Change unpacking of TF mobilebert inputs to use decorator (#16110)

* Change unpacking of TF mobilebert inputs to use decorator

* Move unpack_inputs as the top decorator

* make fixup
Co-authored-by: ChienVM <chien_vm@detomo.co.jp>

src/transformers/models/mobilebert/modeling_tf_mobilebert.py

@@ -49,8 +49,8 @@ from ...modeling_tf_utils import (
     TFSequenceClassificationLoss,
     TFTokenClassificationLoss,
     get_initializer,
-    input_processing,
     keras_serializable,
+    unpack_inputs,
 )
 from ...tf_utils import shape_list
 from ...utils import logging
@@ -679,6 +679,7 @@ class TFMobileBertMainLayer(tf.keras.layers.Layer):
         """
         raise NotImplementedError
 
+    @unpack_inputs
     def call(
         self,
         input_ids=None,
@@ -693,51 +694,29 @@ class TFMobileBertMainLayer(tf.keras.layers.Layer):
         training=False,
         **kwargs,
     ):
-        inputs = input_processing(
-            func=self.call,
-            config=self.config,
-            input_ids=input_ids,
-            attention_mask=attention_mask,
-            token_type_ids=token_type_ids,
-            position_ids=position_ids,
-            head_mask=head_mask,
-            inputs_embeds=inputs_embeds,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-            training=training,
-            kwargs_call=kwargs,
-        )
-
-        if inputs["input_ids"] is not None and inputs["inputs_embeds"] is not None:
+        if input_ids is not None and inputs_embeds is not None:
             raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
-        elif inputs["input_ids"] is not None:
-            input_shape = shape_list(inputs["input_ids"])
-        elif inputs["inputs_embeds"] is not None:
-            input_shape = shape_list(inputs["inputs_embeds"])[:-1]
+        elif input_ids is not None:
+            input_shape = shape_list(input_ids)
+        elif inputs_embeds is not None:
+            input_shape = shape_list(inputs_embeds)[:-1]
         else:
             raise ValueError("You have to specify either input_ids or inputs_embeds")
 
-        if inputs["attention_mask"] is None:
-            inputs["attention_mask"] = tf.fill(input_shape, 1)
+        if attention_mask is None:
+            attention_mask = tf.fill(input_shape, 1)
 
-        if inputs["token_type_ids"] is None:
-            inputs["token_type_ids"] = tf.fill(input_shape, 0)
+        if token_type_ids is None:
+            token_type_ids = tf.fill(input_shape, 0)
 
-        embedding_output = self.embeddings(
-            inputs["input_ids"],
-            inputs["position_ids"],
-            inputs["token_type_ids"],
-            inputs["inputs_embeds"],
-            training=inputs["training"],
-        )
+        embedding_output = self.embeddings(input_ids, position_ids, token_type_ids, inputs_embeds, training=training)
 
         # We create a 3D attention mask from a 2D tensor mask.
         # Sizes are [batch_size, 1, 1, to_seq_length]
         # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
         # this attention mask is more simple than the triangular masking of causal attention
         # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
-        extended_attention_mask = tf.reshape(inputs["attention_mask"], (input_shape[0], 1, 1, input_shape[1]))
+        extended_attention_mask = tf.reshape(attention_mask, (input_shape[0], 1, 1, input_shape[1]))
 
         # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
         # masked positions, this operation will create a tensor which is 0.0 for
@@ -754,25 +733,25 @@ class TFMobileBertMainLayer(tf.keras.layers.Layer):
         # attention_probs has shape bsz x n_heads x N x N
         # input head_mask has shape [num_heads] or [num_hidden_layers x num_heads]
         # and head_mask is converted to shape [num_hidden_layers x batch x num_heads x seq_length x seq_length]
-        if inputs["head_mask"] is not None:
+        if head_mask is not None:
             raise NotImplementedError
         else:
-            inputs["head_mask"] = [None] * self.num_hidden_layers
+            head_mask = [None] * self.num_hidden_layers
 
         encoder_outputs = self.encoder(
             embedding_output,
             extended_attention_mask,
-            inputs["head_mask"],
-            inputs["output_attentions"],
-            inputs["output_hidden_states"],
-            inputs["return_dict"],
-            training=inputs["training"],
+            head_mask,
+            output_attentions,
+            output_hidden_states,
+            return_dict,
+            training=training,
         )
 
         sequence_output = encoder_outputs[0]
         pooled_output = self.pooler(sequence_output) if self.pooler is not None else None
 
-        if not inputs["return_dict"]:
+        if not return_dict:
             return (
                 sequence_output,
                 pooled_output,
@@ -929,6 +908,7 @@ class TFMobileBertModel(TFMobileBertPreTrainedModel):
         super().__init__(config, *inputs, **kwargs)
         self.mobilebert = TFMobileBertMainLayer(config, name="mobilebert")
 
+    @unpack_inputs
     @add_start_docstrings_to_model_forward(MOBILEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
@@ -950,9 +930,7 @@ class TFMobileBertModel(TFMobileBertPreTrainedModel):
         training=False,
         **kwargs,
     ):
-        inputs = input_processing(
-            func=self.call,
-            config=self.config,
+        outputs = self.mobilebert(
             input_ids=input_ids,
             attention_mask=attention_mask,
             token_type_ids=token_type_ids,
@@ -963,19 +941,6 @@ class TFMobileBertModel(TFMobileBertPreTrainedModel):
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
             training=training,
-            kwargs_call=kwargs,
-        )
-        outputs = self.mobilebert(
-            input_ids=inputs["input_ids"],
-            attention_mask=inputs["attention_mask"],
-            token_type_ids=inputs["token_type_ids"],
-            position_ids=inputs["position_ids"],
-            head_mask=inputs["head_mask"],
-            inputs_embeds=inputs["inputs_embeds"],
-            output_attentions=inputs["output_attentions"],
-            output_hidden_states=inputs["output_hidden_states"],
-            return_dict=inputs["return_dict"],
-            training=inputs["training"],
         )
 
         return outputs
@@ -1013,6 +978,7 @@ class TFMobileBertForPreTraining(TFMobileBertPreTrainedModel):
         warnings.warn("The method get_prefix_bias_name is deprecated. Please use `get_bias` instead.", FutureWarning)
         return self.name + "/" + self.predictions.name + "/" + self.predictions.predictions.name
 
+    @unpack_inputs
     @add_start_docstrings_to_model_forward(MOBILEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @replace_return_docstrings(output_type=TFMobileBertForPreTrainingOutput, config_class=_CONFIG_FOR_DOC)
     def call(
@@ -1044,10 +1010,8 @@ class TFMobileBertForPreTraining(TFMobileBertPreTrainedModel):
         >>> outputs = model(input_ids)
         >>> prediction_scores, seq_relationship_scores = outputs[:2]
         ```"""
-        inputs = input_processing(
-            func=self.call,
-            config=self.config,
-            input_ids=input_ids,
+        outputs = self.mobilebert(
+            input_ids,
             attention_mask=attention_mask,
             token_type_ids=token_type_ids,
             position_ids=position_ids,
@@ -1057,26 +1021,13 @@ class TFMobileBertForPreTraining(TFMobileBertPreTrainedModel):
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
             training=training,
-            kwargs_call=kwargs,
-        )
-        outputs = self.mobilebert(
-            inputs["input_ids"],
-            attention_mask=inputs["attention_mask"],
-            token_type_ids=inputs["token_type_ids"],
-            position_ids=inputs["position_ids"],
-            head_mask=inputs["head_mask"],
-            inputs_embeds=inputs["inputs_embeds"],
-            output_attentions=inputs["output_attentions"],
-            output_hidden_states=inputs["output_hidden_states"],
-            return_dict=inputs["return_dict"],
-            training=inputs["training"],
         )
 
         sequence_output, pooled_output = outputs[:2]
         prediction_scores = self.predictions(sequence_output)
         seq_relationship_score = self.seq_relationship(pooled_output)
 
