Full rework of the TF input/output embeddings and bias resizing (#9193)

* Start rework resizing

* Rework bias/decoder resizing

* Full resizing rework

* Full resizing rework

* Start to update the models with the new approach

* Finish to update the models

* Update all the tests

* Update the template

* Fix tests

* Fix tests

* Test a new approach

* Refactoring

* Refactoring

* Refactoring

* New rework

* Rework BART

* Rework bert+blenderbot

* Rework CTRL

* Rework Distilbert

* Rework DPR

* Rework Electra

* Rework Flaubert

* Rework Funnel

* Rework GPT2

* Rework Longformer

* Rework Lxmert

* Rework marian+mbart

* Rework mobilebert

* Rework mpnet

* Rework openai

* Rework pegasus

* Rework Roberta

* Rework T5

* Rework xlm+xlnet

* Rework template

* Fix TFT5EncoderOnly + DPRs

* Restore previous methods

* Fix Funnel

* Fix CTRL and TransforXL

* Apply style

* Apply Sylvain's comments

* Restore a test in DPR

* Address the comments

* Fix bug

* Apply style

* remove unused import

* Fix test

* Forgot a method

* missing test

* Trigger CI

* naming update

* Rebase

* Trigger CI

src/transformers/models/roberta/modeling_tf_roberta.py

@@ -15,6 +15,8 @@
 # limitations under the License.
 """ TF 2.0 RoBERTa model. """
 
+import warnings
+
 import tensorflow as tf
 
 from ...activations_tf import get_tf_activation
@@ -502,7 +504,7 @@ class TFRobertaMainLayer(tf.keras.layers.Layer):
     # Copied from transformers.models.bert.modeling_tf_bert.TFBertMainLayer.set_input_embeddings
     def set_input_embeddings(self, value):
         self.embeddings.word_embeddings = value
-        self.embeddings.vocab_size = value.shape[0]
+        self.embeddings.vocab_size = shape_list(value)[0]
 
     # Copied from transformers.models.bert.modeling_tf_bert.TFBertMainLayer._prune_heads
     def _prune_heads(self, heads_to_prune):
@@ -827,6 +829,20 @@ class TFRobertaLMHead(tf.keras.layers.Layer):
 
         super().build(input_shape)
 
+    def get_output_embeddings(self):
+        return self.decoder
+
+    def set_output_embeddings(self, value):
+        self.decoder.word_embeddings = value
+        self.decoder.vocab_size = shape_list(value)[0]
+
+    def get_bias(self):
+        return {"bias": self.bias}
+
+    def set_bias(self, value):
+        self.bias = value["bias"]
+        self.vocab_size = shape_list(value["bias"])[0]
+
     def call(self, hidden_states):
         hidden_states = self.dense(hidden_states)
         hidden_states = self.act(hidden_states)
@@ -849,13 +865,11 @@ class TFRobertaForMaskedLM(TFRobertaPreTrainedModel, TFMaskedLanguageModelingLos
         self.roberta = TFRobertaMainLayer(config, add_pooling_layer=False, name="roberta")
         self.lm_head = TFRobertaLMHead(config, self.roberta.embeddings, name="lm_head")
 
-    def get_output_embeddings(self):
-        return self.lm_head.decoder
-
-    def get_output_layer_with_bias(self):
+    def get_lm_head(self):
         return self.lm_head
 
     def get_prefix_bias_name(self):
+        warnings.warn("The method get_prefix_bias_name is deprecated. Please use `get_bias` instead.", FutureWarning)
         return self.name + "/" + self.lm_head.name
 
     @add_start_docstrings_to_model_forward(ROBERTA_INPUTS_DOCSTRING.format("batch_size, sequence_length"))