Skip to content

feat: support qwen3-vlm dense/moe model convert - #115

Merged
chaokunyang merged 1 commit into
inclusionAI:mainfrom
sitabulaixizawaluduo:feat/qwen3-vl
Aug 14, 2026
Merged

feat: support qwen3-vlm dense/moe model convert#115
chaokunyang merged 1 commit into
inclusionAI:mainfrom
sitabulaixizawaluduo:feat/qwen3-vl

Conversation

@sitabulaixizawaluduo

Copy link
Copy Markdown
Contributor

What does this PR do?

This PR adds end-to-end AWEX colocate support for Qwen3-VL Dense and MoE models.

It covers the full colocate lifecycle, including:

  • Training and inference weight metadata exchange.
  • Vision tower and language model weight synchronization.
  • Tensor-parallel vision weight handling.
  • Composite Qwen3-VL configuration propagation without flattening text_config and vision_config.

Related issues

Does this PR introduce any user-facing change?

  • Does this PR introduce any public API change?
  • Does this PR introduce any binary protocol compatibility change?

@chaokunyang
chaokunyang merged commit 5c6e802 into inclusionAI:main Aug 14, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants