer 代理此请求。WebRTC 仅支持 VAD 模式（`server_vad` 或 `semantic_vad`），不支持手动模式。 + +### [多模态](../concepts/multimodal.md)开发套件（multimodal-dialog）接入 + +WebRTC 接入端点格式为： +``` +POST https://{workspace_id}.{region}.maas.aliyuncs.com/api/v1/webrtc/inference?model=multimodal-dialog +``` +`workspace_id` 和 `region` 从百炼控制台获取。DataChannel 就绪后向服务端发送 `run-task` 消息启动会话。 + +## 会话配置（session.update） + +连接建立后通过 DataChannel 或 AOQ data track 发送 `session.update` 事件配置会话： + +```json +{ + "type": "session.update", + "session": { + "modalities": ["text", "audio"], + "voice": "Ethan", + "input_audio_format": "pcm", + "output_audio_format": "pcm", + "instructions": "系统提示词", + "turn_detection": { + "type": "semantic_vad", + "threshold": 0.5, + "silence_duration_ms": 800 + } + } +} +``` + +- `input_audio_format` / `output_audio_format`：当前仅支持 `pcm`；输入 16 kHz，输出 24 kHz +- `turn_detection.type`：`server_vad` 或 `semantic_vad`；使用 qwen3.5-omni-realtime 系列模型推荐 `semantic_vad`；设为 `null` 表示手动模式（WebRTC 不支持） + +## AOQ SDK 核心 API + +### 媒体流发送控制 + +`enableSendMediaStream(trackType, enable)` 控制音频/视频发送开关，支持独立控制。必须在 `createEngine` 之后调用；未调用时建连成功后 SDK 立即开始发送。 + +### 音频功能 + +AOQ SDK 提供完整音频能力，包含： +- **采集**：`startAudioCapture(config)`，支持内部采集（默认）和外部采集（`isExternal=true`） +- **播放**：`startAudioPlayer(config)`，支持暂停/恢复（带淡入淡出 `fadeMs` 参数）和打断当前轮 +- **编解码**：`setAudioEncoderConfig` / `setAudioDecoderConfig`，支持 PCM（AudioPCM）和 Opus（AudioOpus）两种格式，PCM 支持 8K/16K/32K/48K 采样率 +- **扬声器管理**：`enableSpeakerphone(enable)`，需在 VoIP 模式下才可切换 +- **自定义音频播放**：设置 `isExternal=true` 启动播放器后，通过 `setAudioFrameObserver` + `enableAudioFrameObserver` 接收 PCM 帧回调，自行渲染 +- **自定义音频采集**：通过 `addAudioExternalStream` + `pushAudioExternalStreamData` 向 SDK 注入 PCM 数据；硬件采集建议 10ms 一帧，文件解析建议 40ms 一帧；缓冲区满时返回错误码 110，需等待重试 +- **音频文件混音**：`startAudioFile(fileId, config)` 将本地音频文件混入推流 + +### 视频功能 + +- **采集**：`startVideoCapture(config)`，支持内部摄像头采集和外部帧输入（`isExternal=true`） +- **渲染**：`setLocalView` / `setRemoteView`，渲染模式支持 Auto / Stretch / Fill / Crop +- **编码**：`setVideoEncoderConfig`，编码格式 H.264（默认）或 JPEG（外部编码帧模式） +- **自定义视频输入**： + - 原始帧模式：`pushExternalVideoCapturedFrame`，支持 BGRA、I420、NV12/NV21、RGBA；iOS 支持 CVPixelBuffer 零拷贝 + - 编码帧模式：`pushExternalVideoEncodedFrame`，目前支持 JPEG 格式，跳过 SDK 内部编码器 + - 两种模式不可混用 + +### 连接状态管理 + +SDK 连接状态为四态：`Disconnected(0)` → `Connecting(1)` → `Connected(2)` / `Failed(3)`。`Failed` 是瞬态，SDK 会自动迁移到 `Disconnected`，业务层无需手动调用 `disconnect`。`disconnect` 后引擎不会自动释放，可重新调用 `connect` 重连。 + +## 注意事项与限制 + +- AOQ SDK 不支持浏览器环境，仅支持 Android / iOS / HarmonyOS 原生应用 +- WebRTC 协议不提供官方 SDK，Web 端通过原生 WebRTC API 接入，其他端通过开源项目或第三方 RTC 服务商接入 +- 音视频轨道数据（AOQ）通过媒体轨道直接传输，无需发送 `input_audio_buffer.append` 或 `input_image_buffer.append` 事件 +- 模型并发[限流](../concepts/rate-limit.md)条件参见百炼控制台的[限流](../concepts/rate-limit.md)文档；模型名称、快照版本、价格等信息以控制台展示为准 +- AOQ SDK 异常处理优先内部恢复，仅在物理限制（网络、设备、资源）或外部因素（token 无效）无法恢复时才回调错误 +- `onPlaybackAudioFrame` 回调在 SDK 内部线程触发，回调中的数据指针仅在回调期间有效，异步使用需自行拷贝 + +## 来源文档 + +- [Realtime API简介](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-overview.md) +- [SDK下载](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-sdk-download.md) +- [Token鉴权](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-token-authentication.md) +- [实现接通模型/应用](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-quick-start-guide/realtime-connect-model.md) +- [通过WebRTC使用qwen3.5-omni-plus-realtime实现实时通话](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-omni-realtime.md) +- [通过WebRTC使用多模态交互套件实现实时通话](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-webrtc-multimodal-dialog.md) +- [通过AOQ使用qwen3.5-omni-plus-realtime实现实时通话](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-best-practices/best-practice-aoq-omni-realtime.md) +- [AOQ SDK简介](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-desc.md) +- [媒体流发送管理](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-media-stream-control.md) +- [音频常用功能介绍](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-audio-features.md) +- [自定义音频播放](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-playback.md) +- [自定义音频采集](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-audio-capture.md) +- [连接状态管理](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-connection-management.md) +- [视频常用功能介绍](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-video-features.md) +- [自定义视频输入](../../raw/model-user-guide/realtime-api-user-guide/realtime-api-aoq-api/realtime-api-aoq-sdk-function/aoq-custom-video-input.md) + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/release-notes.md b/skills/bailian-docs-llm-wiki/wiki/guides/release-notes.md index f686e492..0ac4fcfb 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/release-notes.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/release-notes.md @@ -1,41 +1,61 @@ # release notes -阿里云百炼平台的 release notes 由两条互补的时间线组成：一是**平台功能更新**（计费、部署、调优、知识库 RAG、SDK/接入工具、API 等能力的演进），二是**模型上下架与更新**（各模型的上架时间、服务部署范围、模型规格与能力说明）。开发者可据此追踪能力可用性、模型可调用状态与计费/下线变更，避免因模型下线或网关变更导致线上调用中断。 +百炼平台的 release notes 由两条主线组成：一是**平台功能更新**（计费调整、API 变更、新模块上线、下线公告等），二是**模型上下架与更新**（新模型上架、快照版本发布、价格调整、模型下线）。本页汇总两类动态的查阅入口、近期重要变更和开发者需要关注的迁移事项。 -## 两类更新的定位 +## 平台功能更新 -release notes 覆盖两个正交的维度，查询时应按需求选择对应文档： +平台功能动态按年月组织，每条记录包含日期、功能模块、功能点和说明。完整时间线见[模型平台功能更新](../../raw/model-user-guide/release-notes/model-release-notes.md)。页面顶部还维护一组**公告通知**，涉及降价、免费额度调整、资源包活动和模型下线机制等，计费相关问题建议优先查阅该区域。 -- **平台功能动态**：记录平台侧能力的上线与调整，按 `年 → 月 → 日` 组织，每条含"功能模块 / 功能点 / 功能说明"。详见 [模型平台功能更新](../../raw/model-user-guide/release-notes/model-release-notes.md)。 -- **模型上架清单**：按地域（如华北2（北京））列出每个模型的上架时间、服务部署范围、模型规格（如 `qwen3.7-max`、`kimi/kimi-k2.7-code`）及能力说明。详见 [模型上下架与更新](../../raw/model-user-guide/release-notes/newly-released-models.md)。 +### 近期重点功能（2026 年） -> **注意**：模型**上架**信息在上表中维护，而模型**下线**规则与清单不在此列，需单独参考 [模型下线机制说明](https://help.aliyun.com/zh/model-studio/model-depreciation)。功能更新文档中也频繁出现"部分老旧模型下线通知""网关变更通告"等公告，建议在集成前定期核对，防止依赖的模型或域名失效。 +| 模块 | 变更 | 说明 | +| --- | --- | --- | +| 智能体托管 | Managed Agent 商业化 / 智能体托管运行时 API 上线 | 平台托管会话与工具执行；2026-07 起商业化计费 | +| 知识库 RAG | 知识检索服务、知识问答服务上线 | 支持多知识库联合检索与混合排序；检索调用可投递 SLS 日志 | +| 记忆库 | Memory 2.0 上线并商业化 | 自动提取记忆片段与用户画像，多应用可共享同一记忆库 | +| Responses API | 新增[异步调用](../concepts/async-invocation.md) | 通过 `background=true` 提交长耗时任务并轮询结果 | +| API 鉴权 | 临时 [API Key](../concepts/api-key.md) 文档上线 | 不可信环境下避免永久 [API Key](../concepts/api-key.md) 泄露 | +| 模型部署 | PTU 新增长输入与前缀缓存 | 预置吞吐部署支持长输入与前缀缓存能力 | +| 接入工具 | 新增 Codex、Kilo CLI 等客户端接入 | 支持 [Token](../concepts/token.md) Plan / Coding Plan / 按量计费三种方式 | +| 模型调优 | 强化学习（RL）训练（邀约制）、图像/视频/VL 模型类型支持 | 调优支持的模型类型持续扩展 | +| 模型评测 | 排行榜（Leaderboard）与综合评测 | 支持 BLEU_4、ROUGE 等评分方法，多模型横向对比 | +| 计费 | 团队版共享用量包、Coding Plan Pro 首月特惠 | 跨坐席共享 Credits 弹性用量包可抵扣超额用量 | -## 平台功能更新的关键脉络 +2026 年 7 月还有多条**下线/退市类公告**：企业知识库（旧）下线、部分老旧模型及长尾模型下线、析言 GBI 商品退市、qwen-turbo 资源包启动退市等，详见[模型平台功能更新](../../raw/model-user-guide/release-notes/model-release-notes.md)的公告与功能动态区。 -从 [模型平台功能更新](../../raw/model-user-guide/release-notes/model-release-notes.md) 可提炼出几条对开发者影响较大的主线： +### 模型调优与部署能力演进（2025 年） -- **[模型调优与部署](../concepts/fine-tuning-and-deployment.md)**：陆续新增视觉理解（VL）、视频生成、图像生成模型的定制训练支持；DPO 偏好训练、强化学习（RL，邀约制）训练；以及按模型单元（MU）时长计费的部署方式与预置模型部署（qwen-flash / qwen-plus 等）。 -- **API 能力**：文本生成 API 入口聚合 OpenAI Responses、Anthropic Messages 等分类；Responses API 新增 `background=true` 异步调用；异步任务支持事件总线 EventBridge 回调，无需轮询。 -- **知识库 RAG**：上线知识检索服务、知识问答服务，检索调用全量投递 SLS 日志，Retrieve 接口新增排序模型与指令干预模式。 -- **接入与计费**：新增 Codex、Kilo CLI 等客户端接入；Token Plan 团队版团队管理与共享用量包；Coding Plan 联网搜索 MCP 升级；API Key 加密存储与业务空间专属推理域名升级。 -- **地域接入**：新增美国、德国、日本等地域与服务部署范围，跨地域部署时需确认目标模型的服务部署范围。 +- 2025-10：新增**按模型单元（MU）的部署方式**（按时间计费），可灵活调整模型性能、成本可预测。 +- 2025-10：千问3-VL-8B-Instruct / Thinking 支持 SFT 调优（全参微调与 LoRA）和调优后部署。 +- 2025-09：千问 3（32B/14B/8B）与千问 2.5（72B/32B/14B/8B）支持 **DPO 偏好训练**，通过负反馈降低幻觉。 +- 2025-08：千问2.5-VL（72B/32B/7B）支持调优并部署。 +- 2025-07：新增**免费额度用完即停**功能，额度耗尽后调用返回错误码 `AllocationQuota.FreeTierOnly`，避免产生额外费用。 -## 模型上架的读法与关键字段 +## 模型上下架与更新 -[模型上下架与更新](../../raw/model-user-guide/release-notes/newly-released-models.md) 的每一行都对应一个可调用的模型规格，集成时重点关注： +模型上架记录按地域组织（当前主要为华北2-北京），每条包含模型类型、时间、模型规格（model id）和功能说明。完整清单见[模型上下架与更新](../../raw/model-user-guide/release-notes/newly-released-models.md)；**模型下线**规则及清单参考官方"模型下线机制说明"文档。 -- **模型规格**：即调用时使用的 `model` 名称，例如推理模型 `qwen3.7-max`、`qwen3.7-plus`、`deepseek-v4-pro`、`kimi/kimi-k2.7-code`、`ZHIPU/GLM-5.1`；文字提取 `qwen3.5-ocr`；语音合成 `qwen-audio-3.0-tts-plus/flash` 等。三方厂商模型通常带 `厂商/` 前缀（如 `vidu/`、`pixverse/`、`stepfun/`、`xiaomi/`）。 -- **快照版本**：形如 `qwen3.7-max-2026-05-20`、`wan2.7-t2v-2026-04-25` 的带日期后缀模型为快照版，能力与主版本一致但版本锁定，适合对稳定性敏感的生产环境。 -- **服务部署范围**：当前上架清单多为"中国内地"，跨地域调用前务必核对。 -- **模态与模式限制**：部分模型有明确约束，例如 `qwen3.6-max-preview` 仅支持纯文本输入、不支持图像与视频输入且默认开启思考模式；`kimi/kimi-k2.7-code` 仅支持思考模式。 +### 近期上架的代表性模型（2026 年） -## 限制与注意事项 +- **文本/推理**：`qwen3.7-flash`（[多模态](../concepts/multimodal.md)理解与 Agent 能力全面提升）、`qwen3.7-max` / `qwen3.7-max-2026-06-08`（新增视觉模态）、`qwen3.7-plus`（[多模态](../concepts/multimodal.md)交互混合智能体）、`qwen3.6-flash`、`qwen3.6-27b`、`qwen3.6-max-preview`（仅纯文本输入、默认思考模式）。 +- **第三方直供**：`kimi/kimi-k3`（2.8 万亿参数、100 万 token 上下文）、`kimi/kimi-k2.7-code` 及其 highspeed 版本（速度提升 5~6 倍）、`deepseek-v4-pro` / `deepseek-v4-flash`（阿里直供）、`vanchin/deepseek-v4-pro`（快手万擎直供）、`ZHIPU/GLM-5.1`、`stepfun/step-3.7-flash`、`xiaomi/mimo-v2.5-pro`。 +- **图像生成**：`qwen-image-3.0-pro`（邀测，长文本输入与密集排版）、`qwen-image-2.0-pro-2026-06-22`（文字渲染增强，支持最长 1k token 指令）。 +- **视频生成/编辑**：万相 2.7 系列（`wan2.7-t2v` / `wan2.7-i2v` / `wan2.7-r2v` / `wan2.7-videoedit`）、HappyHorse 1.0/1.1 系列（支持有声视频，3~15 秒、720P/1080P）、爱诗 C1/V6 系列、Vidu 系列。 +- **语音/音频**：`qwen-audio-3.0-realtime-plus/flash`（端到端实时语音对话）、`qwen-audio-3.0-tts-plus/flash`（语音合成）、`fun-asr`（30 语种、七大方言）、`fun-music-v1`（音乐生成）。 +- **其他**：`qwen3.5-ocr`（128K 上下文文字提取）、`qwen3.5-livetranslate-flash-realtime`（60 语种识别、29 语种实时翻译）、Tripo 3D 生成、pixverse 视频对口型/动作模仿/超清。 -- **计费与优惠会随时间调整**：如 `deepseek-v4-pro` 的 `cached_token` 单价曾调整为 1 元/百万 token（标准 `input_token` 不变），GLM-5.2 Fast mode 降价、上下文缓存降价等均以对应公告为准，release notes 中的价格描述可能滞后。 -- **模型可能延期或提前下线**：功能更新中同时存在"部分老旧模型下线通知"与"部分老旧模型延期下线通知"，同一批模型的下线时间可能被修订，务必以最新公告为准。 -- **两份文档存在时间粒度差异**：功能更新文档到具体功能点，模型清单文档到具体模型规格；排查某能力是否可用时，建议交叉比对 [模型平台功能更新](../../raw/model-user-guide/release-notes/model-release-notes.md) 与 [模型上下架与更新](../../raw/model-user-guide/release-notes/newly-released-models.md) 两处。 -- **以控制台/API 实际返回为准**：release notes 是变更记录而非实时状态，最终的可调用模型列表、配额与地域支持应以控制台模型广场和 API 返回为准。 +### 价格调整示例 + +`deepseek-v4-pro` 自 2026-04-29 起 `cached_token` 单价调整为 **1 元/百万 token**，标准 `input_token` 单价不变（上下文缓存计费）。更多降价公告见[模型平台功能更新](../../raw/model-user-guide/release-notes/model-release-notes.md)顶部的公告通知区。 + +## 使用方式与注意事项 + +- **如何跟进变更**：功能动态按月归档，模型上架按地域列表维护；接入前建议核对所用模型的 model id 是否为最新快照版本（如 `qwen3.7-max-2026-06-08` 与稳定别名 `qwen3.7-max` 并存）。 +- **下线风险**：平台会定期下线老旧/长尾模型并发布延期下线通知，生产环境应关注公告通知区与"模型下线机制说明"，避免依赖即将退市的模型或资源包（如 qwen-turbo 资源包已启动退市）。 +- **计费变更**：降价、资源包退市、商业化（记忆库、Managed Agent）等公告直接影响账单，升级集成前先核对最新计费规则。 +- **思考模式差异**：部分新模型默认开启思考模式或仅支持思考模式（如 `qwen3.7-max`、`kimi/kimi-k2.7-code`），调用参数需按对应模型文档配置。 + +> **注意**：[模型上下架与更新](../../raw/model-user-guide/release-notes/newly-released-models.md)中华北2（北京）地域的 `wan2.7-t2v-2026-06-12` 快照条目日期标注为 2026-07-01，与相邻条目（2026-04-26 的 `wan2.7-t2v-2026-04-25`）在时间排序上不一致，引用该快照版本日期时请以官方页面为准。 ## 来源文档 @@ -43,3 +63,5 @@ release notes 覆盖两个正交的维度，查询时应按需求选择对应文 - [模型上下架与更新](../../raw/model-user-guide/release-notes/newly-released-models.md) + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/security-and-compliance.md b/skills/bailian-docs-llm-wiki/wiki/guides/security-and-compliance.md index f9850221..a253e527 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/security-and-compliance.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/security-and-compliance.md @@ -1,182 +1,132 @@ # security and compliance -阿里云百炼围绕"身份权限、内容安全、合规备案、传输加密、私网与安全存储"五个维度构建安全合规体系，覆盖从控制台到 API 调用、从公网到 VPC 私网的完整链路。本文面向开发者，按"权限与身份—内容安全—合规备案—传输加密—[私网访问](../concepts/vpc-private-access.md)—安全存储"的顺序梳理关键能力、参数与注意事项。 +阿里云百炼的安全与合规能力覆盖身份权限、内容安全、备案材料、数据隐私、传输加密、私网访问和安全存储等环节。开发者应先按地域和[业务空间](../concepts/workspace.md)规划权限边界，再根据应用形态选择 AI 安全护栏、加密调用、PrivateLink 或安全存储空间等能力，最后结合上架平台要求准备备案与合规材料。 -## 身份与权限管理 +## 能力范围与适用场景 -百炼的权限管理基于[业务空间](../concepts/workspace.md)（workspace）这一最小管理单元，支持控制台页面级与模型级的多维度权限控制，满足多地域、多用户的复杂组织架构需求，详见 [权限管理](../../raw/model-user-guide/security-and-compliance/permission-management-overview.md)。 +### 权限与[业务空间](../concepts/workspace.md)隔离 -### 三种角色 +百炼以[业务空间](../concepts/workspace.md)作为精细化权限管理和账单分账的最小单元，单个[业务空间](../concepts/workspace.md)不能跨地域存在；不同地域的默认[业务空间](../concepts/workspace.md)也是不同空间。权限体系主要包括超级管理员、[业务空间](../concepts/workspace.md)管理员和普通用户三类角色，详见[权限管理](../../raw/model-user-guide/security-and-compliance/permission-management-overview.md)。 -- **超级管理员**：阿里云主账号，或拥有 `AliyunBailianFullAccess` 系统策略的 RAM 用户，可跨空间统一管理用户权限、空间可用模型、模型限流和 [API Key](../concepts/api-key.md)。建议 AI 安全护栏、模型监控、应用观测等功能使用主账号一次性开通。 -- **[业务空间](../concepts/workspace.md)管理员**：拥有访问某个[业务空间](../concepts/workspace.md)"权限管理"页面的 RAM 用户，可管理该空间内的用户与资源，"管理员"权限包含该空间所有页面的访问权限。 -- **普通用户**：根据分配的权限使用资源，不能管理用户或模型授权。 +核心控制点包括： -[业务空间](../concepts/workspace.md)按地理区域划分，**单个[业务空间](../concepts/workspace.md)不能跨地域存在**，各地域的默认[业务空间](../concepts/workspace.md)也是不同的空间。 +- **模型调用授权与[限流](../concepts/rate-limit.md)**：超级管理员可控制某个模型是否允许在指定[业务空间](../concepts/workspace.md)通过控制台或 API 调用，并可设置请求数[限流](../concepts/rate-limit.md)和 [Token](../concepts/token.md) [限流](../concepts/rate-limit.md)。 +- **模型训练与部署权限**：可限制[业务空间](../concepts/workspace.md)是否允许调优或部署特定模型。 +- **控制台页面权限**：控制 RAM 用户可访问哪些[业务空间](../concepts/workspace.md)页面；该权限不影响其通过 [API Key](../concepts/api-key.md) 调用模型。 +- **[API Key](../concepts/api-key.md) 管理**：[API Key](../concepts/api-key.md) 归属一个地域内的一个[业务空间](../concepts/workspace.md)和一个用户，其可调用功能和限流继承所属[业务空间](../concepts/workspace.md)权限。 +- **OpenAPI 权限**：RAM 用户调用百炼应用的数据、[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)、Prompt 工程、长期记忆等 OpenAPI 时，需要主账号在 RAM 控制台授予 `AliyunBailianDataFullAccess` 或 `AliyunBailianDataReadOnlyAccess`。 -### 模型与 [API Key](../concepts/api-key.md) 权限 +生产环境建议按开发、测试、预发、生产划分独立[业务空间](../concepts/workspace.md)，并为生产空间预留更高限流配额和更严格的用户权限。 -[业务空间](../concepts/workspace.md)内可对模型进行三类精细化授权，**默认[业务空间](../concepts/workspace.md)无法设置这些限制**（所有模型均可调用、调优、部署）： +> **注意**：默认[业务空间](../concepts/workspace.md)无法设置模型调用、训练、部署限制，也无法设置空间级限流；如果需要精细化隔离，应创建独立[业务空间](../concepts/workspace.md)。 -| 权限项 | 控制范围 | 配置入口 | -| --- | --- | --- | -| 限制模型调用 | 是否可调用（控制台 & API）+ 请求数限流 + [Token](../concepts/token.md) 限流 | 模型列表 → 模型调用列开关 + 当前空间限流列 | -| 限制模型训练 | 是否可调优（控制台 & API）及调优后部署 | 模型列表 → 模型授权 → 模型训练列 | -| 限制[模型部署](../concepts/model-deployment.md) | 是否可直接部署 | 模型列表 → 模型授权 → [模型部署](../concepts/model-deployment.md)列 | +### 输入输出内容安全 -单个 [API Key](../concepts/api-key.md) 只能归属一个地域内的一个[业务空间](../concepts/workspace.md)和一个用户，且不能转移。[API Key](../concepts/api-key.md) 的可调用功能与模型限流与**归属[业务空间](../concepts/workspace.md)**的权限保持一致，不受用户控制台权限影响，也无需为不同模型（文生文、文生图、语音合成）创建不同 [API Key](../concepts/api-key.md)。自 2026 年 3 月 25 日起，华北2（北京）地域所有新创建的 [API Key](../concepts/api-key.md) 均归属主账号，并支持设置 IP 访问白名单。 +调用大模型时，百炼会根据模型自动匹配对应的 AI 安全护栏服务，目前文档说明支持文本和图片类型模型。接入方式是在请求头中增加 `X-DashScope-DataInspection`，指定对输入和输出进行内容检查，具体配置见[输⼊输出AI安全护栏](../../raw/model-user-guide/security-and-compliance/content-security.md)。 -> **注意**：[API Key](../concepts/api-key.md) 的有效性受账号操作影响——将 RAM 账号移出[业务空间](../concepts/workspace.md)会使其 [API Key](../concepts/api-key.md) 失效（重新加入后恢复），而在 RAM 控制台删除账号/角色则会使 [API Key](../concepts/api-key.md) 永久失效、不可恢复。 - -### OpenAPI 接口权限 - -RAM 用户默认无权调用百炼应用的数据、[知识库](../concepts/knowledge-base.md)、Prompt 工程、长期记忆等 Open API。需阿里云主账号在 RAM 控制台添加以下系统策略之一： - -- `AliyunBailianDataFullAccess`：可调用应用 API 目录下的所有 API。 -- `AliyunBailianDataReadOnlyAccess`：仅可调用只读类 API（如 `DescribeFile`、`GetIndexJobStatus`）。 - -### 生产环境实践 - -- **空间规划**：推荐按环境（dev/test/prod）划分[业务空间](../concepts/workspace.md)实现隔离，或按业务线划分便于权限与成本管理。 -- **限流策略**：将主账号总配额按比例分配给各[业务空间](../concepts/workspace.md)并预留缓冲。例如总配额 1000 QPM，可分配 prod 600 / test 200 / dev 100，预留 100。 - -## 内容安全：AI 安全护栏 - -大模型输入输出可能包含涉黄、涉政、广告等敏感内容。除模型自有合规检查外，百炼支持接入 AI 安全护栏服务，进一步识别违规信息，保障安全合规，详见 [输⼊输出AI安全护栏](../../raw/model-user-guide/security-and-compliance/content-security.md)。 - -### 开通步骤 - -1. **开通内容审核服务**：访问 AI 安全护栏购买页面，创建服务关联角色并购买。 -2. **授权内容安全设置**：在百炼"安全管理"页面单击"去授权"开启内容安全设置。若已显示"已开通（不可取消）"及《自建安全机制承诺函》全文，可跳过此步。 -3. **设置请求头**：调用百炼时在请求头设置 `X-DashScope-DataInspection`，开启输入输出检测： +关键请求头示例： ```json -{"X-DashScope-DataInspection": {"input": "cip", "output": "cip"}} +{ + "X-DashScope-DataInspection": { + "input": "cip", + "output": "cip" + } +} ``` -目前支持文本和图片类型的模型，模型与护栏服务的对应关系及[计费](../concepts/billing.md)请参见官方说明。命中护栏时返回 `data_inspection_failed`（DashScope 为 `DataInspectionFailed`，HTTP 400），提示输入可能包含不当内容。 - -## 合规资质与备案 - -### 资质与隐私 - -百炼以无保留意见通过 SOC 2 审计，安全、可用性、保密性控制符合国际标准。阿里云承诺**不会将您的数据用于模型训练**，应用构建与模型训练过程中的传输数据均经过 AES-256 加密。根据法律法规，百炼会存储模型与应用调用产生的数据，详见《阿里云百炼服务协议》中的数据处理、隐私和安全条款，参考 [合规资质与隐私说明](../../raw/model-user-guide/security-and-compliance/privacy-notice.md)。 - -### 模型备案信息公示 - -百炼公示所接入大模型的算法备案号与大模型备案号，详见 [模型备案信息公示](../../raw/model-user-guide/security-and-compliance/model-filing-information-publicity.md)。部分示例如下： - -| 模型 | 算法名称 | 算法备案号 | -| --- | --- | --- | -| 千问 | 达摩院交互式多能型合成算法 | 网信算备330110507206401230035号 | -| 万相 | 达摩院图像合成算法 | 网信算备330110507206401230027号 | -| 万相 | 通义万相视频生成算法 | 网信算备330106003156001240091号 | -| DeepSeek | DeepSeek 大语言模型算法 | 网信算备110108970550101240011号 | -| 智谱 AI | 智谱交互式内容生成算法 | 网信算备110108105858001230027号 | - -大模型备案号（如通义千问 `ZheJiang-TongYiQianWen-20230901`、DeepSeek `Beijing-DeepseekChat-202404280016` 等）同样公示。第三方模型备案信息由提供方负责，阿里云百炼不作额外承诺。 - -### 应用上架及合规备案 - -接入千问、万相等模型的应用/小程序上架前，需依《生成式人工智能服务管理暂行办法》完成合规备案，详见 [千问大模型应用上架及合规备案](../../raw/model-user-guide/security-and-compliance/compliance-and-launch-filing-guide-for-ai-apps-powered-by-the-tongyi-model.md)。 +OpenAI 兼容 SDK、[DashScope SDK](../concepts/dashscope-sdk.md)、Java、Node.js 等调用方式都可以通过额外请求头接入。命中高风险内容时，接口通常返回 `data_inspection_failed`、`DataInspectionFailed` 或 400 错误，调用方应将其作为业务可预期异常处理，而不是重试绕过。 -典型场景与所需材料： +### 备案与应用上架材料 -| 场景 | 是否面向 C 端 | 舆论/动员能力 | 主要材料 | -| --- | --- | --- | --- | -| 场景 1 | 是 | 不具有 | 大模型算法备案信息 + 应用主体与阿里云的合作协议 | -| 场景 2 | 是 | 具有 | 场景 1 全部材料 + 企业自主安全评估报告 + 企业自主算法备案 | -| 场景 3 | 企业内部 | — | 关注数据安全、保密等内部合规要求 | +如果应用、小程序通过百炼接入千问、万相等模型并计划上架到应用市场或小程序平台，开发者需要根据自身是否面向 C 端、是否具有舆论属性或社会动员能力等因素准备备案材料。备案号和算法备案信息可参考[模型备案信息公示](../../raw/model-user-guide/security-and-compliance/model-filing-information-publicity.md)与[千问大模型应用上架及合规备案](../../raw/model-user-guide/security-and-compliance/compliance-and-launch-filing-guide-for-ai-apps-powered-by-the-tongyi-model.md)。 -算法备案信息查询步骤：打开互联网信息服务算法备案系统（beian.cac.gov.cn），以"备案编号"为搜索类别输入对应备案号（如千问 `网信算备330110507206401230035号`），截图保存查询页面完整内容。合作协议（需含算法名称、应用产品或备案编号）请联系商务经理获取。 +常见材料包括： -> **注意**：即使使用阿里云提供的模型及备案信息，应用/小程序开发者仍是法规定义的"服务提供者"，需独立履行内容审核、用户保护、数据安全、标识规范等全部法定义务。备案号应以算法备案系统实时查询结果为准，建议定期核验。 +- 大模型算法备案信息或算法备案系统截图。 +- 应用主体与阿里云主体的合作协议。 +- 对具有舆论属性或社会动员能力的场景，可能还需要安全评估报告和企业自主算法备案。 -## 传输加密：以加密方式接入模型推理 +备案信息页面公示了千问、万相、智谱 AI、DeepSeek、Moonshot、Xiaomi MiMo、可灵 AI、MiniMax、阶跃星辰、Vidu、Tripo、PixVerse 等模型或算法的备案编号；第三方模型备案信息由提供方负责。 -当请求涉及敏感信息或经公网传输时，可对请求体 `input` 字段值加密，防止传输过程中被窃听或篡改，详见 [以加密的方式接入模型推理功能](../../raw/model-user-guide/security-and-compliance/transmission-security/encrypted-access-to-model-inference.md)。 +> **注意**：备案信息、监管政策和平台审核要求可能变化。文档中的备案编号可作为材料准备入口，但实际提交前应以监管系统实时查询结果和应用平台最新要求为准。 -### 加解密机制 +### 合规资质与隐私保护 -采用混合加密：数据由 AES 对称算法加密，AES 密钥通过 RSA 非对称加密实现安全传输。流程为：生成 AES 密钥 → 用 AES 密钥加密 `input` → 用 RSA 公钥加密 AES 密钥 → 将加密后的 `input` 与密钥信息（封装在 `X-DashScope-EncryptionKey` 请求头）传入百炼 → 平台推理链路全程加密、解密数据并用相同 AES 密钥加密答案返回 → 用户侧用 AES 密钥解密响应。 +百炼通过 SOC 2 审计，在安全、可用性和保密性方面建立控制体系。隐私说明中明确：阿里云不会将用户数据用于模型训练；构建应用或训练大模型过程中传输的数据会经过 AES-256 加密。更多说明见[合规资质与隐私说明](../../raw/model-user-guide/security-and-compliance/privacy-notice.md)。 -### 两种接入方式 +同时，百炼会根据法律法规和服务协议存储模型与应用调用时产生的数据。开发者在设计系统时，应将调用日志、审计、数据保留和用户告知纳入自身合规方案。 -**1. [DashScope SDK](../concepts/dashscope-sdk.md)（自动加密·开箱即用）** +## 传输安全：加密调用与私网访问 -仅支持 Java 和 Python，不支持自定义密钥： +### DashScope 推理请求加密 -- Java SDK：`GenerationParam.builder().enableEncrypt(true)` -- Python SDK：`dashscope.Generation.call(..., enable_encryption=True)` +当请求内容涉及敏感信息或经公网传输时，可以对请求体中的 `input` 字段加密。百炼采用混合加密机制：使用 AES 加密 `input` 内容，使用 RSA 公钥加密 AES 密钥，并通过 `X-DashScope-EncryptionKey` 请求头传输密钥信息，完整流程见[以加密的方式接入模型推理功能](../../raw/model-user-guide/security-and-compliance/transmission-security/encrypted-access-to-model-inference.md)。 -SDK 自动完成加解密，响应为明文，无需手动处理。 +接入方式有两类： -**2. HTTP 调用（手动密钥管理）** +- **[DashScope SDK](../concepts/dashscope-sdk.md) 自动加密**：Java SDK 设置 `enableEncrypt(true)`，Python SDK 设置 `enable_encryption=True`。SDK 自动完成加解密，响应内容为明文。 +- **HTTP 手动密钥管理**：调用方自行生成 AES 密钥和 IV，加密 `input`，再通过 RSA 公钥加密 AES 密钥，并在响应后自行解密。 -需额外完成三步：添加 `X-DashScope-EncryptionKey` 请求头、对 `input` 内容加密、对响应数据解密。该请求头为 JSON 字符串，含 `public_key_id`（公钥 ID）、`encrypt_key`（RSA 公钥加密后的 AES 密钥）、`iv`（初始向量）三个字段。AES 密钥长度支持 128/192/256 位（32 字节），长度越长安全性越高但开销越大。 +手动模式需要先通过 `GET /api/v1/public-keys/latest` 获取 `public_key_id` 和 `public_key`，接口说明见[获取RSA的公钥](../../raw/model-user-guide/security-and-compliance/transmission-security/model-interface-aes-encryption.md)。 -> **注意**：手动加密调用仅适用于 DashScope 的 Endpoint，OpenAI 兼容（Chat Completions API 和 Responses API）的 Endpoint **不支持**此加密机制。 +关键参数包括： -### 获取 RSA 公钥 +- `input`：加密前为模型请求核心对象；手动加密后变为密文字符串。 +- `X-DashScope-EncryptionKey`：JSON 字符串，包含 `public_key_id`、`encrypt_key`、`iv`。 +- `Authorization`：获取 RSA 公钥与调用模型时均需使用 [API Key 鉴权](../concepts/api-key.md)。 -加密前需调用接口获取当前最新公钥 ID 及其值，详见 [获取RSA的公钥](../../raw/model-user-guide/security-and-compliance/transmission-security/model-interface-aes-encryption.md)。 +> **注意**：文档明确说明 HTTP 手动加密机制适用于 DashScope Endpoint；OpenAI 兼容的 Chat Completions API 和 Responses API 不支持该加密机制。 -- 接口：`GET /api/v1/public-keys/latest` -- 鉴权：`Authorization: Bearer ` -- 返回：`request_id`、`data.public_key`（RSA 公钥值）、`data.public_key_id`（公钥 ID） -- 前提：已开通百炼服务并获得 API-KEY，建议配置到环境变量以降低泄漏风险。 +### PrivateLink 私网访问模型或应用 API -## [私网访问](../concepts/vpc-private-access.md)：通过 PrivateLink 终端节点访问 API +如果需要在 VPC 内调用百炼模型或应用 API，并确保流量不经过公网，可创建接口终端节点，通过 PrivateLink 建立从 VPC 到百炼的单向私网连接。该能力详见[通过终端节点私网访问阿里云百炼模型或应用 API](../../raw/model-user-guide/security-and-compliance/transmission-security/access-model-studio-through-privatelink.md)。 -为在 VPC 内直接调用百炼模型/应用 API 且流量不经公网，可创建私网终端节点（PrivateLink），将通信限制在阿里云内网，详见 [通过终端节点私网访问阿里云百炼模型或应用 API](../../raw/model-user-guide/security-and-compliance/transmission-security/access-model-studio-through-privatelink.md)。 +基本步骤： -### 工作原理与地域 +1. 在百炼服务所在地域创建接口终端节点，终端节点服务选择 `com.aliyuncs.dashscope`。 +2. 选择访问百炼服务的 VPC、交换机和安全组，安全组入方向需允许 80/443。 +3. 获取默认服务域名或自定义服务域名。 +4. 将 API 调用中的 `base_url` 从公网域名替换为终端节点服务域名。 -终端节点连接为**单向设计**，仅允许 VPC 内资源主动访问百炼，百炼无法反向访问 VPC 内资源。百炼公共云服务所在地域：华北2（北京）、新加坡。**美国（弗吉尼亚）地域暂不支持[私网访问](../concepts/vpc-private-access.md)**。 +地域限制与访问方式： -### 配置步骤 +- 公共云百炼服务地域包括华北2（北京）和新加坡。 +- 美国（弗吉尼亚）地域文档标注暂不支持私网访问。 +- 同境内或同境外跨地域访问可使用跨地域端点；跨境跨地域访问可通过 CEN 做 VPC 互通。 -1. **创建接口终端节点**：在终端节点控制台创建，终端节点服务选择"阿里云服务"并筛选 `com.aliyuncs.dashscope`，开启"自定义服务域名"。建议至少选择两个不同可用区的交换机以实现高可用；安全组入方向需允许 80（http）、443（https）。 -2. **获取终端节点服务域名**：默认服务域名格式 `ep-{实例ID}.privatelink.aliyuncs.com`（仅 HTTP）；自定义服务域名格式 `vpc-{实例ID}.{地域ID}.dashscope.aliyuncs.com`（支持 HTTPS）。 -3. **调用验证**：将 API base_url 中的域名替换为终端节点服务域名后在对应 VPC 发起调用。支持 HTTP/curl、OpenAI Python SDK、DashScope Python SDK（建议 ≥1.14.0）、DashScope Java SDK（建议 ≥2.12.0）。 - -### 跨地域访问 - -- **同境内或同境外跨地域**（如华东1杭州 VPC → 华北2北京百炼）：推荐启用跨地域端点。 -- **跨境跨地域**（中国内地与境外之间，如华北2北京 VPC → 新加坡百炼）：通过 CEN 跨地域 VPC 互通。 +> **注意**：默认服务域名仅支持 HTTP；如需 HTTPS，应使用自定义服务域名。 ## 安全存储[业务空间](../concepts/workspace.md) -安全存储[业务空间](../concepts/workspace.md)通过反向终端节点与 VPC 私网连接，让部署其中的应用访问同 VPC 下的 ElasticSearch、AnalyticDB（ADB）、OSS 等云组件，避免公网访问风险。该能力需联系商务人员申请开通。 +安全存储业务空间适用于需要将百炼应用程序的数据读写限制在特定私有网络资源中的场景，例如使用私网内的 OSS、AnalyticDB PostgreSQL、Elasticsearch 等组件，避免通过公网访问外部服务。相关配置由一组文档串联完成，包括[配置终端节点并发起连接](../../raw/model-user-guide/security-and-compliance/secure-storage/configure-an-endpoint-and-initiate-a-connection.md)、[配置可用区IP](../../raw/model-user-guide/security-and-compliance/secure-storage/configure-zone-ip.md)、[配置私有网络中的资源](../../raw/model-user-guide/security-and-compliance/secure-storage/configure-resources-in-private-network.md)和[配置MSE云原生网关](../../raw/model-user-guide/security-and-compliance/secure-storage/configure-mse.md)。 + +### 配置流程 -### 配置流程总览 +1. **创建安全存储业务空间**：在业务空间管理中新增空间，空间类型选择安全存储空间。 +2. **创建反向终端节点**：在华北2（北京）等支持地域创建反向终端节点，并在百炼侧确认连接。 +3. **配置可用区 IP**：创建或复用 MSE 云原生网关，获取 NLB 的可用区 VIP 与交换机网段，并将 VIP 配置到百炼安全存储空间。 +4. **配置私有网络资源**：按需配置 OSS、ADB、ES，并完成百炼访问授权、标签、白名单、账号密码等配置。 +5. **配置 MSE 云原生网关**：将私网资源登记为网关服务，创建路由并发布。 +6. **激活安全存储业务空间**：确认 Elasticsearch、ADB、OSS 都配置无误后激活空间。 -| 步骤 | 说明 | 参考 | -| --- | --- | --- | -| 1. 创建安全存储[业务空间](../concepts/workspace.md) | [业务空间](../concepts/workspace.md)管理 → 新增[业务空间](../concepts/workspace.md)，空间类型选"安全存储空间" | [配置终端节点并发起连接](../../raw/model-user-guide/security-and-compliance/secure-storage/configure-an-endpoint-and-initiate-a-connection.md) | -| 2. 创建反向终端节点 | 终端节点控制台创建反向终端节点，终端节点服务选描述为"百炼公共云生产环境-北京站点-安全存储空间专网通道接入点"的服务（VPC NAT 网关、反向、IPv4），配置 VPC/安全组/可用区与交换机 | 同上 | -| 3. 在百炼确认连接 | [业务空间](../concepts/workspace.md)管理 → 管理安全存储空间 → 选择终端节点 → 连接，等待状态变为"已连接" | 同上 | -| 4. 配置可用区 IP | 创建 MSE 云原生网关（2核4G、2 节点、启用 TLS 硬件加速、私网、至少两可用区），获取 NLB 各可用区 VIP 与交换机网段，在百炼配置对应可用区 IP，并将 VIP 加入反向终端节点安全组入方向（全部端口） | [配置可用区IP](../../raw/model-user-guide/security-and-compliance/secure-storage/configure-zone-ip.md) | -| 5. 配置私有网络资源 | 配置 OSS（Bucket 名 `bailian-safe-workspace-oss-access`、特定标签、CORS 来源 `*bailian.console.aliyun.com`）、ADB（6.0 标准版、高可用版、开启向量引擎优化）、ElasticSearch（7.10、内核增强版、两可用区，交换机网段加入 VPC 私网访问白名单） | [配置私有网络中的资源](../../raw/model-user-guide/security-and-compliance/secure-storage/configure-resources-in-private-network.md) | -| 6. 配置 MSE 云原生网关 | 在网关创建 DNS 域名服务（指向 ES 私网地址/端口，TLS 关闭）、创建路由（域名为 ES 域名、路径 `/`、单服务），然后回百炼"资源配置"页激活安全存储[业务空间](../concepts/workspace.md) | [配置MSE云原生网关](../../raw/model-user-guide/security-and-compliance/secure-storage/configure-mse.md) | +### 资源配置要点 -### 关键约束与注意事项 +- **OSS**：Bucket 需要设置特定标签 `bailian-safe-workspace-oss-access=ReadAndWrite`，并配置允许百炼控制台来源的 CORS 规则。 +- **ADB**：首次配置需要授权服务角色 `AliyunServiceRoleForSFMAccessADB`。 +- **ES**：需要将交换机网段加入 VPC 私网访问白名单，并在百炼配置实例 ID、账号和密码。 +- **MSE**：服务来源选择 DNS 域名，路由路径通常为 `/`，后端服务指向已创建的私网资源服务。 -- 专有网络地域须为华北2（北京），可用区须在 G/H/L（或 ADB 的 G/H/I）中按要求选择，每个可用区至少一个交换机；反向终端节点的安全组不要放入其他云组件、无需配置出入网规则。 -- 阿里云百炼只能访问客户授权过且带特定标签（标签名 `bailian-safe-workspace-oss-access`，标签值 `ReadAndWrite`）的 OSS Bucket。 -- ADB 配置需授权服务关联角色 `AliyunServiceRoleForSFMAccessADB`（策略 `AliyunServiceRolePolicyForSFMAccessADB`）。 -- **资源不可中断**：OSS Bucket 停止服务会导致安全存储空间、[知识库](../concepts/knowledge-base.md)、审计日志、历史记录等模块不可用，恢复 Bucket 后可恢复；但 Bucket 被释放会造成安全存储空间不可用且**无法恢复**，需重建。ES 停止[计费](../concepts/billing.md)/被释放的后果与 OSS 相同。 -- 激活前安全存储空间不可用，激活成功后才可用。 +> **注意**：OSS Bucket 停止服务会导致安全存储空间、[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)、审计日志、历史记录等模块不可用；Bucket 释放或 ES 释放可能导致安全存储空间不可恢复，需要重新创建。 -## 限制与注意事项汇总 +## 开发者落地建议 -- 默认[业务空间](../concepts/workspace.md)无法设置模型调用/训练/部署限制，所有模型均可调用、调优、部署，且无法限流。 -- [API Key](../concepts/api-key.md) 不可跨地域、跨业务空间、跨用户转移；账号移出空间会使其 [API Key](../concepts/api-key.md) 失效（重新加入恢复），删除账号/角色则永久失效。 -- AI 安全护栏目前仅支持文本和图片类型模型。 -- [DashScope SDK](../concepts/dashscope-sdk.md) 自动加密仅支持 Java/Python 且不支持自定义密钥；HTTP 手动加密仅适用于 DashScope Endpoint，OpenAI 兼容 Endpoint 不支持。 -- PrivateLink 私网访问美国（弗吉尼亚）地域暂不支持；跨地域访问需区分同境内/同境外与跨境两种方式。 -- 安全存储业务空间的 OSS/ES 等底层资源一旦释放，安全存储空间不可恢复，需重建。 -- 模型与应用的合规备案信息应以算法备案系统实时查询结果为准，建议定期核验；开发者作为"服务提供者"需独立承担全部法律责任。 +- **最小权限优先**：为不同环境创建独立业务空间，只向 RAM 用户授予必要页面权限、[API Key](../concepts/api-key.md) 权限和 OpenAPI 权限。 +- **[API Key](../concepts/api-key.md) 安全**：优先使用环境变量保存 [API Key](../concepts/api-key.md)，不要硬编码到代码仓库；根据地域能力设置 IP 访问白名单。 +- **内容安全前置**：面向外部用户的应用建议默认接入 AI 安全护栏，并把安全拦截错误纳入前端提示和审计日志。 +- **敏感输入加密**：对包含个人信息、商业秘密或内部数据的 `input` 启用 [DashScope SDK](../concepts/dashscope-sdk.md) 加密或 HTTP 手动加密。 +- **网络隔离**：生产环境优先使用 PrivateLink 或安全存储空间，将公网暴露面降到最低。 +- **合规材料留痕**：保存备案系统查询截图、合作协议、模型备案编号、内容审核策略和用户协议版本，便于上架审核和后续审计。 ## 来源文档 @@ -197,19 +147,4 @@ SDK 自动完成加解密，响应为明文，无需手动处理。 - - - - - - - - - - - - - - - diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/skill.md b/skills/bailian-docs-llm-wiki/wiki/guides/skill.md index c3b17ca0..87f75e4d 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/skill.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/skill.md @@ -126,6 +126,19 @@ description: "Use this skill any time a spreadsheet file is the primary input or + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/start-using.md b/skills/bailian-docs-llm-wiki/wiki/guides/start-using.md index 93ce31e7..766c1171 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/start-using.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/start-using.md @@ -7,8 +7,8 @@ 借助百炼的[智能体应用](../concepts/agent-application.md)构建能力，可在约 5 分钟内零代码完成一个能回答私有领域问题的大模型问答应用，完整流程见 [0代码构建私有知识问答应用](../../raw/application-user-guide/start-using/build-knowledge-base-qa-assistant-without-coding.md)。整体分为三步： 1. **构建第一个[智能体应用](../concepts/agent-application.md)（约 1 分钟）**：在应用管理页面创建空白[智能体应用](../concepts/agent-application.md)，选择大语言模型（建议千问-Max），编写 System Prompt 定义角色与任务，并配置欢迎语和预设问题。此阶段由于缺少私有知识，回答较为笼统甚至可能无中生有。 -2. **构建[知识库](../concepts/knowledge-base.md)（约 3 分钟）**：在数据连接页面创建文件类型连接器并上传知识文档（如 docx），等待 1~6 分钟解析完成；随后在[知识库](../concepts/knowledge-base.md)页面创建标准版知识库，选择默认类目与智能切分策略，等待 1~2 分钟完成解析。智能切分为系统预置策略，经[评测](../concepts/evaluation.md)对多数文档可获得最佳检索效果。 -3. **添加知识库并发布应用（约 1 分钟）**：进入应用配置界面，通过「技能 > 知识库」旁的「+」按钮为应用挂载知识库，验证检索增强效果后点击「发布」。 +2. **构建[知识库](../concepts/knowledge-base.md)（约 3 分钟）**：在数据连接页面创建文件类型连接器并上传知识文档（如 docx），等待 1~6 分钟解析完成；随后在[知识库](../concepts/knowledge-base.md)页面创建标准版[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)，选择默认类目与智能切分策略，等待 1~2 分钟完成解析。智能切分为系统预置策略，经[评测](../concepts/evaluation.md)对多数文档可获得最佳检索效果。 +3. **添加[知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)并发布应用（约 1 分钟）**：进入应用配置界面，通过「技能 > [知识库](../concepts/[knowledge](../api/knowledge.md)-base.md)」旁的「+」按钮为应用挂载知识库，验证检索增强效果后点击「发布」。 > **注意**：使用大模型会产生[计费](../concepts/billing.md)，百炼提供限时免费额度，可在模型广场查看；知识库服务自 2026 年 1 月 4 日起正式[计费](../concepts/billing.md)，费用由规格费用和模型调用费用两部分组成。 @@ -17,15 +17,15 @@ 百炼应用支持多种模型系列，详见 [应用功能动态](../../raw/application-user-guide/start-using/application-release-notes.md)： - **千问系列**：千问-Max 为构建问答应用的推荐模型；[智能体应用](../concepts/agent-application.md)与[工作流](../concepts/workflow.md)应用均支持 QwQ 系列（具备强推理能力，先输出思考过程再输出回答，数学/代码能力达 DeepSeek-R1 满血版水平，但不包括插件、流程、音视频交互能力）；视觉模型支持 qwen-vl-plus-latest、qwen-vl-plus-0125（Qwen2.5-VL 系列，128k 上下文）以及 qwen-vl-max/plus 用于图片解析。 -- **DeepSeek 系列**：[智能体应用](../concepts/agent-application.md)与[工作流](../concepts/workflow.md)应用均可集成 DeepSeek 系列模型，结合知识库、长期记忆和 Prompt 模板构建私有知识问答应用。 -- **嵌入模型**：知识库支持 text-embedding-v3、v4 模型，v4 在语种支持、代码片段向量化效果和向量维度选择上较 v3 全面升级。 +- **DeepSeek 系列**：[智能体应用](../concepts/agent-application.md)与[工作流](../concepts/workflow.md)应用均可集成 DeepSeek 系列模型，结合知识库、[长期记忆](../concepts/long-term-memory.md)和 Prompt 模板构建私有知识问答应用。 +- **嵌入模型**：知识库支持 text-embedding-v3、v4 模型，v4 在语种支持、代码片段[向量化](../concepts/embedding.md)效果和向量维度选择上较 v3 全面升级。 ## 应用类型与关键能力 百炼提供多种应用类型以适配不同场景： - **[智能体应用](../concepts/agent-application.md)**：2025 年 12 月 26 日上线新版[智能体应用](../concepts/agent-application.md)（Agent 2.0），将知识库、MCP 统一为工具，由智能体自主规划调用时机与顺序，并完整展示模型思考与工具调用全过程。文件问答支持全文引用、切片检索和自定义处理三种模式。 -- **工作流应用**：支持批量节点、[多模态](../concepts/multimodal.md)生成节点（生成图像/视频/音频）、异步运行模式（文本生成模式下后台执行并返回 Task ID）、Dify 工作流一键导入、[多模态](../concepts/multimodal.md)数据节点（文档/图片/视频/音频解析）等。 +- **[工作流](../concepts/workflow.md)应用**：支持批量节点、[多模态](../concepts/multimodal.md)生成节点（生成图像/视频/音频）、异步运行模式（文本生成模式下后台执行并返回 Task ID）、Dify [工作流](../concepts/workflow.md)一键导入、[多模态](../concepts/multimodal.md)数据节点（文档/图片/视频/音频解析）等。 - **高代码应用**：2025 年 9 月 24 日上线，支持基于 Python 项目结构部署 AI 后端服务，内置自动化运维、可观测性及日志服务等企业级能力。 - **MCP 服务**：2025 年 4 月 9 日新增 MCP 市场与 MCP 管理功能，可开通预置 MCP 服务或部署自定义 MCP 服务；8 月 13 日新增外部调用功能，支持一键配置到第三方应用或通过 MCP SDK 调用。 @@ -40,10 +40,10 @@ ## 应用调用与发布 -- **API 调用**：2025 年 11 月 3 日起支持通过 Responses API 调用百炼应用，提供同步调用 API（实时交互，可复用 OpenAI 代码库）与[异步调用](../concepts/async-invocation.md) API（设置 `background=true` 立即返回任务 ID）。调用工作流和[智能体编排](../concepts/agent-orchestration.md)应用时需传入自定义参数。 +- **API 调用**：2025 年 11 月 3 日起支持通过 Responses API 调用百炼应用，提供同步调用 API（实时交互，可复用 OpenAI 代码库）与[异步调用](../concepts/async-invocation.md) API（设置 `background=true` 立即返回任务 ID）。调用[工作流](../concepts/workflow.md)和[智能体编排](../concepts/agent-orchestration.md)应用时需传入自定义参数。 - **发布渠道**：支持微信、钉钉分享渠道（创建钉钉 AI 机器人或微信公众号 AI 机器人）；支持音视频实时互动（将图文对话应用转为音视频实时互动应用，提供 H5/APP 调试窗口，通过音视频 SDK 发布到 WEB/iOS/Android）。 - **应用观测**：2024 年 10 月 24 日新增应用观测能力，支持端到端查看应用处理流程；2026 年 2 月 6 日上线新版应用[评测](../concepts/evaluation.md)，支持智能体、工作流和自定义三种类型[评测](../concepts/evaluation.md)集。 -- **长期记忆**：2026 年 1 月 31 日上线新版长期记忆与用户画像管理 API，支持多应用共享同一记忆库、自动提取关键信息、语义检索优化及完整用户画像管理。 +- **[长期记忆](../concepts/long-term-memory.md)**：2026 年 1 月 31 日上线新版[长期记忆](../concepts/long-term-memory.md)与用户画像管理 API，支持多应用共享同一记忆库、自动提取关键信息、语义检索优化及完整用户画像管理。 ## 限制与注意事项 @@ -51,7 +51,7 @@ - 大模型调用产生[计费](../concepts/billing.md)，平台提供限时免费额度，可在模型广场查看各模型系列详情。 - QwQ 系列模型在[智能体应用](../concepts/agent-application.md)中不支持插件、流程、音视频交互能力。 - 文档解析耗时与文档大小相关，知识文档导入通常 1~6 分钟，知识库解析通常 1~2 分钟，需耐心等待。 -- [智能体编排](../concepts/agent-orchestration.md)应用已于 2025 年 8 月 12 日随工作流应用界面升级而下线，相关需求请使用新版智能体应用或工作流应用。 +- [智能体编排](../concepts/agent-orchestration.md)应用已于 2025 年 8 月 12 日随工作流应用界面升级而下线，相关需求请使用新版[智能体应用](../concepts/agent-application.md)或工作流应用。 - Assistant API 处于下线中状态，如需全代码开发高度定制化 RAG 应用请关注官方公告。 ## 来源文档 @@ -72,6 +72,19 @@ + + + + + + + + + + + + + diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/support.md b/skills/bailian-docs-llm-wiki/wiki/guides/support.md deleted file mode 100644 index e235c257..00000000 --- a/skills/bailian-docs-llm-wiki/wiki/guides/support.md +++ /dev/null @@ -1,106 +0,0 @@ -# support - -本页汇总阿里云百炼平台的常见问题解答、服务协议与技术支持渠道，帮助开发者快速定位使用中遇到的问题并找到对应解决方案。内容涵盖[计费](../concepts/billing.md)、API/SDK、模型训练、模型幻觉处理以及平台相关协议。 - -## [计费](../concepts/billing.md)常见问题 - -百炼平台采用按量后付费模式（分钟级出账、按月结算），部分模型支持预付费（节省计划与资源包）。关键要点： - -- 模型调用价格与模型部署/训练[计费](../concepts/billing.md)分开计算，具体单价参见百炼控制台模型市场 -- 开通服务要求阿里云账户余额不小于 0 元 -- 万相会员与百炼 API [计费](../concepts/billing.md)体系相互独立，会员权益不适用于 API 调用 -- 费用明细与发票申请通过阿里云费用与成本控制台操作 - -详细[计费](../concepts/billing.md)说明参见[常见问题](../../raw/model-user-guide/support/faq-about-alibaba-cloud-model-studio.md)中的[计费](../concepts/billing.md)相关章节。 - -## API/SDK 使用 - -百炼支持 Java 和 Python SDK，API 调用返回标准状态码标识结果。开发者常遇问题： - -| 问题 | 解决方式 | -|------|----------| -| Completion API 报错 100004（参数缺失） | 检查必填参数是否完整、格式是否正确（注意 JSON body 字段名大小写） | -| doc_reference_type 不生效 | 该参数仅旧版应用有效；新版应用需在控制台开启"展示回答来源"开关 | -| Assistant API 不支持多函数串行调用 | 当前限制，可创建多个 Assistant 分别处理 | -| Assistant API 无 memory 能力 | 当前暂不支持 | - -错误码完整列表与 SDK 安装指引请参见[常见问题](../../raw/model-user-guide/support/faq-about-alibaba-cloud-model-studio.md)中的 API/SDK 相关章节。 - -## 模型训练与选型 - -### 模型选择 - -- **qwen-turbo**：侧重速度与资源效率，费用更低，适合对响应速度要求高的场景 -- **qwen-max**：侧重顶级性能与全面知识，适合对精度和复杂任务处理能力要求严格的场景 -- 千问系列支持 14 种语言（中文、英文、阿拉伯语、西班牙语、法语等） -- qwen-vl-plus 已支持图片训练微调 - -### 训练注意事项 - -- 仅使用垂直领域数据做 SFT 可能导致模型遗忘通用知识 -- 训练数据需保证：任务定义清晰、数据质量高（准确简洁）、数据多样性（同一语义多种 [prompt](prompt.md) 表达） -- 循环次数与数据量无固定规律，需通过实验确定；不应仅通过 loss 判断是否过拟合，最终效果以人工评估为准 -- 训练后的模型不支持导出；本地训练的模型不支持上传 - -### 模型幻觉处理 - -降低幻觉的主要手段（按实施难度排序）： - -1. **选择更强模型**：Max > Plus > Turbo -2. **[提示词工程](../concepts/prompt-engineering.md)**：限定回答范围、要求引用来源、分步骤引导 -3. **RAG（[检索增强生成](../concepts/rag.md)）**：让模型基于检索到的知识回答，严格限制范围 -4. **插件/MCP**：数值计算等任务通过工具完成，避免模型直接处理 -5. **参数调优**：降低 temperature/top_k/top_p，降低 max_tokens 防止过度生成 -6. **后处理验证**：通过 AI 二次校验回复正确性（增加成本和延迟） - -## 产品使用要点 - -- 百炼服务需**分地域开通**，使用主账号在控制台切换目标地域后自动开通 -- 服务开通后暂不支持关闭；删除 API-Key 即可停止调用 -- 数据隔离通过[业务空间](../concepts/workspace.md)权限管理实现，不同子账号分配不同空间权限 -- 阿里云不会将用户数据用于模型训练，传输数据经 AES-256 加密；根据法规要求会存储调用数据 -- 控制台最多展示 100 条历史对话记录，不设时间限制 -- 模型生成速度非固定值，受服务负载和请求并发影响；限流触发后等待时间取决于具体限流值（如 120 RPM 则约等待 0.8 秒） - -## 服务协议 - -百炼平台涉及的主要协议包括： - -- 阿里云百炼服务协议（平台总协议） -- 阿里云百炼模型推理服务等级协议（SLA） -- 阿里云百炼服务特别说明 -- 开源模型协议条款说明 -- 三方模型服务协议和使用条款清单 - -完整协议链接请参见[相关协议](../../raw/model-user-guide/support/related-agreements.md)。 - -## 技术支持渠道 - -| 需求类型 | 联系方式 | -|----------|----------| -| 业务合作/售前咨询 | 服务热线 4008013260 或官网售前咨询 | -| 产品使用问题/售后 | 官网售后服务 | -| 合作协议申请 | 提交阿里云工单 | - -## 来源文档 - -- [常见问题](../../raw/model-user-guide/support/faq-about-alibaba-cloud-model-studio.md) -- [相关协议](../../raw/model-user-guide/support/related-agreements.md) - - - - - - - - - - - - - - - - - - diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/test-1.md b/skills/bailian-docs-llm-wiki/wiki/guides/test-1.md index 6f95466a..3f88e57d 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/test-1.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/test-1.md @@ -1,71 +1,92 @@ # test 1 -本页汇总阿里云百炼平台在计费与成本方面的核心规则，涵盖新人免费额度、模型调用价格、训练与部署计费、节省计划与资源包，以及账单查询与成本管理。面向开发者，帮助你在调用、微调、部署模型时准确预估费用并规避意外扣费。 +本页汇总阿里云百炼平台的计费与成本管理体系，覆盖新人免费额度、模型调用（推理）价格、模型训练与部署计费、账单查询与分账，以及节省计划与资源包等优惠方案。面向开发者的核心问题是：调用会不会花钱、花多少钱、如何抵扣、如何查账与止损。 -## 计费优先级与整体逻辑 +## 计费体系总览 -无论使用哪种付费方式，系统在实时调用时按固定优先级自动抵扣，无需手动指定： +百炼的费用来源可分为四类： -**免费额度 > 资源包 > 其他模型节省计划 > AI 通用型节省计划 > 按量付费** +| 费用类型 | 计费方式 | 出账粒度 | +| --- | --- | --- | +| 模型推理（调用） | 按输入/输出 Token 按量计费，部分模型阶梯计价 | 分钟级（约 2~10 分钟） | +| 模型训练（调优） | 按训练 [Token 计费](../concepts/token-billing.md) | 小时级 | +| 模型部署 | 按使用时长计费（预置吞吐 TPM 或模型单元） | 部署为"运行中"即持续计费 | +| 附加功能 | 如联网搜索按调用次数单独计费 | 与推理费用分开出账 | -- 免费额度仅抵扣模型**实时推理**费用，不抵扣 Batch 调用、模型调优、模型部署等场景。 -- 超出各类额度后的用量自动转为按量付费，从阿里云账户余额扣除。 - -> **注意**：账户欠费（可用额度 < 0）时，即使某模型仍有免费额度也无法调用；请提前配置余额预警或消费限额。 +费用抵扣优先级固定为：**免费额度 > 资源包 > 其他模型节省计划 > AI 通用型节省计划 > 按量付费**，系统自动按序抵扣，无需手动设置。 ## 新人免费额度 -首次开通阿里云百炼时，平台自动为各模型发放新人专属免费额度，详见 [新人免费额度](../../raw/model-user-guide/test-1/new-free-quota.md)。关键规则： +首次开通百炼时自动发放各模型的免费额度（通常每模型 100 万 Token），详见[新人免费额度](../../raw/model-user-guide/test-1/new-free-quota.md)。关键规则： + +- **有效期 90 天**：从开通百炼、模型发布或申请通过之日起算（取较晚者），到期即作废，不补发、不延期；同一实名主体重新注册无法再领。 +- **仅抵扣实时推理**：不抵扣 Batch 调用、模型调优、模型部署及自定义模型。 +- **额度隔离**：不同模型（含同一模型的不同快照版本，如 `qwen-max` 与 `qwen-max-2026-05-17`）额度相互独立，用完不会自动切换，需手动改 `model` 参数。 +- **主账号与 RAM 子账号共享**同一模型额度。 +- **Token Plan / Coding Plan 专属 API Key 不消耗免费额度**，直接按量付费；如需用免费额度请改用通用 API Key。 -- **有效期**：30～90 天，自开通或模型申请通过之日起计算。自 2025 年 9 月 8 日 11 点起首次开通的用户统一调整为 90 天。到期或耗尽后自动失效，不支持补发、延期或重置。 -- **地域限制**：仅华北2（北京）地域且服务部署范围为中国内地的模型享有免费额度（新加坡国际部署同理），其他地域和部署范围无免费额度。 -- **额度独立**：每个模型（含不同快照版本，如 `qwen-max` 与 `qwen-max-2026-05-17`）拥有独立额度（通常 100 万 Token），不可跨模型合并或转移；主账号与 RAM 子账号共享同一模型的额度。 -- **免费额度用完即停**：开启后额度耗尽即停止响应并返回 `AllocationQuota.FreeTierOnly`，不再扣费。全新未认证用户默认额度耗尽后无法继续调用，需完成认证并充值。 +> **注意**：免费额度的适用地域在文档中存在不一致表述——开头称"仅华北2（北京）地域模型享有免费额度，其他地域无免费额度"，而 FAQ 部分称"仅华北2（北京）地域的模型以及仅新加坡地域的模型享有免费额度"。以控制台免费额度页面实际显示为准。 -> **注意**：Token Plan / Coding Plan 专属 API Key **不消耗免费额度**，会直接按量付费；如需使用免费额度请改用通用 API Key。 +**防止意外扣费**：已认证用户额度耗尽后继续调用会直接扣费，可开启**免费额度用完即停**功能（耗尽时返回错误码 `AllocationQuota.FreeTierOnly`）。未认证用户额度耗尽后调用会直接失败，需实名认证并充值。 -## 模型调用价格（按量付费） +## 模型调用价格 -模型调用默认按量计费，价格表见 [模型调用价格](../../raw/model-user-guide/test-1/model-pricing.md)。要点： +模型调用默认按量计费，按输入/输出 Token 分别计价，完整价目见[模型调用价格](../../raw/model-user-guide/test-1/model-pricing.md)。要点： -- **阶梯计费**：部分模型的单价取决于单次请求的输入 Token 总量，该请求的所有 Token 均按落入的阶梯单价结算（K=1,000，M=1,000,000）。例如 `qwen3-max` 分 0 **注意**：免费额度和节省计划均**不抵扣**模型训练与部署费用；后付费部署在账户欠费后仍会继续保留并计费 24 小时。 +### 部署计费 + +- **预置吞吐（TPM）**：`费用 = 使用时长 × (输入 TPM 单价 × 输入 TPM + 输出 TPM 单价 × 输出 TPM)`；后付费按小时、预付费按天。超出购买 TPM 时按创建时选择的溢出策略处理：「自动溢出」切换为按量付费（响应 Header 带 `x-dashscope-ptu-overflow:true`），「仅使用 PTU 容量」返回 429。 +- **模型单元**：`费用 = 使用时长（小时）× 模型单元数量 × 单元单价`，预付费按月；首月提前退订按日单价 1.2 倍结算。PD 分离模式（Prefill/Decode 分节点）可降低首 Token 延迟、提高吞吐，价格更高。 +- **欠费行为**：后付费部署欠费后资源保留并继续计费 24 小时，超时后停止计费并释放底层资源，但部署任务保留；补足欠费后自动恢复（恢复即恢复计费）。不再使用务必**删除部署任务**。 + +> **注意**：模型部署按时长计费，部署状态为**运行中**即持续扣费，与是否有 API 调用无关。这是"没调用也产生费用"的最常见原因。 ## 节省计划与资源包 -成本优化方案见 [节省计划与资源包](../../raw/model-user-guide/test-1/savings-plan-and-resource-package.md)。 +用于优化按量付费成本，详见[节省计划与资源包](../../raw/model-user-guide/test-1/savings-plan-and-resource-package.md)。选型建议： -- **AI 通用型节省计划（推荐）**：承诺每月消费金额换取阶梯折扣，最高 5.3 折，可抵扣阿里直供的全部模型。以**动态月**为周期发放额度（非自然月），当月未用完自动清零、不累积。支持抵扣模型调用、原生工具调用、上下文缓存、批量推理；不抵扣调优、部署、联网搜索插件、MCP 广场等。 -- **其他模型节省计划**：一次性购买固定金额，抵扣特定模型系列（如大语言模型、语音模型、向量排序、万相），适合用量集中场景，折扣一般不如通用型。 -- **资源包**：预购具体 Token / 张数等资源量，抵扣单个特定模型超出免费额度后的实时推理用量，到期作废。 +- **AI 通用型节省计划（首选）**：承诺月消费金额（1000 元起）换取阶梯折扣，最高 5.3 折；可抵扣阿里直供全部模型的调用、原生工具调用、上下文缓存、批量推理费用。额度按**动态月**发放，当月未用完自动清零、不累积。不抵扣模型调优/部署费用及联网搜索等插件费用。 +- **其他模型节省计划**：针对特定模型系列（大语言模型、千问语音、向量及排序、万相），一次性购买固定金额，折扣普遍弱于通用型。 +- **资源包**：预购具体 Token 量，仅抵扣单个特定模型超出免费额度后的实时推理用量，到期剩余作废。 -> **注意**：若开启了**免费额度用完即停**（安心模式），免费额度耗尽后服务停止，节省计划将无法继续抵扣；需手动关闭该功能后才能切换到节省计划。 +> **注意**：开启"免费额度用完即停"后，额度耗尽即停止服务、不产生按量账单，节省计划将无法抵扣。需要节省计划生效时须手动关闭该开关。 ## 账单查询与成本管理 -账单查询、分账与欠费处理见 [账单查询与成本管理](../../raw/model-user-guide/test-1/bill-query-and-cost-management.md)。 +详见[账单查询与成本管理](../../raw/model-user-guide/test-1/bill-query-and-cost-management.md)。 + +- **查账入口**：控制台"费用概览"仅展示大模型推理费用；训练、知识库等需到阿里云"账单详情"页，产品名称筛选**大模型服务平台百炼**。 +- **账单实例 ID 解读**：推理账单以分号分隔，格式为 `ApiKeyID;业务空间ID;模型名称;输入/输出类型;调用渠道;免费额度用完即停标识`（无 ApiKeyID 通常为控制台调用）；训练账单以感叹号分隔，格式为 `业务空间ID!地域!训练任务标识`。调用渠道 `app`=代码调用、`bmp`=控制台模型体验、`assistant-api`=Assistant API。 +- **分账**：给[业务空间](../concepts/workspace.md)绑定标签并在费用标签页启用，T+1 天后可在账单"实例标签"列按部门/项目归集费用。 +- **欠费影响**：可用额度 < 0 即欠费，按量付费服务暂停；**欠费期间免费额度、节省计划、资源包即使有余量也无法调用**，Coding Plan / Token Plan 套餐可继续使用但自动续费会失败。 +- **扣款模式**：按量付费非实时扣款，采用"预占 + 月结"，次月初出最终账单实际扣款。 +- **止损手段**：删除不再使用的 API Key、下线不用的模型部署、检查代码中的 `enable_search` 参数、设置高额消费预警与预算管理。 -- **出账时效**：大模型推理分钟级出账（通常 2~10 分钟），批量推理、模型训练、知识库等小时级出账。 -- **账单详情**：核心是"实例 ID（出账粒度）"字段，格式为 `ApiKeyID;业务空间ID;模型名称;输入/输出类型;调用渠道;免费额度用完即停标识`，可据此定位产生费用的模型与渠道（`app` 代码调用、`bmp` 控制台体验、`assistant-api`）。 -- **分账管理**：给业务空间绑定标签，按部门/项目归集费用，配置后 T+1 生效。 -- **欠费与停止计费**：欠费按商品维度判定，Coding Plan / Token Plan 套餐额度独立于余额、欠费期间仍可用。停止计费需停止调用、下线部署模型、删除 API Key 或退订套餐。 +## 限制与注意事项 -> **注意**：按量付费采用"预占+月结"模式，并非实时扣款；系统先冻结额度，月账期结束后（次月初）生成最终账单实际扣款。此外，`enable_search` 等联网搜索附加功能按次单独计费，可能在未主动操作时仍产生费用。 +- 账单仅在调用**结束**后生成：推理约 2~10 分钟，训练/批量推理/知识库为小时级，高峰期可能延迟。 +- 控制台免费额度显示为分钟级更新且需手动刷新，页面显示有余量不代表实际未耗尽。 +- 预付费（预置吞吐）订单不可提前终止；退订按 1.5 倍系数结算已使用部分。 +- 公测、邀测（非商业化）模型不产生账单记录。 +- 代金券不计入账户可用额度中的现金余额；后付费场景余额须大于当月冻结金额才能正常调用。 ## 来源文档 diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/token-plan-guide.md b/skills/bailian-docs-llm-wiki/wiki/guides/token-plan-guide.md index 3bb3d847..26ffdd6d 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/token-plan-guide.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/token-plan-guide.md @@ -1,102 +1,128 @@ # token plan guide -Token Plan 团队版与 Coding Plan 是百炼面向 AI 编程/智能体工具的两类订阅服务：Token Plan 团队版以 Credits 统一计量、按 Token 消耗抵扣，支持文本与图像生成模型并提供团队管理后台；Coding Plan 面向个人开发场景，按模型调用次数计费并设有请求限额。两者的 API Key 与 Base URL 完全隔离、互不相通，接入前需先明确使用的是哪种套餐。 +Token Plan 是阿里云百炼推出的 AI 大模型订阅服务，以 Credits 统一计量，一份订阅即可在 Claude Code、Cursor、Qwen Code、Qoder、OpenClaw 等主流 AI 编程和智能体工具中使用。产品分为面向个人开发者的个人版和面向企业团队的团队版，覆盖文本生成、图像生成、视频生成等模型以及联网搜索、代码解释器等 Harness 工具。本页汇总产品定位、套餐限额、接入方式、[多模态](../concepts/multimodal.md)/Harness 扩展及常见排错要点。 -## 两种套餐对比 +> **注意**：Token Plan 目前仅支持**华北2（北京）**地域，需在百炼控制台左上角切换地域后购买使用。Coding Plan 与 Token Plan 是两个独立订阅产品，不可迁移或升级；Coding Plan Lite 已于 2026 年 3 月 20 日停止新购、4 月 13 日停止续费与升级，官方推荐迁移到 Token Plan，详见 [Token Plan 概述](../../raw/model-user-guide/token-plan-guide/token-plan-overview.md)。 -| 维度 | Token Plan 团队版 | Coding Plan | +## 版本与套餐 + +### 个人版 + +个人版提供 Lite / Standard / Pro 三个档位（限时价 39 / 139 / 499 元/月），采用 **5 小时 + 7 天两层固定窗口限额**（单位 Credits）： + +| 档位 | 5 小时限额 | 7 天限额 | 并发 Agent | +| --- | --- | --- | --- | +| Lite | 700 Credits | 2,500 Credits | 1-2 个 | +| Standard | 3,000 Credits | 10,000 Credits | 3-4 个 | +| Pro | 12,000 Credits | 40,000 Credits | 6-8 个 | + +- 任一层限额触顶即暂停服务，可等待窗口结束重置、使用额度重置功能，或购买**用量包**（100 元/2 万 Credits，不受窗口限额约束，需先有有效订阅，最多持有 5 个，有效期 1 个月）。 +- 窗口自首次调用起计时，7 天限额**不是**固定日历日重置。 +- 窗口内未用完的额度不结转。 +- 支持升配（补差价、限额立即提升），不支持降配和退订。 + +详见 [概述（个人版）](../../raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-overview.md)。 + +### 团队版 + +团队版以**席位（坐席）**为最小订阅单位，采用**月度总额度制**，无 5 小时/7 天窗口限额： + +| 坐席 | 限时价 | 月度额度 | | --- | --- | --- | -| 适用场景 | 一人公司/团队/企业日常办公 | 个人开发场景 | -| 支持模型 | 文本生成 + 图像生成 | 文本生成模型 | -| 计费方式 | 按 Token 消耗抵扣 Credits | 按模型调用次数 | -| 使用频次 | 无每 5 小时/每周限额 | 有每 5 小时/每周/每月限额 | -| 高峰期性能 | 多租户隔离，不排队 | 高峰期可能排队 | -| 数据安全 | 承诺不使用数据训练模型 | 用户数据授权用于服务改进 | +| 标准坐席 | 150 元/坐席/月 | 25,000 Credits | +| 高级坐席 | 550 元/坐席/月 | 100,000 Credits | +| 尊享坐席 | 1,398 元/坐席/月 | 250,000 Credits | +| 共享用量包 | 5,000 元/个 | 625,000 Credits（有效期 1 个月，跨坐席共享） | -> **注意**：两个计划互相独立，不支持互转（即使补差价也不行），可同时订阅、各自计费。详见 [Token Plan（团队版）概述](../../raw/model-user-guide/token-plan-guide/token-plan-overview.md) 与 [Coding Plan概述](../../raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan.md)。 +团队版额外提供：席位分配/回收、成员用量分析、SSO（SAML 2.0）/钉钉登录、**承诺不使用对话数据训练模型**、多租户隔离高峰不排队。个人版与团队版可同时购买，独立计费、额度不共享。管理操作详见 [团队管理](../../raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-management.md)。 ## 支持的模型 -Token Plan 团队版的模型清单为**精确字符串白名单**，必须逐字符完全匹配，版本号/子型号任何差异均视为不支持，禁止版本兼容推理。 +- **文本/推理模型**：qwen3.8-max-preview（预览版，限时 Credits 1 折；个人版另有夜间 22:00-08:00 折上折）、qwen3.7-max、qwen3.7-plus、qwen3.6-plus/flash、deepseek-v4 系列、glm-5 系列、kimi-k2.5 及以上、MiniMax-M2.5 等（团队版模型列表更全）。 +- **图片生成**：qwen-image-2.0（团队版）、wan2.7-image、wan2.7-image-pro。 +- **视频生成**：happyhorse-1.1-t2v / i2v / r2v。 -- **千问**：qwen3.7-max（限时活动）、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash、qwen-image-2.0、qwen-image-2.0-pro -- **万相**：wan2.7-image、wan2.7-image-pro -- **DeepSeek**：deepseek-v4-pro、deepseek-v4-flash、deepseek-v3.2 -- **月之暗面**：kimi-k2.7-code、kimi-k2.6、kimi-k2.5 -- **智谱 AI**：glm-5.2、glm-5.1、glm-5 -- **MiniMax**：MiniMax-M2.5 +> **注意**：个人版与团队版支持的模型列表不完全一致（例如 kimi、glm 多版本、qwen-image-2.0 仅出现在团队版列表中），且预览模型权益可能随运营调整，购买前以控制台模型列表页为准。 -Coding Plan Pro 套餐的推荐模型为 qwen3.7-plus、qwen3.6-plus、kimi-k2.5（均支持图片理解）、glm-5、MiniMax-M2.5，更多模型见 [Coding Plan概述](../../raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan.md)。 +## 接入方式（关键配置） -> **注意**：两套清单的白名单不完全一致，且 Coding Plan Lite 套餐已于 2026 年 3 月 20 日停止新购、4 月 13 日停止续费与升级。调用时务必以对应套餐控制台的实时清单为准。 +三步接入：订阅套餐 → 获取 API Key → 配置 AI 工具，详见 [快速开始（个人版）](../../raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-quick-start.md) 与 [快速开始（团队版）](../../raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-quickstart.md)。 -## 快速接入（三步） +- **API Key**：Token Plan 专属 Key 以 `sk-sp-` 开头，仅在生成/重置时完整显示一次。 +- **Base URL**： + - OpenAI 兼容：`https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` + - Anthropic 兼容：`https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` +- **三套体系完全隔离，必须配套使用**： + - Token Plan：`sk-sp-` Key + `token-plan.cn-beijing.maas.aliyuncs.com` + - Coding Plan：专属 Key + `https://coding.dashscope.aliyuncs.com/v1`（或 `/apps/anthropic`） + - 按量付费：通用 `sk-` Key + `https://dashscope.aliyuncs.com/compatible-mode/v1` -以 Token Plan 团队版为例，详见 [快速开始](../../raw/model-user-guide/token-plan-guide/token-plan-quickstart.md)： + 混用会导致 401/403 鉴权失败，或走按量计费通道产生意外扣费。 +- **RAM 用户**：需主账号授予 `AliyunTokenPlanReadOnlyAccess` 或 `AliyunTokenPlanFullAccess` + BSS 相关策略，并在百炼控制台账号管理中分配管理员或订阅套餐权限。 -1. **订阅**：在购买页选择坐席类型、数量和订阅周期。RAM 子账号订阅前需主账号授予 `AliyunBailianFullAccess` 权限。 -2. **获取 API Key 和 Base URL**：分配席位后为成员生成专属 API Key（Token Plan 以 `sk-sp-` 开头，与通用 `sk-` 不可混用；仅首次显示一次，需立即保存）。 -3. **接入 AI 工具**：支持 Claude Code、Qwen Code、OpenCode、OpenClaw、Cursor、Codex、Qoder、Cline、Kilo CLI 等。 +## 扩展能力 -### Base URL 对照 +### [多模态](../concepts/multimodal.md)生成模型 -| 套餐 / 协议 | Base URL | -| --- | --- | -| Token Plan · OpenAI 兼容 | `https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` | -| Token Plan · Anthropic 兼容 | `https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` | -| Coding Plan · OpenAI 兼容 | `https://coding.dashscope.aliyuncs.com/v1` | -| Coding Plan · Anthropic 兼容 | `https://coding.dashscope.aliyuncs.com/apps/anthropic` | -| 按量付费 · OpenAI 兼容 | `https://dashscope.aliyuncs.com/compatible-mode/v1` | +图像/视频生成模型使用独立接口（`/api/v1/services/aigc/...`），不能通过文本模型的 Base URL 直接调用，需通过工具的扩展机制（Claude Code 的 Slash Command、Codex/Qwen Code/Qoder 的 Skill、OpenCode 的 Agent）以 curl 方式接入；视频生成为异步接口，流程为"提交任务 → 轮询 `/api/v1/tasks/` → 下载视频"。完整示例见 [接入多模态生成模型](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-multimodal-gen.md)。 -> **注意**：Token Plan、Coding Plan、按量付费三者的 API Key 与 Base URL 必须配套使用。混用会导致走按量计费通道产生意外扣费，或返回 401/403 鉴权失败。 +### Harness 工具 -## 工具调用与扩展能力 +部分 Qwen 模型内置 Harness 工具：联网搜索（web_search）、代码解释器（code_interpreter）、网页抓取（web_extractor）、文搜图（t2i_search）、图搜图（i2i_search）。当前仅 qwen3.7 / qwen3.8 系列支持原生工具调用（qwen3.7-max 不支持搜图类工具），切换模型后对话中直接提问即可自动调用，按成功调用次数从 Credits 抵扣。仅适用于 Token Plan，不适用于 Coding Plan，详见 [接入 Harness 工具](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-harness-tool.md)。 -- **模型内置工具**：qwen3.7-max、qwen3.7-plus、qwen3.6-plus、qwen3.6-flash 通过 Responses API 内置联网搜索、代码解释器、网页抓取、以图搜图、文搜图 5 个工具，不额外收费，token 消耗统一从套餐 Credits 抵扣。 -- **MCP 服务**：其他模型（如 deepseek-v3.2、glm-5）通过百炼 MCP 广场接入工具。联网搜索 MCP 前 2000 次调用免费，之后按 29 元/千次计费。接入 MCP 用的是**百炼通用 API Key（`sk-xxx`）**，而非套餐专属 Key。 -- **图像生成模型**：不在文本模型清单展示，需通过工具的 Skill / Slash Command / Agent 机制调用 `multimodal-generation` API，详见 [接入多模态生成模型](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-multimodal-gen.md)。 -- **视觉理解**：qwen3.6-plus、qwen3.5-plus、kimi-k2.5 原生支持视觉；glm-5、MiniMax-M2.5 等纯文本模型可通过 Skill/Agent 辅助获得视觉能力。OpenCode/OpenClaw 需在配置中显式声明 `modalities`/`input` 为 `["text","image"]`。 +### 视觉理解 -## Credits 计费与额度 +qwen3.8-max-preview、qwen3.7-plus、qwen3.6-plus、kimi-k2.5 等模型原生支持视觉，直接传图即可；glm-5、MiniMax-M2.5、qwen3-coder 系列等纯文本模型可通过本地 Skill/Agent（转调 qwen3.7-plus）间接获得视觉能力。OpenCode 需在配置中显式声明 `modalities.input: ["text", "image"]` 才能启用视觉。详见 [添加视觉理解能力](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/add-vision-skill.md)。 -Token Plan 团队版单次请求消耗的 Credits **并非固定值**，由模型类型、Token 用量、思考模式及工具调用动态决定。多轮对话中上下文持续累积，消耗会随之上升；部分模型按上下文长度阶梯计费，长上下文可能进入更高价位档。 +### 联网搜索 MCP -抵扣顺序：坐席套餐月度额度 → 共享用量包（多个时优先扣最近到期的）→ 全部用尽后服务暂停至下一计费周期。 +除 Harness 原生搜索外，也可通过百炼 MCP 广场的联网搜索 MCP（Streamable HTTP：`https://dashscope.aliyuncs.com/api/v1/mcps/WebSearch/mcp`）为 Qwen Code、Claude Code、OpenClaw、OpenCode 等工具扩展搜索能力。 -> **注意**：续费/续订只延长有效期或预定下期额度，**不会叠加补充到当前计费周期**。当期额度用尽需立即恢复时，应购买共享用量包、升级坐席或加购坐席（加购后需分配给成员才能使用）。 +> **注意**：MCP 服务鉴权使用**百炼通用 API Key（`sk-` 开头）**而非 Token Plan 专属 Key，前 2000 次调用免费，之后按 29 元/千次计费，详见 [联网搜索](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/web-search-mcp.md)。 -控制消耗建议：任务切换时及时开启新会话、清理无关历史；对长文档/大代码库按需拆分输入；在控制台订阅页用量明细关注实时消耗趋势。 +## 使用限制与注意事项 + +1. **严禁 API 调用**：仅限在编程/智能体工具中交互式使用，禁止用于自动化脚本、应用后端或批量调用；违规可能导致订阅暂停或 API Key 封禁。 +2. **数据授权差异**：个人版（及 Coding Plan）的输入输出会用于服务改进与模型优化；团队版承诺不使用对话数据训练模型。 +3. **账号规范**：套餐/席位为个人专享，禁止共享 API Key。 +4. **不支持降配**；个人版不支持退订；团队版可按席位退订（已有用量消耗的席位除外）。 +5. **续费仅延长有效期**，不叠加当前周期额度；订阅到期重购或退订重购后 API Key 会变更，需在工具中重新配置。 +6. **[限流](../concepts/rate-limit.md)按主账号维度计算**，RAM 子账号、[业务空间](../concepts/workspace.md)和 API Key 的调用量合并计算。 ## 常见报错速查 -- **401 Invalid API-key / invalid access token**：误用了通用 Key 或其他套餐的 Key/Base URL、订阅过期、或 Key 复制不完整含空格。核对套餐专属 Key 与配套 Base URL，必要时重置。 -- **404 model not found / model not [support](support.md)ed**：模型名拼写或大小写错误，或不在套餐白名单内。 -- **400 url error / Range of input length**：Base URL 路径与协议不匹配（Anthropic 端点以 `/apps/anthropic` 结尾，OpenAI 端点以 `/compatible-mode/v1` 或 `/v1` 结尾），或输入超出上下文长度（新建会话或切换更长上下文模型）。 -- **429 quota exceeded**：套餐额度用尽（加购/等待重置）或触发 TPS/TPM 限流（限流按主账号维度合并计算，等待约一分钟后平滑重试）。 -- **Coding Plan 限额类**：`hour/week/month allocated quota exceeded` 分别对应每 5 小时（滚动恢复）、每周一 00:00 重置、每月订阅日重置。 +| 报错 | 典型原因 | 处理 | +| --- | --- | --- | +| 401 InvalidApiKey / invalid access token | 混用了通用 Key、Coding Plan Key 或错误 Base URL；订阅过期 | 使用 `sk-sp-` Key + Token Plan 专属 Base URL | +| 404 model not found or not supported | 模型名拼写/大小写错误或不在支持列表 | 对照套餐支持的模型 ID（精确匹配） | +| 400 Range of input length | 输入超出上下文长度 | 新建会话或压缩上下文 | +| 400 调用图像/视频模型报错 | [多模态](../concepts/multimodal.md)模型不能走文本 Base URL | 通过 Skill/Slash Command/Agent 扩展机制接入 | +| 429 Requests rate limit exceeded | 请求过密触发[限流](../concepts/rate-limit.md) | 等待后重试、降低频率 | +| 429 Allocated quota exceeded | 个人版窗口限额或团队版月额度用尽 | 购买用量包/共享用量包，或等待额度重置 | -完整报错表见 [Token Plan 常见问题](../../raw/model-user-guide/token-plan-guide/token-plan-faq.md) 与 [Coding Plan 常见问题](../../raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan-faq.md)。 +更多排错细节见 [常见问题（个人版）](../../raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-faq.md)、[常见问题（团队版）](../../raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-faq.md) 以及 Coding Plan 的 [常见问题](../../raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan-faq.md)（含 OpenCode `thinking_budget`、Claude Code 初始化等工具级问题）。 -## 团队管理与使用限制 +## 与 Coding Plan 的关系 -- **角色**：所有者、管理员（权限同所有者，可被移除/降级）、成员（仅使用分配的 Key 调用）。 -- **成员接入**：支持手动添加（仅供 API 调用）、SAML 2.0（SSO）、钉钉登录三种方式。 -- **席位操作**：分配后自动生成 API Key；回收后席位释放、原 Key 失效；加购/升级按剩余时长折算费用；退订按席位维度，已消耗用量的席位不可退订。 -- **使用范围**：仅限在兼容的 AI 编程和智能体工具中**交互式**使用，禁止用于自动化脚本或应用后端，违规可能导致订阅暂停或 API Key 封禁。 +Coding Plan 是早期的编程模型订阅产品（Pro 套餐 200 元/月，按请求次数限额：每 5 小时 6,000 次 / 每周 45,000 次 / 每月 90,000 次），仅支持精确白名单内的文本模型，Base URL 为 `coding.dashscope.aliyuncs.com`，详见 [Coding Plan概述](../../raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan.md)。 -> **注意**：Token Plan 团队版目前仅支持**华北2（北京）**地域；每个阿里云账号限购一个订阅，共享用量包需先订阅坐席套餐后才能购买、有效期 1 个月且到期清零。 +> **注意**：Coding Plan 按"模型调用次数"计量，Token Plan 按 Credits（Token 用量）计量，两者额度机制不同、Key 与 Base URL 不互通、无法互相迁移。新用户应直接选择 Token Plan。 ## 来源文档 -- [Token Plan（团队版）概述](../../raw/model-user-guide/token-plan-guide/token-plan-overview.md) -- [快速开始](../../raw/model-user-guide/token-plan-guide/token-plan-quickstart.md) -- [团队管理](../../raw/model-user-guide/token-plan-guide/token-plan-team.md) -- [常见问题](../../raw/model-user-guide/token-plan-guide/token-plan-faq.md) -- [工具调用](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-tool.md) +- [Token Plan 概述](../../raw/model-user-guide/token-plan-guide/token-plan-overview.md) - [接入多模态生成模型](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-multimodal-gen.md) - [Coding Plan概述](../../raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan.md) -- [联网搜索](../../raw/model-user-guide/token-plan-guide/coding-plan-guide/web-search-for-coding-plan.md) -- [添加视觉理解能力](../../raw/model-user-guide/token-plan-guide/coding-plan-guide/add-vision-skill.md) +- [添加视觉理解能力](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/add-vision-skill.md) - [常见问题](../../raw/model-user-guide/token-plan-guide/coding-plan-guide/coding-plan-faq.md) +- [概述](../../raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-overview.md) +- [快速开始](../../raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-quick-start.md) +- [概述](../../raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-overview.md) +- [常见问题](../../raw/model-user-guide/token-plan-guide/token-plan-personal/token-plan-personal-faq.md) +- [接入 Harness 工具](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/token-plan-harness-tool.md) +- [联网搜索](../../raw/model-user-guide/token-plan-guide/token-plan-best-practice/web-search-mcp.md) +- [快速开始](../../raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-quickstart.md) +- [常见问题](../../raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-faq.md) +- [团队管理](../../raw/model-user-guide/token-plan-guide/token-plan-team-edition/token-plan-team-management.md) diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/use-cases.md b/skills/bailian-docs-llm-wiki/wiki/guides/use-cases.md index c59a6d33..2153227d 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/use-cases.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/use-cases.md @@ -1,56 +1,100 @@ # use cases -本页汇总阿里云百炼平台的典型使用场景与实践指南，覆盖三大方向：Prompt 设计技巧（文生文、文生图、文生视频）、第三方/多供应商模型接入（DeepSeek、Kimi、GLM、MiniMax、MiMo、Stepfun 等），以及工程化最佳实践（RAG、限流应对、显式缓存、模型调优、端到端解决方案）。面向开发者，以下内容按主题组织，便于快速定位到对应的参数、调用方式和注意事项。 +本页汇总阿里云百炼平台的典型使用场景与最佳实践，涵盖三大类内容：Prompt 工程指南（文生文/文生图/文生视频）、工程化最佳实践（[限流](../concepts/rate-limit.md)应对、显式缓存、RAG 构建、模型调优）、以及第三方模型接入教程（DeepSeek、Kimi、GLM、MiniMax、MiMo、Stepfun 等）。这些实践面向开发者，帮助从模型调用到完整 AI 应用落地的全流程提效。 -## Prompt 设计与生成类场景 +## Prompt 工程指南 -针对不同模态，百炼提供了结构化的提示词方法论： +### 文生文 -- **文生文**：推荐使用「背景 / 目的 / 风格 / 语气 / 受众 / 输出」六要素的 Prompt 框架，任务描述越清晰具体，模型表现越贴近预期。控制台还提供 Prompt「自动优化」工具，可自动扩写和补充细节（该功能调用大模型，按推理费用计费）。详见 [文生文Prompt指南](../../raw/model-user-guide/use-cases/prompt-engineering-guide.md)。 -- **文生图**：核心参数为正向提示词 `prompt`、反向提示词 `negative_prompt`；文生图 V2 额外支持 `prompt_extend`（默认 `true`，开启大模型智能改写）。提示词公式分基础版（主体 + 场景 + 风格）与进阶版（增加镜头语言、氛围词、细节修饰），并配有景别、视角、风格、光线等提示词词典。详见 [文生图Prompt指南](../../raw/model-user-guide/use-cases/text-to-image-prompt.md)。 -- **文生视频 / 图生视频**：正向提示词描述画面内容与运动过程。基础公式为「主体 + 场景 + 运动」，进阶公式增加「美学控制 + 风格化」，图生视频则以「运动 + 运镜」为主。较新的 wan2.7 / wan2.6 还支持声音公式（人声/音效/BGM）、多镜头公式（镜头序号 + 时间戳 + 分镜内容）和参考生视频公式。详见 [文生视频/图生视频Prompt指南](../../raw/model-user-guide/use-cases/text-to-video-prompt.md)；第三方视频模型可参考 [Vidu视频生成Prompt指南](../../raw/model-user-guide/use-cases/third-party-model-integration-tutorial/vidu-video-generation-prompt-guide.md)（含大动态、运镜、风格等触发关键词词典）。 +[文生文Prompt指南](../../raw/model-user-guide/use-cases/prompt-engineering-guide.md) 的核心原则是构建清晰、具体、无歧义的任务描述，并推荐使用 Prompt 框架系统化组织提示词： -> **注意**：wan2.7 模型不再支持通过 `shot_type` 指定单镜头/多镜头，改由模型结合提示词自行发挥；如需一镜到底，中文写「生成单镜头」、英文写「Generate single shot.」。 +- **框架六要素**：背景、目的、风格、语气、受众、输出格式。 +- 平台提供 **Prompt 一键优化工具**（控制台 Prompt 页面的"自动优化"），可自动扩写和添加细节。注意该功能通过调用大模型实现，按模型推理费用计费。 -## 第三方与多供应商模型接入 +### 文生图 -多篇教程介绍了在百炼平台通过 **[OpenAI 兼容接口](../concepts/openai-compatible-interface.md)** 或 **DashScope SDK** 调用第三方模型，通用要点如下： +[文生图Prompt指南](../../raw/model-user-guide/use-cases/text-to-image-prompt.md) 适用于万相-文生图 V1/V2，关键参数： -- **前置条件**：先[获取 API Key](https://help.aliyun.com/zh/model-studio/get-api-key) 并配置到环境变量；部分模型需在控制台模型广场「立即开通」后才能调用。 -- **思考模式**：多数模型通过 `enable_thinking` 参数控制是否输出推理过程（`reasoning_content`）。注意 `enable_thinking` 非 OpenAI 标准参数——OpenAI Python SDK 需通过 `extra_body` 传入，Node.js SDK 作为顶层参数传入。 -- **地域差异**：不同地域的 Base URL 不同，部分供应商（硅基流动、快手万擎、月之暗面、智谱、MiniMax、小米、阶跃星辰）仅限特定地域（多为华北2（北京））。详见 [DeepSeek-阿里云](../../raw/model-user-guide/use-cases/third-party-model-integration-tutorial/deepseek-api.md) 与 [Kimi](../../raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api.md)。 +- `prompt`：正向提示词，支持中英文。 +- `negative_prompt`：反向提示词，描述不希望出现的内容。 +- `prompt_extend`（仅 V2）：是否开启大模型智能改写，默认 `true`，推荐保持默认。 -各供应商的默认思考模式行为并不一致，接入时需按模型区分： +提示词公式：基础版为"主体 + 场景 + 风格"；进阶版追加主体/场景描述、镜头语言、氛围词与细节修饰。文档还提供景别、视角、风格、光线等维度的提示词词典。 -- **默认开启思考**：`mimo-v2.5-pro`（[MiMo-小米](../../raw/model-user-guide/use-cases/third-party-model-integration-tutorial/mimo.md)）、`kimi/kimi-k2.6`/`kimi-k2.5` 默认开启，可关闭。 -- **仅思考模型**：`kimi/kimi-k2.7-code` 系列 `enable_thinking` 始终为 `true`，无法关闭；`kimi-k2.7-code-highspeed` 功能与 `kimi-k2.7-code` 一致但速度提升 5~6 倍（见 [Kimi-月之暗面](../../raw/model-user-guide/use-cases/third-party-model-integration-tutorial/kimi-api-by-moonshot-ai.md)）。 -- **默认关闭思考**：`stepfun/step-3.7-flash` 默认关闭，需显式开启，并可用 `reasoning_effort`（`low`/`medium`/`high`）控制深度。 -- **供应商差异**：同为 DeepSeek，硅基流动供应商支持更长上下文；阿里云百炼供应商限流更宽松，并支持联网搜索与上下文缓存。GLM 智谱直供的 `glm-5.2` 支持 1M 上下文，并可用 `reasoning_effort`（`max`/`high`/`none`）。 +### 文生视频 / 图生视频 -> **注意**：多篇教程标注 deepseek-v3/v3.1/v3.2/r1 系列、`MoonshotKimi-K2` 与 `kimi-k2-thinking`、`glm-4.6`/`glm-4.7`、`MiniMax-M2.1` 等模型将于 **2026年7月9日** 下架，推荐转用 `qwen3.7-plus` / `qwen3.7-max` / `qwen3.6-flash`。同时不同教程示例中出现的模型版本号存在差异（如 deepseek-v3.2 与 deepseek-v4-pro、MiniMax-M2.5 与 MiniMax-M2.7、kimi-k2.5 与 kimi-k2.7），以模型广场实际可用列表为准。 +[文生视频/图生视频Prompt指南](../../raw/model-user-guide/use-cases/text-to-video-prompt.md) 针对万相视频模型，按能力分层： -## RAG 与知识库 +- **基础公式**：主体 + 场景 + 运动；**进阶公式**追加美学控制与风格化。 +- **图生视频**：图像已确定主体与风格，提示词只需描述"运动 + 运镜"。 +- **声音公式**（wan2.7/wan2.6/wan2.5）：支持人声（内容+情绪+语调+语速+音色+口音）、音效、背景音乐三类描述。 +- **多镜头公式**（wan2.7/wan2.6）：总体描述 + 镜头序号 + 时间戳 + 分镜内容，可生成多镜头连贯叙事视频。 +- **参考生视频**（wan2.7/wan2.6）：用"图n"/"视频n"指代参考素材，可参考主体外观、动态特征、音色和背景。 -[基于LlamaIndex构建RAG应用](../../raw/model-user-guide/use-cases/build-rag-applications-based-on-llamaindex.md) 演示了在 LlamaIndex 中使用百炼检索增强服务的完整链路： +> **注意**：wan2.7 不再支持 `shot_type` 参数指定单镜头/多镜头，改由模型结合提示词决定；需要一镜到底时在提示词中写"生成单镜头"。 -- 安装 `llama-index-core`、`llama-index-llms-dashscope`、`llama-index-indices-managed-dashscope`（Python 版本要求 >=3.8 且 <=3.12）。 -- 使用 `DashScopeParse` 在线解析 .doc/.docx/.pdf 文件（单文件 <100M、页数 <1000），再通过 `DashScopeCloudIndex.from_documents` 创建知识库，`index.as_retriever()` / `index.as_query_engine()` 获取检索器与查询引擎。 +第三方 Vidu 模型有独立的提示词体系（见 Vidu视频生成Prompt指南），支持"大动态/中动态/小动态"等触发关键词、运镜控制与导演风格。 ## 工程化最佳实践 -- **限流应对**：百炼 API 按 RPM/TPM（分钟级）、RPS/TPS（瞬时）、Traffic Burst（增速）三种规则限流，按主账号维度、模型独立计算，触发后通常 1 分钟恢复。方案按改动成本由低到高分为平台配置（服务端排队等待、提升额度、PTU、Batch API）、客户端流控（重试、令牌桶、平滑限速、自适应拥塞控制）、架构兜底（模型降级、MQ 削峰）。针对突发限流推荐首选在请求头添加 `X-DashScope-Wait-Timeout`（建议 3~120 秒），并相应调大客户端超时时间。详见 [限流应对最佳实践](../../raw/model-user-guide/use-cases/rate-limiting-best-practices.md)。 -- **显式缓存**：通过在请求中添加缓存标记实现 100% 确定性命中，适合高频复用相同 Prompt、长上下文 Agent 等场景。首次写入约产生标准价格 25% 的额外开销，后续命中可节省约 90% 成本。Claude Code、OpenCode、OpenClaw 等工具通过 Anthropic 兼容端点（`/apps/anthropic`）接入时原生支持。详见 [显式缓存最佳实践](../../raw/model-user-guide/use-cases/explicit-cache-guide.md)。 -- **自定义模型**：创建自定义模型分为模型调优、模型部署、模型评测三个主步骤，配套训练数据准备、评测模板设计、调整训练策略。数据需编排为「Prompt-Completion」格式，建议至少准备 500 条并做脱敏处理。注意**调优后的模型必须先部署才能调用和评测**。详见 [自定义模型调优、部署与评测](../../raw/model-user-guide/use-cases/model-training-best-practices.md)。 +### [限流](../concepts/rate-limit.md)应对 -## 端到端解决方案 +百炼 API 按主账号 + 模型维度[限流](../concepts/rate-limit.md)，分为 RPM/TPM（分钟级配额）、RPS/TPS（瞬时频率）、Traffic Burst（增速限制）三类规则。[限流应对最佳实践](../../raw/model-user-guide/use-cases/rate-limiting-best-practices.md) 按改动成本从低到高给出三层方案： -多篇实践方案展示了如何组合百炼模型能力构建完整应用，多数基于函数计算 FC、开箱即用并提供免费试用额度： +1. **平台配置**：服务端排队等待（请求头 `X-DashScope-Wait-Timeout`，建议 3~120 秒，仅对 Throttling.BurstRate 生效）、提升限流额度、PTU、Batch API。使用排队等待时需相应上调客户端超时时间（非流式：原超时 + Wait-Timeout 值）。 +2. **客户端流控**：从基础重试、令牌桶/并发信号量、双重令牌桶/平滑限速器到自适应拥塞控制，按工程复杂度递进。 +3. **架构兜底**：模型降级（Fallback）与基于消息队列的削峰填谷。 -- **文档转视频**：结合大模型与多模态技术，将文档自动切片、生成演示文稿、语音字幕并合成视频，依赖 FFmpeg 与 Marp 工具，提供完整代码包。详见 [借助大模型将文档转换为视频](../../raw/model-user-guide/use-cases/use-llm-to-convert-document-to-video.md)。 -- **AI 智能体与工作流**：以 AI 电商客服为例，覆盖智能问答、RAG、自主决策 Agent、对话流四种应用形态（[高效搭建 AI 智能体与工作流应用](../../raw/model-user-guide/use-cases/build-ai-applications-based-on-alibaba-cloud-model-studio.md)）。 -- **视觉创作平台**：集成 Wan2.7 图像生成与 HappyHorse 视频生成，提供节点式编排、AI 导演与在线剪辑（[HappyHorse 打造一站式影视创作平台](../../raw/model-user-guide/use-cases/infinite-canvas.md)）。 -- **深度研究报告**：Qwen-Deep-Research 自动规划检索路径、多源交叉验证并生成结构化洞察报告（[深度研究：生成你的独家洞察报告](../../raw/model-user-guide/use-cases/deep-research.md)）。 -- **AI 解题批改**：基于 Qwen3-VL 视觉模型实现拍照解题与作业自动批改，支持 33 种语言（[AI 解题 + 批改：推动课程教学智变](../../raw/model-user-guide/use-cases/ai-homework-helper.md)）。 +错误码对照：`Throttling.RateQuota`（RPM/RPS 超限）、`Throttling.AllocationQuota`（TPM/TPS 超限）、`Throttling.BurstRate`（增速超限）。 + +### 显式缓存 + +显式缓存通过在请求中添加缓存标记实现 **100% 确定性命中**：首次写入产生标准价格 25% 的额外开销，后续命中节省 90% 成本。适用于高频复用相同 Prompt、Agent 长上下文管理等场景。Claude Code、OpenCode、OpenClaw 等工具通过 Anthropic 兼容端点（`/apps/anthropic`）接入百炼后自动携带 `cache_control` 标记，无需额外配置。 + +### RAG 与模型定制 + +- **基于 LlamaIndex 构建 RAG**：使用 `DashScopeParse` 解析文档（支持 .doc/.docx/.pdf，单文件 ≤100MB 且 ≤1000 页），通过 `DashScopeCloudIndex` 创建/读取知识库，`as_retriever()` / `as_query_engine()` 构建检索与问答链路。Python 版本要求 >=3.8 且 <=3.12。 +- **自定义模型调优、部署与评测**：流程为模型调优 → 模型部署 → 模型评测；调优完成的模型**必须部署后才能调用和评测**。训练数据需整理为 Prompt-Completion 格式，建议至少 500 条。 +- **文档转视频**：借助大模型完成文档切片、演示文稿生成、语音字幕合成与视频剪辑，依赖 FFmpeg 与 Marp 工具。 + +## 第三方模型接入 + +百炼模型广场提供多家第三方模型，统一通过 [OpenAI 兼容接口](../concepts/openai-compatible-interface.md)或 DashScope SDK 调用，通用模式为： + +- Base URL 采用[业务空间](../concepts/workspace.md)专属域名：`https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1`（各地域不同）。 +- 直供模型（硅基流动、快手万擎、月之暗面、智谱、MiniMax 稀宇、小米、阶跃星辰）通常**仅限华北2（北京）地域**，且需先在控制台模型广场开通并授权。 +- 思考模式普遍通过 `enable_thinking` 控制（非 OpenAI 标准参数，Python SDK 经 `extra_body` 传入），推理过程经 `reasoning_content` 字段返回；部分模型（Kimi、GLM-智谱、Stepfun）另支持 `reasoning_effort` 控制推理深度。 + +各模型要点： + +| 供应商 | 代表模型 | 特点 | +| --- | --- | --- | +| DeepSeek-阿里云 | deepseek-v4-pro | 多地域可用；限流更宽松，支持联网搜索与上下文缓存 | +| DeepSeek-硅基流动 | siliconflow/deepseek-v3.2 | 支持更长上下文；仅北京地域 | +| DeepSeek-快手万擎 | vanchin/deepseek-v4-pro | 仅北京地域 | +| Kimi | kimi/kimi-k3、kimi-k2.7-code | 多地域；k3 与 k2.7-code 系列为仅思考模型 | +| GLM-智谱 | ZHIPU/GLM-5.2 | 支持 1M 上下文；仅北京地域 | +| MiniMax | MiniMax/MiniMax-M2.7 | 仅北京地域 | +| MiMo-小米 | xiaomi/mimo-v2.5-pro | 默认开启思考模式；仅北京地域 | +| Stepfun-阶跃星辰 | stepfun/step-3.7-flash | [多模态](../concepts/multimodal.md)推理，**默认关闭**思考模式；仅北京地域 | + +> **注意**：多款第三方模型已公布下架计划——deepseek-v3/v3.1/v3.2/r1 系列将于 2026年10月10日 下架；Moonshot-Kimi-K2-Instruct、kimi-k2-thinking、glm-4.6/4.7、MiniMax-M2.1 将于 2026年7月9日 下架。官方推荐迁移至 qwen3.7-plus、qwen3.7-max 或 qwen3.6-flash，新项目应避免依赖这些即将下架的模型。 + +## 解决方案模板 + +平台提供多个可一键部署的场景化方案（多基于函数计算 + 百炼模型，按量付费）： + +- **HappyHorse 影视创作平台**：基于 Wan2.7 图像生成与视频生成能力的节点式无限画布创作平台。 +- **AI 智能体与工作流应用**：以 AI 电商客服为例，覆盖智能问答、RAG、Agent、对话流四种应用形态。 +- **深度研究（Qwen-Deep-Research）**：自动规划研究路径、多源交叉验证并生成结构化决策报告。 +- **AI 解题 + 批改**：基于 qwen3-vl-plus 视觉模型的拍照解题与作业自动批改，支持 33 种语言。 + +## 限制与注意事项 + +- 直供第三方模型多数仅在华北2（北京）地域可用，需使用该地域的 API Key，并将 Base URL 中的 `{WorkspaceId}` 替换为真实[业务空间](../concepts/workspace.md) ID。 +- `enable_thinking`、`reasoning_effort` 等均为非 OpenAI 标准参数，各 SDK 传入方式不同（Python 用 `extra_body`，Node.js 作为顶层参数）。 +- Prompt 优化工具、模型调优/部署/评测均产生费用，使用前确认计费规则与账号余额。 +- 服务端排队等待仅缓解突发限流，对 RPM/TPM 绝对配额超限无效，需配合客户端流控或提额。 ## 来源文档 @@ -80,5 +124,3 @@ - [AI 解题 + 批改：推动课程教学智变](../../raw/model-user-guide/use-cases/ai-homework-helper.md) - - diff --git a/skills/bailian-docs-llm-wiki/wiki/guides/use-chat-client-or-development-tool.md b/skills/bailian-docs-llm-wiki/wiki/guides/use-chat-client-or-development-tool.md index 6f73a29b..099878a0 100644 --- a/skills/bailian-docs-llm-wiki/wiki/guides/use-chat-client-or-development-tool.md +++ b/skills/bailian-docs-llm-wiki/wiki/guides/use-chat-client-or-development-tool.md @@ -1,65 +1,77 @@ # use chat client or development tool -阿里云百炼支持将平台上的模型接入各类第三方 AI 聊天客户端、编程工具与应用开发平台。这些工具本身不由百炼提供，接入方式统一为「填入 Base URL + API Key + 模型 ID」，通过 **OpenAI 兼容协议**或 **Anthropic 兼容协议**访问百炼网关。本文汇总不同工具的接入要点、共用的凭证规则以及常见限制。 +阿里云百炼支持将平台模型接入各类 AI 编程工具、聊天客户端与应用开发平台，包括 Claude Code、Codex、Cursor、Cline、Qwen Code、Cherry Studio、Chatbox、Dify 等。接入方式统一为「选择计费方案 → 获取对应 API Key → 在工具中配置 Base URL 与模型」三步。本文汇总各工具的接入要点、端点地址与常见问题。 -## 支持的工具类型 +## 计费方案与凭证 -按形态大致分为三类： +所有工具均围绕四种计费方案接入，各方案的 API Key **互不通用**： -- **终端 / CLI 编程工具**：[Hermes Agent](../../raw/model-user-guide/use-chat-client-or-development-tool/hermes-agent.md)、[Claude Code](../../raw/model-user-guide/use-chat-client-or-development-tool/claude-code.md)、[OpenCode](../../raw/model-user-guide/use-chat-client-or-development-tool/opencode.md)、[Codex](../../raw/model-user-guide/use-chat-client-or-development-tool/codex.md)、[Qwen Code](../../raw/model-user-guide/use-chat-client-or-development-tool/qwen-code.md)、[Kilo CLI](../../raw/model-user-guide/use-chat-client-or-development-tool/kilo-cli.md)、Qoder CLI。 -- **IDE / 编辑器插件**：[Cursor](../../raw/model-user-guide/use-chat-client-or-development-tool/cursor.md)、[Cline](../../raw/model-user-guide/use-chat-client-or-development-tool/cline.md)（VSCode）、Qoder（IDE / JetBrains 插件）、Qoder CN（原 Lingma）。 -- **桌面 / 跨平台聊天客户端与助手**：[Cherry Studio](../../raw/model-user-guide/use-chat-client-or-development-tool/cherry-studio.md)、[Chatbox](../../raw/model-user-guide/use-chat-client-or-development-tool/chatbox.md)、[OpenClaw](../../raw/model-user-guide/use-chat-client-or-development-tool/openclaw.md)、QwenPaw。 -- **应用开发 / 工作流平台**：[Dify](../../raw/model-user-guide/use-chat-client-or-development-tool/dify.md)。 +| 方案 | 计费方式 | 说明 | +| --- | --- | --- | +| Token Plan 个人版 | 按 token 消耗抵扣个人 Credits | 个人订阅 | +| Token Plan 团队版 | 按坐席订阅，按 token 消耗抵扣 Credits | 团队订阅 | +| Coding Plan | 固定月费，按模型调用次数计量 | 编程场景专用 | +| 按量计费 | 按实际调用量后付费 | API Key 与地域绑定 | -此外，任何兼容 OpenAI / Anthropic 协议且支持自定义服务端点的工具（如 Trae）都可参照[更多工具](../../raw/model-user-guide/use-chat-client-or-development-tool/more-tools.md)接入。若只想快速验证图像/视频生成 API，可用 [Postman 或 cURL](../../raw/model-user-guide/use-chat-client-or-development-tool/first-call-to-image-and-video-api.md) 直接调用。 +Token Plan 个人版可用模型包括 qwen3.8-max-preview、qwen3.7-max、qwen3.7-plus、qwen3.6-flash、glm-5.2、deepseek-v4-pro。 -## 三种计费方案与凭证 +## 统一端点（Base URL） -绝大多数工具的接入差异只在「Base URL 属于哪个方案」。百炼提供三种计费方案，各自有独立的 API Key，**互不通用**： +各方案提供 OpenAI 兼容与 Anthropic 兼容两类协议端点，完整对照见[更多工具](../../raw/model-user-guide/use-chat-client-or-development-tool/more-tools.md)： -| 方案 | 说明 | OpenAI 兼容 Base URL | Anthropic 兼容 Base URL | -| --- | --- | --- | --- | -| Token Plan 团队版 | 按坐席订阅，按 token 消耗抵扣 Credits | `https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` | `https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` | -| Coding Plan | 固定月费订阅，按模型调用次数计量 | `https://coding.dashscope.aliyuncs.com/v1` | `https://coding.dashscope.aliyuncs.com/apps/anthropic` | -| 按量计费（华北2·北京） | 按实际调用量后付费 | `https://dashscope.aliyuncs.com/compatible-mode/v1` | `https://dashscope.aliyuncs.com/apps/anthropic` | +| 方案 | OpenAI 兼容 | Anthropic 兼容 | +| --- | --- | --- | +| Token Plan（个人/团队） | `https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` | `https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic` | +| Coding Plan | `https://coding.dashscope.aliyuncs.com/v1` | `https://coding.dashscope.aliyuncs.com/apps/anthropic` | +| 按量计费（北京） | `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1` | `https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/apps/anthropic` | +| 按量计费（新加坡） | `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1` | `https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/apps/anthropic` | +| 按量计费（美国弗吉尼亚） | `https://dashscope-us.aliyuncs.com/compatible-mode/v1` | — | -按量计费还支持多地域，需保证 API Key 与 Base URL 地域一致： +按量计费需将 `{WorkspaceId}` 替换为真实[业务空间](../concepts/workspace.md) ID，且 API Key 必须与端点地域一致。 -- 新加坡：`https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1`（`WorkspaceId` 替换为真实值） -- 美国（弗吉尼亚）：`https://dashscope-us.aliyuncs.com/compatible-mode/v1` +## 工具接入方式一览 -> **注意**：OpenAI 协议的 Base URL 以 `/compatible-mode/v1`（或 `/v1`）结尾，Anthropic 协议以 `/apps/anthropic` 结尾。部分工具（如 OpenCode、Kilo CLI）要求在 Anthropic 端点后再追加 `/v1`。以各工具原文为准。 +### 终端 CLI 类 -## 协议选择与配置形态 +- **Claude Code**：在 `~/.claude/settings.json` 的 `env` 中配置 `ANTHROPIC_AUTH_TOKEN` / `ANTHROPIC_BASE_URL` / `ANTHROPIC_MODEL` 等变量，走 Anthropic 兼容端点；需先在 `~/.claude.json` 设置 `hasCompletedOnboarding: true` 跳过官方登录。详见 [Claude Code](../../raw/model-user-guide/use-chat-client-or-development-tool/claude-code.md)。 +- **Codex**：编辑 `~/.codex/config.toml` 定义 `model_providers`，并设置 `OPENAI_API_KEY` 环境变量。qwen3.8-max-preview、qwen3.7-max/plus、qwen3.6-plus/flash 支持 `wire_api = "responses"`（最新版 Codex）；其他模型需 `wire_api = "chat"` 并降级安装 `@openai/codex@0.80.0`。自定义模型还需配置 `model-catalog.local.json` 元数据文件。 +- **Qwen Code**：启动后输入 `/auth` 可视化配置，或编辑 `~/.qwen/settings.json` 的 `modelProviders`；思考模型需在 `generationConfig.extra_body` 中设置 `enable_thinking: true`。 +- **Hermes Agent** / **OpenCode** / **Kilo CLI** / **Qoder CLI**：均支持四种方案。Hermes 用 `hermes config set` 写 `~/.hermes/config.yaml`（默认 Anthropic 协议，`api_mode: anthropic_messages`）；OpenCode 配置 `~/.config/opencode/opencode.json`（注意其 Anthropic Base URL 末尾多一级 `/v1`）；Kilo CLI 配置 `~/.config/kilo/config.json`。 -不同工具选用的协议和配置载体各异： +### IDE / 桌面客户端类 -- **Anthropic 协议**：[Claude Code](../../raw/model-user-guide/use-chat-client-or-development-tool/claude-code.md) 通过 `~/.claude/settings.json` 的 `ANTHROPIC_BASE_URL` / `ANTHROPIC_AUTH_TOKEN` 环境变量配置；[Hermes Agent](../../raw/model-user-guide/use-chat-client-or-development-tool/hermes-agent.md) 默认使用 Anthropic 协议（`api_mode: anthropic_messages`），也可切到 OpenAI 协议。 -- **OpenAI 协议**：[Cursor](../../raw/model-user-guide/use-chat-client-or-development-tool/cursor.md)、[Cline](../../raw/model-user-guide/use-chat-client-or-development-tool/cline.md)、Cherry Studio、Chatbox 等在 GUI 中选择「OpenAI Compatible / 兼容」并填入 Base URL、API Key、模型 ID。 -- **配置文件**：Hermes（`~/.hermes/config.yaml`）、OpenCode（`~/.config/opencode/opencode.json`）、Kilo CLI（`~/.config/kilo/config.json`）、Qwen Code（`~/.qwen/settings.json`）、Codex（`~/.codex/config.toml` + `OPENAI_API_KEY` 环境变量）。 -- **原生下拉选择**：Qoder / Qoder CN 在设置中选择「阿里云百炼 - 国内」提供商 + 计费方案「类型」，仅需填 API Key。 +- **Cursor**：在 **Cursor Settings > Models** 开启 OpenAI API Key 与 Override OpenAI Base URL。免费版仅支持 Auto 模式，调用自定义模型需 Cursor Pro 及以上。部分模型名与内置模型冲突，需用别名（如 kimi-k2.6 写为 `kimi-k2-6`、glm-5.2 写为 `glm-5-2`、glm-5 写为 `glm-5-0`）。详见 [Cursor](../../raw/model-user-guide/use-chat-client-or-development-tool/cursor.md)。 +- **Cline**（VSCode 插件）：API Provider 选 **OpenAI Compatible**；使用 Qwen3 思考模式或 QwQ 模型时需勾选 **Enable R1 messages format**，否则报 400 InternalError.Algo.InvalidParameter。 +- **Qoder / Qoder CN（原 Lingma）**：内置「阿里云百炼 - 国内」提供商，直接选类型（Token Plan / Coding Plan / 按量付费）并填 API Key，仅支持文本生成模型。Qoder CN 企业版不支持接入。 +- **Cherry Studio / Chatbox**：添加自定义提供商，API 模式选 OpenAI 兼容，填入对应方案的 Base URL 与 API Key，手动新建模型 ID。 -## 关键参数与注意事项 +### 平台 / 助手类 -- **思考模式**：许多模型（如 Qwen3 思考模式、QwQ）需显式开启思考。OpenCode / Kilo CLI 用 `thinking.budgetTokens`，Qwen Code 用 `extra_body.enable_thinking: true`，Cline 需勾选 **Enable R1 messages format**。若报错 `enable_thinking parameter is restricted to True`，说明该模型仅支持思考模式运行，需在客户端开启。 -- **模型名称别名**：[Cursor](../../raw/model-user-guide/use-chat-client-or-development-tool/cursor.md) 因内置模型名冲突，需改写模型名，如 `kimi-k2.6` → `kimi-k2-6`、`glm-5` → `glm-5-0`。其他工具一般直接使用原始模型 ID。 -- **上下文窗口**：Claude Code 默认 200K，可通过 `CLAUDE_CODE_MAX_CONTEXT_TOKENS=1000000` 或模型名后缀 `[1m]` 扩展到 1M（需模型支持）。 -- **Codex 版本差异**：仅 qwen3.7-max/plus、qwen3.6-plus/flash 支持 Responses API（可用最新版 Codex）；其他模型需用 Chat/Completions API，须安装旧版本（如 `@openai/codex@0.80.0`）。 -- **401 认证失败**：几乎都是「API Key 与 Base URL 不属于同一方案」或「按量计费 Key 与地域不匹配」，逐项核对即可。 +- **Dify**：安装通义千问插件（Dify 官方维护，非阿里云提供）或 OpenAI-API-compatible 插件接入；Qwen-Omni / Qwen-Audio / Qwen-OCR 不支持直接配置，需通过 HTTP 节点接入；万相文生图/视频通过工作流模板实现。 +- **QwenPaw / OpenClaw**：开源个人 AI 助手，分别通过 Console 设置页和 `~/.openclaw/openclaw.json` 配置内置的百炼提供商。 -## 套餐使用范围限制 +### 直接调 HTTP（Postman / cURL） -> **注意**：Token Plan 团队版与 Coding Plan **仅限**在 AI 编程工具和 OpenClaw 类 Agent 中使用。以下类型不支持接入，误用可能导致订阅暂停或 API Key 被封禁（详见[更多工具](../../raw/model-user-guide/use-chat-client-or-development-tool/more-tools.md)）： -> -> - 工作流/自动化平台：如 Dify、n8n、Coze 等； -> - API 测试工具：如 Postman、Insomnia 等； -> - 自定义应用程序：脚本或后端代码中直接调用 API。 +图像与视频生成 API 采用**异步机制**：先 POST 创建任务拿到 `task_id`（有效期 24 小时），再 GET 轮询直到 `task_status` 为 `SUCCEEDED`；生成结果 URL 有效期同样为 24 小时。此方式仅适合快速测试，生产环境建议用 SDK。详见[使用Postman或cURL调用图像/视频生成API](../../raw/model-user-guide/use-chat-client-or-development-tool/first-call-to-image-and-video-api.md)。 -因此 [Dify](../../raw/model-user-guide/use-chat-client-or-development-tool/dify.md) 这类应用开发平台只能通过**按量计费**（`get-api-key` 获取的 API Key）接入，且在 Dify 中通过安装「通义千问」或「OpenAI-API-compatible」插件配置。免费额度仅适用于华北2（北京）地域，且各模型额度独立、不可跨模型共享。 +> **注意**：「更多工具」文档明确 Token Plan 与 Coding Plan 的 API Key **仅限 AI 编程工具和 OpenClaw 类 Agent 使用**，不支持工作流/自动化平台（Dify、n8n、Coze）、API 测试工具（Postman）和自定义应用后端；而 Dify 文档又给出了通过 OpenAI-API-compatible 插件接入 Token Plan 个人版的步骤，两处存在冲突。请以套餐使用条款为准，违规使用可能导致订阅暂停或 API Key 封禁。 -## 快速验证 +## 关键参数与限制 -配置完成后统一用一句问候验证连通性，例如：`claude "你好"`、`hermes chat -q "你好"`，或在 GUI 客户端对话框发送「你好」。模型正常返回响应即表示接入成功。若为 RAM 子账号，需确保在业务空间中已获得目标模型的调用权限。 +- **qwen3.8-max-preview 思考模式**（多篇文档一致强调）： + - `thinking` 始终开启，不支持关闭； + - `temperature` 思考模式下默认 0.6，传入小于 0.6 会被自动调整为 0.6； + - `reasoning_effort` 可选 `xhigh` / `medium` / `low`，默认 `xhigh`； + - 上下文窗口 983616，最大输出 131072 tokens。 +- **模型名称别名**：Cursor 中含 `.` 的模型名需改写（kimi-k2.5 → `kimi-k2-5` 等），其他工具无需调整。 +- **地域绑定**：按量计费的 API Key、Base URL、模型部署地域三者必须匹配，免费额度仅限华北2（北京）地域。 + +## 常见问题排查 + +- **401 Incorrect API key provided / Unauthorized**：API Key 与 Base URL 不属于同一计费方案，或按量计费的 Key 与端点地域不一致；确认套餐未过期、Key 复制完整。 +- **enable_thinking 参数报错**：模型仅支持思考模式，需在客户端开启 thinking。 +- **上下文超限**：在工具的生成参数中调低 `max_tokens` 或调整上下文配置。 +- 各方案错误码详情：按量计费见错误码文档，Coding Plan / Token Plan 见对应常见问题文档。 ## 来源文档 @@ -82,4 +94,3 @@ - [更多工具](../../raw/model-user-guide/use-chat-client-or-development-tool/more-tools.md) - diff --git a/skills/bailian-docs-llm-wiki/wiki/index.md b/skills/bailian-docs-llm-wiki/wiki/index.md index f339c226..2d4da121 100644 --- a/skills/bailian-docs-llm-wiki/wiki/index.md +++ b/skills/bailian-docs-llm-wiki/wiki/index.md @@ -8,11 +8,11 @@ - [application monitoring](guides/application-monitoring.md) — 1 篇源文档 - [application permission management](guides/application-permission-management.md) — 1 篇源文档 - [application publishing and sharing](guides/application-publishing-and-sharing.md) — 3 篇源文档 -- [application support](guides/application-support.md) — 2 篇源文档 +- [application support](guides/application-support.md) — 3 篇源文档 - [application use cases](guides/application-use-cases.md) — 5 篇源文档 - [bailian application calling](guides/bailian-application-calling.md) — 3 篇源文档 - [data connection overview](guides/data-connection-overview.md) — 1 篇源文档 -- [fine tuning](guides/fine-tuning.md) — 7 篇源文档 +- [fine tuning](guides/fine-tuning.md) — 8 篇源文档 - [get started with models](guides/get-started-with-models.md) — 6 篇源文档 - [knowledge base](guides/knowledge-base.md) — 8 篇源文档 - [llm application](guides/llm-application.md) — 6 篇源文档 @@ -20,7 +20,7 @@ - [memory library overview](guides/memory-library-overview.md) — 3 篇源文档 - [model compression](guides/model-compression.md) — 1 篇源文档 - [model context protocol](guides/model-context-protocol.md) — 5 篇源文档 -- [model data overview](guides/model-data-overview.md) — 2 篇源文档 +- [model data overview](guides/model-data-overview.md) — 3 篇源文档 - [model deployment 1](guides/model-deployment-1.md) — 4 篇源文档 - [model evaluation introduction](guides/model-evaluation-introduction.md) — 2 篇源文档 - [model experience](guides/model-experience.md) — 11 篇源文档 @@ -28,13 +28,13 @@ - [model monitoring](guides/model-monitoring.md) — 2 篇源文档 - [plug in](guides/plug-in.md) — 3 篇源文档 - [prompt](guides/prompt.md) — 5 篇源文档 +- [realtime api user guide](guides/realtime-api-user-guide.md) — 15 篇源文档 - [release notes](guides/release-notes.md) — 2 篇源文档 - [security and compliance](guides/security-and-compliance.md) — 12 篇源文档 - [skill](guides/skill.md) — 1 篇源文档 - [start using](guides/start-using.md) — 2 篇源文档 -- [support](guides/support.md) — 2 篇源文档 - [test 1](guides/test-1.md) — 5 篇源文档 -- [token plan guide](guides/token-plan-guide.md) — 10 篇源文档 +- [token plan guide](guides/token-plan-guide.md) — 14 篇源文档 - [use cases](guides/use-cases.md) — 24 篇源文档 - [use chat client or development tool](guides/use-chat-client-or-development-tool.md) — 17 篇源文档 @@ -42,10 +42,10 @@ - [3d generation](api/3d-generation.md) — 1 篇源文档 - [application call](api/application-call.md) — 5 篇源文档 -- [application component api reference](api/application-component-api-reference.md) — 57 篇源文档 +- [application component api reference](api/application-component-api-reference.md) — 58 篇源文档 - [file management api](api/file-management-api.md) — 1 篇源文档 - [frameworks](api/frameworks.md) — 3 篇源文档 -- [image generation](api/image-generation.md) — 26 篇源文档 +- [image generation](api/image-generation.md) — 27 篇源文档 - [knowledge](api/knowledge.md) — 1 篇源文档 - [long term memory new](api/long-term-memory-new.md) — 1 篇源文档 - [managed agents api](api/managed-agents-api.md) — 7 篇源文档 @@ -63,22 +63,21 @@ ## 横切概念 - [API Key 鉴权](concepts/api-key.md) — 关联 6 个主题 -- [MCP 与工具扩展](concepts/mcp-and-tools.md) — 关联 5 个主题 - [OpenAI 兼容接口](concepts/openai-compatible-interface.md) — 关联 6 个主题 -- [Token 与计费](concepts/token-and-billing.md) — 关联 6 个主题 -- [业务空间（Workspace）](concepts/workspace.md) — 关联 5 个主题 -- [函数调用（Function Calling）](concepts/function-calling.md) — 关联 4 个主题 -- [异步调用与任务轮询](concepts/async-invocation.md) — 关联 5 个主题 +- [Token 计费](concepts/token-billing.md) — 关联 5 个主题 +- [业务空间](concepts/workspace.md) — 关联 6 个主题 +- [地域与接入点](concepts/region.md) — 关联 5 个主题 +- [多模态](concepts/multimodal.md) — 关联 6 个主题 +- [异步调用](concepts/async-invocation.md) — 关联 5 个主题 - [检索增强生成（RAG）](concepts/rag.md) — 关联 6 个主题 -- [模型调优与部署](concepts/fine-tuning-and-deployment.md) — 关联 5 个主题 -- [流式输出](concepts/streaming-output.md) — 关联 3 个主题 +- [流式输出](concepts/streaming.md) — 关联 5 个主题 +- [限流](concepts/rate-limit.md) — 关联 4 个主题 ## 对比分析 -- [图像、视频与 3D 生成对比](comparisons/image-vs-video-vs-3d-generation.md) — 对比 3 个主题 -- [应用评估与应用监控对比](comparisons/app-evaluation-vs-monitoring.md) — 对比 2 个主题 -- [托管智能体：指南与 API 对比](comparisons/managed-agents-guide-vs-api.md) — 对比 2 个主题 -- [模型微调、压缩与部署对比](comparisons/fine-tuning-vs-compression-vs-deployment.md) — 对比 3 个主题 -- [模型评估与模型监控对比](comparisons/model-evaluation-vs-monitoring.md) — 对比 2 个主题 -- [知识库与记忆库对比](comparisons/knowledge-base-vs-memory-library.md) — 对比 2 个主题 +- [图像、视频与3D生成对比](comparisons/multimodal-generation-comparison.md) — 对比 3 个主题 +- [模型微调、压缩与高速推理对比](comparisons/model-optimization-comparison.md) — 对比 3 个主题 +- [模型监控与应用监控对比](comparisons/model-vs-application-monitoring.md) — 对比 2 个主题 +- [模型评测与应用评测对比](comparisons/model-vs-application-evaluation.md) — 对比 2 个主题 +- [知识库与记忆体系对比](comparisons/knowledge-vs-memory-comparison.md) — 对比 3 个主题