From e80bca309eb866e6d62e081e6ad80f3e10c113e9 Mon Sep 17 00:00:00 2001 From: Ning Xie Date: Sat, 30 Aug 2025 21:42:25 +0800 Subject: [PATCH] [Refactor] refactor freezing_value/cuda_event initialize outside try finally (#23758) Signed-off-by: Andy Xie --- vllm/v1/worker/cpu_model_runner.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/vllm/v1/worker/cpu_model_runner.py b/vllm/v1/worker/cpu_model_runner.py index 226d7792a42f7..360a626979e54 100644 --- a/vllm/v1/worker/cpu_model_runner.py +++ b/vllm/v1/worker/cpu_model_runner.py @@ -128,8 +128,8 @@ def _torch_cuda_wrapper(): self.record = lambda: None self.synchronize = lambda: None + cuda_event = torch.cuda.Event try: - cuda_event = torch.cuda.Event torch.cuda.Event = _EventPlaceholder yield finally: @@ -141,9 +141,9 @@ def _set_global_compilation_settings(config: VllmConfig): import torch._inductor.config inductor_config = config.compilation_config.inductor_compile_config + # Note: The MKLDNN and CPPGEMM backend requires freezing parameters. + freezing_value = torch._inductor.config.freezing try: - # Note: The MKLDNN and CPPGEMM backend requires freezing parameters. - freezing_value = torch._inductor.config.freezing if inductor_config.get("max_autotune", False): torch._inductor.config.freezing = True yield