Merge remote-tracking branch 'huggingface/master'

pytorch_transformers/tests/tokenization_dilbert_test.py

+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from __future__ import absolute_import, division, print_function, unicode_literals
+
+import os
+import unittest
+from io import open
+
+from pytorch_transformers.tokenization_distilbert import (DistilBertTokenizer)
+
+from .tokenization_tests_commons import CommonTestCases
+from .tokenization_bert_test import BertTokenizationTest
+
+class DistilBertTokenizationTest(BertTokenizationTest):
+
+    tokenizer_class = DistilBertTokenizer
+
+    def get_tokenizer(self):
+        return DistilBertTokenizer.from_pretrained(self.tmpdirname)
+
+    def test_sequence_builders(self):
+        tokenizer = DistilBertTokenizer.from_pretrained("distilbert-base-uncased")
+
+        text = tokenizer.encode("sequence builders")
+        text_2 = tokenizer.encode("multi-sequence build")
+
+        encoded_sentence = tokenizer.add_special_tokens_single_sentence(text)
+        encoded_pair = tokenizer.add_special_tokens_sentences_pair(text, text_2)
+
+        assert encoded_sentence == [101] + text + [102]
+        assert encoded_pair == [101] + text + [102] + text_2 + [102]
+
+if __name__ == '__main__':
+    unittest.main()

pytorch_transformers/tokenization_bert.py

@@ -125,6 +125,9 @@ class BertTokenizer(PreTrainedTokenizer):
         super(BertTokenizer, self).__init__(unk_token=unk_token, sep_token=sep_token,
                                             pad_token=pad_token, cls_token=cls_token,
                                             mask_token=mask_token, **kwargs)
+        self.max_len_single_sentence = self.max_len - 2  # take into account special tokens
+        self.max_len_sentences_pair = self.max_len - 3  # take into account special tokens
+
         if not os.path.isfile(vocab_file):
             raise ValueError(
                 "Can't find a vocabulary file at path '{}'. To load the vocabulary from a Google pretrained "
@@ -187,6 +190,8 @@ class BertTokenizer(PreTrainedTokenizer):
         index = 0
         if os.path.isdir(vocab_path):
             vocab_file = os.path.join(vocab_path, VOCAB_FILES_NAMES['vocab_file'])
+        else:
+            vocab_file = vocab_path
         with open(vocab_file, "w", encoding="utf-8") as writer:
             for token, token_index in sorted(self.vocab.items(), key=lambda kv: kv[1]):
                 if index != token_index:

pytorch_transformers/tokenization_distilbert.py

+# coding=utf-8
+# Copyright 2018 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Tokenization classes for DistilBERT."""
+
+from __future__ import absolute_import, division, print_function, unicode_literals
+
+import collections
+import logging
+import os
+import unicodedata
+from io import open
+
+from .tokenization_bert import BertTokenizer
+
+logger = logging.getLogger(__name__)
+
+VOCAB_FILES_NAMES = {'vocab_file': 'vocab.txt'}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    'vocab_file':
+    {
+        'distilbert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-vocab.txt",
+        'distilbert-base-uncased-distilled-squad': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-large-uncased-vocab.txt",
+    }
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    'distilbert-base-uncased': 512,
+    'distilbert-base-uncased-distilled-squad': 512,
+}
+
+
+class DistilBertTokenizer(BertTokenizer):
+    r"""
+    Constructs a DistilBertTokenizer.
+    :class:`~pytorch_transformers.DistilBertTokenizer` is identical to BertTokenizer and runs end-to-end tokenization: punctuation splitting + wordpiece
+
+    Args:
+        vocab_file: Path to a one-wordpiece-per-line vocabulary file
+        do_lower_case: Whether to lower case the input. Only has an effect when do_wordpiece_only=False
+        do_basic_tokenize: Whether to do basic tokenization before wordpiece.
+        max_len: An artificial maximum length to truncate tokenized sequences to; Effective maximum length is always the
+            minimum of this value (if specified) and the underlying BERT model's sequence length.
+        never_split: List of tokens which will never be split during tokenization. Only has an effect when
+            do_wordpiece_only=False
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES

pytorch_transformers/tokenization_gpt2.py

@@ -45,17 +45,20 @@ PRETRAINED_VOCAB_FILES_MAP = {
     {
         'gpt2': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-vocab.json",
         'gpt2-medium': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-medium-vocab.json",
+        'gpt2-large': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-large-vocab.json",
     },
     'merges_file':
     {
         'gpt2': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-merges.txt",
         'gpt2-medium': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-medium-merges.txt",
+        'gpt2-large': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-large-merges.txt",
     },
 }
 
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     'gpt2': 1024,
     'gpt2-medium': 1024,
+    'gpt2-large': 1024,
 }
 
 @lru_cache()
@@ -105,6 +108,8 @@ class GPT2Tokenizer(PreTrainedTokenizer):
     def __init__(self, vocab_file, merges_file, errors='replace', unk_token="<|endoftext|>",
                  bos_token="<|endoftext|>", eos_token="<|endoftext|>", **kwargs):
         super(GPT2Tokenizer, self).__init__(bos_token=bos_token, eos_token=eos_token, unk_token=unk_token, **kwargs)
+        self.max_len_single_sentence = self.max_len # no default special tokens - you can update this value if you add special tokens
+        self.max_len_sentences_pair = self.max_len # no default special tokens - you can update this value if you add special tokens
 
         self.encoder = json.load(open(vocab_file))
         self.decoder = {v:k for k,v in self.encoder.items()}

pytorch_transformers/tokenization_openai.py

@@ -87,10 +87,14 @@ class OpenAIGPTTokenizer(PreTrainedTokenizer):
     def __init__(self, vocab_file, merges_file, unk_token="<unk>", **kwargs):
         super(OpenAIGPTTokenizer, self).__init__(unk_token=unk_token, **kwargs)
 
+        self.max_len_single_sentence = self.max_len # no default special tokens - you can update this value if you add special tokens
+        self.max_len_sentences_pair = self.max_len # no default special tokens - you can update this value if you add special tokens
+
         try:
             import ftfy
-            import spacy
-            self.nlp = spacy.load('en', disable=['parser', 'tagger', 'ner', 'textcat'])
+            from spacy.lang.en import English
+            _nlp = English()
+            self.nlp = _nlp.Defaults.create_tokenizer(_nlp)
             self.fix_text = ftfy.fix_text
         except ImportError:
             logger.warning("ftfy or spacy is not installed using BERT BasicTokenizer instead of SpaCy & ftfy.")

pytorch_transformers/tokenization_roberta.py

@@ -77,6 +77,9 @@ class RobertaTokenizer(PreTrainedTokenizer):
                                                sep_token=sep_token, cls_token=cls_token, pad_token=pad_token,
                                                mask_token=mask_token, **kwargs)
 
+        self.max_len_single_sentence = self.max_len - 2  # take into account special tokens
+        self.max_len_sentences_pair = self.max_len - 4  # take into account special tokens
+
         self.encoder = json.load(open(vocab_file, encoding="utf-8"))
         self.decoder = {v: k for k, v in self.encoder.items()}
         self.errors = errors  # how to handle errors in decoding
@@ -163,14 +166,14 @@ class RobertaTokenizer(PreTrainedTokenizer):
     def add_special_tokens_single_sentence(self, token_ids):
         """
         Adds special tokens to a sequence for sequence classification tasks.
-        A RoBERTa sequence has the following format: [CLS] X [SEP]
+        A RoBERTa sequence has the following format: <s> X </s>
         """
         return [self._convert_token_to_id(self.cls_token)] + token_ids + [self._convert_token_to_id(self.sep_token)]
 
     def add_special_tokens_sentences_pair(self, token_ids_0, token_ids_1):
         """
         Adds special tokens to a sequence pair for sequence classification tasks.
-        A RoBERTa sequence pair has the following format: [CLS] A [SEP][SEP] B [SEP]
+        A RoBERTa sequence pair has the following format: <s> A </s></s> B </s>
         """
         sep = [self._convert_token_to_id(self.sep_token)]
         cls = [self._convert_token_to_id(self.cls_token)]

