diff --git a/vllm/v1/worker/gpu/cudagraph_utils.py b/vllm/v1/worker/gpu/cudagraph_utils.py index 384f1192435..0648de29859 100644 --- a/vllm/v1/worker/gpu/cudagraph_utils.py +++ b/vllm/v1/worker/gpu/cudagraph_utils.py @@ -299,7 +299,6 @@ class CudaGraphManager: self.breakable_cg_runner = BreakableCUDAGraphWrapper( model, self.vllm_config ) - self.breakable_cg_runner.graph_pool = self.pool def run_pw_graph(self, model: nn.Module, model_inputs: dict[str, Any]) -> Any: if not self.use_breakable_cg: