[Bugfix] Fix Mistral Large 3 NVFP4 TRTLLM MoE

[elvischenv]

Motivation

TRTLLM MoE refactoring PR(#15151) broke Mistral Large 3 NVFP4 MoE support(#15049), this PR is trying to fix the issue.

Modifications

Accuracy Tests

|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|
|-----|------:|----------------|-----:|-----------|---|-----:|---|-----:|
|gsm8k|      3|flexible-extract|     8|exact_match|↑  |0.9174|±  |0.0076|
|     |       |strict-match    |     8|exact_match|↑  |0.7415|±  |0.0121|

Benchmarking and Profiling

Same with results in #15049

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review Process

1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Summary of Changes

Hello @elvischenv, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request resolves a critical bug that had impacted Mistral Large 3 NVFP4 MoE support, a regression introduced by a prior TRTLLM MoE refactoring. The fix involves consolidating the Mixture-of-Experts (MoE) application logic within the CompressedTensorsMoEMethod class. By integrating the flashinfer_trtllm execution path directly into the main apply method, the PR ensures that the affected model operates correctly while maintaining compatibility with alternative MoE implementations.

Highlights

• Consolidated MoE Application Logic: The primary apply method within CompressedTensorsMoEMethod has been refactored to conditionally handle both flashinfer_trtllm and cutlass_moe_fp4 execution paths, streamlining the Mixture-of-Experts (MoE) processing.
• Removed Redundant Method: The apply_with_router_logits method has been eliminated, as its functionality was integrated directly into the refactored apply method, simplifying the class structure.
• Bugfix for Mistral Large 3 NVFP4 MoE: This change specifically addresses and resolves a regression that had broken Mistral Large 3 NVFP4 MoE support, restoring its correct functionality within the TRTLLM framework.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request fixes a bug in Mistral Large 3 NVFP4 MoE support by refactoring the apply method in CompressedTensorsW4A4Nvfp4MoEMethod. The logic from the now-removed apply_with_router_logits method has been correctly integrated into apply, conditioned on self.use_flashinfer_trtllm. This change unifies the two execution paths and ensures consistent return types. The fix is sound. I have one suggestion to further improve maintainability by splitting the large apply method into smaller, more focused private helper methods.

[b8zhong]

Thanks. Sorry, we will be more careful in the future.

[b8zhong]

/tag-and-rerun-ci