xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-10 07:40:13 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Attention] Register FLASHMLA_SPARSE (#26441)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni 2025-10-08 18:28:52 -04:00 committed by GitHub
parent da364615fc
commit 2a03f93de9
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/attention/backends/registry.py
View File

@ -21,6 +21,7 @@ class _Backend(enum.Enum):
TRITON_MLA = enum.auto() TRITON_MLA = enum.auto()
CUTLASS_MLA = enum.auto() CUTLASS_MLA = enum.auto()
FLASHMLA = enum.auto() FLASHMLA = enum.auto()
FLASHMLA_SPARSE = enum.auto()
FLASH_ATTN_MLA = enum.auto() FLASH_ATTN_MLA = enum.auto()
PALLAS = enum.auto() PALLAS = enum.auto()
IPEX = enum.auto() IPEX = enum.auto()
@ -43,6 +44,7 @@ BACKEND_MAP = {
_Backend.TRITON_MLA: "vllm.v1.attention.backends.mla.triton_mla.TritonMLABackend", # noqa: E501 _Backend.TRITON_MLA: "vllm.v1.attention.backends.mla.triton_mla.TritonMLABackend", # noqa: E501
_Backend.CUTLASS_MLA: "vllm.v1.attention.backends.mla.cutlass_mla.CutlassMLABackend", # noqa: E501 _Backend.CUTLASS_MLA: "vllm.v1.attention.backends.mla.cutlass_mla.CutlassMLABackend", # noqa: E501
_Backend.FLASHMLA: "vllm.v1.attention.backends.mla.flashmla.FlashMLABackend", # noqa: E501 _Backend.FLASHMLA: "vllm.v1.attention.backends.mla.flashmla.FlashMLABackend", # noqa: E501
_Backend.FLASHMLA_SPARSE: "vllm.v1.attention.backends.mla.flashmla_sparse.FlashMLASparseBackend", # noqa: E501
_Backend.FLASH_ATTN_MLA: "vllm.v1.attention.backends.mla.flashattn_mla.FlashAttnMLABackend", # noqa: E501 _Backend.FLASH_ATTN_MLA: "vllm.v1.attention.backends.mla.flashattn_mla.FlashAttnMLABackend", # noqa: E501
_Backend.PALLAS: "vllm.v1.attention.backends.pallas.PallasAttentionBackend", # noqa: E501 _Backend.PALLAS: "vllm.v1.attention.backends.pallas.PallasAttentionBackend", # noqa: E501
_Backend.FLEX_ATTENTION: "vllm.v1.attention.backends.flex_attention.FlexAttentionBackend", # noqa: E501 _Backend.FLEX_ATTENTION: "vllm.v1.attention.backends.flex_attention.FlexAttentionBackend", # noqa: E501

2
vllm/v1/attention/backends/mla/flashmla_sparse.py
View File

@ -55,7 +55,7 @@ class FlashMLASparseBackend(AttentionBackend):
@staticmethod @staticmethod
def get_name() -> str: def get_name() -> str:
return "FLASHMLA_SPARSE_VLLM_V1" return "FLASHMLA_SPARSE"
@staticmethod @staticmethod
def get_metadata_cls() -> type[AttentionMetadata]: def get_metadata_cls() -> type[AttentionMetadata]: