Merge remote-tracking branch 'refs/remotes/huggingface/master'

transformers/tests/tokenization_roberta_test.py

@@ -21,6 +21,7 @@ from io import open
 
 from transformers.tokenization_roberta import RobertaTokenizer, VOCAB_FILES_NAMES
 from .tokenization_tests_commons import CommonTestCases
+from .utils import slow
 
 
 class RobertaTokenizationTest(CommonTestCases.CommonTokenizerTester):
@@ -70,19 +71,20 @@ 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]
         )
 
+    @slow
     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)
@@ -102,14 +102,48 @@ class CommonTestCases:
             with TemporaryDirectory() as tmpdirname:
 
                 filename = os.path.join(tmpdirname, u"tokenizer.bin")
-                pickle.dump(tokenizer, open(filename, "wb"))
+                with open(filename, "wb") as handle:
+                    pickle.dump(tokenizer, handle)
 
-                tokenizer_new = pickle.load(open(filename, "rb"))
+                with open(filename, "rb") as handle:
+                    tokenizer_new = pickle.load(handle)
 
             subwords_loaded = tokenizer_new.tokenize(text)
 
             self.assertListEqual(subwords, subwords_loaded)
 
+        def test_added_tokens_do_lower_case(self):
+            tokenizer = self.get_tokenizer(do_lower_case=True)
+
+            special_token = tokenizer.all_special_tokens[0]
+
+            text = special_token + " aaaaa bbbbbb low cccccccccdddddddd l " + special_token
+            text2 = special_token + " AAAAA BBBBBB low CCCCCCCCCDDDDDDDD l " + special_token
+
+            toks0 = tokenizer.tokenize(text)  # toks before adding new_toks
+
+            new_toks = ["aaaaa bbbbbb", "cccccccccdddddddd", 'AAAAA BBBBBB', 'CCCCCCCCCDDDDDDDD']
+            added = tokenizer.add_tokens(new_toks)
+            self.assertEqual(added, 2)
+
+            toks = tokenizer.tokenize(text)
+            toks2 = tokenizer.tokenize(text2)
+
+            self.assertEqual(len(toks), len(toks2))
+            self.assertNotEqual(len(toks), len(toks0))  # toks0 should be longer
+            self.assertListEqual(toks, toks2)
+
+            tokenizer = self.get_tokenizer(do_lower_case=False)
+
+            added = tokenizer.add_tokens(new_toks)
+            self.assertEqual(added, 4)
+
+            toks = tokenizer.tokenize(text)
+            toks2 = tokenizer.tokenize(text2)
+
+            self.assertEqual(len(toks), len(toks2))  # Length should still be the same
+            self.assertNotEqual(len(toks), len(toks0))
+            self.assertNotEqual(toks[1], toks2[1])  # But at least the first non-special tokens should differ
 
         def test_add_tokens_tokenizer(self):
             tokenizer = self.get_tokenizer()
@@ -130,7 +164,7 @@ 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 +182,8 @@ 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)
@@ -159,6 +194,26 @@ class CommonTestCases:
             self.assertEqual(tokens[0], tokenizer.eos_token_id)
             self.assertEqual(tokens[-2], tokenizer.pad_token_id)
 
+        def test_add_special_tokens(self):
+            tokenizer = self.get_tokenizer()
+            input_text, output_text = self.get_input_output_texts()
+
+            special_token = "[SPECIAL TOKEN]"
+
+            tokenizer.add_special_tokens({"cls_token": special_token})
+            encoded_special_token = tokenizer.encode(special_token, add_special_tokens=False)
+            assert len(encoded_special_token) == 1
+
+            text = " ".join([input_text, special_token, output_text])
+            encoded = tokenizer.encode(text, add_special_tokens=False)
+
+            input_encoded = tokenizer.encode(input_text, add_special_tokens=False)
+            output_encoded = tokenizer.encode(output_text, add_special_tokens=False)
+            special_token_id = tokenizer.encode(special_token, add_special_tokens=False)
+            assert encoded == input_encoded + special_token_id + output_encoded
+
+            decoded = tokenizer.decode(encoded, skip_special_tokens=True)
+            assert special_token not in decoded
 
         def test_required_methods_tokenizer(self):
             tokenizer = self.get_tokenizer()
@@ -166,7 +221,7 @@ 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 +261,7 @@ 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)
@@ -219,10 +274,14 @@ 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)
+            information = tokenizer.encode_plus(seq_0,
+                                                max_length=total_length - 2,
+                                                add_special_tokens=True,
+                                                stride=stride,
+                                                return_overflowing_tokens=True)
 
             truncated_sequence = information["input_ids"]
             overflowing_tokens = information["overflowing_tokens"]
@@ -239,20 +298,22 @@ 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,
-                                                stride=stride, truncation_strategy='only_second')
+                                                stride=stride, truncation_strategy='only_second',
+                                                return_overflowing_tokens=True)
             information_first_truncated = tokenizer.encode_plus(seq_0, seq_1, max_length=len(sequence) - 2,
                                                                 add_special_tokens=True, stride=stride,
-                                                                truncation_strategy='only_first')
+                                                                truncation_strategy='only_first',
+                                                                return_overflowing_tokens=True)
 
             truncated_sequence = information["input_ids"]
             overflowing_tokens = information["overflowing_tokens"]
@@ -283,8 +344,8 @@ class CommonTestCases:
             sequence_1 = "This one too please."
 
             # Testing single inputs
-            encoded_sequence = tokenizer.encode(sequence_0)
-            encoded_sequence_dict = tokenizer.encode_plus(sequence_0, add_special_tokens=True)
+            encoded_sequence = tokenizer.encode(sequence_0, add_special_tokens=False)
+            encoded_sequence_dict = tokenizer.encode_plus(sequence_0, add_special_tokens=True, return_special_tokens_mask=True)
             encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
             special_tokens_mask = encoded_sequence_dict["special_tokens_mask"]
             self.assertEqual(len(special_tokens_mask), len(encoded_sequence_w_special))
@@ -294,8 +355,10 @@ class CommonTestCases:
             self.assertEqual(encoded_sequence, filtered_sequence)
 
             # Testing inputs pairs
-            encoded_sequence = tokenizer.encode(sequence_0) + tokenizer.encode(sequence_1)
-            encoded_sequence_dict = tokenizer.encode_plus(sequence_0, sequence_1, add_special_tokens=True)
+            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,
+                                                          return_special_tokens_mask=True)
             encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
             special_tokens_mask = encoded_sequence_dict["special_tokens_mask"]
             self.assertEqual(len(special_tokens_mask), len(encoded_sequence_w_special))
@@ -307,9 +370,98 @@ class CommonTestCases:
             # Testing with already existing special tokens
             if tokenizer.cls_token_id == tokenizer.unk_token_id and tokenizer.cls_token_id == tokenizer.unk_token_id:
                 tokenizer.add_special_tokens({'cls_token': '</s>', 'sep_token': '<s>'})
-            encoded_sequence_dict = tokenizer.encode_plus(sequence_0, add_special_tokens=True)
+            encoded_sequence_dict = tokenizer.encode_plus(sequence_0,
+                                                          add_special_tokens=True,
+                                                          return_special_tokens_mask=True)
             encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
             special_tokens_mask_orig = encoded_sequence_dict["special_tokens_mask"]
             special_tokens_mask = tokenizer.get_special_tokens_mask(encoded_sequence_w_special, already_has_special_tokens=True)
             self.assertEqual(len(special_tokens_mask), len(encoded_sequence_w_special))
             self.assertEqual(special_tokens_mask_orig, special_tokens_mask)
+
+        def test_padding_to_max_length(self):
+            tokenizer = self.get_tokenizer()
+
+            sequence = "Sequence"
+            padding_size = 10
+            padding_idx = tokenizer.pad_token_id
+
+            # RIGHT PADDING - Check that it correctly pads when a maximum length is specified along with the padding flag set to True
+            tokenizer.padding_side = "right"
+            encoded_sequence = tokenizer.encode(sequence)
+            sequence_length = len(encoded_sequence)
+            padded_sequence = tokenizer.encode(sequence, max_length=sequence_length + padding_size, pad_to_max_length=True)
+            padded_sequence_length = len(padded_sequence)
+            assert sequence_length + padding_size == padded_sequence_length
+            assert encoded_sequence + [padding_idx] * padding_size == padded_sequence
+
+            # LEFT PADDING - Check that it correctly pads when a maximum length is specified along with the padding flag set to True
+            tokenizer.padding_side = "left"
+            encoded_sequence = tokenizer.encode(sequence)
+            sequence_length = len(encoded_sequence)
+            padded_sequence = tokenizer.encode(sequence, max_length=sequence_length + padding_size, pad_to_max_length=True)
+            padded_sequence_length = len(padded_sequence)
+            assert sequence_length + padding_size == padded_sequence_length
+            assert [padding_idx] * padding_size + encoded_sequence == padded_sequence
+
+            # RIGHT & LEFT PADDING - Check that nothing is done when a maximum length is not specified
+            encoded_sequence = tokenizer.encode(sequence)
+            sequence_length = len(encoded_sequence)
+
+            tokenizer.padding_side = "right"
+            padded_sequence_right = tokenizer.encode(sequence, pad_to_max_length=True)
+            padded_sequence_right_length = len(padded_sequence_right)
+
+            tokenizer.padding_side = "left"
+            padded_sequence_left = tokenizer.encode(sequence, pad_to_max_length=True)
+            padded_sequence_left_length = len(padded_sequence_left)
+
+            assert sequence_length == padded_sequence_right_length
+            assert encoded_sequence == padded_sequence_right
+            assert sequence_length == padded_sequence_left_length
+            assert encoded_sequence == padded_sequence_left
+
+        def test_encode_plus_with_padding(self):
+            tokenizer = self.get_tokenizer()
+
+            sequence = "Sequence"
+            padding_size = 10
+            padding_idx = tokenizer.pad_token_id
+            token_type_padding_idx = tokenizer.pad_token_type_id
+
+            encoded_sequence = tokenizer.encode_plus(sequence, return_special_tokens_mask=True)
+            input_ids = encoded_sequence['input_ids']
+            token_type_ids = encoded_sequence['token_type_ids']
+            attention_mask = encoded_sequence['attention_mask']
+            special_tokens_mask = encoded_sequence['special_tokens_mask']
+            sequence_length = len(input_ids)
+
+            # Test right padding
+            tokenizer.padding_side = "right"
+            padded_sequence = tokenizer.encode_plus(sequence, max_length=sequence_length + padding_size, pad_to_max_length=True, return_special_tokens_mask=True)
+            padded_input_ids = padded_sequence['input_ids']
+            padded_token_type_ids = padded_sequence['token_type_ids']
+            padded_attention_mask = padded_sequence['attention_mask']
+            padded_special_tokens_mask = padded_sequence['special_tokens_mask']
+            padded_sequence_length = len(padded_input_ids)
+
+            assert sequence_length + padding_size == padded_sequence_length
+            assert input_ids + [padding_idx] * padding_size == padded_input_ids
+            assert token_type_ids + [token_type_padding_idx] * padding_size == padded_token_type_ids
+            assert attention_mask + [0] * padding_size == padded_attention_mask 
+            assert special_tokens_mask + [1] * padding_size == padded_special_tokens_mask 
+
+            # Test left padding
+            tokenizer.padding_side = "left"
+            padded_sequence = tokenizer.encode_plus(sequence, max_length=sequence_length + padding_size, pad_to_max_length=True, return_special_tokens_mask=True)
+            padded_input_ids = padded_sequence['input_ids']
+            padded_token_type_ids = padded_sequence['token_type_ids']
+            padded_attention_mask = padded_sequence['attention_mask']
+            padded_special_tokens_mask = padded_sequence['special_tokens_mask']
+            padded_sequence_length = len(padded_input_ids)
+
+            assert sequence_length + padding_size == padded_sequence_length
+            assert [padding_idx] * padding_size + input_ids == padded_input_ids
+            assert [token_type_padding_idx] * padding_size + token_type_ids == padded_token_type_ids
+            assert [0] * padding_size + attention_mask == padded_attention_mask 
+            assert [1] * padding_size + special_tokens_mask == padded_special_tokens_mask 
\ No newline at end of file

transformers/tests/tokenization_transfo_xl_test.py

@@ -16,7 +16,6 @@ from __future__ import absolute_import, division, print_function, unicode_litera
 
 import os
 import unittest
-import pytest
 from io import open
 
 from transformers import is_torch_available
@@ -24,11 +23,12 @@ from transformers import is_torch_available
 if is_torch_available():
     import torch
     from transformers.tokenization_transfo_xl import TransfoXLTokenizer, VOCAB_FILES_NAMES
-else:
-    pytestmark = pytest.mark.skip("Require Torch")  # TODO: untangle Transfo-XL tokenizer from torch.load and torch.save
 
 from .tokenization_tests_commons import CommonTestCases
+from .utils import require_torch
 
+
+@require_torch
 class TransfoXLTokenizationTest(CommonTestCases.CommonTokenizerTester):
 
     tokenizer_class = TransfoXLTokenizer if is_torch_available() else None

transformers/tests/tokenization_utils_test.py

@@ -22,7 +22,10 @@ import six
 from transformers import PreTrainedTokenizer
 from transformers.tokenization_gpt2 import GPT2Tokenizer
 
+from .utils import slow
+
 class TokenizerUtilsTest(unittest.TestCase):
+
     def check_tokenizer_from_pretrained(self, tokenizer_class):
         s3_models = list(tokenizer_class.max_model_input_sizes.keys())
         for model_name in s3_models[:1]:
@@ -39,6 +42,7 @@ class TokenizerUtilsTest(unittest.TestCase):
                 special_tok_id = tokenizer.convert_tokens_to_ids(special_tok)
                 self.assertIsInstance(special_tok_id, int)
 
+    @slow
     def test_pretrained_tokenizers(self):
         self.check_tokenizer_from_pretrained(GPT2Tokenizer)
 

transformers/tests/tokenization_xlm_test.py

@@ -21,6 +21,7 @@ import json
 from transformers.tokenization_xlm import XLMTokenizer, VOCAB_FILES_NAMES
 
 from .tokenization_tests_commons import CommonTestCases
+from .utils import slow
 
 class XLMTokenizationTest(CommonTestCases.CommonTokenizerTester):
 
@@ -66,11 +67,12 @@ class XLMTokenizationTest(CommonTestCases.CommonTokenizerTester):
         self.assertListEqual(
             tokenizer.convert_tokens_to_ids(input_tokens), input_bpe_tokens)
 
+    @slow
     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

@@ -20,6 +20,7 @@ import unittest
 from transformers.tokenization_xlnet import (XLNetTokenizer, SPIECE_UNDERLINE)
 
 from .tokenization_tests_commons import CommonTestCases
+from .utils import slow
 
 SAMPLE_VOCAB = os.path.join(os.path.dirname(os.path.abspath(__file__)),
                     'fixtures/test_sentencepiece.model')
@@ -89,11 +90,12 @@ class XLNetTokenizationTest(CommonTestCases.CommonTokenizerTester):
                                       u'9', u'2', u'0', u'0', u'0', u',', SPIECE_UNDERLINE + u'and', SPIECE_UNDERLINE + u'this',
                                       SPIECE_UNDERLINE + u'is', SPIECE_UNDERLINE + u'f', u'al', u'se', u'.'])
 
+    @slow
     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/tests/utils.py

+import os
+import unittest
+
+from distutils.util import strtobool
+
+from transformers.file_utils import _tf_available, _torch_available
+
+
+SMALL_MODEL_IDENTIFIER = "julien-c/bert-xsmall-dummy"
+
+
+def parse_flag_from_env(key, default=False):
+    try:
+        value = os.environ[key]
+    except KeyError:
+        # KEY isn't set, default to `default`.
+        _value = default
+    else:
+        # KEY is set, convert it to True or False.
+        try:
+            _value = strtobool(value)
+        except ValueError:
+            # More values are supported, but let's keep the message simple.
+            raise ValueError("If set, {} must be yes or no.".format(key))
+    return _value
+
+_run_slow_tests = parse_flag_from_env("RUN_SLOW", default=False)
+_run_custom_tokenizers = parse_flag_from_env("RUN_CUSTOM_TOKENIZERS", default=False)
+
+
+def slow(test_case):
+    """
+    Decorator marking a test as slow.
+
+    Slow tests are skipped by default. Set the RUN_SLOW environment variable
+    to a truthy value to run them.
+
+    """
+    if not _run_slow_tests:
+        test_case = unittest.skip("test is slow")(test_case)
+    return test_case
+
+
+def custom_tokenizers(test_case):
+    """
+    Decorator marking a test for a custom tokenizer.
+
+    Custom tokenizers require additional dependencies, and are skipped
+    by default. Set the RUN_CUSTOM_TOKENIZERS environment variable
+    to a truthy value to run them.
+    """
+    if not _run_custom_tokenizers:
+        test_case = unittest.skip("test of custom tokenizers")(test_case)
+    return test_case
+
+
+def require_torch(test_case):
+    """
+    Decorator marking a test that requires PyTorch.
+
+    These tests are skipped when PyTorch isn't installed.
+
+    """
+    if not _torch_available:
+        test_case = unittest.skip("test requires PyTorch")(test_case)
+    return test_case
+
+
+def require_tf(test_case):
+    """
+    Decorator marking a test that requires TensorFlow.
+
+    These tests are skipped when TensorFlow isn't installed.
+
+    """
+    if not _tf_available:
+        test_case = unittest.skip("test requires TensorFlow")(test_case)
+    return test_case
+
+
+if _torch_available:
+    # Set the USE_CUDA environment variable to select a GPU.
+    torch_device = "cuda" if parse_flag_from_env("USE_CUDA") else "cpu"
+else:
+    torch_device = None

