[Fix] Set fa3 as default MHA backend on Hopper

[Fridge003]

Motivation

After flashinfer is upgraded to 0.6.1, the prefill kernel on Hopper caused some performance drops (#17411)
So we fallback to fa3 by default

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.
• 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 @Fridge003, 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 critical performance fix for Hopper GPU architectures. Following a recent upgrade of the flashinfer library to version 0.6.1, a significant performance drop was identified in the prefill kernel. To counteract this, the system will now default to using the fa3 Multi-Head Attention backend on Hopper devices running with CUDA 12.3, ensuring optimal performance until the underlying flashinfer kernel issue is resolved.

Highlights

• Default MHA Backend for Hopper: The default Multi-Head Attention (MHA) backend for Hopper architectures running with CUDA 12.3 is now explicitly set to fa3.
• Performance Regression Fix: This change addresses a performance regression observed in the prefill kernel after upgrading flashinfer to version 0.6.1 on Hopper GPUs.
• Simplified Backend Selection Logic: The condition for enabling the fa3 backend on Hopper has been simplified by removing the is_fa3_default_architecture check, making fa3 the default under the specified hardware and software configuration.

🧠 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. ↩

[Fridge003]

/tag-and-rerun-ci

[gemini-code-assist]

Code Review

This pull request addresses a performance regression on Hopper GPUs by setting fa3 as the default Multi-Head Attention (MHA) backend. The changes are straightforward and well-justified. The removal of the is_fa3_default_architecture check correctly implements the fallback for all MHA models on Hopper, as intended. The added code comment clearly explains the reason for this temporary fix. The code looks good and I have no further suggestions.

[ispobock]

We need to add it back once flashinfer issue fixed.