biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Attention] Add ROCM_AITER_MLA_SPARSE to attention backend registry (#29103)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni
2025-11-20 23:24:43 -05:00
committed by GitHub
parent 8c25f9cfb6
commit 11857a00b0
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/attention/backends/registry.py
View File

@@ -52,6 +52,9 @@ class AttentionBackendEnum(Enum, metaclass=_AttentionBackendEnumMeta):
ROCM_AITER_FA = ( ROCM_AITER_FA = (
"vllm.v1.attention.backends.rocm_aiter_fa.AiterFlashAttentionBackend" "vllm.v1.attention.backends.rocm_aiter_fa.AiterFlashAttentionBackend"
) )
ROCM_AITER_MLA_SPARSE = (
"vllm.v1.attention.backends.mla.rocm_aiter_mla_sparse.ROCMAiterMLASparseBackend"
)
TORCH_SDPA = "" # this tag is only used for ViT TORCH_SDPA = "" # this tag is only used for ViT
FLASHINFER = "vllm.v1.attention.backends.flashinfer.FlashInferBackend" FLASHINFER = "vllm.v1.attention.backends.flashinfer.FlashInferBackend"
FLASHINFER_MLA = ( FLASHINFER_MLA = (

5
vllm/platforms/rocm.py
View File

@@ -233,10 +233,7 @@ class RocmPlatform(Platform):
"Sparse MLA backend on ROCm only supports block size 1 for now." "Sparse MLA backend on ROCm only supports block size 1 for now."
) )
logger.info_once("Using Sparse MLA backend on V1 engine.") logger.info_once("Using Sparse MLA backend on V1 engine.")
return ( return AttentionBackendEnum.ROCM_AITER_MLA_SPARSE.get_path()
"vllm.v1.attention.backends.mla.rocm_aiter_mla_sparse."
"ROCMAiterMLASparseBackend"
)
if use_mla: if use_mla:
if selected_backend is None: if selected_backend is None: