Move unnecessary input_addr capture under debug mode flag for speed-up (#13690)

2025-11-22 11:42:26 -08:00
parent 3990b84bd3
commit b29769f3b6
5 changed files with 24 additions and 8 deletions
@@ -170,6 +170,7 @@ class PiecewiseCudaGraphRunner:
        self.compile_config = CompilationConfig(
            self.model_runner.server_args.piecewise_cuda_graph_tokens,
            self.model_runner.server_args.piecewise_cuda_graph_compiler,
+            self.model_runner.server_args.enable_torch_compile_debug_mode,
        )
        self.quant_config = getattr(self.model_runner.model, "quant_config", None)