GPT-2 PyTorch models + better tips for BERT

docs/source/model_doc/bert.rst

@@ -27,7 +27,13 @@ Tips:
 
 - BERT is a model with absolute position embeddings so it's usually advised to pad the inputs on
   the right rather than the left.
-
+- BERT was trained with a masked language modeling (MLM) objective. It is therefore efficient at predicting masked
+  tokens and at NLU in general, but is not optimal for text generation. Models trained with a causal language
+  modeling (CLM) objective are better in that regard.
+- Alongside MLM, BERT was trained using a next sentence prediction (NSP) objective using the [CLS] token as a sequence
+  approximate. The user may use this token (the first token in a sequence built with special tokens) to get a sequence
+  prediction rather than a token prediction. However, averaging over the sequence may yield better results than using
+  the [CLS] token.
 
 BertConfig
 ~~~~~~~~~~~~~~~~~~~~~

docs/source/model_doc/gpt2.rst

 OpenAI GPT2
 ----------------------------------------------------
 
+Overview
+~~~~~~~~~~~~~~~~~~~~~
+
+OpenAI GPT-2 model was proposed in
+`Language Models are Unsupervised Multitask Learners`_
+by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
+It's a causal (unidirectional) transformer pre-trained using  language modeling on a very large
+corpus of ~40 GB of text data.
+
+The abstract from the paper is the following:
+
+*GPT-2 is a large transformer-based language model with 1.5 billion parameters, trained on a dataset[1]
+of 8 million web pages. GPT-2 is trained with a simple objective: predict the next word, given all of the previous
+words within some text. The diversity of the dataset causes this simple goal to contain naturally occurring
+demonstrations of many tasks across diverse domains. GPT-2 is a direct scale-up of GPT, with more than 10X
+the parameters and trained on more than 10X the amount of data.*
+
+Tips:
+
+- GPT-2 is a model with absolute position embeddings so it's usually advised to pad the inputs on
+  the right rather than the left.
+- GPT-2 was trained with a causal language modeling (CLM) objective and is therefore powerful at predicting the next
+  token in a sequence. Leveraging this feature allows GPT-2 to generate syntactically coherent text as
+  it can be observed in the `run_generation.py` example script.
+- The PyTorch models can take the `past` as input, which is the previously computed key/value attention pairs. Using
+  this `past` value prevents the model from re-computing pre-computed values in the context of text generation.
+  See `reusing the past in generative models <../quickstart.html#using-the-past>`_ for more information on the usage
+  of this argument.
+
+
 ``GPT2Config``
 ~~~~~~~~~~~~~~~~~~~~~
 

src/transformers/modeling_gpt2.py

@@ -25,7 +25,7 @@ import torch.nn as nn
 from torch.nn import CrossEntropyLoss
 
 from .configuration_gpt2 import GPT2Config
-from .file_utils import add_start_docstrings
+from .file_utils import add_start_docstrings, add_start_docstrings_to_callable
 from .modeling_utils import Conv1D, PreTrainedModel, SequenceSummary, prune_conv1d_layer
 
 
@@ -265,12 +265,7 @@ class GPT2PreTrainedModel(PreTrainedModel):
             module.weight.data.fill_(1.0)
 
 
-GPT2_START_DOCSTRING = r"""    OpenAI GPT-2 model was proposed in
-    `Language Models are Unsupervised Multitask Learners`_
-    by Alec Radford*, Jeffrey Wu*, Rewon Child, David Luan, Dario Amodei** and Ilya Sutskever**.
-    It's a causal (unidirectional) transformer pre-trained using  language modeling on a very large
-    corpus of ~40 GB of text data.
-
+GPT2_START_DOCSTRING = r"""    
     This model is a PyTorch `torch.nn.Module`_ sub-class. Use it as a regular PyTorch Module and
     refer to the PyTorch documentation for all matter related to general usage and behavior.
 
@@ -286,36 +281,43 @@ GPT2_START_DOCSTRING = r"""    OpenAI GPT-2 model was proposed in
             Check out the :meth:`~transformers.PreTrainedModel.from_pretrained` method to load the model weights.
 """
 
-GPT2_INPUTS_DOCSTRING = r"""    Inputs:
-        **input_ids**: ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
-            Indices of input sequence tokens in the vocabulary.
-            GPT-2 is a model with absolute position embeddings so it's usually advised to pad the inputs on
-            the right rather than the left.
+GPT2_INPUTS_DOCSTRING = r"""    
+    Args:
+        input_id (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. 
+            
             Indices can be obtained using :class:`transformers.GPT2Tokenizer`.
             See :func:`transformers.PreTrainedTokenizer.encode` and
-            :func:`transformers.PreTrainedTokenizer.convert_tokens_to_ids` for details.
-        **past**:
-            list of ``torch.FloatTensor`` (one for each layer):
-            that contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model
+            :func:`transformers.PreTrainedTokenizer.encode_plus` for details.
+            
+            `What are input IDs? <../glossary.html#input-ids>`__
+        past (:obj:`List[torch.FloatTensor]` of length :obj:`config.n_layers`):
+            Contains pre-computed hidden-states (key and values in the attention blocks) as computed by the model
             (see `past` output below). Can be used to speed up sequential decoding. The token ids which have their past given to this model
             should not be passed as input ids as they have already been computed.
-        **attention_mask**: (`optional`) ``torch.FloatTensor`` of shape ``(batch_size, sequence_length)``:
+        attention_mask (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
             Mask to avoid performing attention on padding token indices.
             Mask values selected in ``[0, 1]``:
             ``1`` for tokens that are NOT MASKED, ``0`` for MASKED tokens.
-        **token_type_ids**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
-            A parallel sequence of tokens (can be used to indicate various portions of the inputs).
-            The embeddings from these tokens will be summed with the respective token embeddings.
-            Indices are selected in the vocabulary (unlike BERT which has a specific vocabulary for segment indices).
-        **position_ids**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
+            
+            `What are attention masks? <../glossary.html#attention-mask>`__
+        token_type_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`): 
+            Segment token indices to indicate first and second portions of the inputs.
+            Indices are selected in ``[0, 1]``: ``0`` corresponds to a `sentence A` token, ``1``
+            corresponds to a `sentence B` token
+            
+            `What are token type IDs? <../glossary.html#token-type-ids>`_
+        position_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
             Indices of positions of each input sequence tokens in the position embeddings.
             Selected in the range ``[0, config.max_position_embeddings - 1]``.
-        **head_mask**: (`optional`) ``torch.FloatTensor`` of shape ``(num_heads,)`` or ``(num_layers, num_heads)``:
+            
+            `What are position IDs? <../glossary.html#position-ids>`_
+        head_mask (:obj:`torch.FloatTensor` of shape :obj:`(num_heads,)` or :obj:`(num_layers, num_heads)`, `optional`, defaults to :obj:`None`):
             Mask to nullify selected heads of the self-attention modules.
             Mask values selected in ``[0, 1]``:
-            ``1`` indicates the head is **not masked**, ``0`` indicates the head is **masked**.
-        **inputs_embeds**: (`optional`) ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, embedding_dim)``:
-            Optionally, instead of passing ``input_ids`` you can choose to directly pass an embedded representation.
+            :obj:`1` indicates the head is **not masked**, :obj:`0` indicates the head is **masked**.
+        input_embeds (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`, `optional`, defaults to :obj:`None`):
+            Optionally, instead of passing :obj:`input_ids` you can choose to directly pass an embedded representation.
             This is useful if you want more control over how to convert `input_ids` indices into associated vectors
             than the model's internal embedding lookup matrix.
 """
@@ -324,35 +326,8 @@ GPT2_INPUTS_DOCSTRING = r"""    Inputs:
 @add_start_docstrings(
     "The bare GPT2 Model transformer outputting raw hidden-states without any specific head on top.",
     GPT2_START_DOCSTRING,
-    GPT2_INPUTS_DOCSTRING,
 )
 class GPT2Model(GPT2PreTrainedModel):
-    r"""
-    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
-        **last_hidden_state**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, hidden_size)``
-            Sequence of hidden-states at the last layer of the model.
-        **past**:
-            list of ``torch.FloatTensor`` (one for each layer) of shape ``(2, batch_size, num_heads, sequence_length, embed_size_per_head)``:
-            that contains pre-computed hidden-states (key and values in the attention blocks).
-            Can be used (see `past` input) to speed up sequential decoding. The token ids which have their past given to this model
-            should not be passed as input ids as they have already been computed.
-        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
-            list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
-            of shape ``(batch_size, sequence_length, hidden_size)``:
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        **attentions**: (`optional`, returned when ``config.output_attentions=True``)
-            list of ``torch.FloatTensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
-            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
-
-    Examples::
-
-        tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
-        model = GPT2Model.from_pretrained('gpt2')
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
-        outputs = model(input_ids)
-        last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple
-
-    """
 
