Merge pull request #1601 from huggingface/clean-roberta

Clean roberta model & all tokenizers now add special tokens by default (breaking change)

examples/benchmarks.py

@@ -309,7 +309,7 @@ def _compute_pytorch(model_names, dictionary, average_over, device, torchscript)
         model = AutoModel.from_pretrained(model_name, config=config)
         tokenizer = AutoTokenizer.from_pretrained(model_name)
 
-        tokenized_sequence = tokenizer.encode(input_text)
+        tokenized_sequence = tokenizer.encode(input_text, add_special_tokens=False)
 
         max_input_size = tokenizer.max_model_input_sizes[model_name]
         batch_sizes = [1, 2, 4, 8]
@@ -353,7 +353,7 @@ def _compute_tensorflow(model_names, dictionary, average_over):
         model = TFAutoModel.from_pretrained(model_name, config=config)
         tokenizer = AutoTokenizer.from_pretrained(model_name)
 
-        tokenized_sequence = tokenizer.encode(input_text)
+        tokenized_sequence = tokenizer.encode(input_text, add_special_tokens=False)
 
         max_input_size = tokenizer.max_model_input_sizes[model_name]
         batch_sizes = [1, 2, 4, 8]

examples/distillation/scripts/binarized_data.py

@@ -68,7 +68,7 @@ def main():
     start = time.time()
     for text in data:
         text = f'{bos} {text.strip()} {sep}'
-        token_ids = tokenizer.encode(text)
+        token_ids = tokenizer.encode(text, add_special_tokens=False)
         rslt.append(token_ids)
 
         iter += 1

examples/run_generation.py

@@ -223,7 +223,7 @@ def main():
         if args.model_type in ["transfo-xl", "xlnet"]:
             # Models with memory likes to have a long prompt for short inputs.
             raw_text = (args.padding_text if args.padding_text else PADDING_TEXT) + raw_text
-        context_tokens = tokenizer.encode(raw_text)
+        context_tokens = tokenizer.encode(raw_text, add_special_tokens=False)
         if args.model_type == "ctrl":
             if not any(context_tokens[0] == x for x in tokenizer.control_codes.values()):
                 logger.info("WARNING! You are not starting your generation from a control code so you won't get good results")

transformers/modeling_roberta.py

@@ -169,18 +169,6 @@ class RobertaModel(BertModel):
         self.embeddings = RobertaEmbeddings(config)
         self.init_weights()
 
-    def forward(self, input_ids, attention_mask=None, token_type_ids=None, position_ids=None, head_mask=None):
-        if input_ids[:, 0].sum().item() != 0:
-            logger.warning("A sequence with no special tokens has been passed to the RoBERTa model. "
-                           "This model requires special tokens in order to work. "
-                           "Please specify add_special_tokens=True in your tokenize.encode()"
-                           "or tokenizer.convert_tokens_to_ids().")
-        return super(RobertaModel, self).forward(input_ids,
-                                                 attention_mask=attention_mask,
-                                                 token_type_ids=token_type_ids,
-                                                 position_ids=position_ids,
-                                                 head_mask=head_mask)
-
 
 @add_start_docstrings("""RoBERTa Model with a `language modeling` head on top. """,
     ROBERTA_START_DOCSTRING, ROBERTA_INPUTS_DOCSTRING)

transformers/modeling_tf_roberta.py

@@ -65,22 +65,6 @@ class TFRobertaMainLayer(TFBertMainLayer):
         super(TFRobertaMainLayer, self).__init__(config, **kwargs)
         self.embeddings = TFRobertaEmbeddings(config, name='embeddings')
 
-    def call(self, inputs, **kwargs):
-        # Check that input_ids starts with control token
-        if isinstance(inputs, (tuple, list)):
-            input_ids = inputs[0]
-        elif isinstance(inputs, dict):
-            input_ids = inputs.get('input_ids')
-        else:
-            input_ids = inputs
-
-        if tf.not_equal(tf.reduce_sum(input_ids[:, 0]), 0):
-            tf.print("A sequence with no special tokens has been passed to the RoBERTa model. "
-                           "This model requires special tokens in order to work. "
-                           "Please specify add_special_tokens=True in your encoding.")
-
-        return super(TFRobertaMainLayer, self).call(inputs, **kwargs)
-
 
 class TFRobertaPreTrainedModel(TFPreTrainedModel):
     """ An abstract class to handle weights initialization and

transformers/tests/tokenization_bert_test.py

@@ -128,8 +128,8 @@ class BertTokenizationTest(CommonTestCases.CommonTokenizerTester):
     def test_sequence_builders(self):
         tokenizer = self.tokenizer_class.from_pretrained("bert-base-uncased")
 
-        text = tokenizer.encode("sequence builders")
-        text_2 = tokenizer.encode("multi-sequence build")
+        text = tokenizer.encode("sequence builders", add_special_tokens=False)
+        text_2 = tokenizer.encode("multi-sequence build", add_special_tokens=False)
 
         encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
         encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)

transformers/tests/tokenization_distilbert_test.py

@@ -33,8 +33,8 @@ class DistilBertTokenizationTest(BertTokenizationTest):
     def test_sequence_builders(self):
         tokenizer = DistilBertTokenizer.from_pretrained("distilbert-base-uncased")
 
-        text = tokenizer.encode("sequence builders")
-        text_2 = tokenizer.encode("multi-sequence build")
+        text = tokenizer.encode("sequence builders", add_special_tokens=False)
+        text_2 = tokenizer.encode("multi-sequence build", add_special_tokens=False)
 
         encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
         encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)

transformers/tests/tokenization_roberta_test.py

@@ -70,19 +70,19 @@ class RobertaTokenizationTest(CommonTestCases.CommonTokenizerTester):
         tokenizer = self.get_tokenizer()
 
         self.assertListEqual(
-            tokenizer.encode('Hello world!'),
+            tokenizer.encode('Hello world!', add_special_tokens=False),
             [0, 31414, 232, 328, 2]
         )
         self.assertListEqual(
-            tokenizer.encode('Hello world! cécé herlolip 418'),
+            tokenizer.encode('Hello world! cécé herlolip 418', add_special_tokens=False),
             [0, 31414, 232, 328, 740, 1140, 12695, 69, 46078, 1588, 2]
         )
 
     def test_sequence_builders(self):
         tokenizer = RobertaTokenizer.from_pretrained("roberta-base")
 
-        text = tokenizer.encode("sequence builders")
-        text_2 = tokenizer.encode("multi-sequence build")
+        text = tokenizer.encode("sequence builders", add_special_tokens=False)
+        text_2 = tokenizer.encode("multi-sequence build", add_special_tokens=False)
 
         encoded_text_from_decode = tokenizer.encode("sequence builders", add_special_tokens=True)
         encoded_pair_from_decode = tokenizer.encode("sequence builders", "multi-sequence build", add_special_tokens=True)

transformers/tests/tokenization_tests_commons.py

@@ -79,13 +79,13 @@ class CommonTestCases:
             # Now let's start the test
             tokenizer = self.get_tokenizer(max_len=42)
 
-            before_tokens = tokenizer.encode(u"He is very happy, UNwant\u00E9d,running")
+            before_tokens = tokenizer.encode(u"He is very happy, UNwant\u00E9d,running", add_special_tokens=False)
 
             with TemporaryDirectory() as tmpdirname:
                 tokenizer.save_pretrained(tmpdirname)
                 tokenizer = self.tokenizer_class.from_pretrained(tmpdirname)
 
-                after_tokens = tokenizer.encode(u"He is very happy, UNwant\u00E9d,running")
+                after_tokens = tokenizer.encode(u"He is very happy, UNwant\u00E9d,running", add_special_tokens=False)
                 self.assertListEqual(before_tokens, after_tokens)
 
                 self.assertEqual(tokenizer.max_len, 42)
@@ -130,7 +130,7 @@ class CommonTestCases:
             self.assertEqual(added_toks, len(new_toks))
             self.assertEqual(all_size_2, all_size + len(new_toks))
 
-            tokens = tokenizer.encode("aaaaa bbbbbb low cccccccccdddddddd l")
+            tokens = tokenizer.encode("aaaaa bbbbbb low cccccccccdddddddd l", add_special_tokens=False)
             out_string = tokenizer.decode(tokens)
 
             self.assertGreaterEqual(len(tokens), 4)
@@ -148,7 +148,8 @@ class CommonTestCases:
             self.assertEqual(added_toks_2, len(new_toks_2))
             self.assertEqual(all_size_3, all_size_2 + len(new_toks_2))
 
-            tokens = tokenizer.encode(">>>>|||<||<<|<< aaaaabbbbbb low cccccccccdddddddd <<<<<|||>|>>>>|> l")
+            tokens = tokenizer.encode(">>>>|||<||<<|<< aaaaabbbbbb low cccccccccdddddddd <<<<<|||>|>>>>|> l",
+                                      add_special_tokens=False)
             out_string = tokenizer.decode(tokens)
 
             self.assertGreaterEqual(len(tokens), 6)
@@ -166,7 +167,7 @@ class CommonTestCases:
 
             tokens = tokenizer.tokenize(input_text)
             ids = tokenizer.convert_tokens_to_ids(tokens)
-            ids_2 = tokenizer.encode(input_text)
+            ids_2 = tokenizer.encode(input_text, add_special_tokens=False)
             self.assertListEqual(ids, ids_2)
 
             tokens_2 = tokenizer.convert_ids_to_tokens(ids)
@@ -206,7 +207,7 @@ class CommonTestCases:
             seq_0 = "Test this method."
             seq_1 = "With these inputs."
 
-            sequences = tokenizer.encode(seq_0, seq_1)
+            sequences = tokenizer.encode(seq_0, seq_1, add_special_tokens=False)
             attached_sequences = tokenizer.encode(seq_0, seq_1, add_special_tokens=True)
 
             # Method is implemented (e.g. not GPT-2)
@@ -219,7 +220,7 @@ class CommonTestCases:
             seq_0 = "This is a sentence to be encoded."
             stride = 2
 
-            sequence = tokenizer.encode(seq_0)
+            sequence = tokenizer.encode(seq_0, add_special_tokens=False)
             num_added_tokens = tokenizer.num_added_tokens()
             total_length = len(sequence) + num_added_tokens
             information = tokenizer.encode_plus(seq_0, max_length=total_length - 2, add_special_tokens=True, stride=stride)
@@ -239,13 +240,13 @@ class CommonTestCases:
             seq_1 = "This is another sentence to be encoded."
             stride = 2
 
-            sequence_0_no_special_tokens = tokenizer.encode(seq_0)
-            sequence_1_no_special_tokens = tokenizer.encode(seq_1)
+            sequence_0_no_special_tokens = tokenizer.encode(seq_0, add_special_tokens=False)
+            sequence_1_no_special_tokens = tokenizer.encode(seq_1, add_special_tokens=False)
 
             sequence = tokenizer.encode(seq_0, seq_1, add_special_tokens=True)
             truncated_second_sequence = tokenizer.build_inputs_with_special_tokens(
-                tokenizer.encode(seq_0),
-                tokenizer.encode(seq_1)[:-2]
+                tokenizer.encode(seq_0, add_special_tokens=False),
+                tokenizer.encode(seq_1, add_special_tokens=False)[:-2]
             )
 
             information = tokenizer.encode_plus(seq_0, seq_1, max_length=len(sequence) - 2, add_special_tokens=True,
@@ -283,7 +284,7 @@ class CommonTestCases:
             sequence_1 = "This one too please."
 
             # Testing single inputs
-            encoded_sequence = tokenizer.encode(sequence_0)
+            encoded_sequence = tokenizer.encode(sequence_0, add_special_tokens=False)
             encoded_sequence_dict = tokenizer.encode_plus(sequence_0, add_special_tokens=True)
             encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
             special_tokens_mask = encoded_sequence_dict["special_tokens_mask"]
@@ -294,7 +295,8 @@ class CommonTestCases:
             self.assertEqual(encoded_sequence, filtered_sequence)
 
             # Testing inputs pairs
-            encoded_sequence = tokenizer.encode(sequence_0) + tokenizer.encode(sequence_1)
+            encoded_sequence = tokenizer.encode(sequence_0, add_special_tokens=False) + tokenizer.encode(sequence_1,
+                                                                                                         add_special_tokens=False)
             encoded_sequence_dict = tokenizer.encode_plus(sequence_0, sequence_1, add_special_tokens=True)
             encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
             special_tokens_mask = encoded_sequence_dict["special_tokens_mask"]

transformers/tests/tokenization_xlm_test.py

@@ -69,8 +69,8 @@ class XLMTokenizationTest(CommonTestCases.CommonTokenizerTester):
     def test_sequence_builders(self):
         tokenizer = XLMTokenizer.from_pretrained("xlm-mlm-en-2048")
 
-        text = tokenizer.encode("sequence builders")
-        text_2 = tokenizer.encode("multi-sequence build")
+        text = tokenizer.encode("sequence builders", add_special_tokens=False)
+        text_2 = tokenizer.encode("multi-sequence build", add_special_tokens=False)
 
         encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
         encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)

transformers/tests/tokenization_xlnet_test.py

@@ -92,8 +92,8 @@ class XLNetTokenizationTest(CommonTestCases.CommonTokenizerTester):
     def test_sequence_builders(self):
         tokenizer = XLNetTokenizer.from_pretrained("xlnet-base-cased")
 
-        text = tokenizer.encode("sequence builders")
-        text_2 = tokenizer.encode("multi-sequence build")
+        text = tokenizer.encode("sequence builders", add_special_tokens=False)
+        text_2 = tokenizer.encode("multi-sequence build", add_special_tokens=False)
 
         encoded_sentence = tokenizer.build_inputs_with_special_tokens(text)
         encoded_pair = tokenizer.build_inputs_with_special_tokens(text, text_2)

transformers/tokenization_utils.py

@@ -689,14 +689,14 @@ class PreTrainedTokenizer(object):
         raise NotImplementedError
 
     def encode(self,
-                text,
-                text_pair=None,
-                add_special_tokens=False,
-                max_length=None,
-                stride=0,
-                truncation_strategy='longest_first',
-                return_tensors=None,
-                **kwargs):
+               text,
+               text_pair=None,
+               add_special_tokens=True,
+               max_length=None,
+               stride=0,
+               truncation_strategy='longest_first',
+               return_tensors=None,
+               **kwargs):
         """
         Converts a string in a sequence of ids (integer), using the tokenizer and vocabulary.
 
@@ -739,7 +739,7 @@ class PreTrainedTokenizer(object):
     def encode_plus(self,
                     text,
                     text_pair=None,
-                    add_special_tokens=False,
+                    add_special_tokens=True,
                     max_length=None,
                     stride=0,
                     truncation_strategy='longest_first',
@@ -794,7 +794,7 @@ class PreTrainedTokenizer(object):
                                       truncation_strategy=truncation_strategy,
                                       return_tensors=return_tensors)
 
-    def prepare_for_model(self, ids, pair_ids=None, max_length=None, add_special_tokens=False, stride=0,
+    def prepare_for_model(self, ids, pair_ids=None, max_length=None, add_special_tokens=True, stride=0,
                           truncation_strategy='longest_first', return_tensors=None):
         """
         Prepares a sequence of input id, or a pair of sequences of inputs ids so that it can be used by the model.