[Perf] Enable Flashinfer autotune by default (#14357)
This commit is contained in:
@@ -2219,11 +2219,15 @@ class ModelRunner:
|
||||
|
||||
def _should_run_flashinfer_autotune(self) -> bool:
|
||||
"""Check if flashinfer autotune should be run."""
|
||||
if not self.server_args.enable_flashinfer_autotune:
|
||||
if self.server_args.disable_flashinfer_autotune:
|
||||
return False
|
||||
|
||||
backend_str = self.server_args.attention_backend
|
||||
if backend_str not in ["flashinfer", "trtllm_mla", "trtllm_mha"]:
|
||||
backend_str = self.server_args.moe_runner_backend
|
||||
if backend_str not in [
|
||||
"flashinfer_trtllm",
|
||||
"flashinfer_cutlass",
|
||||
"flashinfer_mxfp4",
|
||||
]:
|
||||
return False
|
||||
|
||||
major, _ = torch.cuda.get_device_capability()
|
||||
|
||||
Reference in New Issue
Block a user