docs(user): add task-oriented feature handbook and overview - #2165
Open
YunjiQin wants to merge 1 commit into
Open
docs(user): add task-oriented feature handbook and overview#2165YunjiQin wants to merge 1 commit into
YunjiQin wants to merge 1 commit into
Conversation
Introduce a user-facing documentation structure under docs/{en,zh-cn}/user/
covering the three pillars requested in hw-native-sys#2126 (precision localization,
performance-tuning syntax, DFX features) plus an overview and reference
scaffolding for hw-native-sys#2120.
- 00-overview.md: design philosophy, execution model (with diagram), memory
hierarchy, compilation pipeline, feature map, and a master documentation TOC.
- handbook/: symptom-to-tool index; precision (torch-golden / pass-IR bisection
/ selective dump, incl. the lost-WAR-dependency gotcha, hw-native-sys#2058); perf, grouped
by in-core vs inter-core axis (loop/pipeline, mixed kernel, dependency &
dispatch, memory placement, measuring impact incl. benchmark & swimlane); dfx
(flag matrix, swimlane, diagnostics render, replay, logging).
- distributed/, api-reference/, troubleshooting.md, glossary.md scaffolding.
- Renumber existing user docs to free 00 for the overview; move the standalone
torch-codegen debug guide into handbook/precision/; fix affected cross-links
in dev/language docs.
English is authoritative with a zh-cn mirror.
📝 WalkthroughWalkthroughThis PR expands bilingual PyPTO documentation with new overview and operation-reference pages, reorganized navigation, precision-localization workflows, performance-tuning guidance, DFX documentation, distributed-programming outlines, glossaries, and troubleshooting scaffolds. ChangesPyPTO documentation expansion
Estimated code review effort: 3 (Moderate) | ~25 minutes Possibly related issues
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6635bd1323
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| from pypto.runtime import benchmark | ||
|
|
||
| stats = benchmark(compiled, args, rounds=100, warmup=3) | ||
| print(stats.device_us_mean(), stats.device_us_median(), stats.device_us_stdev()) |
There was a problem hiding this comment.
In this newly added benchmark example, device_us_mean, device_us_median, and device_us_stdev are @property accessors on BenchmarkStats, not methods, so copying the snippet will fail with a 'float' object is not callable TypeError after the first access. Please show these as stats.device_us_mean, etc., and mirror the same fix in the zh-cn page. .claude/rules/documentation.mdL104-L109
Useful? React with 👍 / 👎.
| ## 逐步操作 | ||
|
|
||
| 1. **建立黄金值。** 构造输入与参考输出（如一个 torch 函数）。 | ||
| [章节 01](01-torch-golden.md#build-inputs-and-golden-output) 有一份完整的 |
There was a problem hiding this comment.
These Chinese workflow links use English heading fragments (#build-inputs-and-golden-output, #1-codegen-directly-from-program-ir, and #2-codegen-after-passmanagerdefault), but the target zh-cn page has translated headings, so the anchors do not exist and readers land at the wrong place when following the steps. Please use the localized fragments or add explicit stable anchors for all three links. .claude/rules/documentation.mdL89-L96
Useful? React with 👍 / 👎.
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
docs/en/user/handbook/perf/01-loop-pipeline.md (1)
42-45: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick winUse in-core profiling to verify pipeline overlap. Mixed C→V→C pipelining occurs inside one task and is not observable in the per-task swimlane.
docs/en/user/handbook/perf/01-loop-pipeline.md#L42-L45: point verification to in-core profiling; use the swimlane for inter-task/SPMD effects.docs/zh-cn/user/handbook/perf/01-loop-pipeline.md#L38-L40: apply the same verification distinction in the Chinese mirror.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/en/user/handbook/perf/01-loop-pipeline.md` around lines 42 - 45, Update the “Verifying the effect” section in docs/en/user/handbook/perf/01-loop-pipeline.md at lines 42-45 to direct readers to in-core profiling for mixed C→V→C pipeline overlap, while retaining the Swimlane link for inter-task/SPMD effects. Apply the same verification distinction in docs/zh-cn/user/handbook/perf/01-loop-pipeline.md at lines 38-40.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/en/user/00-overview.md`:
- Around line 117-121: Add the BackendType import to the overview compile
examples so the existing BackendType.Ascend910B references resolve when copied.
