update testing

Author: jiqing-feng

optimum/intel/ipex/modeling_base.py

@@ -193,7 +193,7 @@ def forward(
             inputs["token_type_ids"] = token_type_ids
 
         outputs = self._call_model(**inputs)
-        return ModelOutput(**outputs) if isinstance(outputs, dict) else ModelOutput(logits=outputs[0])
+        return ModelOutput(**outputs) if isinstance(outputs, dict) else ModelOutput(last_hidden_state=outputs[0])
 
     def eval(self):
         self.model.eval()
@@ -282,7 +282,7 @@ def forward(
             inputs["attention_mask"] = attention_mask
 
         outputs = self._call_model(**inputs)
-        return ModelOutput(**outputs) if isinstance(outputs, dict) else ModelOutput(last_hidden_state=outputs[0])
+        return ModelOutput(**outputs) if isinstance(outputs, dict) else ModelOutput(logits=outputs[0])
 
 
 class IPEXModelForQuestionAnswering(IPEXModel):

tests/ipex/test_modeling.py

@@ -126,9 +126,10 @@ def test_compare_to_transformers(self, model_arch):
         with torch.no_grad():
             transformers_outputs = transformers_model(**tokens)
         outputs = ipex_model(**tokens)
-        self.assertTrue(
-            torch.allclose(outputs["last_hidden_state"], transformers_outputs["last_hidden_state"], atol=1e-4)
-        )
+        # Compare tensor outputs
+        for output_name in {"logits", "last_hidden_state"}:
+            if output_name in transformers_outputs:
+                self.assertTrue(torch.allclose(outputs[output_name], transformers_outputs[output_name], atol=1e-4))
 
     @parameterized.expand(SUPPORTED_ARCHITECTURES)
     def test_pipeline(self, model_arch):