[https://nvbugs/6086538][fix] suppress misleading skip-softmax FMHA warning in generation

[bobboli]

Summary

Suppress the FMHA "Consider using numInstsQ = 2" INFO log for skip-softmax generation kernels.

The warning is misleading for this path because skip-softmax generation intentionally uses a 1x1 numInsts configuration, while the existing INFO suggests 2x1 would be preferable.

Validation

• Rebuilt build_wheel_targets successfully
• Verified from local MiniMax-M2 repro logs that the warning spam is aligned with generation warmup, not context/prefill warmup
• Kept the change scoped to generation skip-softmax so context warnings remain intact

Summary by CodeRabbit

• Chores
  • Performance optimization for generation kernels enabling conditional softmax computation skipping to improve efficiency in specific scenarios.

[coderabbitai]

📝 Walkthrough

Walkthrough

A configuration flag isGenerationSkipSoftmax was introduced to conditionally gate an informational tuning check in a FMHA kernel traits header. The check now only executes when generation skip softmax is disabled, effectively bypassing the recommendation for applicable generation kernels.

Changes

Cohort / File(s)	Summary
FMHA Kernel Configuration cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/trtllmGen_fmha_export/KernelTraits.h	Introduced isGenerationSkipSoftmax flag derived from options.mSkipsSoftmaxWhenPossible and context kernel status; guarded informational tuning check to skip recommendation for generation kernels that can skip softmax.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description check	✅ Passed	The PR description explains the issue and solution clearly, includes validation steps performed, and identifies the scope of changes. However, it lacks a dedicated 'Test Coverage' section explicitly listing relevant tests.
Title check	✅ Passed	The title directly addresses the main change: suppressing a misleading FMHA skip-softmax warning in generation kernels, which aligns with the code modifications that guard the tuning check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[bobboli]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #44187 [ run ] triggered by Bot. Commit: 10effe6 Link to invocation

[tensorrt-cicd]

PR_Github #44187 [ run ] completed with state SUCCESS. Commit: 10effe6
/LLM/main/L0_MergeRequest_PR pipeline #34613 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

Link to invocation

[bobboli]

/bot run --disable-fail-fast --reuse-test

[tensorrt-cicd]

PR_Github #44210 [ run ] triggered by Bot. Commit: 10effe6 Link to invocation

[tensorrt-cicd]

PR_Github #44210 [ run ] completed with state SUCCESS. Commit: 10effe6
/LLM/main/L0_MergeRequest_PR pipeline #34634 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

Link to invocation

[bobboli]

/bot run --disable-fail-fast --reuse-test

[tensorrt-cicd]

PR_Github #44277 [ run ] triggered by Bot. Commit: 10effe6 Link to invocation

[tensorrt-cicd]

PR_Github #44277 [ run ] completed with state FAILURE. Commit: 10effe6
/LLM/main/L0_MergeRequest_PR pipeline #34698 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

Link to invocation

[bobboli]

/bot run --disable-fail-fast --reuse-test

[tensorrt-cicd]

PR_Github #44295 [ run ] triggered by Bot. Commit: 10effe6 Link to invocation

[tensorrt-cicd]

PR_Github #44295 [ run ] completed with state FAILURE. Commit: 10effe6
/LLM/main/L0_MergeRequest_PR pipeline #34716 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

Link to invocation

[bobboli]

/bot run --disable-fail-fast --reuse-test

[tensorrt-cicd]

PR_Github #44416 [ run ] triggered by Bot. Commit: 10effe6 Link to invocation

[tensorrt-cicd]

PR_Github #44416 [ run ] completed with state SUCCESS. Commit: 10effe6
/LLM/main/L0_MergeRequest_PR pipeline #34828 completed with status: 'SUCCESS'

CI Report

Link to invocation