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

[Model Runner V2] Support FLASHINFER_MLA backend (#32709)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon
2026-01-20 10:26:17 -08:00
committed by GitHub
parent 04a9e064db
commit 9ab4388cd3
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/v1/worker/gpu/attn_utils.py
View File

@@ -57,7 +57,7 @@ def init_attn_backend(
)
attn_metadata_builders.append(attn_metadata_builder) # type: ignore
if "FLASHINFER" in attn_backend.get_name():
if attn_backend.get_name() == "FLASHINFER":
if flashinfer_workspace is None:
flashinfer_workspace = attn_metadata_builder._get_workspace_buffer()
else:

2
vllm/v1/worker/gpu/model_runner.py
View File

@@ -248,7 +248,7 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
)
# TODO(woosuk): Support other backends.
supported_backends = ("FLASH_ATTN", "FLASHINFER")
supported_backends = ("FLASH_ATTN", "FLASHINFER", "FLASHINFER_MLA")
for backend in self.attn_backends.values():
backend_name = backend.get_name()
if backend_name not in supported_backends: