diff --git a/tests/conftest.py b/tests/conftest.py index e3e412f90af5d..66106d1bf7792 100644 --- a/tests/conftest.py +++ b/tests/conftest.py @@ -732,7 +732,7 @@ class VllmRunner: set_default_torch_num_threads(default_torch_num_threads)) if not kwargs.get("compilation_config", None): - kwargs["compilation_config"] = {"cudagraph_capture_sizes": [8]} + kwargs["compilation_config"] = {"cudagraph_capture_sizes": [4]} with init_ctx: self.llm = LLM(