forked from Karylab-cklius/vllm
@@ -70,8 +70,6 @@ class TokenspeedMLAPrefillBackend(MLAPrefillBackend):
|
||||
qk_rope_head_dim: int,
|
||||
v_head_dim: int,
|
||||
vllm_config: "VllmConfig",
|
||||
device: torch.device,
|
||||
layer_names: list[str] | None = None,
|
||||
) -> None:
|
||||
super().__init__(
|
||||
num_heads=num_heads,
|
||||
@@ -81,8 +79,6 @@ class TokenspeedMLAPrefillBackend(MLAPrefillBackend):
|
||||
qk_rope_head_dim=qk_rope_head_dim,
|
||||
v_head_dim=v_head_dim,
|
||||
vllm_config=vllm_config,
|
||||
device=device,
|
||||
layer_names=layer_names,
|
||||
)
|
||||
|
||||
# Pre-JIT the kernel for both BF16 and FP8 prefill shapes so the first
|
||||
|
||||
Reference in New Issue
Block a user