Turn on flashinfer by default (#578)
This commit is contained in:
@@ -26,7 +26,7 @@ class RadixAttention(nn.Module):
|
||||
|
||||
from sglang.srt.managers.controller.model_runner import global_server_args_dict
|
||||
|
||||
if global_server_args_dict.get("enable_flashinfer", False):
|
||||
if not global_server_args_dict.get("disable_flashinfer", False):
|
||||
self.prefill_forward = self.prefill_forward_flashinfer
|
||||
self.extend_forward = self.prefill_forward_flashinfer
|
||||
self.decode_forward = self.decode_forward_flashinfer
|
||||
|
||||
Reference in New Issue
Block a user