[2/3] Optimize Slime Update Weights: Avoid GPU-to-CPU Device Sync when update expert weights

[hebiao064]

Motivation

• 1/3: [1/3] Optimize Slime Update Weights: Remove QWen3MOE Load Weight Overhead #8751 （23s => 22s)

• 2/3: [2/3] Optimize Slime Update Weights: Avoid GPU-to-CPU Device Sync when update expert weights #8753 (22s => 12s)

• 3/3: [3/3] Optimize Slime Update Weights: Load Weight in non-blocking mode #8754 (12s => 10s)

• This PR[2/3] Optimize Slime Update Weights: Avoid GPU-to-CPU Device Sync when update expert weights

Below graph shows a unnecessary device sync from Device to Host

This PR maps to optimization 2

Modifications

Accuracy Test

Benchmark & Profiling

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://slack.sglang.ai to discuss your PR.

[gemini-code-assist]

Summary of Changes

Hello @hebiao064, 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!

I've implemented a performance optimization aimed at reducing the overhead associated with updating expert weights. This change specifically targets the elimination of frequent and expensive GPU-to-CPU device synchronization, which was identified as a bottleneck. By caching necessary data on the CPU, I've streamlined the process of accessing expert location information, leading to more efficient weight updates.

Highlights

• Performance Optimization for Expert Weight Updates: I've introduced a caching mechanism within the logical_to_all_physical method. By creating a CPU copy (_logical_to_all_physical_map_cpu) of the logical_to_all_physical_map tensor, we can avoid repeated and costly GPU-to-CPU data transfers, which are particularly expensive during expert weight updates.

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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

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 issue 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 or fill out our survey to provide feedback.

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 introduces an optimization to avoid GPU-to-CPU synchronization in logical_to_all_physical by caching a CPU copy of the logical_to_all_physical_map tensor. While the optimization is a good idea, the current implementation introduces a critical bug where the cached data can become stale. The update method modifies the GPU tensor in-place but does not invalidate the new CPU cache, which will lead to incorrect expert routing. I've added a comment with details on the issue and how to resolve it.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!