[None][perf] Add CUTEDSL FC2 N-tile tuning override

[peihu-nv]

@coderabbitai summary

Description

Small per-expert FC2 GEMMs in WideEP can favor a different N tile size than larger MoE shapes. The existing CuTeDSL autotuner evaluates N128 and N256 tactics, but it has no way to pin one candidate for controlled tuning and deployment.

Add a strict TRTLLM_CUTEDSL_FC2_N_TILE_SIZE override to the FC2 finalize-fusion runner. Values 128 and 256 select the corresponding candidate; leaving the variable unset preserves the existing autotuned behavior. The selected value is included in the runner identity so tuning caches cannot collide, and unsupported values fail fast.

For DeepSeek-R1 FP4 WideEP EP32 decode on B200, selecting N128 improved the position-balanced MBS32 point by 1.82% and the full-Pareto geometric mean by 1.22%.

Test Coverage

• Added focused unit coverage for default N128/N256 discovery, both valid overrides, tuning-cache identity separation, and invalid values.
• Homebrew Python 3.12 syntax compilation passes for the changed source and test.
• Changed-file pre-commit checks pass.
• Structural validation confirms the override is scoped to the FC2 finalize-fusion runner.

PR Checklist

Please review the following before submitting your PR:

• PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.

• PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

• Test cases are provided for new code paths (see test instructions)

• If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

• Any new dependencies have been scanned for license and vulnerabilities

• CODEOWNERS updated if ownership changes

• Documentation updated as needed

• Update tava architecture diagram if there is a significant design change in PR.

• The reviewers assigned automatically/manually are appropriate for the PR.

• Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.

[peihu-nv]

/bot run

[tensorrt-cicd]

PR_Github #62584 [ run ] triggered by Bot. Commit: 0e6fb4e Link to invocation

[tensorrt-cicd]

PR_Github #62584 [ run ] completed with state SUCCESS. Commit: 0e6fb4e
/LLM/main/L0_MergeRequest_PR pipeline #50726 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

CI Agent Failure Analysis

Link to invocation