[FA/Chore] Bump vllm-flash-attention

[LucasWilkinson]

Bump FA to pickup

vllm-project/flash-attention#94
vllm-project/flash-attention#91
vllm-project/flash-attention#87

[gemini-code-assist]

Code Review

This pull request updates the vllm-flash-attention dependency to a newer commit to pick up several bug fixes. The change is straightforward and correct. I've added one suggestion to improve maintainability by documenting the reason for pinning this specific commit hash.

[gemini-code-assist]

For better maintainability, it's good practice to add a comment explaining what this Git commit hash corresponds to. This helps other developers understand why this specific commit is being used without having to look it up. You could mention the PRs this commit includes, as you did in the pull request description.

          # Includes fixes for FA3 build on CUDA 12.3, 12.4, and 12.5.
          # See https://github.com/vllm-project/flash-attention/pull/94
          GIT_TAG 4695e6bed5366c41e28c06cd86170166e4f43d00

[jmkuebler]

@LucasWilkinson would you mind bumping directly to 07602adc2baa7a3ea037321ef2ca71c83096e5d5 to include vllm-project/flash-attention#96 as well?

[LucasWilkinson]

@LucasWilkinson would you mind bumping directly to 07602adc2baa7a3ea037321ef2ca71c83096e5d5 to include vllm-project/flash-attention#96 as well?

It takes a while to get it through CI (need to populate sccache) so ill try to create a fast follow that includes:

vllm-project/flash-attention#96
vllm-project/flash-attention#95

👍 apologies for the the cumbersome process