Switch `return_dict` to `True` by default. (#8530)

* Use the CI to identify failing tests

* Remove from all examples and tests

* More default switch

* Fixes

* More test fixes

* More fixes

* Last fixes hopefully

* Use the CI to identify failing tests

* Remove from all examples and tests

* More default switch

* Fixes

* More test fixes

* More fixes

* Last fixes hopefully

* Run on the real suite

* Fix slow tests

tests/test_modeling_tf_xlm_roberta.py

@@ -39,7 +39,7 @@ class TFFlaubertModelIntegrationTest(unittest.TestCase):
             "attention_mask": tf.convert_to_tensor([[1, 1, 1, 1, 1, 1]], dtype=tf.int32),
         }
 
-        output = model(features, return_dict=True)["last_hidden_state"]
+        output = model(features)["last_hidden_state"]
         expected_shape = tf.TensorShape((1, 6, 768))
         self.assertEqual(output.shape, expected_shape)
         # compare the actual values for a slice.

tests/test_modeling_tf_xlnet.py

@@ -111,7 +111,6 @@ class TFXLNetModelTester:
             bos_token_id=self.bos_token_id,
             pad_token_id=self.pad_token_id,
             eos_token_id=self.eos_token_id,
-            return_dict=True,
         )
 
         return (

tests/test_modeling_transfo_xl.py

@@ -78,7 +78,6 @@ class TransfoXLModelTester:
             div_val=self.div_val,
             n_layer=self.num_hidden_layers,
             eos_token_id=self.eos_token_id,
-            return_dict=True,
         )
 
         return (config, input_ids_1, input_ids_2, lm_labels)

tests/test_modeling_xlm.py

@@ -116,7 +116,6 @@ class XLMModelTester:
             use_proj=self.use_proj,
             num_labels=self.num_labels,
             bos_token_id=self.bos_token_id,
-            return_dict=True,
         )
 
         return (

tests/test_modeling_xlm_roberta.py

@@ -32,7 +32,7 @@ if is_torch_available():
 class XLMRobertaModelIntegrationTest(unittest.TestCase):
     @slow
     def test_xlm_roberta_base(self):
-        model = XLMRobertaModel.from_pretrained("xlm-roberta-base", return_dict=True)
+        model = XLMRobertaModel.from_pretrained("xlm-roberta-base")
         input_ids = torch.tensor([[0, 581, 10269, 83, 99942, 136, 60742, 23, 70, 80583, 18276, 2]])
         # The dog is cute and lives in the garden house
 
@@ -51,7 +51,7 @@ class XLMRobertaModelIntegrationTest(unittest.TestCase):
 
     @slow
     def test_xlm_roberta_large(self):
-        model = XLMRobertaModel.from_pretrained("xlm-roberta-large", return_dict=True)
+        model = XLMRobertaModel.from_pretrained("xlm-roberta-large")
         input_ids = torch.tensor([[0, 581, 10269, 83, 99942, 136, 60742, 23, 70, 80583, 18276, 2]])
         # The dog is cute and lives in the garden house
 

tests/test_modeling_xlnet.py

@@ -148,7 +148,6 @@ class XLNetModelTester:
             bos_token_id=self.bos_token_id,
             pad_token_id=self.pad_token_id,
             eos_token_id=self.eos_token_id,
-            return_dict=True,
         )
 
         return (