diff --git a/vllm/v1/worker/gpu_worker.py b/vllm/v1/worker/gpu_worker.py index df1d9de6ae..3a3c5314b6 100644 --- a/vllm/v1/worker/gpu_worker.py +++ b/vllm/v1/worker/gpu_worker.py @@ -838,9 +838,12 @@ class Worker(WorkerBase): # cuda graph capture. kernel_warmup(self) - if self.use_v2_model_runner: - # A workspace resize after capture frees what the graphs point at. - warmup_kernels(self.model_runner, self.execute_model, self.sample_tokens) + if self.use_v2_model_runner: # noqa: SIM102 + if self.vllm_config.kernel_config.enable_jit_warmup: + # A workspace resize after capture frees what the graphs point at. + warmup_kernels( + self.model_runner, self.execute_model, self.sample_tokens + ) cuda_graph_memory_bytes = 0 if not self.model_config.enforce_eager: