use eager mode by default in ut

Signed-off-by: MengqingCao <cmq0113@163.com>

Author: MengqingCao

tests/compile/test_aclgraph.py

@@ -96,5 +96,7 @@ def test_deepseek_raises_error(monkeypatch: pytest.MonkeyPatch) -> None:
         m.setenv("VLLM_USE_MODELSCOPE", "True")
         m.setenv("VLLM_USE_V1", "1")
         with pytest.raises(NotImplementedError) as excinfo:
-            VllmRunner("deepseek-ai/DeepSeek-V2-Lite-Chat", max_model_len=1024)
+            VllmRunner("deepseek-ai/DeepSeek-V2-Lite-Chat",
+                       max_model_len=1024,
+                       enforce_eager=False)
         assert "ACL Graph does not support deepseek" in str(excinfo.value)

tests/conftest.py

@@ -77,7 +77,7 @@ def __init__(
         block_size: int = 16,
         enable_chunked_prefill: bool = False,
         swap_space: int = 4,
-        enforce_eager: Optional[bool] = False,
+        enforce_eager: Optional[bool] = True,
         **kwargs,
     ) -> None:
         self.model = LLM(