Make the test simpler

tests/utils/test_modeling_utils.py

@@ -467,16 +467,11 @@ def is_on_meta(model_id, dtype):
                 model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=dtype)
                 return all(value.device.type == "meta" for value in model.state_dict().values())
 
-        result = {}
-        expected = {}
         model_ids = ("fxmarty/tiny-llama-fast-tokenizer", "fxmarty/small-llama-testing")
         dtypes = (None, "auto", torch.float16)
 
         for model_id, dtype in itertools.product(model_ids, dtypes):
-            result[(model_id, dtype)] = is_on_meta(model_id, dtype)
-            expected[(model_id, dtype)] = True
-
-        assert result == expected
+            self.assertTrue(is_on_meta(model_id, dtype))
 
     def test_model_from_pretrained_torch_dtype(self):
         # test that the model can be instantiated with dtype of either