[Bugfix] Avoid import AttentionMetadata explicitly in Mllama (#10593)
Signed-off-by: Isotr0py <2037008807@qq.com>
This commit is contained in:
@@ -87,6 +87,11 @@ class BlocksparseParams:
|
||||
|
||||
class BlocksparseFlashAttentionBackend(AttentionBackend):
|
||||
|
||||
@staticmethod
|
||||
def get_name() -> str:
|
||||
# For attention layer compatibility
|
||||
return "FLASH_ATTN"
|
||||
|
||||
@staticmethod
|
||||
def get_impl_cls() -> Type["BlocksparseFlashAttentionImpl"]:
|
||||
return BlocksparseFlashAttentionImpl
|
||||
|
||||
Reference in New Issue
Block a user