pytorch_transformers/tokenization_transfo_xl.py

@@ -73,6 +73,10 @@ class TransfoXLTokenizer(PreTrainedTokenizer):
         super(TransfoXLTokenizer, self).__init__(unk_token=unk_token, eos_token=eos_token,
                                                  additional_special_tokens=additional_special_tokens,
                                                  **kwargs)
+
+        self.max_len_single_sentence = self.max_len # no default special tokens - you can update this value if you add special tokens
+        self.max_len_sentences_pair = self.max_len # no default special tokens - you can update this value if you add special tokens
+
         if never_split is None:
             never_split = self.all_special_tokens
         if special is None:

pytorch_transformers/tokenization_utils.py

@@ -166,6 +166,9 @@ class PreTrainedTokenizer(object):
         self._additional_special_tokens = []
 
         self.max_len = max_len if max_len is not None else int(1e12)
+        self.max_len_single_sentence = self.max_len
+        self.max_len_sentences_pair = self.max_len
+
         self.added_tokens_encoder = {}
         self.added_tokens_decoder = {}
 
@@ -193,6 +196,13 @@ class PreTrainedTokenizer(object):
             cache_dir: (`optional`) string:
                 Path to a directory in which a downloaded predefined tokenizer vocabulary files should be cached if the standard cache should not be used.
 
+            force_download: (`optional`) boolean, default False:
+                Force to (re-)download the vocabulary files and override the cached versions if they exists.
+
+            proxies: (`optional`) dict, default None:
+                A dictionary of proxy servers to use by protocol or endpoint, e.g.: {'http': 'foo.bar:3128', 'http://hostname': 'foo.bar:4012'}.
+                The proxies are used on each request.
+
             inputs: (`optional`) positional arguments: will be passed to the Tokenizer ``__init__`` method.
 
             kwargs: (`optional`) keyword arguments: will be passed to the Tokenizer ``__init__`` method. Can be used to set special tokens like ``bos_token``, ``eos_token``, ``unk_token``, ``sep_token``, ``pad_token``, ``cls_token``, ``mask_token``, ``additional_special_tokens``. See parameters in the doc string of :class:`~pytorch_transformers.PreTrainedTokenizer` for details.
@@ -223,6 +233,8 @@ class PreTrainedTokenizer(object):
     @classmethod
     def _from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs):
         cache_dir = kwargs.pop('cache_dir', None)
+        force_download = kwargs.pop('force_download', False)
+        proxies = kwargs.pop('proxies', None)
 
         s3_models = list(cls.max_model_input_sizes.keys())
         vocab_files = {}
@@ -283,8 +295,8 @@ class PreTrainedTokenizer(object):
                 if file_path is None:
                     resolved_vocab_files[file_id] = None
                 else:
-                    resolved_vocab_files[file_id] = cached_path(file_path, cache_dir=cache_dir)
-        except EnvironmentError:
+                    resolved_vocab_files[file_id] = cached_path(file_path, cache_dir=cache_dir, force_download=force_download, proxies=proxies)
+        except EnvironmentError as e:
             if pretrained_model_name_or_path in s3_models:
                 logger.error("Couldn't reach server to download vocabulary.")
             else:
@@ -294,7 +306,7 @@ class PreTrainedTokenizer(object):
                     "at this path or url.".format(
                         pretrained_model_name_or_path, ', '.join(s3_models),
                         pretrained_model_name_or_path, str(vocab_files.keys())))
-            return None
+            raise e
 
         for file_id, file_path in vocab_files.items():
             if file_path == resolved_vocab_files[file_id]:
@@ -477,15 +489,45 @@ class PreTrainedTokenizer(object):
 
