[Attention] Flash Attention 3 - fp8 (#14570)

Signed-off-by: Mickael Seznec <mickael@mistral.ai>
2025-03-20 06:14:20 +01:00
parent ae65f3e237
commit a597a57595
15 changed files with 272 additions and 76 deletions
--- a/cmake/external_projects/vllm_flash_attn.cmake
+++ b/cmake/external_projects/vllm_flash_attn.cmake
@@ -38,7 +38,7 @@ else()
  FetchContent_Declare(
          vllm-flash-attn
          GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
-          GIT_TAG 9bfa9869829d8c593527eb34c5271d0090f7ccc9 
+          GIT_TAG dc9d410b3e2d6534a4c70724c2515f4def670a22
          GIT_PROGRESS TRUE
          # Don't share the vllm-flash-attn build between build types
          BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn