FA cute

[hyhieu]

Motivation

Integrate Flash Attention's implementation in CuTe DSL.

Modifications

• Copy the code into sglang/srt/layers/attention/cute_ops
• Create a new backend blackwell_prefill_attention_backend.py
• Allow --prefill-attention-backend to take the value "fa-cute"

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.

[gemini-code-assist]

Summary of Changes

Hello @hyhieu, 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 integrates a new, highly optimized Flash Attention implementation tailored for NVIDIA Blackwell (SM100) GPUs. It introduces a dedicated prefill attention backend that utilizes a newly added Flash Attention kernel built with NVIDIA's CuTe DSL. This enhancement aims to significantly boost performance for prefill operations, particularly with paged KV cache, by leveraging low-level hardware optimizations.

Highlights

• New Blackwell-Optimized Attention Backend: Introduces a new fa-cute attention backend specifically designed for NVIDIA Blackwell (SM100) GPUs, leveraging their advanced hardware capabilities.
• CuTe DSL-based Flash Attention Kernel: Integrates a highly optimized Flash Attention kernel built using NVIDIA's CuTe DSL, enabling significant performance improvements for attention computations.
• Dedicated Prefill Attention Backend: Adds a dedicated BlackwellPrefillAttentionBackend to handle prefill operations efficiently, utilizing the new Flash Attention kernel and supporting paged KV cache.
• Automatic Page Size Configuration: Ensures compatibility and optimal performance by automatically setting the page_size to 128 when the fa-cute backend is selected, aligning with the kernel's requirements.

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. 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 integrates Flash Attention's implementation in CuTe DSL, adding a new attention backend fa-cute for Blackwell GPUs. The changes are well-structured, including the new backend, supporting kernel code, and necessary updates to the model runner and server arguments. I have a couple of minor suggestions to remove some dead code and a debug print statement to improve code clarity and maintainability.

[yiakwy-xpu-ml-framework-team]

Hi @hyhieu really nice job for integrating new attention backend. Do we have any performance benchmarking against latest triton (triton_kernels), and cutlass implementation ?