[https://nvbugs/6011517][fix] Fix autotuner OOM for trtllmGen MoE runners at large context length

[hyukn]

Summary by CodeRabbit

• Chores
  • Updated model optimization tuning configuration to enhance inference performance.

Description

Cap tune_max_num_tokens=8192 in TuningConfig for FP4BlockScaleMoERunner. Without this cap, the autotuner uses raw input dimensions (e.g. 1M+ tokens at 128k context) to generate profiling shapes, causing OOM or CUDA_ERROR_INVALID_VALUE during tactic profiling.

Test Coverage

• Verify DeepSeek-V3 serving at 128k context length no longer OOMs during autotuning
• Verify autotuner still produces valid tactics at smaller context lengths

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)

• 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.

[hyukn]

/bot run

[coderabbitai]

📝 Walkthrough

Walkthrough

Updated the FP4BlockScaleMoERunner.get_tuning_config() method to pass tune_max_num_tokens=8192 as a parameter to the TuningConfig constructor, modifying the autotuner configuration for this runner.

Changes

Cohort / File(s)	Summary
Autotuner Configuration tensorrt_llm/_torch/custom_ops/trtllm_gen_custom_ops.py	Added tune_max_num_tokens=8192 parameter to TuningConfig constructor in FP4BlockScaleMoERunner.get_tuning_config() method.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: fixing OOM issues in the autotuner for MoE runners at large context lengths by capping tuning parameters.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Description check	✅ Passed	The pull request description is well-structured, clearly explains the issue and solution, includes test coverage details, and confirms all PR checklist items are addressed.

_{✏️ 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.}

[tensorrt-cicd]

PR_Github #40234 [ run ] triggered by Bot. Commit: 37ff110 Link to invocation

[tensorrt-cicd]

PR_Github #40234 [ run ] completed with state SUCCESS. Commit: 37ff110
/LLM/main/L0_MergeRequest_PR pipeline #31369 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

[hyukn]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #40673 [ run ] triggered by Bot. Commit: 08d6add Link to invocation

[tensorrt-cicd]

PR_Github #40673 [ run ] completed with state FAILURE. Commit: 08d6add

Link to invocation

[hyukn]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #40827 [ run ] triggered by Bot. Commit: 08d6add Link to invocation

[tensorrt-cicd]

PR_Github #40827 [ run ] completed with state SUCCESS. Commit: 08d6add
/LLM/main/L0_MergeRequest_PR pipeline #31838 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

[longlee0622]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #41098 [ run ] triggered by Bot. Commit: 08d6add Link to invocation

[tensorrt-cicd]

PR_Github #41098 [ run ] completed with state SUCCESS. Commit: 08d6add
/LLM/main/L0_MergeRequest_PR pipeline #32072 completed with status: 'SUCCESS'

CI Report

Link to invocation