Implements dual-chunk-flash-attn backend for dual chunk attention with sparse attention support (#11844)
This commit is contained in:
@@ -51,6 +51,7 @@ class _Backend(enum.Enum):
|
||||
PALLAS_VLLM_V1 = enum.auto()
|
||||
IPEX = enum.auto()
|
||||
BLOCK_SPARSE_FLASH_ATTN = enum.auto()
|
||||
DUAL_CHUNK_FLASH_ATTN = enum.auto()
|
||||
NO_ATTENTION = enum.auto()
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user