diff --git a/vllm_mindspore/model_executor/models/mf_models/mindformers.py b/vllm_mindspore/model_executor/models/mf_models/mindformers.py index 325584e804a0180a378a6cada3e6825eadfeace7..44989c80e8654cd20ecd0aa6c74fca54f7eeec8b 100644 --- a/vllm_mindspore/model_executor/models/mf_models/mindformers.py +++ b/vllm_mindspore/model_executor/models/mf_models/mindformers.py @@ -450,6 +450,7 @@ class MindFormersForCausalLM(MsModelBase, SupportsPP): else: logits = self.lm_head(hidden_states) logits = logits.view(-1, logits.shape[-1]) + ms.runtime.synchronize() return logits def capture_start_time(self, weights: Iterable[tuple[str, Tensor]]):