unset lazy&verbose env var when disabling mock torch by command line

python/oneflow/mock_torch/__init__.py

@@ -46,11 +46,17 @@ def __getattr__(self, name: str) -> Any:
                 return [attr for attr in dir(self.module) if not attr.startswith("_")]
             new_name = self.module.__name__ + "." + name
             if _importer.lazy:
-                if _importer.verbose:
-                    print(
-                        f"{new_name} is not found in oneflow, use dummy object as fallback."
-                    )
-                return DummyModule(new_name)
+                blacklist = ["scaled_dot_product_attention"]
+                if name in blacklist:
+                    if _importer.verbose:
+                        print(f'"{new_name}" is in blacklist, raise AttributeError')
+                    raise AttributeError(new_name + error_msg)
+                else:
+                    if _importer.verbose:
+                        print(
+                            f'"{new_name}" is not found in oneflow, use dummy object as fallback.'
+                        )
+                    return DummyModule(new_name)
             else:
                 raise AttributeError(new_name + error_msg)
         attr = getattr(self.module, name)

python/oneflow/mock_torch/__main__.py

@@ -41,7 +41,9 @@ def main():
         paths = os.environ["PYTHONPATH"].rstrip(":").split(":")
         paths = [x for x in paths if x != str(torch_env)]
         path = ":".join(paths)
-        print("export PYTHONPATH=" + path)
+        print(
+            f"export PYTHONPATH={path}; unset ONEFLOW_MOCK_TORCH_LAZY; unset ONEFLOW_MOCK_TORCH_VERBOSE"
+        )
 
 
 if __name__ == "__main__":

python/oneflow/test/misc/test_mock_scope.py

@@ -154,6 +154,16 @@ def test_mock_lazy_for_loop(test_case):
             for _ in torch.not_exist:
                 pass
 
+    def test_blacklist(test_case):
+        with mock.enable(lazy=True):
+            import torch
+            import torch.nn.functional as F
+
+            test_case.assertFalse(hasattr(F, "scaled_dot_product_attention"))
+            test_case.assertFalse(
+                hasattr(torch.nn.functional, "scaled_dot_product_attention")
+            )
+
 
 # MUST use pytest to run this test
 def test_verbose(capsys):