Skip to content

[None][chore] fix deepep trtllm backend MXFP4#12219

Merged
xxi-nv merged 2 commits into
NVIDIA:mainfrom
leslie-fang25:leslie/fix_deepep_trtllm_MXFP4
Mar 16, 2026
Merged

[None][chore] fix deepep trtllm backend MXFP4#12219
xxi-nv merged 2 commits into
NVIDIA:mainfrom
leslie-fang25:leslie/fix_deepep_trtllm_MXFP4

Conversation

@leslie-fang25
Copy link
Copy Markdown
Collaborator

@leslie-fang25 leslie-fang25 commented Mar 15, 2026 •
edited by coderabbitai Bot
Loading

Summary by CodeRabbit

  • Tests
    • Updated test utilities to refine quantization algorithm compatibility detection for multi-GPU DeepEP configurations, reducing the set of algorithms flagged as problematic in specific scenarios.

Description

Test Coverage

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.

@leslie-fang25
Copy link
Copy Markdown
Collaborator Author

leslie-fang25 commented Mar 15, 2026

/bot run --disable-fail-fast

@coderabbitai
Copy link
Copy Markdown
Contributor

coderabbitai Bot commented Mar 15, 2026 •
edited
Loading

📝 Walkthrough

Walkthrough

Two quantization algorithms (W4A16_MXFP4 and W4A8_MXFP8) are removed from a test utility set tracking known crash cases in DeepEP+TRTLLM scenarios, reducing the set to only FP8_BLOCK_SCALES.

Changes

Cohort / File(s) Summary
Test Utility Configuration
tests/unittest/_torch/modules/moe/moe_test_utils.py		 Removed two quantization algorithms (W4A16_MXFP4 and W4A8_MXFP8) from the deepep_crash_quant_algos set, reducing the tracked crash candidates under multi-GPU DeepEP mode with TRTLLM to only FP8_BLOCK_SCALES.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name Status Explanation Resolution
Description check ⚠️ Warning The PR description is incomplete and contains only the template boilerplate without any actual content explaining the changes, rationale, or test coverage. Fill in the Description section explaining what the MXFP4 issue is and why removing those quantization algorithms fixes it. Add relevant test coverage details and ensure all checklist items are properly completed before submission.
✅ Passed checks (2 passed)
Check name Status Explanation
Title check ✅ Passed The title clearly specifies a fix to the DeepEP TRTLLM backend MXFP4 functionality, directly matching the code changes that remove two MXFP4 quant algorithms from the crash list.
Docstring Coverage ✅ Passed No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage 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
  • Post copyable unit tests in a comment
📝 Coding Plan
  • Generate coding plan for human review comments

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

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

tensorrt-cicd commented Mar 15, 2026

PR_Github #38965 [ run ] triggered by Bot. Commit: 100bd30 Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

tensorrt-cicd commented Mar 15, 2026

PR_Github #38965 [ run ] completed with state SUCCESS. Commit: 100bd30
/LLM/main/L0_MergeRequest_PR pipeline #30247 completed with status: 'SUCCESS'

CI Report

Link to invocation

@leslie-fang25
Copy link
Copy Markdown
Collaborator Author

leslie-fang25 commented Mar 16, 2026

/bot run --disable-fail-fast

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

tensorrt-cicd commented Mar 16, 2026

PR_Github #39019 [ run ] triggered by Bot. Commit: 5f35f90 Link to invocation

@tensorrt-cicd
Copy link
Copy Markdown
Collaborator

tensorrt-cicd commented Mar 16, 2026

PR_Github #39019 [ run ] completed with state SUCCESS. Commit: 5f35f90
/LLM/main/L0_MergeRequest_PR pipeline #30297 completed with status: 'SUCCESS'

CI Report

Link to invocation

@leslie-fang25 leslie-fang25 requested a review from xxi-nv March 16, 2026 09:07
@xxi-nv xxi-nv merged commit 7686deb into NVIDIA:main Mar 16, 2026
5 checks passed
reasonsolo pushed a commit to reasonsolo/TensorRT-LLM that referenced this pull request Mar 17, 2026
@leslie-fang25 @reasonsolo
[None][chore] fix deepep trtllm backend MXFP4 (NVIDIA#12219)
1888ff6 
Signed-off-by: leslie-fang25 <[email protected]>
limin2021 pushed a commit to limin2021/TensorRT-LLM that referenced this pull request Mar 19, 2026
@leslie-fang25
[None][chore] fix deepep trtllm backend MXFP4 (NVIDIA#12219)
5e2a7b9 
Signed-off-by: leslie-fang25 <[email protected]>
longcheng-nv pushed a commit to longcheng-nv/TensorRT-LLM that referenced this pull request Mar 31, 2026
@leslie-fang25 @longcheng-nv
[None][chore] fix deepep trtllm backend MXFP4 (NVIDIA#12219)
12b3583 
Signed-off-by: leslie-fang25 <[email protected]>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@xxi-nv xxi-nv xxi-nv approved these changes

Assignees

@leslie-fang25 leslie-fang25

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@leslie-fang25 @tensorrt-cicd @xxi-nv