Skip to content

Commit b0b1efa

Browse files
wxsIceywangxiyuan
authored andcommitted
Rebase main
Signed-off-by: wxsIcey <[email protected]> Signed-off-by: wangxiyuan <[email protected]>
1 parent c4b5a53 commit b0b1efa

File tree

1 file changed

+1
-1
lines changed

1 file changed

+1
-1
lines changed

vllm_ascend/worker/model_runner_v1.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2511,7 +2511,7 @@ def profile_run(self) -> None:
25112511
# MC2 will consume additional NPU memory.
25122512
# Therefore, we need to run the MC2 path once here to complete its initialization,
25132513
# allowing vLLM to correctly estimate the maximum memory required.
2514-
if not self.ascend_config.torchair_graph_config.enabled and self._select_moe_comm_method(
2514+
if self._select_moe_comm_method(
25152515
self.mc2_tokens_capacity,
25162516
with_prefill=True) == MoECommType.MC2:
25172517
self._dummy_run(self.mc2_tokens_capacity, with_prefill=True)

0 commit comments

Comments
 (0)