fix kimi k2.5's moe gemm config init

[cicirori]

Motivation

--moe-runner-backend flashinfer_trtllm doesn't work for kimi k2.5

Modifications

Accuracy Tests

Benchmarking and Profiling

• b200 tp8
  before this: : ~80 TPS
  after this: ~180 TPS

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 @cicirori, 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 addresses a critical initialization bug affecting the Mixture-of-Experts (MoE) GEMM configuration for specific models, such as kimi k2.5, when utilizing the flashinfer_trtllm backend. The fix ensures that the MoE configuration is correctly identified and initialized, particularly for multimodal models that may store relevant configuration details under a text_config attribute. This correction not only resolves the functional issue but also leads to substantial performance improvements.

Highlights

• MoE GEMM Configuration Fix: Resolves an issue where the --moe-runner-backend flashinfer_trtllm failed to initialize correctly for kimi k2.5 models.
• Multimodal Model Compatibility: Introduces a more robust initialization logic for Mixture-of-Experts (MoE) GEMM configurations, specifically by checking for num_experts_per_tok within the text_config attribute for multimodal models, if present.
• Performance Improvement: Achieves a significant performance boost on b200 tp8, increasing Throughput Per Second (TPS) from approximately 80 to 180 after applying this fix.

🧠 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 correctly fixes an issue with MoE GEMM configuration initialization for multimodal models by checking for a text_config attribute within the model's Hugging Face configuration. The logic is sound. I've added a suggestion to make the implementation more concise.

[cicirori]

/tag-and-rerun-ci

[yhyang201]

/rerun-failed-ci

[cicirori]

for the last round pr-test, only 1 case failed and not related to this commit
https://github.com/sgl-project/sglang/actions/runs/21561447518/job/62451317474?pr=18064

[yhyang201]

/rerun-failed-ci

[yhyang201]

No need to rebase (merge branch main into fix_kimi_k25_moe_init); just run /rerun-failed-ci. Rebasing will cause all previously passed CI checks to run again.