Update both docs/en/user/00-overview.md (lines 117-121) and
docs/zh-cn/user/00-overview.md (lines 101-105) by importing BackendType from
pypto.backend or qualifying the reference consistently.
In `@docs/en/user/handbook/perf/05-measuring-impact.md`:
- Around line 33-39: Clarify the introductory Compile-Time Profiling
descriptions so ir.compile(profiling=True) is documented as recording
compilation stages only, excluding device execution; note that device_execution
is available through runtime.run(..., compile_profiling=True). Apply the
correction in docs/en/user/handbook/perf/05-measuring-impact.md lines 33-39 and
mirror the equivalent correction in Chinese in
docs/zh-cn/user/handbook/perf/05-measuring-impact.md lines 29-33.
In `@docs/en/user/handbook/precision/01-torch-golden.md`:
- Around line 94-109: Make both handbook versions generic rather than presenting
Qwen3 test values as universal: in
docs/en/user/handbook/precision/01-torch-golden.md lines 94-109 and
docs/zh-cn/user/handbook/precision/01-torch-golden.md lines 94-109, make the
generated entry-point lookup configurable; in lines 131-147 of both files,
remove or explicitly qualify cross-core assertions; and in lines 159-166 of both
files, replace the fixed Qwen3 dump path and "out" key with configurable
placeholders or clearly documented substitutions.
In `@docs/zh-cn/user/03-operation_reference.md`:
- Around line 23-25: Update the unified and Tensor operation tables in
03-operation_reference.md to match the Tile API by adding entries for minimum
and row_min when their dispatchers exist; otherwise explicitly mark both
operations as Tile-only. Preserve the existing table format and descriptions
used for the corresponding maximum and reduction operations.
---
Outside diff comments:
In `@docs/en/user/handbook/perf/01-loop-pipeline.md`:
- Around line 42-45: Update the “Verifying the effect” section in
docs/en/user/handbook/perf/01-loop-pipeline.md at lines 42-45 to direct readers
to in-core profiling for mixed C→V→C pipeline overlap, while retaining the
Swimlane link for inter-task/SPMD effects. Apply the same verification
distinction in docs/zh-cn/user/handbook/perf/01-loop-pipeline.md at lines 38-40.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 1f1c5d3b-4163-408a-9c80-879988da9212
📒 Files selected for processing (50)
docs/en/dev/language/00-python_syntax.mddocs/en/user/00-overview.mddocs/en/user/01-getting_started.mddocs/en/user/02-language_guide.mddocs/en/user/03-operation_reference.mddocs/en/user/api-reference/index.mddocs/en/user/distributed/00-guide.mddocs/en/user/glossary.mddocs/en/user/handbook/00-index.mddocs/en/user/handbook/dfx/00-flag-matrix.mddocs/en/user/handbook/dfx/01-swimlane.mddocs/en/user/handbook/dfx/02-diagnostics-render.mddocs/en/user/handbook/dfx/03-replay.mddocs/en/user/handbook/dfx/04-logging.mddocs/en/user/handbook/perf/00-workflow.mddocs/en/user/handbook/perf/01-loop-pipeline.mddocs/en/user/handbook/perf/02-split-parallel.mddocs/en/user/handbook/perf/03-dependency-dispatch.mddocs/en/user/handbook/perf/04-memory-placement.mddocs/en/user/handbook/perf/05-measuring-impact.mddocs/en/user/handbook/precision/00-workflow.mddocs/en/user/handbook/precision/01-torch-golden.mddocs/en/user/handbook/precision/02-pass-ir-bisection.mddocs/en/user/handbook/precision/03-selective-dump.mddocs/en/user/troubleshooting.mddocs/zh-cn/dev/language/00-python_syntax.mddocs/zh-cn/user/00-overview.mddocs/zh-cn/user/01-getting_started.mddocs/zh-cn/user/02-language_guide.mddocs/zh-cn/user/03-operation_reference.mddocs/zh-cn/user/api-reference/index.mddocs/zh-cn/user/distributed/00-guide.mddocs/zh-cn/user/glossary.mddocs/zh-cn/user/handbook/00-index.mddocs/zh-cn/user/handbook/dfx/00-flag-matrix.mddocs/zh-cn/user/handbook/dfx/01-swimlane.mddocs/zh-cn/user/handbook/dfx/02-diagnostics-render.mddocs/zh-cn/user/handbook/dfx/03-replay.mddocs/zh-cn/user/handbook/dfx/04-logging.mddocs/zh-cn/user/handbook/perf/00-workflow.mddocs/zh-cn/user/handbook/perf/01-loop-pipeline.mddocs/zh-cn/user/handbook/perf/02-split-parallel.mddocs/zh-cn/user/handbook/perf/03-dependency-dispatch.mddocs/zh-cn/user/handbook/perf/04-memory-placement.mddocs/zh-cn/user/handbook/perf/05-measuring-impact.mddocs/zh-cn/user/handbook/precision/00-workflow.mddocs/zh-cn/user/handbook/precision/01-torch-golden.mddocs/zh-cn/user/handbook/precision/02-pass-ir-bisection.mddocs/zh-cn/user/handbook/precision/03-selective-dump.mddocs/zh-cn/user/troubleshooting.md
Comment on lines
+117
to
+121
| You trigger the whole pipeline with a single call: | ||
|
|
||
| ```python | ||
| output_dir = ir.compile(MyProgram, backend_type=BackendType.Ascend910B) | ||
| ``` |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "== docs snippets =="
for f in docs/en/user/00-overview.md docs/zh-cn/user/00-overview.md; do
echo "--- $f"
nl -ba "$f" | sed -n '88,130p'
done
echo
echo "== BackendType definitions/usages in codebase =="
rg -n "class BackendType|enum BackendType|BackendType|backend_type" . --glob '!docs' --glob '!**/*.png' --glob '!**/*.pdf' | head -n 200
echo
echo "== likely ir.py/module definitions =="
fd -a '^ir\.(py|pyi)$|backend|compiler|compile' . | sed 's#^\./##' | head -n 200Repository: hw-native-sys/pypto
Length of output: 249
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "== docs snippets =="
for f in docs/en/user/00-overview.md docs/zh-cn/user/00-overview.md; do
echo "--- $f"
sed -n '88,130=' "$f"
sed -n '88,130p' "$f"
done
echo
echo "== BackendType definitions/usages in codebase =="
rg -n "class BackendType|enum BackendType|BackendType|backend_type" . --glob '!docs' --glob '!**/*.png' --glob '!**/*.pdf' | head -n 200 || true
echo
echo "== likely ir.py/module definitions =="
find . \( -path './docs' -o -path './.git' \) -prune -o \( -name 'ir.py' -o -name 'ir.pyi' -o -name '*backend*' -o -name '*compiler*' -o -name '*compile*' \) -print | sed 's#^\./##' | head -n 200Repository: hw-native-sys/pypto
Length of output: 26442
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "== backend init signature =="
cat -n python/pypto/backend/__init__.py | sed -n '1,70p'
echo
echo "== pypto_core backend stub =="
cat -n python/pypto/pypto_core/backend.pyi | sed -n '1,35p'
echo
echo "== surrounding imports in overview examples =="
sed -n '1,90p' docs/en/user/00-overview.md
echo
sed -n '1,90p' docs/zh-cn/user/00-overview.md
echo
echo "== pypto public imports exposing BackendType =="
rg -n "^from pypto import |from pypto\\.__init__ import |\\bBackendType\\b" python/pypto --glob '*.py' python/bindings --glob '*.cpp' tests/ut/conftest.py examples/models/04_paged_attention.py | head -n 120Repository: hw-native-sys/pypto
Length of output: 15333
Fix the overview compile examples’ BackendType imports.
The snippets call ir.compile(..., backend_type=BackendType.Ascend910B) but only import pl and ir, so copy-pasted examples fail with NameError. Add from pypto.backend import BackendType or qualify the same import in both English and Chinese overview examples.
📍 Affects 2 files
docs/en/user/00-overview.md#L117-L121(this comment)docs/zh-cn/user/00-overview.md#L101-L105
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/en/user/00-overview.md` around lines 117 - 121, Add the BackendType
import to the overview compile examples so the existing BackendType.Ascend910B
references resolve when copied. Update both docs/en/user/00-overview.md (lines
117-121) and docs/zh-cn/user/00-overview.md (lines 101-105) by importing
BackendType from pypto.backend or qualifying the reference consistently.
Comment on lines
+33
to
+39
| ## Compile-Time Profiling | ||
|
|
||
| Records **compilation** wall-clock at each pipeline stage — frontend parse, | ||
| passes (per-pass), codegen (per-kernel + orchestration), and on-device | ||
| execution. Use it to answer *"why is compiling slow"* and *"which pass dominates | ||
| compile time"*, not to measure a kernel's on-chip speed (for that, use in-core | ||
| profiling and the swimlane below). |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Do not conflate compile-time profiling with device execution. ir.compile(profiling=True) records compilation stages only; device_execution is included through runtime.run(..., compile_profiling=True).
docs/en/user/handbook/perf/05-measuring-impact.md#L33-L39: revise the introductory description.docs/zh-cn/user/handbook/perf/05-measuring-impact.md#L29-L33: mirror the same correction in Chinese.
📍 Affects 2 files
docs/en/user/handbook/perf/05-measuring-impact.md#L33-L39(this comment)docs/zh-cn/user/handbook/perf/05-measuring-impact.md#L29-L33
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@docs/en/user/handbook/perf/05-measuring-impact.md` around lines 33 - 39,
Clarify the introductory Compile-Time Profiling descriptions so
ir.compile(profiling=True) is documented as recording compilation stages only,
excluding device execution; note that device_execution is available through
runtime.run(..., compile_profiling=True). Apply the correction in
docs/en/user/handbook/perf/05-measuring-impact.md lines 33-39 and mirror the
equivalent correction in Chinese in
docs/zh-cn/user/handbook/perf/05-measuring-impact.md lines 29-33.
There was a problem hiding this comment.
Caution
Inline review comments failed to post. This is likely due to GitHub's internal server error or limits when posting large numbers of comments. If you are seeing this consistently it is likely a permissions issue. Please check "Moderation" -> "Code review limits" under your organization settings.
Actionable comments posted: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
docs/en/user/handbook/perf/01-loop-pipeline.md (1)
42-45: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick winUse in-core profiling to verify pipeline overlap. Mixed C→V→C pipelining occurs inside one task and is not observable in the per-task swimlane.
docs/en/user/handbook/perf/01-loop-pipeline.md#L42-L45: point verification to in-core profiling; use the swimlane for inter-task/SPMD effects.docs/zh-cn/user/handbook/perf/01-loop-pipeline.md#L38-L40: apply the same verification distinction in the Chinese mirror.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/en/user/handbook/perf/01-loop-pipeline.md` around lines 42 - 45, Update the “Verifying the effect” section in docs/en/user/handbook/perf/01-loop-pipeline.md at lines 42-45 to direct readers to in-core profiling for mixed C→V→C pipeline overlap, while retaining the Swimlane link for inter-task/SPMD effects. Apply the same verification distinction in docs/zh-cn/user/handbook/perf/01-loop-pipeline.md at lines 38-40.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@docs/en/user/00-overview.md`:
- Around line 117-121: Add the BackendType import to the overview compile
examples so the existing BackendType.Ascend910B references resolve when copied.
Update both docs/en/user/00-overview.md (lines 117-121) and
docs/zh-cn/user/00-overview.md (lines 101-105) by importing BackendType from
pypto.backend or qualifying the reference consistently.
In `@docs/en/user/handbook/perf/05-measuring-impact.md`:
- Around line 33-39: Clarify the introductory Compile-Time Profiling
descriptions so ir.compile(profiling=True) is documented as recording
compilation stages only, excluding device execution; note that device_execution
is available through runtime.run(..., compile_profiling=True). Apply the
correction in docs/en/user/handbook/perf/05-measuring-impact.md lines 33-39 and
mirror the equivalent correction in Chinese in
docs/zh-cn/user/handbook/perf/05-measuring-impact.md lines 29-33.