     def __init__(self, config):
         super().__init__(config)
@@ -381,6 +356,7 @@ class GPT2Model(GPT2PreTrainedModel):
         for layer, heads in heads_to_prune.items():
             self.h[layer].attn.prune_heads(heads)
 
+    @add_start_docstrings_to_callable(GPT2_INPUTS_DOCSTRING)
     def forward(
         self,
         input_ids=None,
@@ -391,6 +367,36 @@ class GPT2Model(GPT2PreTrainedModel):
         head_mask=None,
         inputs_embeds=None,
     ):
+        r"""
+    Return:
+        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (config) and inputs:
+        last_hidden_state (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, hidden_size)`):
+            Sequence of hidden-states at the last layer of the model.
+        past (:obj:`List[torch.FloatTensor]` of length :obj:`config.n_layers` with each tensor of shape :obj:`(2, batch_size, num_heads, sequence_length, embed_size_per_head)`):
+            Contains pre-computed hidden-states (key and values in the attention blocks).
+            Can be used (see `past` input) to speed up sequential decoding. The token ids which have their past given to this model
+            should not be passed as input ids as they have already been computed.
+        hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_hidden_states=True``):
+            Tuple of :obj:`torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer)
+            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_attentions=True``):
+            Tuple of :obj:`torch.FloatTensor` (one for each layer) of shape
+            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+
+    Examples::
+
+        tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
+        model = GPT2Model.from_pretrained('gpt2')
+        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
+        outputs = model(input_ids)
+        last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple
+
+        """
         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 input_ids is not None:
@@ -503,51 +509,11 @@ class GPT2Model(GPT2PreTrainedModel):
 
 
 @add_start_docstrings(
-    """The GPT2 Model transformer with a language modeling head on top
-(linear layer with weights tied to the input embeddings). """,
+    """The GPT2 Model transformer with a language modeling head on top 
+    (linear layer with weights tied to the input embeddings). """,
     GPT2_START_DOCSTRING,
-    GPT2_INPUTS_DOCSTRING,
 )
 class GPT2LMHeadModel(GPT2PreTrainedModel):
-    r"""
-        **labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
-            Labels for language modeling.
-            Note that the labels **are shifted** inside the model, i.e. you can set ``lm_labels = input_ids``
-            Indices are selected in ``[-100, 0, ..., config.vocab_size]``
-            All labels set to ``-100`` are ignored (masked), the loss is only
-            computed for labels in ``[0, ..., config.vocab_size]``
-
-    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
-        **loss**: (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
-            Language modeling loss.
-        **prediction_scores**: ``torch.FloatTensor`` of shape ``(batch_size, sequence_length, config.vocab_size)``
-            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
-        **past**:
-            list of ``torch.FloatTensor`` (one for each layer) of shape ``(2, batch_size, num_heads, sequence_length, embed_size_per_head)``:
-            that contains pre-computed hidden-states (key and values in the attention blocks).
-            Can be used (see `past` input) to speed up sequential decoding. The token ids which have their past given to this model
-            should not be passed as input ids as they have already been computed.
-        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
-            list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
-            of shape ``(batch_size, sequence_length, hidden_size)``:
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        **attentions**: (`optional`, returned when ``config.output_attentions=True``)
-            list of ``torch.FloatTensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
-            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
-
-    Examples::
-
-        import torch
-        from transformers import GPT2Tokenizer, GPT2LMHeadModel
-
-        tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
-        model = GPT2LMHeadModel.from_pretrained('gpt2')
-
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
-        outputs = model(input_ids, labels=input_ids)
-        loss, logits = outputs[:2]
-
-    """
 
     def __init__(self, config):
         super().__init__(config)
@@ -568,6 +534,7 @@ class GPT2LMHeadModel(GPT2PreTrainedModel):
         inputs.update(kwargs)
         return inputs
 
+    @add_start_docstrings_to_callable(GPT2_INPUTS_DOCSTRING)
     def forward(
         self,
         input_ids=None,
@@ -579,6 +546,49 @@ class GPT2LMHeadModel(GPT2PreTrainedModel):
         inputs_embeds=None,
         labels=None,
     ):
+        r"""
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
+            Labels for language modeling.
+            Note that the labels **are shifted** inside the model, i.e. you can set ``lm_labels = input_ids``
+            Indices are selected in ``[-100, 0, ..., config.vocab_size]``
+            All labels set to ``-100`` are ignored (masked), the loss is only
+            computed for labels in ``[0, ..., config.vocab_size]``
+
+    Return:
+        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:obj:`~transformers.GPT2Config`) and inputs:
+        loss (:obj:`torch.FloatTensor` of shape `(1,)`, `optional`, returned when ``labels`` is provided)
+            Language modeling loss.
+        prediction_scores (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        past (:obj:`List[torch.FloatTensor]` of length :obj:`config.n_layers` with each tensor of shape :obj:`(2, batch_size, num_heads, sequence_length, embed_size_per_head)`):
+            Contains pre-computed hidden-states (key and values in the attention blocks).
+            Can be used (see `past` input) to speed up sequential decoding. The token ids which have their past given to this model
+            should not be passed as input ids as they have already been computed.
+        hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_hidden_states=True``):
+            Tuple of :obj:`torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer)
+            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
+        attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_attentions=True``):
+            Tuple of :obj:`torch.FloatTensor` (one for each layer) of shape
+            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+
+    Examples::
+
+        import torch
+        from transformers import GPT2Tokenizer, GPT2LMHeadModel
+
+        tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
+        model = GPT2LMHeadModel.from_pretrained('gpt2')
+
+        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
+        outputs = model(input_ids, labels=input_ids)
+        loss, logits = outputs[:2]
+
+        """
         transformer_outputs = self.transformer(
             input_ids,
             past=past,
@@ -607,50 +617,80 @@ class GPT2LMHeadModel(GPT2PreTrainedModel):
 
 @add_start_docstrings(
     """The GPT2 Model transformer with a language modeling and a multiple-choice classification
-head on top e.g. for RocStories/SWAG tasks. The two heads are two linear layers.
-The language modeling head has its weights tied to the input embeddings,
-the classification head takes as input the input of a specified classification token index in the input sequence).
+    head on top e.g. for RocStories/SWAG tasks. The two heads are two linear layers.
+    The language modeling head has its weights tied to the input embeddings,
+    the classification head takes as input the input of a specified classification token index in the input sequence).
 """,
     GPT2_START_DOCSTRING,
-    GPT2_INPUTS_DOCSTRING,
 )
 class GPT2DoubleHeadsModel(GPT2PreTrainedModel):
-    r"""
-        **mc_token_ids**: (`optional`, default to index of the last token of the input) ``torch.LongTensor`` of shape ``(batch_size, num_choices)``:
+
+    def __init__(self, config):
+        super().__init__(config)
+        config.num_labels = 1
+        self.transformer = GPT2Model(config)
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
+        self.multiple_choice_head = SequenceSummary(config)
+
+        self.init_weights()
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    @add_start_docstrings_to_callable(GPT2_INPUTS_DOCSTRING)
+    def forward(
+        self,
+        input_ids=None,
+        past=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        mc_token_ids=None,
+        lm_labels=None,
+        mc_labels=None,
+    ):
+        r"""
+        mc_token_ids (:obj:`torch.LongTensor` of shape :obj:`(batch_size, num_choices)`, `optional`, default to index of the last token of the input)
             Index of the classification token in each input sequence.
             Selected in the range ``[0, input_ids.size(-1) - 1[``.
-        **lm_labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size, sequence_length)``:
+        lm_labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`)
             Labels for language modeling.
             Note that the labels **are shifted** inside the model, i.e. you can set ``lm_labels = input_ids``
             Indices are selected in ``[-1, 0, ..., config.vocab_size]``
             All labels set to ``-100`` are ignored (masked), the loss is only
             computed for labels in ``[0, ..., config.vocab_size]``
-        **mc_labels**: (`optional`) ``torch.LongTensor`` of shape ``(batch_size)``:
+        mc_labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size)`, `optional`, defaults to :obj:`None`)
             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)
 
-    Outputs: `Tuple` comprising various elements depending on the configuration (config) and inputs:
-        **lm_loss**: (`optional`, returned when ``lm_labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
+    Return:
+        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:obj:`~transformers.GPT2Config`) and inputs:
+        lm_loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, returned when ``lm_labels`` is provided):
             Language modeling loss.
-        **mc_loss**: (`optional`, returned when ``multiple_choice_labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
+        mc_loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, returned when :obj:`multiple_choice_labels` is provided):
             Multiple choice classification loss.
-        **lm_prediction_scores**: ``torch.FloatTensor`` of shape ``(batch_size, num_choices, sequence_length, config.vocab_size)``
+        lm_prediction_scores (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, num_choices, sequence_length, config.vocab_size)`):
             Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
-        **mc_prediction_scores**: ``torch.FloatTensor`` of shape ``(batch_size, num_choices)``
-            Prediction scores of the multiplechoice classification head (scores for each choice before SoftMax).
-        **past**:
-            list of ``torch.FloatTensor`` (one for each layer) of shape ``(2, batch_size, num_heads, sequence_length, embed_size_per_head)``:
-            that contains pre-computed hidden-states (key and values in the attention blocks).
+        mc_prediction_scores (:obj:`torch.FloatTensor` of shape :obj:`(batch_size, num_choices)`):
+            Prediction scores of the multiple choice classification head (scores for each choice before SoftMax).
+        past (:obj:`List[torch.FloatTensor]` of length :obj:`config.n_layers` with each tensor of shape :obj:`(2, batch_size, num_heads, sequence_length, embed_size_per_head)`):
+            Contains pre-computed hidden-states (key and values in the attention blocks).
             Can be used (see `past` input) to speed up sequential decoding. The token ids which have their past given to this model
             should not be passed as input ids as they have already been computed.
-        **hidden_states**: (`optional`, returned when ``config.output_hidden_states=True``)
-            list of ``torch.FloatTensor`` (one for the output of each layer + the output of the embeddings)
-            of shape ``(batch_size, sequence_length, hidden_size)``:
+        hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_hidden_states=True``):
+            Tuple of :obj:`torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer)
+            of shape :obj:`(batch_size, sequence_length, hidden_size)`.
+
             Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        **attentions**: (`optional`, returned when ``config.output_attentions=True``)
-            list of ``torch.FloatTensor`` (one for each layer) of shape ``(batch_size, num_heads, sequence_length, sequence_length)``:
-            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
+        attentions (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_attentions=True``):
+            Tuple of :obj:`torch.FloatTensor` (one for each layer) of shape
+            :obj:`(batch_size, num_heads, sequence_length, sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
 
     Examples::
 
@@ -675,33 +715,7 @@ class GPT2DoubleHeadsModel(GPT2PreTrainedModel):
         outputs = model(input_ids, mc_token_ids=mc_token_ids)
         lm_prediction_scores, mc_prediction_scores = outputs[:2]
 
-    """
-
-    def __init__(self, config):
-        super().__init__(config)
-        config.num_labels = 1
-        self.transformer = GPT2Model(config)
-        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
-        self.multiple_choice_head = SequenceSummary(config)
-
-        self.init_weights()
-
-    def get_output_embeddings(self):
-        return self.lm_head
-
-    def forward(
-        self,
-        input_ids=None,
-        past=None,
-        attention_mask=None,
-        token_type_ids=None,
-        position_ids=None,
-        head_mask=None,
-        inputs_embeds=None,
-        mc_token_ids=None,
-        lm_labels=None,
-        mc_labels=None,
-    ):
+        """
         transformer_outputs = self.transformer(
             input_ids,
             past=past,