xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-15 09:15:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix FA2 fallback for Blackwell V1 (#19781)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin 2025-06-19 10:53:55 +09:00 committed by GitHub
parent dfada85eee
commit 36239f79dd
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/platforms/cuda.py
View File

@ -255,7 +255,7 @@ class CudaPlatformBase(Platform):
"install FlashInfer for better performance.") "install FlashInfer for better performance.")
pass pass
# FlashAttention is the default for SM 8.0+ GPUs # FlashAttention is the default for SM 8.0+ GPUs
elif cls.has_device_capability(80): if cls.has_device_capability(80):
logger.info_once("Using Flash Attention backend on V1 engine.") logger.info_once("Using Flash Attention backend on V1 engine.")
return ("vllm.v1.attention.backends." return ("vllm.v1.attention.backends."
"flash_attn.FlashAttentionBackend") "flash_attn.FlashAttentionBackend")