[Tests] Add Common Test for Training + Fix a couple of bugs (#8415)

* add training tests

* correct longformer

* fix docs

* fix some tests

* fix some more train tests

* remove ipdb

* fix multiple edge case model training

* fix funnel and prophetnet

* clean gpt models

* undo renaming of albert

tests/test_modeling_xlm.py

@@ -351,6 +351,21 @@ class XLMModelTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
         (XLMWithLMHeadModel,) if is_torch_available() else ()
     )  # TODO (PVP): Check other models whether language generation is also applicable
 
+    # XLM has 2 QA models -> need to manually set the correct labels for one of them here
+    def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):
+        inputs_dict = super()._prepare_for_class(inputs_dict, model_class, return_labels=return_labels)
+
+        if return_labels:
+            if model_class.__name__ == "XLMForQuestionAnswering":
+                inputs_dict["start_positions"] = torch.zeros(
+                    self.model_tester.batch_size, dtype=torch.long, device=torch_device
+                )
+                inputs_dict["end_positions"] = torch.zeros(
+                    self.model_tester.batch_size, dtype=torch.long, device=torch_device
+                )
+
+        return inputs_dict
+
     def setUp(self):
         self.model_tester = XLMModelTester(self)
         self.config_tester = ConfigTester(self, config_class=XLMConfig, emb_dim=37)