aiter attention-backend (default enabled on AMD/ROCm) (#6381)
This commit is contained in:
@@ -957,6 +957,7 @@ class ServerArgs:
|
||||
"--attention-backend",
|
||||
type=str,
|
||||
choices=[
|
||||
"aiter",
|
||||
"flashinfer",
|
||||
"triton",
|
||||
"torch_native",
|
||||
|
||||
Reference in New Issue
Block a user