In `@docs/en/user/handbook/precision/01-torch-golden.md`:
- Around line 94-109: Make both handbook versions generic rather than presenting
Qwen3 test values as universal: in
docs/en/user/handbook/precision/01-torch-golden.md lines 94-109 and
docs/zh-cn/user/handbook/precision/01-torch-golden.md lines 94-109, make the
generated entry-point lookup configurable; in lines 131-147 of both files,
remove or explicitly qualify cross-core assertions; and in lines 159-166 of both
files, replace the fixed Qwen3 dump path and "out" key with configurable
placeholders or clearly documented substitutions.
In `@docs/zh-cn/user/03-operation_reference.md`:
- Around line 23-25: Update the unified and Tensor operation tables in
03-operation_reference.md to match the Tile API by adding entries for minimum
and row_min when their dispatchers exist; otherwise explicitly mark both
operations as Tile-only. Preserve the existing table format and descriptions
used for the corresponding maximum and reduction operations.
---
Outside diff comments:
In `@docs/en/user/handbook/perf/01-loop-pipeline.md`:
- Around line 42-45: Update the “Verifying the effect” section in
docs/en/user/handbook/perf/01-loop-pipeline.md at lines 42-45 to direct readers
to in-core profiling for mixed C→V→C pipeline overlap, while retaining the
Swimlane link for inter-task/SPMD effects. Apply the same verification
distinction in docs/zh-cn/user/handbook/perf/01-loop-pipeline.md at lines 38-40.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 1f1c5d3b-4163-408a-9c80-879988da9212
📒 Files selected for processing (50)
docs/en/dev/language/00-python_syntax.mddocs/en/user/00-overview.mddocs/en/user/01-getting_started.mddocs/en/user/02-language_guide.mddocs/en/user/03-operation_reference.mddocs/en/user/api-reference/index.mddocs/en/user/distributed/00-guide.mddocs/en/user/glossary.mddocs/en/user/handbook/00-index.mddocs/en/user/handbook/dfx/00-flag-matrix.mddocs/en/user/handbook/dfx/01-swimlane.mddocs/en/user/handbook/dfx/02-diagnostics-render.mddocs/en/user/handbook/dfx/03-replay.mddocs/en/user/handbook/dfx/04-logging.mddocs/en/user/handbook/perf/00-workflow.mddocs/en/user/handbook/perf/01-loop-pipeline.mddocs/en/user/handbook/perf/02-split-parallel.mddocs/en/user/handbook/perf/03-dependency-dispatch.mddocs/en/user/handbook/perf/04-memory-placement.mddocs/en/user/handbook/perf/05-measuring-impact.mddocs/en/user/handbook/precision/00-workflow.mddocs/en/user/handbook/precision/01-torch-golden.mddocs/en/user/handbook/precision/02-pass-ir-bisection.mddocs/en/user/handbook/precision/03-selective-dump.mddocs/en/user/troubleshooting.mddocs/zh-cn/dev/language/00-python_syntax.mddocs/zh-cn/user/00-overview.mddocs/zh-cn/user/01-getting_started.mddocs/zh-cn/user/02-language_guide.mddocs/zh-cn/user/03-operation_reference.mddocs/zh-cn/user/api-reference/index.mddocs/zh-cn/user/distributed/00-guide.mddocs/zh-cn/user/glossary.mddocs/zh-cn/user/handbook/00-index.mddocs/zh-cn/user/handbook/dfx/00-flag-matrix.mddocs/zh-cn/user/handbook/dfx/01-swimlane.mddocs/zh-cn/user/handbook/dfx/02-diagnostics-render.mddocs/zh-cn/user/handbook/dfx/03-replay.mddocs/zh-cn/user/handbook/dfx/04-logging.mddocs/zh-cn/user/handbook/perf/00-workflow.mddocs/zh-cn/user/handbook/perf/01-loop-pipeline.mddocs/zh-cn/user/handbook/perf/02-split-parallel.mddocs/zh-cn/user/handbook/perf/03-dependency-dispatch.mddocs/zh-cn/user/handbook/perf/04-memory-placement.mddocs/zh-cn/user/handbook/perf/05-measuring-impact.mddocs/zh-cn/user/handbook/precision/00-workflow.mddocs/zh-cn/user/handbook/precision/01-torch-golden.mddocs/zh-cn/user/handbook/precision/02-pass-ir-bisection.mddocs/zh-cn/user/handbook/precision/03-selective-dump.mddocs/zh-cn/user/troubleshooting.md
🛑 Comments failed to post (2)
docs/en/user/handbook/precision/01-torch-golden.md (1)
94-109: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Avoid presenting test-specific values as a generic workflow.
Both language versions hardcode the
scope3entry point, Qwen3 pass-dump path,"out"output key, and cross-core assertions. Parameterize these values or explicitly identify the snippets as Qwen3-specific and document the required substitutions.
docs/en/user/handbook/precision/01-torch-golden.md#L94-L109: make the generated entry-point lookup configurable instead of assumingscope3.docs/en/user/handbook/precision/01-torch-golden.md#L131-L147: remove or qualify cross-core assertions that are not valid for every program.docs/en/user/handbook/precision/01-torch-golden.md#L159-L166: replace the fixed Qwen3 path and"out"key with placeholders or configurable values.docs/zh-cn/user/handbook/precision/01-torch-golden.md#L94-L109: mirror the generic entry-point guidance.docs/zh-cn/user/handbook/precision/01-torch-golden.md#L131-L147: mirror the cross-core assertion guidance.docs/zh-cn/user/handbook/precision/01-torch-golden.md#L159-L166: mirror the configurable dump path and output-key guidance.📍 Affects 2 files
docs/en/user/handbook/precision/01-torch-golden.md#L94-L109(this comment)docs/en/user/handbook/precision/01-torch-golden.md#L131-L147docs/en/user/handbook/precision/01-torch-golden.md#L159-L166docs/zh-cn/user/handbook/precision/01-torch-golden.md#L94-L109docs/zh-cn/user/handbook/precision/01-torch-golden.md#L131-L147docs/zh-cn/user/handbook/precision/01-torch-golden.md#L159-L166🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/en/user/handbook/precision/01-torch-golden.md` around lines 94 - 109, Make both handbook versions generic rather than presenting Qwen3 test values as universal: in docs/en/user/handbook/precision/01-torch-golden.md lines 94-109 and docs/zh-cn/user/handbook/precision/01-torch-golden.md lines 94-109, make the generated entry-point lookup configurable; in lines 131-147 of both files, remove or explicitly qualify cross-core assertions; and in lines 159-166 of both files, replace the fixed Qwen3 dump path and "out" key with configurable placeholders or clearly documented substitutions.docs/zh-cn/user/03-operation_reference.md (1)
23-25: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win
Keep the unified and Tensor operation tables consistent with the Tile API.
minimumandrow_minare documented for Tiles (Lines 127 and 167) but are absent from the corresponding unified/Tensor tables, whilemaximumand the other reductions are listed. Add the unified/Tensor entries if those dispatchers exist; otherwise explicitly document that these operations are Tile-only.Also applies to: 31-38
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@docs/zh-cn/user/03-operation_reference.md` around lines 23 - 25, Update the unified and Tensor operation tables in 03-operation_reference.md to match the Tile API by adding entries for minimum and row_min when their dispatchers exist; otherwise explicitly mark both operations as Tile-only. Preserve the existing table format and descriptions used for the corresponding maximum and reduction operations.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds a user-facing documentation structure under
docs/{en,zh-cn}/user/— an overview plus a task-oriented feature handbook — addressing the three pillars in #2126 and the reference scaffolding in #2120.00-overview.md— design philosophy, execution model (with diagram), memory hierarchy, compilation pipeline, feature map, and a master documentation TOC.handbook/— a symptom→tool index and three pillars:pl.submit(..., deps=[...])workaround.distributed/,api-reference/(for the [Docs] Docstring-driven API reference for the frontend DSL (pl.*) — full audit + auto-generated, always-in-sync reference #2120 generated reference),troubleshooting.md,glossary.md.00for the overview; move the standalone torch-codegen debug guide intohandbook/precision/; fix affected cross-links indev/languagedocs.English is authoritative with a
zh-cnmirror (EN/ZH parity checked).Directory Structure
docs/zh-cn/user/mirrors this tree 1:1.Testing
Related Issues
Addresses #2126 (feature handbook) and lays the
api-reference/groundwork for #2120; documents the #2058 WAR-dependency workaround.