Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Flash Attention 3 (FA3) Support #12429

Open
3 tasks
mgoin opened this issue Jan 25, 2025 · 1 comment
Open
3 tasks

Flash Attention 3 (FA3) Support #12429

mgoin opened this issue Jan 25, 2025 · 1 comment

Comments

@mgoin
Copy link
Member

mgoin commented Jan 25, 2025

As of #12093 Flash Attention 3 is now supported in vLLM for Hopper GPUs (SM 9.0).

It can also be enabled for SM 8.0 and 8.7 using VLLM_FLASH_ATTN_VERSION=3.

For 8.6 and 8.9 its fully disabled since they don't have enough shared memory for the current implementation, some work needs to be done here.

This issue tracks the remaining features that have yet to be implemented

Hardware Support

  • SM 8.9 Ada Lovelace (L4, L40s) Support
  • SM 8.6 Ampere (A6000) Support

Optimizations

  • FP8 Attention
@mgoin mgoin changed the title FlashAttention3 Flash Attention 3 Support Jan 25, 2025
@mgoin mgoin changed the title Flash Attention 3 Support Flash Attention 3 (FA3) Support Jan 25, 2025
@zhangjiekui
Copy link

As of #12093 Flash Attention 3 is now supported in vLLM for Hopper GPUs (SM 9.0).

It can also be enabled for SM 8.0 and 8.7 using VLLM_FLASH_ATTN_VERSION=3.

For 8.6 and 8.9 its fully disabled since they don't have enough shared memory for the current implementation, some work needs to be done here.

This issue tracks the remaining features that have yet to be implemented

Hardware Support

  • SM 8.9 Ada Lovelace (L4, L40s) Support[ ] SM 8.6 Ampere (A6000) Support

Optimizations

  • FP8 Attention

Watch

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants