Merge branch 'master' into generation_sampler

transformers/modeling_bert.py

@@ -48,6 +48,12 @@ BERT_PRETRAINED_MODEL_ARCHIVE_MAP = {
     'bert-base-cased-finetuned-mrpc': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-cased-finetuned-mrpc-pytorch_model.bin",
     'bert-base-german-dbmdz-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-german-dbmdz-cased-pytorch_model.bin",
     'bert-base-german-dbmdz-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-german-dbmdz-uncased-pytorch_model.bin",
+    'bert-base-japanese': "https://s3.amazonaws.com/models.huggingface.co/bert/cl-tohoku/bert-base-japanese-pytorch_model.bin",
+    'bert-base-japanese-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/cl-tohoku/bert-base-japanese-whole-word-masking-pytorch_model.bin",
+    'bert-base-japanese-char': "https://s3.amazonaws.com/models.huggingface.co/bert/cl-tohoku/bert-base-japanese-char-pytorch_model.bin",
+    'bert-base-japanese-char-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/cl-tohoku/bert-base-japanese-char-whole-word-masking-pytorch_model.bin",
+    'bert-base-finnish-cased-v1': "https://s3.amazonaws.com/models.huggingface.co/bert/TurkuNLP/bert-base-finnish-cased-v1/pytorch_model.bin",
+    'bert-base-finnish-uncased-v1': "https://s3.amazonaws.com/models.huggingface.co/bert/TurkuNLP/bert-base-finnish-uncased-v1/pytorch_model.bin",
 }
 
 

transformers/modeling_ctrl.py

@@ -268,7 +268,7 @@ class CTRLModel(CTRLPreTrainedModel):
 
         tokenizer = CTRLTokenizer.from_pretrained('ctrl')
         model = CTRLModel.from_pretrained('ctrl')
-        input_ids = torch.tensor(tokenizer.encode("Links Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        input_ids = torch.tensor(tokenizer.encode("Links 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
 
@@ -458,7 +458,7 @@ class CTRLLMHeadModel(CTRLPreTrainedModel):
         tokenizer = CTRLTokenizer.from_pretrained('ctrl')
         model = CTRLLMHeadModel.from_pretrained('ctrl')
 
-        input_ids = torch.tensor(tokenizer.encode("Links Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        input_ids = torch.tensor(tokenizer.encode("Links 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]
 

transformers/modeling_distilbert.py

@@ -415,7 +415,7 @@ class DistilBertModel(DistilBertPreTrainedModel):
 
         tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')
         model = DistilBertModel.from_pretrained('distilbert-base-uncased')
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        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
 
@@ -511,7 +511,7 @@ class DistilBertForMaskedLM(DistilBertPreTrainedModel):
 
         tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')
         model = DistilBertForMaskedLM.from_pretrained('distilbert-base-uncased')
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
         outputs = model(input_ids, masked_lm_labels=input_ids)
         loss, prediction_scores = outputs[:2]
 
@@ -581,7 +581,7 @@ class DistilBertForSequenceClassification(DistilBertPreTrainedModel):
 
         tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')
         model = DistilBertForSequenceClassification.from_pretrained('distilbert-base-uncased')
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
         labels = torch.tensor([1]).unsqueeze(0)  # Batch size 1
         outputs = model(input_ids, labels=labels)
         loss, logits = outputs[:2]
@@ -656,7 +656,7 @@ class DistilBertForQuestionAnswering(DistilBertPreTrainedModel):
 
         tokenizer = DistilBertTokenizer.from_pretrained('distilbert-base-uncased')
         model = DistilBertForQuestionAnswering.from_pretrained('distilbert-base-uncased')
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True)).unsqueeze(0)  # Batch size 1
         start_positions = torch.tensor([1])
         end_positions = torch.tensor([3])
         outputs = model(input_ids, start_positions=start_positions, end_positions=end_positions)

transformers/modeling_encoder_decoder.py

@@ -61,12 +61,14 @@ class PreTrainedEncoderDecoder(nn.Module):
             encoder_pretrained_model_name_or_path: information necessary to initiate the encoder. Either:
 
                 - a string with the `shortcut name` of a pre-trained model to load from cache or download, e.g.: ``bert-base-uncased``.
+                - a string with the `identifier name` of a pre-trained model that was user-uploaded to our S3, e.g.: ``dbmdz/bert-base-german-cased``.
                 - a path to a `directory` containing model weights saved using :func:`~transformers.PreTrainedModel.save_pretrained`, e.g.: ``./my_model_directory/encoder``.
                 - a path or url to a `tensorflow index checkpoint file` (e.g. `./tf_model/model.ckpt.index`). In this case, ``from_tf`` should be set to True and a configuration object should be provided as ``config`` argument. This loading path is slower than converting the TensorFlow checkpoint in a PyTorch model using the provided conversion scripts and loading the PyTorch model afterwards.
 
             decoder_pretrained_model_name_or_path: information necessary to initiate the decoder. Either:
 
                 - a string with the `shortcut name` of a pre-trained model to load from cache or download, e.g.: ``bert-base-uncased``.
+                - a string with the `identifier name` of a pre-trained model that was user-uploaded to our S3, e.g.: ``dbmdz/bert-base-german-cased``.
                 - a path to a `directory` containing model weights saved using :func:`~transformers.PreTrainedModel.save_pretrained`, e.g.: ``./my_model_directory/decoder``.
                 - a path or url to a `tensorflow index checkpoint file` (e.g. `./tf_model/model.ckpt.index`). In this case, ``from_tf`` should be set to True and a configuration object should be provided as ``config`` argument. This loading path is slower than converting the TensorFlow checkpoint in a PyTorch model using the provided conversion scripts and loading the PyTorch model afterwards.
 
@@ -165,7 +167,39 @@ class PreTrainedEncoderDecoder(nn.Module):
 
         We save the encoder' and decoder's parameters in two separate directories.
         """
+
+        # If the root output directory does not exist, create it 
+        if not os.path.exists(save_directory):
+            os.mkdir(save_directory)
+
+        # Check whether the output directory is empty or not
+        sub_directories = [directory for directory in os.listdir(save_directory)
+            if os.path.isdir(os.path.join(save_directory, directory))]
+
+        if len(sub_directories) > 0:
+            if "encoder" in sub_directories and "decoder" in sub_directories:
+                print("WARNING: there is an older version of encoder-decoder saved in" +\
+                    " the output directory. The default behaviour is to overwrite them.")
+
+            # Empty the output directory
+            for directory_to_remove in sub_directories:
+                # Remove all files into the subdirectory
+                files_to_remove = os.listdir(os.path.join(save_directory, directory_to_remove))
+                for file_to_remove in files_to_remove:
+                    os.remove(os.path.join(save_directory, directory_to_remove, file_to_remove))
+                # Remove the subdirectory itself
+                os.rmdir(os.path.join(save_directory, directory_to_remove))
+
+            assert(len(os.listdir(save_directory)) == 0) # sanity check
+
+        # Create the "encoder" directory inside the output directory and save the encoder into it
+        if not os.path.exists(os.path.join(save_directory, "encoder")):
+            os.mkdir(os.path.join(save_directory, "encoder"))
         self.encoder.save_pretrained(os.path.join(save_directory, "encoder"))
+
+        # Create the "encoder" directory inside the output directory and save the decoder into it
+        if not os.path.exists(os.path.join(save_directory, "decoder")):
+            os.mkdir(os.path.join(save_directory, "decoder"))
         self.decoder.save_pretrained(os.path.join(save_directory, "decoder"))
 
     def forward(self, encoder_input_ids, decoder_input_ids, **kwargs):
@@ -236,7 +270,6 @@ class PreTrainedEncoderDecoder(nn.Module):
             }
         )
         decoder_kwargs["encoder_attention_mask"] = encoder_kwargs.get("attention_mask", None)
-
         return encoder_kwargs, decoder_kwargs
 
 

transformers/modeling_gpt2.py

@@ -345,7 +345,7 @@ class GPT2Model(GPT2PreTrainedModel):
 
         tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
         model = GPT2Model.from_pretrained('gpt2')
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        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
 
@@ -523,7 +523,7 @@ class GPT2LMHeadModel(GPT2PreTrainedModel):
         tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
         model = GPT2LMHeadModel.from_pretrained('gpt2')
 
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        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]
 
