xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-14 00:55:26 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Attention][Bugfix] Fix FA sink support (#28660)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni 2025-11-13 12:20:01 -06:00 committed by GitHub
parent 119c4927b3
commit f9f3b596f3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/v1/attention/backends/flash_attn.py
View File

@ -130,6 +130,12 @@ class FlashAttentionBackend(AttentionBackend):
return flash_attn_supports_fp8() return flash_attn_supports_fp8()
return kv_cache_dtype in ["auto"] return kv_cache_dtype in ["auto"]
@classmethod
def supports_sink(cls) -> bool:
if not is_flash_attn_varlen_func_available():
return False
return flash_attn_supports_sinks()
@classmethod @classmethod
def supports_compute_capability(cls, capability: DeviceCapability) -> bool: def supports_compute_capability(cls, capability: DeviceCapability) -> bool:
return capability >= DeviceCapability(8, 0) return capability >= DeviceCapability(8, 0)