Apply back moe_sum_reduce for fused_marlin_moe

[ispobock]

Motivation

Previously we applied this change in #12888 but revert it to debug IMA crash issue.
Now the marlin kernel crash issue is solved, so we can apply this change back.

Modifications

Accuracy Tests

Accuracy: 0.941
Invalid: 0.000
Latency: 88.846 s
Output throughput: 1451.369 token/s

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.
• Follow the SGLang code style guidance.
• Work with maintainers to merge your PR. See the PR Merge Process

[gemini-code-assist]

Summary of Changes

Hello @ispobock, 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 reintroduces a performance optimization for the fused_marlin_moe component, specifically by integrating a custom moe_sum_reduce kernel. This change was previously rolled back due to a kernel stability issue, which has since been addressed. The primary goal is to enhance the efficiency of Mixture-of-Experts (MoE) computations by replacing a multi-step summation and scaling process with a single, optimized kernel call.

Highlights

• Re-application of moe_sum_reduce: This pull request re-applies the moe_sum_reduce optimization for the fused_marlin_moe operation, which was previously reverted due to a kernel crash issue that has now been resolved.
• Optimized Summation: The standard torch.sum operation and subsequent scaling logic have been replaced with a call to the specialized moe_sum_reduce kernel, which integrates both the summation and scaling into a single, potentially more efficient, operation.
• Dependency Update: The moe_sum_reduce function is now explicitly imported from the sgl_kernel module in fused_marlin_moe.py.

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 re-introduces the moe_sum_reduce fused kernel in fused_marlin_moe for performance optimization. The change replaces torch.sum followed by a multiplication with a single, more efficient kernel call. The logic correctly handles the routed_scaling_factor by defaulting it to 1.0 if not provided. The implementation is sound and consistent with other parts of the codebase. I approve of this change.

[ispobock]

/rerun-stage unit-test-backend-8-gpu-h200

[ispobock]

/rerun-stage unit-test-backend-8-gpu-h200

[ispobock]

/tag-run-ci-label

[ispobock]

/rerun-stage unit-test-backend-8-gpu-h200