diff --git a/tests/conftest.py b/tests/conftest.py index a50985a465e6c..e3e412f90af5d 100644 --- a/tests/conftest.py +++ b/tests/conftest.py @@ -731,6 +731,9 @@ class VllmRunner: init_ctx = (nullcontext() if default_torch_num_threads is None else set_default_torch_num_threads(default_torch_num_threads)) + if not kwargs.get("compilation_config", None): + kwargs["compilation_config"] = {"cudagraph_capture_sizes": [8]} + with init_ctx: self.llm = LLM( model=model_name,