fix trtllm mla spec

[b8zhong]

In theory

test/srt/test_deepseek_v3_fp4_4gpu.py should pass

[gemini-code-assist]

Summary of Changes

Hello @b8zhong, 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 implements a targeted fix within the TensorRT-LLM (TRTLLM) Multi-Layer Attention (MLA) backend by eliminating a specific Triton kernel caching mechanism. The removal of the cached_triton_kernel decorator and its associated import is intended to correct an underlying issue that was causing a DeepSeek v3 FP4 4-GPU test to fail, thereby improving the stability and correctness of the TRTLLM integration.

Highlights

• Triton Kernel Caching Removal: The cached_triton_kernel import and its application as a decorator to two Triton kernels, pad_draft_extend_query_kernel and unpad_draft_extend_output_kernel, have been removed.
• TRTLLM MLA Specification Fix: This change specifically addresses an issue with TensorRT-LLM Multi-Layer Attention (MLA) specifications, aiming to resolve a problem that prevented a DeepSeek v3 FP4 4-GPU test from passing.

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 addresses a bug in the TRT-LLM MLA speculative decoding path by removing the @cached_triton_kernel decorator from pad_draft_extend_query_kernel and unpad_draft_extend_output_kernel. This custom caching layer appears to have been causing issues, and its removal is a clean and safe fix, falling back to Triton's standard JIT caching mechanism. Since the cache key was based on a constant BLOCK_SIZE, there should be no performance regression. The code is now simpler and more robust. The changes look good.

[Qiaolin-Yu]

Why delete this? This cache is important and was added a very long time ago. Or perhaps the previous usage of this cache here was incorrect? Could you please explain more? 👀 I just took a close look at the code, and I think you’re right.

[hnyls2002]

@Qiaolin-Yu @b8zhong Please track the CI status. After the test/srt/test_deepseek_v3_fp4_4gpu.py passes, add a read-to-merge label

[b8zhong]

/tag-and-rerun-ci