             Take care of added tokens.
         """
+        def split_on_token(tok, text):
+            result = []
+            split_text = text.split(tok)
+            for i, sub_text in enumerate(split_text):
+                sub_text = sub_text.strip()
+                if i == 0 and not sub_text:
+                    result += [tok]
+                elif i == len(split_text) - 1:
+                    if sub_text:
+                        result += [sub_text]
+                    else:
+                        pass
+                else:
+                    if sub_text:
+                        result += [sub_text]
+                    result += [tok]
+            return result
+
         def split_on_tokens(tok_list, text):
             if not text:
                 return []
             if not tok_list:
                 return self._tokenize(text, **kwargs)
-            tok = tok_list[0]
-            split_text = text.split(tok)
-            return sum((split_on_tokens(tok_list[1:], sub_text.strip()) + [tok] \
-                        for sub_text in split_text), [])[:-1]
+
+            tokenized_text = []
+            text_list = [text]
+            for tok in tok_list:
+                tokenized_text = []
+                for sub_text in text_list:
+                    if sub_text not in self.added_tokens_encoder \
+                            and sub_text not in self.all_special_tokens:
+                        tokenized_text += split_on_token(tok, sub_text)
+                    else:
+                        tokenized_text += [sub_text]
+                text_list = tokenized_text
+
+            return sum((self._tokenize(token, **kwargs) if token not \
+                    in self.added_tokens_encoder and token not in self.all_special_tokens \
+                    else [token] for token in tokenized_text), [])
 
         added_tokens = list(self.added_tokens_encoder.keys()) + self.all_special_tokens
         tokenized_text = split_on_tokens(added_tokens, text)
@@ -551,10 +593,12 @@ class PreTrainedTokenizer(object):
             return first_sentence_tokens, second_sentence_tokens
 
     def add_special_tokens_single_sentence(self, token_ids):
-        raise NotImplementedError
+        logger.warning("This tokenizer does not make use of special tokens. The sequence has been returned with no modification.")
+        return token_ids
 
     def add_special_tokens_sentences_pair(self, token_ids_0, token_ids_1):
-        raise NotImplementedError
+        logger.warning("This tokenizer does not make use of special tokens. The two sequences have been concatenated.")
+        return token_ids_0 + token_ids_1
 
     def convert_ids_to_tokens(self, ids, skip_special_tokens=False):
         """ Converts a single index or a sequence of indices (integers) in a token "
@@ -597,9 +641,9 @@ class PreTrainedTokenizer(object):
         filtered_tokens = self.convert_ids_to_tokens(token_ids, skip_special_tokens=skip_special_tokens)
         text = self.convert_tokens_to_string(filtered_tokens)
 
-        if self.sep_token is not None and self.sep_token in text:
-            text = text.replace(self.cls_token, self.sep_token)
-            split_text = list(filter(lambda sentence: len(sentence) > 0, text.split(self.sep_token)))
+        if self._sep_token is not None and self._sep_token in text:
+            text = text.replace(self._cls_token, self._sep_token)
+            split_text = list(filter(lambda sentence: len(sentence) > 0, text.split(self._sep_token)))
             if clean_up_tokenization_spaces:
                 clean_text = [self.clean_up_tokenization(text) for text in split_text]
                 return clean_text

pytorch_transformers/tokenization_xlm.py

@@ -122,10 +122,15 @@ class XLMTokenizer(PreTrainedTokenizer):
                                            cls_token=cls_token, mask_token=mask_token,
                                            additional_special_tokens=additional_special_tokens,
                                            **kwargs)
+
+        self.max_len_single_sentence = self.max_len - 2  # take into account special tokens
+        self.max_len_sentences_pair = self.max_len - 3  # take into account special tokens
+
         try:
             import ftfy
-            import spacy
-            self.nlp = spacy.load('en', disable=['parser', 'tagger', 'ner', 'textcat'])
+            from spacy.lang.en import English
+            _nlp = English()
+            self.nlp = _nlp.Defaults.create_tokenizer(_nlp)
             self.fix_text = ftfy.fix_text
         except ImportError:
             logger.warning("ftfy or spacy is not installed using BERT BasicTokenizer instead of SpaCy & ftfy.")

pytorch_transformers/tokenization_xlnet.py

@@ -71,6 +71,10 @@ class XLNetTokenizer(PreTrainedTokenizer):
                                              pad_token=pad_token, cls_token=cls_token,
                                              mask_token=mask_token, additional_special_tokens=
                                              additional_special_tokens, **kwargs)
+
+        self.max_len_single_sentence = self.max_len - 2  # take into account special tokens
+        self.max_len_sentences_pair = self.max_len - 3  # take into account special tokens
+
         try:
             import sentencepiece as spm
         except ImportError: