triton-npu-repair-guide
Heuristic fixes for Ascend Triton compile/JIT/kernel errors and numerical precision mismatches when editing or converting the operator. Does not replace ascend-npu-gen-test, ascend-npu-gen-bench, or harness specs.
- License 许可
- Apache-2.0
- 所属仓库
- Piagari/helix
- 最近更新
- 2026-07-16
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- triton-npu-repair-guide
- description
- Heuristic fixes for Ascend Triton compile/JIT/kernel errors and numerical precision mismatches when editing or converting the operator. Does not replace ascend-npu-gen-test, ascend-npu-gen-bench, or harness specs.
02 资源文件 2
- 参考文档output.md1 B
- 参考文档references/repair-experience.md6.7 KB
03 SKILL.md 正文
Triton repair experience (Ascend)
Use this skill when repairing the operator after Triton/Ascend compilation,
JIT, kernel-side failures, or numerical mismatches vs the torch baseline
— especially during ascend-npu-gen-eval-suite, convert, optimize, or any flow
that exercises the real Triton path.
Patterns and code hints live in references/repair-experience.md. Match the error or symptom, apply a minimal change, then re-run validation.
Relationship to other skills
- Does not replace
ascend-npu-gen-test,ascend-npu-gen-bench, or normative harness specs. - Complements
ascend-npu-run-eval(re-validate after applying a heuristic). - When a generation-only workflow such as
ascend-npu-gen-testorascend-npu-gen-benchstages this skill, use it as a diagnostic reference for compile, JIT, launch, kernel-side, or numerical symptoms.
How to apply
- Open references/repair-experience.md and match error text or symptom to a section.
- Apply the smallest change; re-run validation through the
ascend-npu-run-evalskill, usingrun-test-baselineorrun-test-optimize/run-benchas appropriate, and prefer--verboseonrun-testwhile diagnosing failures. - If nothing fits, do not force a heuristic—fall back to logs, IR skills, or deeper debugging.
Append-Only Repair Log
If you later fix the operator successfully with a new pattern not covered above, append a short entry to output.md.
Start each new block with:
----- <short title> ----
Then add a few lines covering the symptom, the fix, and how you verified it. Append only—do not delete or rewrite older blocks.
04 同仓库更多 Skills
triton-npu-optimize
Iteratively optimize a Triton Ascend NPU operator with correctness and performance gates. Use for operator optimization tasks that need repeated correctness validation, benchmark validation, multi-round experiment tracking, reusable optimization notes, and profiler-backed performance analysis when benchmark results need deeper explanation.
triton-npu-optimize-knowledge
Generic reference-only optimize knowledge for pattern triage and evidence-backed symptom routing. This skill does not define optimize workflow or own round artifacts.
triton-npu-optimize-knowledge-v3
Generic reference-only optimize knowledge (v3 working copy) for pattern triage and evidence-backed symptom routing. This skill does not define optimize workflow or own round artifacts; prefer this tree for ongoing knowledge updates.
triton-npu-optimize-knowledge
Generic reference-only optimize knowledge for pattern triage and evidence-backed symptom routing. This skill does not define optimize workflow or own round artifacts.
triton-npu-convert-pytorch-operator
Convert one PyTorch operator into a Triton NPU-backed PyTorch operator, preserve the trailing input-helper block, and validate the converted output through standalone or differential testing.
triton-npu-cann-ext-api-patterns
A5-only specialized optimization pattern references for Triton Ascend NPU kernels that use CANN Triton extension APIs. This skill does not define the optimize workflow; it only provides pattern material for optimize to consume.