Split LMBert model in two (#4874)

* Split LMBert model in two

* Fix example

* Remove lm_labels

* Adapt tests, refactor prepare_for_generation

* Fix merge

* Hide BeartLMHeadModel

src/transformers/modeling_bert.py

@@ -873,11 +873,13 @@ def forward(
         return outputs  # (loss), prediction_scores, seq_relationship_score, (hidden_states), (attentions)
 
 
-# TODO: Split with a different BertWithLMHead to get rid of `lm_labels` here and in encoder_decoder.
-@add_start_docstrings("""Bert Model with a `language modeling` head on top. """, BERT_START_DOCSTRING)
-class BertForMaskedLM(BertPreTrainedModel):
+@add_start_docstrings(
+    """Bert Model with a `language modeling` head on top for CLM fine-tuning. """, BERT_START_DOCSTRING
+)
+class BertLMHeadModel(BertPreTrainedModel):
     def __init__(self, config):
         super().__init__(config)
+        assert config.is_decoder, "If you want to use `BertLMHeadModel` as a standalone, add `is_decoder=True`."
 
         self.bert = BertModel(config)
         self.cls = BertOnlyMLMHead(config)
@@ -899,18 +901,119 @@ def forward(
         labels=None,
         encoder_hidden_states=None,
         encoder_attention_mask=None,
-        lm_labels=None,
         output_attentions=None,
         **kwargs
     ):
         r"""
         labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-            Labels for computing the masked language modeling loss.
+            Labels for computing the left-to-right language modeling loss (next word prediction).
             Indices should be in ``[-100, 0, ..., 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
             in ``[0, ..., config.vocab_size]``
-        lm_labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-            Labels for computing the left-to-right language modeling loss (next word prediction).
+        kwargs (:obj:`Dict[str, any]`, optional, defaults to `{}`):
+            Used to hide legacy arguments that have been deprecated.
+
+    Returns:
+        :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.BertConfig`) and inputs:
+        ltr_lm_loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, returned when :obj:`labels` is provided):
+            Next token prediction 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).
+        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::
+
+            from transformers import BertTokenizer, BertLMHeadModel
+            import torch
+
+            tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
+            model = BertLMHeadModel.from_pretrained('bert-base-uncased', is_decoder=True)
+
+            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, prediction_scores = outputs[:2]
+
+        """
+
+        outputs = self.bert(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            output_attentions=output_attentions,
+        )
+
+        sequence_output = outputs[0]
+        prediction_scores = self.cls(sequence_output)
+
+        outputs = (prediction_scores,) + outputs[2:]  # Add hidden states and attention if they are here
+
+        if labels is not None:
+            # we are doing next-token prediction; shift prediction scores and input ids by one
+            prediction_scores = prediction_scores[:, :-1, :].contiguous()
+            labels = labels[:, 1:].contiguous()
+            loss_fct = CrossEntropyLoss()
+            ltr_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
+            outputs = (ltr_lm_loss,) + outputs
+
+        return outputs  # (ltr_lm_loss), prediction_scores, (hidden_states), (attentions)
+
+    def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **model_kwargs):
+        input_shape = input_ids.shape
+
+        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
+        if attention_mask is None:
+            attention_mask = input_ids.new_ones(input_shape)
+
+        return {"input_ids": input_ids, "attention_mask": attention_mask}
+
+
+@add_start_docstrings("""Bert Model with a `language modeling` head on top. """, BERT_START_DOCSTRING)
+class BertForMaskedLM(BertPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.bert = BertModel(config)
+        self.cls = BertOnlyMLMHead(config)
+
+        self.init_weights()
+
+    def get_output_embeddings(self):
+        return self.cls.predictions.decoder
+
+    @add_start_docstrings_to_callable(BERT_INPUTS_DOCSTRING.format("(batch_size, sequence_length)"))
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        output_attentions=None,
+        **kwargs
+    ):
+        r"""
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
+            Labels for computing the masked language modeling loss.
             Indices should be in ``[-100, 0, ..., 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
             in ``[0, ..., config.vocab_size]``
@@ -921,8 +1024,6 @@ def forward(
         :obj:`tuple(torch.FloatTensor)` comprising various elements depending on the configuration (:class:`~transformers.BertConfig`) and inputs:
         masked_lm_loss (`optional`, returned when ``labels`` is provided) ``torch.FloatTensor`` of shape ``(1,)``:
             Masked language modeling loss.
-        ltr_lm_loss (:obj:`torch.FloatTensor` of shape :obj:`(1,)`, `optional`, returned when :obj:`lm_labels` is provided):
-                Next token prediction 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).
         hidden_states (:obj:`tuple(torch.FloatTensor)`, `optional`, returned when ``config.output_hidden_states=True``):
@@ -957,6 +1058,7 @@ def forward(
                 DeprecationWarning,
             )
             labels = kwargs.pop("masked_lm_labels")
+        assert "lm_labels" not in kwargs, "Use `BertWithLMHead` for autoregressive language modeling task."
         assert kwargs == {}, f"Unexpected keyword arguments: {list(kwargs.keys())}."
 
