Disable PCG for draft worker (#16354)
This commit is contained in:
@@ -1517,6 +1517,9 @@ class ModelRunner(ModelRunnerKVCacheMixin):
|
|||||||
return self.mamba2_config or self.hybrid_gdn_config or self.kimi_linear_config
|
return self.mamba2_config or self.hybrid_gdn_config or self.kimi_linear_config
|
||||||
|
|
||||||
def can_run_piecewise_cuda_graph(self):
|
def can_run_piecewise_cuda_graph(self):
|
||||||
|
if self.is_draft_worker:
|
||||||
|
return False
|
||||||
|
|
||||||
if self.server_args.enable_torch_compile:
|
if self.server_args.enable_torch_compile:
|
||||||
log_info_on_rank0(
|
log_info_on_rank0(
|
||||||
logger,
|
logger,
|
||||||
|
|||||||
Reference in New Issue
Block a user