-        if not inputs["return_dict"]:
+        if not return_dict:
             return (prediction_scores, seq_relationship_score) + outputs[2:]
 
         return TFMobileBertForPreTrainingOutput(
@@ -1120,6 +1071,7 @@ class TFMobileBertForMaskedLM(TFMobileBertPreTrainedModel, TFMaskedLanguageModel
         warnings.warn("The method get_prefix_bias_name is deprecated. Please use `get_bias` instead.", FutureWarning)
         return self.name + "/" + self.mlm.name + "/" + self.mlm.predictions.name
 
+    @unpack_inputs
     @add_start_docstrings_to_model_forward(MOBILEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
@@ -1148,10 +1100,8 @@ class TFMobileBertForMaskedLM(TFMobileBertPreTrainedModel, TFMaskedLanguageModel
             config.vocab_size]` (see `input_ids` docstring) Tokens with indices set to `-100` are ignored (masked), the
             loss is only computed for the tokens with labels
         """
-        inputs = input_processing(
-            func=self.call,
-            config=self.config,
-            input_ids=input_ids,
+        outputs = self.mobilebert(
+            input_ids,
             attention_mask=attention_mask,
             token_type_ids=token_type_ids,
             position_ids=position_ids,
@@ -1160,28 +1110,14 @@ class TFMobileBertForMaskedLM(TFMobileBertPreTrainedModel, TFMaskedLanguageModel
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            labels=labels,
             training=training,
-            kwargs_call=kwargs,
-        )
-        outputs = self.mobilebert(
-            inputs["input_ids"],
-            attention_mask=inputs["attention_mask"],
-            token_type_ids=inputs["token_type_ids"],
-            position_ids=inputs["position_ids"],
-            head_mask=inputs["head_mask"],
-            inputs_embeds=inputs["inputs_embeds"],
-            output_attentions=inputs["output_attentions"],
-            output_hidden_states=inputs["output_hidden_states"],
-            return_dict=inputs["return_dict"],
-            training=inputs["training"],
         )
         sequence_output = outputs[0]
-        prediction_scores = self.predictions(sequence_output, training=inputs["training"])
+        prediction_scores = self.predictions(sequence_output, training=training)
 
-        loss = None if inputs["labels"] is None else self.hf_compute_loss(inputs["labels"], prediction_scores)
+        loss = None if labels is None else self.hf_compute_loss(labels, prediction_scores)
 
-        if not inputs["return_dict"]:
+        if not return_dict:
             output = (prediction_scores,) + outputs[2:]
             return ((loss,) + output) if loss is not None else output
 
@@ -1224,6 +1160,7 @@ class TFMobileBertForNextSentencePrediction(TFMobileBertPreTrainedModel, TFNextS
         self.mobilebert = TFMobileBertMainLayer(config, name="mobilebert")
         self.cls = TFMobileBertOnlyNSPHead(config, name="seq_relationship___cls")
 
+    @unpack_inputs
     @add_start_docstrings_to_model_forward(MOBILEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @replace_return_docstrings(output_type=TFNextSentencePredictorOutput, config_class=_CONFIG_FOR_DOC)
     def call(
@@ -1259,10 +1196,8 @@ class TFMobileBertForNextSentencePrediction(TFMobileBertPreTrainedModel, TFNextS
 
         >>> logits = model(encoding["input_ids"], token_type_ids=encoding["token_type_ids"])[0]
         ```"""
-        inputs = input_processing(
-            func=self.call,
-            config=self.config,
-            input_ids=input_ids,
+        outputs = self.mobilebert(
+            input_ids,
             attention_mask=attention_mask,
             token_type_ids=token_type_ids,
             position_ids=position_ids,
@@ -1271,32 +1206,18 @@ class TFMobileBertForNextSentencePrediction(TFMobileBertPreTrainedModel, TFNextS
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            next_sentence_label=next_sentence_label,
             training=training,
-            kwargs_call=kwargs,
-        )
-        outputs = self.mobilebert(
-            inputs["input_ids"],
-            attention_mask=inputs["attention_mask"],
-            token_type_ids=inputs["token_type_ids"],
-            position_ids=inputs["position_ids"],
-            head_mask=inputs["head_mask"],
-            inputs_embeds=inputs["inputs_embeds"],
-            output_attentions=inputs["output_attentions"],
-            output_hidden_states=inputs["output_hidden_states"],
-            return_dict=inputs["return_dict"],
-            training=inputs["training"],
         )
         pooled_output = outputs[1]
         seq_relationship_scores = self.cls(pooled_output)
 
         next_sentence_loss = (
             None
-            if inputs["next_sentence_label"] is None
-            else self.hf_compute_loss(labels=inputs["next_sentence_label"], logits=seq_relationship_scores)
+            if next_sentence_label is None
+            else self.hf_compute_loss(labels=next_sentence_label, logits=seq_relationship_scores)
         )
 
-        if not inputs["return_dict"]:
+        if not return_dict:
             output = (seq_relationship_scores,) + outputs[2:]
             return ((next_sentence_loss,) + output) if next_sentence_loss is not None else output
 
@@ -1345,6 +1266,7 @@ class TFMobileBertForSequenceClassification(TFMobileBertPreTrainedModel, TFSeque
             config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="classifier"
         )
 
+    @unpack_inputs
     @add_start_docstrings_to_model_forward(MOBILEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
@@ -1373,10 +1295,8 @@ class TFMobileBertForSequenceClassification(TFMobileBertPreTrainedModel, TFSeque
             config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
             `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
         """
-        inputs = input_processing(
-            func=self.call,
-            config=self.config,
-            input_ids=input_ids,
+        outputs = self.mobilebert(
+            input_ids,
             attention_mask=attention_mask,
             token_type_ids=token_type_ids,
             position_ids=position_ids,
@@ -1385,30 +1305,16 @@ class TFMobileBertForSequenceClassification(TFMobileBertPreTrainedModel, TFSeque
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            labels=labels,
             training=training,
-            kwargs_call=kwargs,
-        )
-        outputs = self.mobilebert(
-            inputs["input_ids"],
-            attention_mask=inputs["attention_mask"],
-            token_type_ids=inputs["token_type_ids"],
-            position_ids=inputs["position_ids"],
-            head_mask=inputs["head_mask"],
-            inputs_embeds=inputs["inputs_embeds"],
-            output_attentions=inputs["output_attentions"],
-            output_hidden_states=inputs["output_hidden_states"],
-            return_dict=inputs["return_dict"],
-            training=inputs["training"],
         )
         pooled_output = outputs[1]
 
-        pooled_output = self.dropout(pooled_output, training=inputs["training"])
+        pooled_output = self.dropout(pooled_output, training=training)
         logits = self.classifier(pooled_output)
 
-        loss = None if inputs["labels"] is None else self.hf_compute_loss(inputs["labels"], logits)
+        loss = None if labels is None else self.hf_compute_loss(labels, logits)
 
-        if not inputs["return_dict"]:
+        if not return_dict:
             output = (logits,) + outputs[2:]
             return ((loss,) + output) if loss is not None else output
 
@@ -1453,6 +1359,7 @@ class TFMobileBertForQuestionAnswering(TFMobileBertPreTrainedModel, TFQuestionAn
             config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="qa_outputs"
         )
 
+    @unpack_inputs
     @add_start_docstrings_to_model_forward(MOBILEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
@@ -1486,10 +1393,8 @@ class TFMobileBertForQuestionAnswering(TFMobileBertPreTrainedModel, TFQuestionAn
             Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
             are not taken into account for computing the loss.
         """
-        inputs = input_processing(
-            func=self.call,
-            config=self.config,
-            input_ids=input_ids,
+        outputs = self.mobilebert(
+            input_ids,
             attention_mask=attention_mask,
             token_type_ids=token_type_ids,
             position_ids=position_ids,
@@ -1498,22 +1403,7 @@ class TFMobileBertForQuestionAnswering(TFMobileBertPreTrainedModel, TFQuestionAn
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            start_positions=start_positions,
-            end_positions=end_positions,
             training=training,
-            kwargs_call=kwargs,
-        )
-        outputs = self.mobilebert(
-            inputs["input_ids"],
-            attention_mask=inputs["attention_mask"],
-            token_type_ids=inputs["token_type_ids"],
-            position_ids=inputs["position_ids"],
-            head_mask=inputs["head_mask"],
-            inputs_embeds=inputs["inputs_embeds"],
-            output_attentions=inputs["output_attentions"],
-            output_hidden_states=inputs["output_hidden_states"],
-            return_dict=inputs["return_dict"],
-            training=inputs["training"],
         )
         sequence_output = outputs[0]
 
@@ -1523,12 +1413,11 @@ class TFMobileBertForQuestionAnswering(TFMobileBertPreTrainedModel, TFQuestionAn
         end_logits = tf.squeeze(end_logits, axis=-1)
 
         loss = None
-        if inputs["start_positions"] is not None and inputs["end_positions"] is not None:
-            labels = {"start_position": inputs["start_positions"]}
-            labels["end_position"] = inputs["end_positions"]
+        if start_positions is not None and end_positions is not None:
+            labels = {"start_position": start_positions, "end_position": end_positions}
             loss = self.hf_compute_loss(labels, (start_logits, end_logits))
 
-        if not inputs["return_dict"]:
+        if not return_dict:
             output = (start_logits, end_logits) + outputs[2:]
             return ((loss,) + output) if loss is not None else output
 
@@ -1586,6 +1475,7 @@ class TFMobileBertForMultipleChoice(TFMobileBertPreTrainedModel, TFMultipleChoic
         """
         return {"input_ids": tf.constant(MULTIPLE_CHOICE_DUMMY_INPUTS)}
 
+    @unpack_inputs
     @add_start_docstrings_to_model_forward(
         MOBILEBERT_INPUTS_DOCSTRING.format("batch_size, num_choices, sequence_length")
     )
@@ -1615,43 +1505,20 @@ class TFMobileBertForMultipleChoice(TFMobileBertPreTrainedModel, TFMultipleChoic
             Labels for computing the multiple choice classification loss. Indices should be in `[0, ..., num_choices]`
             where `num_choices` is the size of the second dimension of the input tensors. (See `input_ids` above)
         """
-        inputs = input_processing(
-            func=self.call,
-            config=self.config,
-            input_ids=input_ids,
-            attention_mask=attention_mask,
-            token_type_ids=token_type_ids,
-            position_ids=position_ids,
-            head_mask=head_mask,
-            inputs_embeds=inputs_embeds,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-            labels=labels,
-            training=training,
-            kwargs_call=kwargs,
-        )
-
-        if inputs["input_ids"] is not None:
-            num_choices = shape_list(inputs["input_ids"])[1]
-            seq_length = shape_list(inputs["input_ids"])[2]
+        if input_ids is not None:
+            num_choices = shape_list(input_ids)[1]
+            seq_length = shape_list(input_ids)[2]
         else:
-            num_choices = shape_list(inputs["inputs_embeds"])[1]
-            seq_length = shape_list(inputs["inputs_embeds"])[2]
+            num_choices = shape_list(inputs_embeds)[1]
+            seq_length = shape_list(inputs_embeds)[2]
 