         outputs = self.bert(
@@ -976,46 +1078,24 @@ def forward(
 
         outputs = (prediction_scores,) + outputs[2:]  # Add hidden states and attention if they are here
 
-        # Although this may seem awkward, BertForMaskedLM supports two scenarios:
-        # 1. If a tensor that contains the indices of masked labels is provided,
-        #    the cross-entropy is the MLM cross-entropy that measures the likelihood
-        #    of predictions for masked words.
-        # 2. If `lm_labels` is provided we are in a causal scenario where we
-        #    try to predict the next token for each input in the decoder.
         if labels is not None:
             loss_fct = CrossEntropyLoss()  # -100 index = padding token
             masked_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), labels.view(-1))
             outputs = (masked_lm_loss,) + outputs
 
-        if lm_labels is not None:
-            # we are doing next-token prediction; shift prediction scores and input ids by one
-            prediction_scores = prediction_scores[:, :-1, :].contiguous()
-            lm_labels = lm_labels[:, 1:].contiguous()
-            loss_fct = CrossEntropyLoss()
-            ltr_lm_loss = loss_fct(prediction_scores.view(-1, self.config.vocab_size), lm_labels.view(-1))
-            outputs = (ltr_lm_loss,) + outputs
-
-        return outputs  # (ltr_lm_loss), (masked_lm_loss), prediction_scores, (hidden_states), (attentions)
+        return outputs  # (masked_lm_loss), prediction_scores, (hidden_states), (attentions)
 
     def prepare_inputs_for_generation(self, input_ids, attention_mask=None, **model_kwargs):
         input_shape = input_ids.shape
         effective_batch_size = input_shape[0]
 
-        # if model is used as a decoder in encoder-decoder model, the decoder attention mask is created on the fly
-        if attention_mask is None:
-            attention_mask = input_ids.new_ones(input_shape)
-
-        # if model is does not use a causal mask then add a dummy token
-        if self.config.is_decoder is False:
-            assert self.config.pad_token_id is not None, "The PAD token should be defined for generation"
-            attention_mask = torch.cat(
-                [attention_mask, attention_mask.new_zeros((attention_mask.shape[0], 1))], dim=-1
-            )
-
-            dummy_token = torch.full(
-                (effective_batch_size, 1), self.config.pad_token_id, dtype=torch.long, device=input_ids.device
-            )
-            input_ids = torch.cat([input_ids, dummy_token], dim=1)
+        #  add a dummy token
+        assert self.config.pad_token_id is not None, "The PAD token should be defined for generation"
+        attention_mask = torch.cat([attention_mask, attention_mask.new_zeros((attention_mask.shape[0], 1))], dim=-1)
+        dummy_token = torch.full(
+            (effective_batch_size, 1), self.config.pad_token_id, dtype=torch.long, device=input_ids.device
+        )
+        input_ids = torch.cat([input_ids, dummy_token], dim=1)
 
         return {"input_ids": input_ids, "attention_mask": attention_mask}
 

src/transformers/modeling_encoder_decoder.py

@@ -192,7 +192,6 @@ def forward(
         decoder_head_mask=None,
         decoder_inputs_embeds=None,
         labels=None,
-        lm_labels=None,
         **kwargs,
     ):
 
@@ -239,11 +238,6 @@ def forward(
                 Indices should be in ``[-100, 0, ..., 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
                 in ``[0, ..., config.vocab_size]``
-            lm_labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size, sequence_length)`, `optional`, defaults to :obj:`None`):
-                Labels for computing the left-to-right language modeling loss (next word prediction) for the decoder.
-                Indices should be in ``[-100, 0, ..., 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
-                in ``[0, ..., config.vocab_size]``
             kwargs: (`optional`) Remaining dictionary of keyword arguments. Keyword arguments come in two flavors:
                 - Without a prefix which will be input as `**encoder_kwargs` for the encoder forward function.
                 - With a `decoder_` prefix which will be input as `**decoder_kwargs` for the decoder forward function.
@@ -293,7 +287,6 @@ def forward(
             encoder_hidden_states=hidden_states,
             encoder_attention_mask=attention_mask,
             head_mask=decoder_head_mask,
-            lm_labels=lm_labels,
             labels=labels,
             **kwargs_decoder,
         )

tests/test_modeling_bert.py

@@ -35,7 +35,7 @@
         BertForTokenClassification,
         BertForMultipleChoice,
     )
-    from transformers.modeling_bert import BERT_PRETRAINED_MODEL_ARCHIVE_LIST
+    from transformers.modeling_bert import BERT_PRETRAINED_MODEL_ARCHIVE_LIST, BertLMHeadModel
 
 
 class BertModelTester:
@@ -211,6 +211,33 @@ def create_and_check_bert_model_as_decoder(
         )
         self.parent.assertListEqual(list(result["pooled_output"].size()), [self.batch_size, self.hidden_size])
 
+    def create_and_check_bert_for_causal_lm(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+        sequence_labels,
+        token_labels,
+        choice_labels,
+        encoder_hidden_states,
+        encoder_attention_mask,
+    ):
+        model = BertLMHeadModel(config=config)
+        model.to(torch_device)
+        model.eval()
+        loss, prediction_scores = model(
+            input_ids, attention_mask=input_mask, token_type_ids=token_type_ids, labels=token_labels
+        )
+        result = {
+            "loss": loss,
+            "prediction_scores": prediction_scores,
+        }
+        self.parent.assertListEqual(
+            list(result["prediction_scores"].size()), [self.batch_size, self.seq_length, self.vocab_size]
+        )
+        self.check_loss_output(result)
+
     def create_and_check_bert_for_masked_lm(
         self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
     ):
@@ -229,7 +256,7 @@ def create_and_check_bert_for_masked_lm(
         )
         self.check_loss_output(result)
 
-    def create_and_check_bert_model_for_masked_lm_as_decoder(
+    def create_and_check_bert_model_for_causal_lm_as_decoder(
         self,
         config,
         input_ids,
@@ -241,7 +268,7 @@ def create_and_check_bert_model_for_masked_lm_as_decoder(
         encoder_hidden_states,
         encoder_attention_mask,
     ):
-        model = BertForMaskedLM(config=config)
+        model = BertLMHeadModel(config=config)
         model.to(torch_device)
         model.eval()
         loss, prediction_scores = model(
@@ -461,13 +488,17 @@ def test_bert_model_as_decoder_with_default_input_mask(self):
             encoder_attention_mask,
         )
 
+    def test_for_causal_lm(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
+        self.model_tester.create_and_check_bert_for_causal_lm(*config_and_inputs)
+
     def test_for_masked_lm(self):
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_bert_for_masked_lm(*config_and_inputs)
 
-    def test_for_masked_lm_decoder(self):
+    def test_for_causal_lm_decoder(self):
         config_and_inputs = self.model_tester.prepare_config_and_inputs_for_decoder()
-        self.model_tester.create_and_check_bert_model_for_masked_lm_as_decoder(*config_and_inputs)
+        self.model_tester.create_and_check_bert_model_for_causal_lm_as_decoder(*config_and_inputs)
 
     def test_for_multiple_choice(self):
         config_and_inputs = self.model_tester.prepare_config_and_inputs()