@@ -634,6 +634,7 @@ class GPT2DoubleHeadsModel(GPT2PreTrainedModel):
     """
     def __init__(self, config):
         super(GPT2DoubleHeadsModel, self).__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)

transformers/modeling_openai.py

@@ -349,7 +349,7 @@ class OpenAIGPTModel(OpenAIGPTPreTrainedModel):
 
         tokenizer = OpenAIGPTTokenizer.from_pretrained('openai-gpt')
         model = OpenAIGPTModel.from_pretrained('openai-gpt')
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        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
 
@@ -491,7 +491,7 @@ class OpenAIGPTLMHeadModel(OpenAIGPTPreTrainedModel):
 
         tokenizer = OpenAIGPTTokenizer.from_pretrained('openai-gpt')
         model = OpenAIGPTLMHeadModel.from_pretrained('openai-gpt')
-        input_ids = torch.tensor(tokenizer.encode("Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        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]
 
@@ -590,6 +590,7 @@ class OpenAIGPTDoubleHeadsModel(OpenAIGPTPreTrainedModel):
     def __init__(self, config):
         super(OpenAIGPTDoubleHeadsModel, self).__init__(config)
 
+        config.num_labels = 1
         self.transformer = OpenAIGPTModel(config)
         self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
         self.multiple_choice_head = SequenceSummary(config)

transformers/modeling_tf_albert.py

@@ -587,8 +587,8 @@ class TFAlbertModel(TFAlbertPreTrainedModel):
         import tensorflow as tf
         from transformers import AlbertTokenizer, TFAlbertModel
 
-        tokenizer = AlbertTokenizer.from_pretrained('bert-base-uncased')
-        model = TFAlbertModel.from_pretrained('bert-base-uncased')
+        tokenizer = AlbertTokenizer.from_pretrained('albert-base-v1')
+        model = TFAlbertModel.from_pretrained('albert-base-v1')
         input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :]  # Batch size 1
         outputs = model(input_ids)
         last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple

transformers/modeling_tf_ctrl.py

@@ -418,7 +418,7 @@ class TFCTRLModel(TFCTRLPreTrainedModel):
 
         tokenizer = CTRLTokenizer.from_pretrained('ctrl')
         model = TFCTRLModel.from_pretrained('ctrl')
-        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :]  # Batch size 1
+        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True))[None, :]  # Batch size 1
         outputs = model(input_ids)
         last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple
 
@@ -481,7 +481,7 @@ class TFCTRLLMHeadModel(TFCTRLPreTrainedModel):
         tokenizer = CTRLTokenizer.from_pretrained('ctrl')
         model = TFCTRLLMHeadModel.from_pretrained('ctrl')
 
-        input_ids = torch.tensor(tokenizer.encode("Links Hello, my dog is cute")).unsqueeze(0)  # Batch size 1
+        input_ids = torch.tensor(tokenizer.encode("Links 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]
 

transformers/modeling_tf_gpt2.py

@@ -454,7 +454,7 @@ class TFGPT2Model(TFGPT2PreTrainedModel):
 
         tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
         model = TFGPT2Model.from_pretrained('gpt2')
-        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :]  # Batch size 1
+        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True))[None, :]  # Batch size 1
         outputs = model(input_ids)
         last_hidden_states = outputs[0]  # The last hidden-state is the first element of the output tuple
 
@@ -495,7 +495,7 @@ class TFGPT2LMHeadModel(TFGPT2PreTrainedModel):
         tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
         model = TFGPT2LMHeadModel.from_pretrained('gpt2')
 
-        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute"))[None, :]  # Batch size 1
+        input_ids = tf.constant(tokenizer.encode("Hello, my dog is cute", add_special_tokens=True))[None, :]  # Batch size 1
         outputs = model(input_ids)
         logits = outputs[0]
 
@@ -574,6 +574,7 @@ class TFGPT2DoubleHeadsModel(TFGPT2PreTrainedModel):
     """
     def __init__(self, config, *inputs, **kwargs):
         super(TFGPT2DoubleHeadsModel, self).__init__(config, *inputs, **kwargs)
+        config.num_labels = 1
         self.transformer = TFGPT2MainLayer(config, name='transformer')
         self.multiple_choice_head = TFSequenceSummary(config, initializer_range=config.initializer_range, name='multiple_choice_head')