-        flat_input_ids = tf.reshape(inputs["input_ids"], (-1, seq_length)) if inputs["input_ids"] is not None else None
-        flat_attention_mask = (
-            tf.reshape(inputs["attention_mask"], (-1, seq_length)) if inputs["attention_mask"] is not None else None
-        )
-        flat_token_type_ids = (
-            tf.reshape(inputs["token_type_ids"], (-1, seq_length)) if inputs["token_type_ids"] is not None else None
-        )
-        flat_position_ids = (
-            tf.reshape(inputs["position_ids"], (-1, seq_length)) if inputs["position_ids"] is not None else None
-        )
+        flat_input_ids = tf.reshape(input_ids, (-1, seq_length)) if input_ids is not None else None
+        flat_attention_mask = tf.reshape(attention_mask, (-1, seq_length)) if attention_mask is not None else None
+        flat_token_type_ids = tf.reshape(token_type_ids, (-1, seq_length)) if token_type_ids is not None else None
+        flat_position_ids = tf.reshape(position_ids, (-1, seq_length)) if position_ids is not None else None
         flat_inputs_embeds = (
-            tf.reshape(inputs["inputs_embeds"], (-1, seq_length, shape_list(inputs["inputs_embeds"])[3]))
-            if inputs["inputs_embeds"] is not None
+            tf.reshape(inputs_embeds, (-1, seq_length, shape_list(inputs_embeds)[3]))
+            if inputs_embeds is not None
             else None
         )
         outputs = self.mobilebert(
@@ -1659,21 +1526,21 @@ class TFMobileBertForMultipleChoice(TFMobileBertPreTrainedModel, TFMultipleChoic
             flat_attention_mask,
             flat_token_type_ids,
             flat_position_ids,
-            inputs["head_mask"],
+            head_mask,
             flat_inputs_embeds,
-            inputs["output_attentions"],
-            inputs["output_hidden_states"],
-            return_dict=inputs["return_dict"],
-            training=inputs["training"],
+            output_attentions,
+            output_hidden_states,
+            return_dict=return_dict,
+            training=training,
         )
         pooled_output = outputs[1]
-        pooled_output = self.dropout(pooled_output, training=inputs["training"])
+        pooled_output = self.dropout(pooled_output, training=training)
         logits = self.classifier(pooled_output)
         reshaped_logits = tf.reshape(logits, (-1, num_choices))
 
-        loss = None if inputs["labels"] is None else self.hf_compute_loss(inputs["labels"], reshaped_logits)
+        loss = None if labels is None else self.hf_compute_loss(labels, reshaped_logits)
 
-        if not inputs["return_dict"]:
+        if not return_dict:
             output = (reshaped_logits,) + outputs[2:]
             return ((loss,) + output) if loss is not None else output
 
@@ -1738,6 +1605,7 @@ class TFMobileBertForTokenClassification(TFMobileBertPreTrainedModel, TFTokenCla
             config.num_labels, kernel_initializer=get_initializer(config.initializer_range), name="classifier"
         )
 
+    @unpack_inputs
     @add_start_docstrings_to_model_forward(MOBILEBERT_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
     @add_code_sample_docstrings(
         processor_class=_TOKENIZER_FOR_DOC,
@@ -1764,10 +1632,8 @@ class TFMobileBertForTokenClassification(TFMobileBertPreTrainedModel, TFTokenCla
         labels (`tf.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
             Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
         """
-        inputs = input_processing(
-            func=self.call,
-            config=self.config,
-            input_ids=input_ids,
+        outputs = self.mobilebert(
+            input_ids,
             attention_mask=attention_mask,
             token_type_ids=token_type_ids,
             position_ids=position_ids,
@@ -1776,30 +1642,16 @@ class TFMobileBertForTokenClassification(TFMobileBertPreTrainedModel, TFTokenCla
             output_attentions=output_attentions,
             output_hidden_states=output_hidden_states,
             return_dict=return_dict,
-            labels=labels,
             training=training,
-            kwargs_call=kwargs,
-        )
-        outputs = self.mobilebert(
-            inputs["input_ids"],
-            attention_mask=inputs["attention_mask"],
-            token_type_ids=inputs["token_type_ids"],
-            position_ids=inputs["position_ids"],
-            head_mask=inputs["head_mask"],
-            inputs_embeds=inputs["inputs_embeds"],
-            output_attentions=inputs["output_attentions"],
-            output_hidden_states=inputs["output_hidden_states"],
-            return_dict=return_dict,
-            training=inputs["training"],
         )
         sequence_output = outputs[0]
 
-        sequence_output = self.dropout(sequence_output, training=inputs["training"])
+        sequence_output = self.dropout(sequence_output, training=training)
         logits = self.classifier(sequence_output)
 
-        loss = None if inputs["labels"] is None else self.hf_compute_loss(inputs["labels"], logits)
+        loss = None if labels is None else self.hf_compute_loss(labels, logits)
 
-        if not inputs["return_dict"]:
+        if not return_dict:
             output = (logits,) + outputs[2:]
             return ((loss,) + output) if loss is not None else output