diff --git a/tests/conftest.py b/tests/conftest.py index f37c9883..5bbfd87f 100644 --- a/tests/conftest.py +++ b/tests/conftest.py @@ -365,7 +365,7 @@ class HfRunner: cleanup() -@pytest.fixture +@pytest.fixture(scope="session") def hf_runner(): return HfRunner @@ -385,6 +385,7 @@ class VllmRunner: block_size: int = 16, enable_chunked_prefill: bool = False, swap_space: int = 4, + enforce_eager: bool = False, **kwargs, ) -> None: self.model = LLM( @@ -393,6 +394,7 @@ class VllmRunner: trust_remote_code=True, dtype=dtype, swap_space=swap_space, + enforce_eager=enforce_eager, disable_log_stats=disable_log_stats, tensor_parallel_size=tensor_parallel_size, max_model_len=max_model_len,