[AMD] Clean up vllm dependencies in moe_runner/triton.py

[hubertlu-tw]

Motivation

Modifications

Accuracy Tests

Benchmarking and Profiling

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.

CC: @HaiShaw @kkHuang-amd

[gemini-code-assist]

Summary of Changes

Hello @hubertlu-tw, 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 PR refactors dependency imports and usage in python/sglang/srt/layers/moe/moe_runner/triton.py. It unifies the use of sgl_kernel for gelu_and_mul and silu_and_mul across CUDA and HIP, moving away from vllm custom ops for these specific functions on HIP. The environment variable SGLANG_MOE_USE_AITER is renamed to SGLANG_USE_AITER. Additionally, the import of vllm._custom_ops is now more granular, only occurring for HIP when aiter is not used and specifically for moe_sum.

Highlights

• Intent: This pull request aims to clean up and streamline the dependency management for vLLM-related operations within the moe_runner/triton.py file, specifically improving the handling of AMD (HIP) environments.
• Key Changes: The changes unify the use of sgl_kernel for gelu_and_mul and silu_and_mul across both CUDA and HIP, reducing reliance on vllm custom ops for these functions. It also renames an environment variable and refines the conditional import of vllm for moe_sum on HIP.
• Reviewer Activity: The pull request explicitly CC's @HaiShaw and @kkHuang-amd for review.

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 aims to clean up vllm dependencies for AMD/HIP builds, which is a good goal. The changes correctly switch to using sgl_kernel for gelu_and_mul and silu_and_mul on HIP platforms.

However, I've found a critical issue in the import logic that makes a code block unreachable, which would break moe_sum functionality on HIP. Please see my detailed comment below.

[HaiShaw]

One fix seems necessary from gemini review.

[yctseng0211]

These three failures will be fixed by #16675

and this one

has been resolved by #16535