transformers/tokenization_albert.py

+# coding=utf-8
+# Copyright 2018 Google AI, Google Brain and 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 ALBERT model."""
+from __future__ import (absolute_import, division, print_function,
+                        unicode_literals)
+
+from .tokenization_utils import PreTrainedTokenizer
+import logging
+import unicodedata
+import six
+import os
+from shutil import copyfile
+
+logger = logging.getLogger(__name__)
+VOCAB_FILES_NAMES = {'vocab_file': 'spiece.model'}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    'vocab_file':
+    {
+        'albert-base-v1': "https://s3.amazonaws.com/models.huggingface.co/bert/albert-base-spiece.model",
+        'albert-large-v1': "https://s3.amazonaws.com/models.huggingface.co/bert/albert-large-spiece.model",
+        'albert-xlarge-v1': "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xlarge-spiece.model",
+        'albert-xxlarge-v1': "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xxlarge-spiece.model",
+        'albert-base-v2': "https://s3.amazonaws.com/models.huggingface.co/bert/albert-base-v2-spiece.model",
+        'albert-large-v2': "https://s3.amazonaws.com/models.huggingface.co/bert/albert-large-v2-spiece.model",
+        'albert-xlarge-v2': "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xlarge-v2-spiece.model",
+        'albert-xxlarge-v2': "https://s3.amazonaws.com/models.huggingface.co/bert/albert-xxlarge-v2-spiece.model",
+    }
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    'albert-base-v1': 512,
+    'albert-large-v1': 512,
+    'albert-xlarge-v1': 512,
+    'albert-xxlarge-v1': 512,
+    'albert-base-v2': 512,
+    'albert-large-v2': 512,
+    'albert-xlarge-v2': 512,
+    'albert-xxlarge-v2': 512,
+}
+
+SPIECE_UNDERLINE = u'▁'
+
+class AlbertTokenizer(PreTrainedTokenizer):
+    """
+        SentencePiece based tokenizer. Peculiarities:
+
+            - requires `SentencePiece <https://github.com/google/sentencepiece>`_
+    """
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+
+    def __init__(self, vocab_file,
+                 do_lower_case=True, remove_space=True, keep_accents=False,
+                 bos_token="[CLS]", eos_token="[SEP]", unk_token="<unk>", sep_token="[SEP]",
+                 pad_token="<pad>", cls_token="[CLS]", mask_token="[MASK]", **kwargs):
+        super(AlbertTokenizer, self).__init__(bos_token=bos_token, eos_token=eos_token,
+                                             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
+
+        try:
+            import sentencepiece as spm
+        except ImportError:
+            logger.warning("You need to install SentencePiece to use AlbertTokenizer: https://github.com/google/sentencepiece"
+                           "pip install sentencepiece")
+
+        self.do_lower_case = do_lower_case
+        self.remove_space = remove_space
+        self.keep_accents = keep_accents
+        self.vocab_file = vocab_file
+
+        self.sp_model = spm.SentencePieceProcessor()
+        self.sp_model.Load(vocab_file)
+
+    @property
+    def vocab_size(self):
+        return len(self.sp_model)
+
+    def __getstate__(self):
+        state = self.__dict__.copy()
+        state["sp_model"] = None
+        return state
+
+    def __setstate__(self, d):
+        self.__dict__ = d
+        try:
+            import sentencepiece as spm
+        except ImportError:
+            logger.warning("You need to install SentencePiece to use AlbertTokenizer: https://github.com/google/sentencepiece"
+                           "pip install sentencepiece")
+        self.sp_model = spm.SentencePieceProcessor()
+        self.sp_model.Load(self.vocab_file)
+
+    def preprocess_text(self, inputs):
+        if self.remove_space:
+            outputs = ' '.join(inputs.strip().split())
+        else:
+            outputs = inputs
+        outputs = outputs.replace("``", '"').replace("''", '"')
+
+        if six.PY2 and isinstance(outputs, str):
+            outputs = outputs.decode('utf-8')
+
+        if not self.keep_accents:
+            outputs = unicodedata.normalize('NFKD', outputs)
+            outputs = ''.join([c for c in outputs if not unicodedata.combining(c)])
+        if self.do_lower_case:
+            outputs = outputs.lower()
+
+        return outputs
+
+    def _tokenize(self, text, return_unicode=True, sample=False):
+        """ Tokenize a string.
+            return_unicode is used only for py2
+        """
+        text = self.preprocess_text(text)
+        # note(zhiliny): in some systems, sentencepiece only accepts str for py2
+        if six.PY2 and isinstance(text, unicode):
+            text = text.encode('utf-8')
+
+        if not sample:
+            pieces = self.sp_model.EncodeAsPieces(text)
+        else:
+            pieces = self.sp_model.SampleEncodeAsPieces(text, 64, 0.1)
+        new_pieces = []
+        for piece in pieces:
+            if len(piece) > 1 and piece[-1] == str(',') and piece[-2].isdigit():
+                cur_pieces = self.sp_model.EncodeAsPieces(
+                    piece[:-1].replace(SPIECE_UNDERLINE, ''))
+                if piece[0] != SPIECE_UNDERLINE and cur_pieces[0][0] == SPIECE_UNDERLINE:
+                    if len(cur_pieces[0]) == 1:
+                        cur_pieces = cur_pieces[1:]
+                    else:
+                        cur_pieces[0] = cur_pieces[0][1:]
+                cur_pieces.append(piece[-1])
+                new_pieces.extend(cur_pieces)
+            else:
+                new_pieces.append(piece)
+
+        # note(zhiliny): convert back to unicode for py2
+        if six.PY2 and return_unicode:
+            ret_pieces = []
+            for piece in new_pieces:
+                if isinstance(piece, str):
+                    piece = piece.decode('utf-8')
+                ret_pieces.append(piece)
+            new_pieces = ret_pieces
+
+        return new_pieces
+
+    def _convert_token_to_id(self, token):
+        """ Converts a token (str/unicode) in an id using the vocab. """
+        return self.sp_model.PieceToId(token)
+
+    def _convert_id_to_token(self, index, return_unicode=True):
+        """Converts an index (integer) in a token (string/unicode) using the vocab."""
+        token = self.sp_model.IdToPiece(index)
+        if six.PY2 and return_unicode and isinstance(token, str):
+            token = token.decode('utf-8')
+        return token
+
+    def convert_tokens_to_string(self, tokens):
+        """Converts a sequence of tokens (strings for sub-words) in a single string."""
+        out_string = ''.join(tokens).replace(SPIECE_UNDERLINE, ' ').strip()
+        return out_string
+
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks
+        by concatenating and adding special tokens.
+        An ALBERT sequence has the following format:
+            single sequence: [CLS] X [SEP]
+            pair of sequences: [CLS] A [SEP] B [SEP]
+        """
+        sep = [self.sep_token_id]
+        cls = [self.cls_token_id]
+        if token_ids_1 is None:
+            return cls + token_ids_0 + sep
+        return cls + token_ids_0 + sep + token_ids_1 + sep
+
+    def get_special_tokens_mask(self, token_ids_0, token_ids_1=None, already_has_special_tokens=False):
+        """
+        Retrieves sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer ``prepare_for_model`` or ``encode_plus`` methods.
+
+        Args:
+            token_ids_0: list of ids (must not contain special tokens)
+            token_ids_1: Optional list of ids (must not contain special tokens), necessary when fetching sequence ids
+                for sequence pairs
+            already_has_special_tokens: (default False) Set to True if the token list is already formated with
+                special tokens for the model
+
+        Returns:
+            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
+        """
+
+        if already_has_special_tokens:
+            if token_ids_1 is not None:
+                raise ValueError("You should not supply a second sequence if the provided sequence of "
+                                 "ids is already formated with special tokens for the model.")
+            return list(map(lambda x: 1 if x in [self.sep_token_id, self.cls_token_id] else 0, token_ids_0))
+
+        if token_ids_1 is not None:
+            return [1] + ([0] * len(token_ids_0)) + [1] + ([0] * len(token_ids_1)) + [1]
+        return [1] + ([0] * len(token_ids_0)) + [1]
+
+    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
+        """
+        Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
+        An ALBERT sequence pair mask has the following format:
+        0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1
+        | first sequence    | second sequence
+
+        if token_ids_1 is None, only returns the first portion of the mask (0's).
+        """
+        sep = [self.sep_token_id]
+        cls = [self.cls_token_id]
+
+        if token_ids_1 is None:
+            return len(cls + token_ids_0 + sep) * [0]
+        return len(cls + token_ids_0 + sep) * [0] + len(token_ids_1 + sep) * [1]
+
+    def save_vocabulary(self, save_directory):
+        """ Save the sentencepiece vocabulary (copy original file) and special tokens file
+            to a directory.
+        """
+        if not os.path.isdir(save_directory):
+            logger.error("Vocabulary path ({}) should be a directory".format(save_directory))
+            return
+        out_vocab_file = os.path.join(save_directory, VOCAB_FILES_NAMES['vocab_file'])
+
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
+            copyfile(self.vocab_file, out_vocab_file)
+
+        return (out_vocab_file,)

transformers/tokenization_auto.py

@@ -19,6 +19,7 @@ from __future__ import absolute_import, division, print_function, unicode_litera
 import logging
 
 from .tokenization_bert import BertTokenizer
+from .tokenization_bert_japanese import BertJapaneseTokenizer
 from .tokenization_openai import OpenAIGPTTokenizer
 from .tokenization_gpt2 import GPT2Tokenizer
 from .tokenization_ctrl import CTRLTokenizer
@@ -27,6 +28,8 @@ from .tokenization_xlnet import XLNetTokenizer
 from .tokenization_xlm import XLMTokenizer
 from .tokenization_roberta import RobertaTokenizer
 from .tokenization_distilbert import DistilBertTokenizer
+from .tokenization_camembert import CamembertTokenizer
+from .tokenization_albert import AlbertTokenizer
 
 logger = logging.getLogger(__name__)
 
@@ -42,14 +45,16 @@ class AutoTokenizer(object):
         The tokenizer class to instantiate is selected as the first pattern matching
         in the `pretrained_model_name_or_path` string (in the following order):
             - contains `distilbert`: DistilBertTokenizer (DistilBert model)
+            - contains `albert`: AlbertTokenizer (ALBERT model)
+            - contains `camembert`: CamembertTokenizer (CamemBERT model)
             - contains `roberta`: RobertaTokenizer (RoBERTa model)
             - contains `bert`: BertTokenizer (Bert model)
             - contains `openai-gpt`: OpenAIGPTTokenizer (OpenAI GPT model)
             - contains `gpt2`: GPT2Tokenizer (OpenAI GPT-2 model)
-            - contains `ctrl`: CTRLTokenizer (Salesforce CTRL model)
             - contains `transfo-xl`: TransfoXLTokenizer (Transformer-XL model)
             - contains `xlnet`: XLNetTokenizer (XLNet model)
             - contains `xlm`: XLMTokenizer (XLM model)
+            - contains `ctrl`: CTRLTokenizer (Salesforce CTRL model)
 
         This class cannot be instantiated using `__init__()` (throw an error).
     """
@@ -65,19 +70,23 @@ class AutoTokenizer(object):
         The tokenizer class to instantiate is selected as the first pattern matching
         in the `pretrained_model_name_or_path` string (in the following order):
             - contains `distilbert`: DistilBertTokenizer (DistilBert model)
-            - contains `roberta`: RobertaTokenizer (XLM model)
+            - contains `albert`: AlbertTokenizer (ALBERT model)
+            - contains `camembert`: CamembertTokenizer (CamemBERT model)
+            - contains `roberta`: RobertaTokenizer (RoBERTa model)
+            - contains `bert-base-japanese`: BertJapaneseTokenizer (Bert model)
             - contains `bert`: BertTokenizer (Bert model)
             - contains `openai-gpt`: OpenAIGPTTokenizer (OpenAI GPT model)
             - contains `gpt2`: GPT2Tokenizer (OpenAI GPT-2 model)
-            - contains `ctrl`: CTRLTokenizer (Salesforce CTRL model)
             - contains `transfo-xl`: TransfoXLTokenizer (Transformer-XL model)
             - contains `xlnet`: XLNetTokenizer (XLNet model)
             - contains `xlm`: XLMTokenizer (XLM model)
+            - contains `ctrl`: CTRLTokenizer (Salesforce CTRL model)
 
         Params:
             pretrained_model_name_or_path: either:
 
                 - a string with the `shortcut name` of a predefined tokenizer to load from cache or download, e.g.: ``bert-base-uncased``.
+                - a string with the `identifier name` of a predefined tokenizer that was user-uploaded to our S3, e.g.: ``dbmdz/bert-base-german-cased``.
                 - a path to a `directory` containing vocabulary files required by the tokenizer, for instance saved using the :func:`~transformers.PreTrainedTokenizer.save_pretrained` method, e.g.: ``./my_model_directory/``.
                 - (not applicable to all derived classes) a path or url to a single saved vocabulary file if and only if the tokenizer only requires a single vocabulary file (e.g. Bert, XLNet), e.g.: ``./my_model_directory/vocab.txt``.
 
@@ -87,6 +96,9 @@ class AutoTokenizer(object):
             force_download: (`optional`) boolean, default False:
                 Force to (re-)download the vocabulary files and override the cached versions if they exists.
 
+            resume_download: (`optional`) boolean, default False:
+                Do not delete incompletely recieved file. Attempt to resume the download if such a file 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.
@@ -97,14 +109,26 @@ class AutoTokenizer(object):
 
         Examples::
 
-            tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')    # Download vocabulary from S3 and cache.
-            tokenizer = AutoTokenizer.from_pretrained('./test/bert_saved_model/')  # E.g. tokenizer was saved using `save_pretrained('./test/saved_model/')`
+            # Download vocabulary from S3 and cache.
+            tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
+
+            # Download vocabulary from S3 (user-uploaded) and cache.
+            tokenizer = AutoTokenizer.from_pretrained('dbmdz/bert-base-german-cased')
+
+            # If vocabulary files are in a directory (e.g. tokenizer was saved using `save_pretrained('./test/saved_model/')`)
+            tokenizer = AutoTokenizer.from_pretrained('./test/bert_saved_model/')
 
         """
         if 'distilbert' in pretrained_model_name_or_path:
             return DistilBertTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
+        elif 'albert' in pretrained_model_name_or_path:
+            return AlbertTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
+        elif 'camembert' in pretrained_model_name_or_path:
+            return CamembertTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
         elif 'roberta' in pretrained_model_name_or_path:
             return RobertaTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
+        elif 'bert-base-japanese' in pretrained_model_name_or_path:
+            return BertJapaneseTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
         elif 'bert' in pretrained_model_name_or_path:
             return BertTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
         elif 'openai-gpt' in pretrained_model_name_or_path:
@@ -121,4 +145,4 @@ class AutoTokenizer(object):
             return CTRLTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)
         raise ValueError("Unrecognized model identifier in {}. Should contains one of "
                          "'bert', 'openai-gpt', 'gpt2', 'transfo-xl', 'xlnet', "
-                         "'xlm', 'roberta', 'ctrl'".format(pretrained_model_name_or_path))
+                         "'xlm', 'roberta', 'distilbert,' 'camembert', 'ctrl', 'albert'".format(pretrained_model_name_or_path))

transformers/tokenization_bert.py

@@ -220,7 +220,7 @@ class BertTokenizer(PreTrainedTokenizer):
                 special tokens for the model
 
         Returns:
-            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
+            A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
         """
 
         if already_has_special_tokens:

transformers/tokenization_bert_japanese.py

+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors and 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."""
+
+from __future__ import absolute_import, division, print_function, unicode_literals
+
+import collections
+import logging
+import os
+import six
+import unicodedata
+from io import open
+
+from .tokenization_bert import BertTokenizer, BasicTokenizer, WordpieceTokenizer, load_vocab
+from .tokenization_utils import PreTrainedTokenizer
+
+logger = logging.getLogger(__name__)
+
+VOCAB_FILES_NAMES = {'vocab_file': 'vocab.txt'}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    'vocab_file':
+    {
+        'bert-base-japanese': "https://s3.amazonaws.com/models.huggingface.co/bert/cl-tohoku/bert-base-japanese-vocab.txt",
+        'bert-base-japanese-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/cl-tohoku/bert-base-japanese-whole-word-masking-vocab.txt",
+        'bert-base-japanese-char': "https://s3.amazonaws.com/models.huggingface.co/bert/cl-tohoku/bert-base-japanese-char-vocab.txt",
+        'bert-base-japanese-char-whole-word-masking': "https://s3.amazonaws.com/models.huggingface.co/bert/cl-tohoku/bert-base-japanese-char-whole-word-masking-vocab.txt"
+    }
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    'bert-base-japanese': 512,
+    'bert-base-japanese-whole-word-masking': 512,
+    'bert-base-japanese-char': 512,
+    'bert-base-japanese-char-whole-word-masking': 512
+}
+
+PRETRAINED_INIT_CONFIGURATION = {
+    'bert-base-japanese': {
+        'do_lower_case': False,
+        'word_tokenizer_type': 'mecab',
+        'subword_tokenizer_type': 'wordpiece'
+    },
+    'bert-base-japanese-whole-word-masking':{
+        'do_lower_case': False,
+        'word_tokenizer_type': 'mecab',
+        'subword_tokenizer_type': 'wordpiece'
+    },
+    'bert-base-japanese-char': {
+        'do_lower_case': False,
+        'word_tokenizer_type': 'mecab',
+        'subword_tokenizer_type': 'character'
+    },
+    'bert-base-japanese-char-whole-word-masking': {
+        'do_lower_case': False,
+        'word_tokenizer_type': 'mecab',
+        'subword_tokenizer_type': 'character'
+    }
+}
+
+
+class BertJapaneseTokenizer(BertTokenizer):
+    """BERT tokenizer for Japanese text"""
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+
+    def __init__(self, vocab_file, do_lower_case=False,
+                 do_word_tokenize=True, do_subword_tokenize=True,
+                 word_tokenizer_type='basic', subword_tokenizer_type='wordpiece',
+                 never_split=None, unk_token='[UNK]', sep_token='[SEP]',
+                 pad_token='[PAD]', cls_token='[CLS]', mask_token='[MASK]', **kwargs):
+        """Constructs a MecabBertTokenizer.
+
+        Args:
+            **vocab_file**: Path to a one-wordpiece-per-line vocabulary file.
+            **do_lower_case**: (`optional`) boolean (default True)
+                Whether to lower case the input.
+                Only has an effect when do_basic_tokenize=True.
+            **do_word_tokenize**: (`optional`) boolean (default True)
+                Whether to do word tokenization.
+            **do_subword_tokenize**: (`optional`) boolean (default True)
+                Whether to do subword tokenization.
+            **word_tokenizer_type**: (`optional`) string (default "basic")
+                Type of word tokenizer.
+            **subword_tokenizer_type**: (`optional`) string (default "wordpiece")
+                Type of subword tokenizer.
+        """
+        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 "
+                "model use `tokenizer = BertTokenizer.from_pretrained(PRETRAINED_MODEL_NAME)`".format(vocab_file))
+        self.vocab = load_vocab(vocab_file)
+        self.ids_to_tokens = collections.OrderedDict(
+            [(ids, tok) for tok, ids in self.vocab.items()])
+
+        self.do_word_tokenize = do_word_tokenize
+        if do_word_tokenize:
+            if word_tokenizer_type == 'basic':
+                self.word_tokenizer = BasicTokenizer(do_lower_case=do_lower_case,
+                                                     never_split=never_split,
+                                                     tokenize_chinese_chars=False)
+            elif word_tokenizer_type == 'mecab':
+                self.word_tokenizer = MecabTokenizer(do_lower_case=do_lower_case,
+                                                     never_split=never_split)
+            else:
+                raise ValueError(
+                    "Invalid word_tokenizer_type '{}' is specified.".format(word_tokenizer_type))
+
+        self.do_subword_tokenize = do_subword_tokenize
+        if do_subword_tokenize:
+            if subword_tokenizer_type == 'wordpiece':
+                self.subword_tokenizer = WordpieceTokenizer(vocab=self.vocab,
+                                                            unk_token=self.unk_token)
+            elif subword_tokenizer_type == 'character':
+                self.subword_tokenizer = CharacterTokenizer(vocab=self.vocab,
+                                                            unk_token=self.unk_token)
+            else:
+                raise ValueError(
+                    "Invalid subword_tokenizer_type '{}' is specified.".format(subword_tokenizer_type))
+
+
+    def _tokenize(self, text):
+        if self.do_word_tokenize:
+            tokens = self.word_tokenizer.tokenize(text,
+                                                  never_split=self.all_special_tokens)
+        else:
+            tokens = [text]
+
+        if self.do_subword_tokenize:
+            split_tokens = [sub_token for token in tokens
+                            for sub_token in self.subword_tokenizer.tokenize(token)]
+        else:
+            split_tokens = tokens
+
+        return split_tokens
+
+
+class MecabTokenizer(object):
+    """Runs basic tokenization with MeCab morphological parser."""
+
+    def __init__(self, do_lower_case=False, never_split=None, normalize_text=True):
+        """Constructs a MecabTokenizer.
+
+        Args:
+            **do_lower_case**: (`optional`) boolean (default True)
+                Whether to lower case the input.
+            **never_split**: (`optional`) list of str
+                Kept for backward compatibility purposes.
+                Now implemented directly at the base class level (see :func:`PreTrainedTokenizer.tokenize`)
+                List of token not to split.
+            **normalize_text**: (`optional`) boolean (default True)
+                Whether to apply unicode normalization to text before tokenization.
+        """
+        self.do_lower_case = do_lower_case
+        self.never_split = never_split if never_split is not None else []
+        self.normalize_text = normalize_text
+
+        import MeCab
+        self.mecab = MeCab.Tagger()
+
+    def tokenize(self, text, never_split=None, **kwargs):
+        """Tokenizes a piece of text."""
+        if self.normalize_text:
+            text = unicodedata.normalize('NFKC', text)
+
+        never_split = self.never_split + (never_split if never_split is not None else [])
+        tokens = []
+
+        if six.PY2:
+            mecab_output = self.mecab.parse(text.encode('utf-8')).decode('utf-8')
+        else:
+            mecab_output = self.mecab.parse(text)
+
+        cursor = 0
+        for line in mecab_output.split('\n'):
+            if line == 'EOS':
+                break
+
+            token, _ = line.split('\t')
+            token_start = text.index(token, cursor)
+            token_end = token_start + len(token)
+            if self.do_lower_case and token not in never_split:
+                token = token.lower()
+
+            tokens.append(token)
+            cursor = token_end
+
+        return tokens
+
+
+class CharacterTokenizer(object):
+    """Runs Character tokenziation."""
+
+    def __init__(self, vocab, unk_token, normalize_text=True):
+        """Constructs a CharacterTokenizer.
+
+        Args:
+            **vocab**:
+                Vocabulary object.
+            **unk_token**: str
+                A special symbol for out-of-vocabulary token.
+            **normalize_text**: (`optional`) boolean (default True)
+                Whether to apply unicode normalization to text before tokenization.
+        """
+        self.vocab = vocab
+        self.unk_token = unk_token
+        self.normalize_text = normalize_text
+
+    def tokenize(self, text):
+        """Tokenizes a piece of text into characters.
+
+        For example:
+            input = "apple"
+            output = ["a", "p", "p", "l", "e"]
+        Args:
+            text: A single token or whitespace separated tokens.
+                This should have already been passed through `BasicTokenizer`.
+        Returns:
+            A list of characters.
+        """
+        if self.normalize_text:
+            text = unicodedata.normalize('NFKC', text)
+
+        output_tokens = []
+        for i, char in enumerate(text):
+            if char not in self.vocab:
+                output_tokens.append(self.unk_token)
+                continue
+
+            output_tokens.append(char)
+
+        return output_tokens

transformers/tokenization_camembert.py

+# coding=utf-8
+# Copyright 2018 Google AI, Google Brain and Carnegie Mellon University Authors and 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 Camembert model."""
+from __future__ import (absolute_import, division, print_function,
+                        unicode_literals)
+
+import logging
+import os
+from shutil import copyfile
+
+import sentencepiece as spm
+from transformers.tokenization_utils import PreTrainedTokenizer
+
+logger = logging.getLogger(__name__)
+
+VOCAB_FILES_NAMES = {'vocab_file': 'sentencepiece.bpe.model'}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    'vocab_file':
+    {
+    'camembert-base': "https://s3.amazonaws.com/models.huggingface.co/bert/camembert-base-sentencepiece.bpe.model",
+    }
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    'camembert-base': None,
+}
+
+class CamembertTokenizer(PreTrainedTokenizer):
+    """
+        Adapted from RobertaTokenizer and XLNetTokenizer
+        SentencePiece based tokenizer. Peculiarities:
+
+            - requires `SentencePiece <https://github.com/google/sentencepiece>`_
+    """
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+
+    def __init__(self, vocab_file, bos_token="<s>", eos_token="</s>", sep_token="</s>",
+                 cls_token="<s>", unk_token="<unk>", pad_token='<pad>', mask_token='<mask>',
+                 additional_special_tokens=['<s>NOTUSED', '</s>NOTUSED'], **kwargs):
+        super(CamembertTokenizer, self).__init__(max_len=512, bos_token=bos_token, eos_token=eos_token, unk_token=unk_token,
+                                                 sep_token=sep_token, cls_token=cls_token, pad_token=pad_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 - 4  # take into account special tokens
+        self.sp_model = spm.SentencePieceProcessor()
+        self.sp_model.Load(str(vocab_file))
+        self.vocab_file = vocab_file
+        # HACK: These tokens were added by fairseq but don't seem to be actually used when duplicated in the actual
+        # sentencepiece vocabulary (this is the case for <s> and </s>
+        self.fairseq_tokens_to_ids = {'<s>NOTUSED': 0, '<pad>': 1, '</s>NOTUSED': 2, '<unk>': 3}
+        self.fairseq_offset = len(self.fairseq_tokens_to_ids)
+        self.fairseq_tokens_to_ids['<mask>'] = len(self.sp_model) + len(self.fairseq_tokens_to_ids)
+        self.fairseq_ids_to_tokens = {v: k for k, v in self.fairseq_tokens_to_ids.items()}
+
+    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks
+        by concatenating and adding special tokens.
+        A RoBERTa sequence has the following format:
+            single sequence: <s> X </s>
+            pair of sequences: <s> A </s></s> B </s>
+        """
+        if token_ids_1 is None:
+            return [self.cls_token_id] + token_ids_0 + [self.sep_token_id]
+        cls = [self.cls_token_id]
+        sep = [self.sep_token_id]
+        return cls + token_ids_0 + sep + sep + token_ids_1 + sep
+
+    def get_special_tokens_mask(self, token_ids_0, token_ids_1=None, already_has_special_tokens=False):
+        """
+        Retrieves sequence ids from a token list that has no special tokens added. This method is called when adding
+        special tokens using the tokenizer ``prepare_for_model`` or ``encode_plus`` methods.
+
+        Args:
+            token_ids_0: list of ids (must not contain special tokens)
+            token_ids_1: Optional list of ids (must not contain special tokens), necessary when fetching sequence ids
+                for sequence pairs
+            already_has_special_tokens: (default False) Set to True if the token list is already formated with
+                special tokens for the model
+
+        Returns:
+            A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
+        """
+        if already_has_special_tokens:
+            if token_ids_1 is not None:
+                raise ValueError("You should not supply a second sequence if the provided sequence of "
+                                 "ids is already formated with special tokens for the model.")
+            return list(map(lambda x: 1 if x in [self.sep_token_id, self.cls_token_id] else 0, token_ids_0))
+
+        if token_ids_1 is None:
+            return [1] + ([0] * len(token_ids_0)) + [1]
+        return [1] + ([0] * len(token_ids_0)) + [1, 1] + ([0] * len(token_ids_1)) + [1]
+
+    def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
+        """
+        Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
+        A RoBERTa sequence pair mask has the following format:
+        0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1
+        | first sequence    | second sequence
+
+        if token_ids_1 is None, only returns the first portion of the mask (0's).
+        """
+        sep = [self.sep_token_id]
+        cls = [self.cls_token_id]
+
+        if token_ids_1 is None:
+            return len(cls + token_ids_0 + sep) * [0]
+        return len(cls + token_ids_0 + sep + sep) * [0] + len(token_ids_1 + sep) * [1]
+
+    @property
+    def vocab_size(self):
+        return len(self.fairseq_tokens_to_ids) + len(self.sp_model)
+
+    def _tokenize(self, text):
+        return self.sp_model.EncodeAsPieces(text)
+
+    def _convert_token_to_id(self, token):
+        """ Converts a token (str/unicode) in an id using the vocab. """
+        if token in self.fairseq_tokens_to_ids:
+            return self.fairseq_tokens_to_ids[token]
+        elif self.sp_model.PieceToId(token) == 0:
+            # Convert sentence piece unk token to fairseq unk token index
+            return self.unk_token_id
+        return self.fairseq_offset + self.sp_model.PieceToId(token)
+
+    def _convert_id_to_token(self, index):
+        """Converts an index (integer) in a token (string/unicode) using the vocab."""
+        if index in self.fairseq_ids_to_tokens:
+            return self.fairseq_ids_to_tokens[index]
+        return self.sp_model.IdToPiece(index - self.fairseq_offset)
+
+    def save_vocabulary(self, save_directory):
+        """ Save the sentencepiece vocabulary (copy original file) and special tokens file
+            to a directory.
+        """
+        if not os.path.isdir(save_directory):
+            logger.error("Vocabulary path ({}) should be a directory".format(save_directory))
+            return
+        out_vocab_file = os.path.join(save_directory, VOCAB_FILES_NAMES['vocab_file'])
+
+        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
+            copyfile(self.vocab_file, out_vocab_file)
+
+        return (out_vocab_file,)

transformers/tokenization_ctrl.py

@@ -46,6 +46,64 @@ PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     'ctrl': 256,
 }
 
+CONTROL_CODES = {
+    "Pregnancy": 168629,
+    "Christianity": 7675,
+    "Explain": 106423,
+    "Fitness": 63440,
+    "Saving": 63163,
+    "Ask": 27171,
+    "Ass": 95985,
+    "Joke": 163509,
+    "Questions": 45622,
+    "Thoughts": 49605,
+    "Retail": 52342,
+    "Feminism": 164338,
+    "Writing": 11992,
+    "Atheism": 192263,
+    "Netflix": 48616,
+    "Computing": 39639,
+    "Opinion": 43213,
+    "Alone": 44967,
+    "Funny": 58917,
+    "Gaming": 40358,
+    "Human": 4088,
+    "India": 1331,
+    "Joker": 77138,
+    "Diet": 36206,
+    "Legal": 11859,
+    "Norman": 4939,
+    "Tip": 72689,
+    "Weight": 52343,
+    "Movies": 46273,
+    "Running": 23425,
+    "Science": 2090,
+    "Horror": 37793,
+    "Confession": 60572,
+    "Finance": 12250,
+    "Politics": 16360,
+    "Scary": 191985,
+    "Support": 12654,
+    "Technologies": 32516,
+    "Teenage": 66160,
+    "Event": 32769,
+    "Learned": 67460,
+    "Notion": 182770,
+    "Wikipedia": 37583,
+    "Books": 6665,
+    "Extract": 76050,
+    "Confessions": 102701,
+    "Conspiracy": 75932,
+    "Links": 63674,
+    "Narcissus": 150425,
+    "Relationship": 54766,
+    "Relationships": 134796,
+    "Reviews": 41671,
+    "News": 4256,
+    "Translation": 26820,
+    "multilingual": 128406,
+}
+
 def get_pairs(word):
     """Return set of symbol pairs in a word.
 
@@ -63,24 +121,23 @@ def get_pairs(word):
 class CTRLTokenizer(PreTrainedTokenizer):
     """
     CTRL BPE tokenizer. Peculiarities:
-        - Byte-level Byte-Pair-Encoding
-        - Requires a space to start the input string => the encoding methods should be called with the
-          ``add_prefix_space`` flag set to ``True``.
-          Otherwise, this tokenizer ``encode`` and ``decode`` method will not conserve
-          the absence of a space at the beginning of a string: `tokenizer.decode(tokenizer.encode("Hello")) = " Hello"`
+        - Byte-Pair-Encoding
     """
     vocab_files_names = VOCAB_FILES_NAMES
     pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
     max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    control_codes = CONTROL_CODES
 
     def __init__(self, vocab_file, merges_file, unk_token="<unk>", **kwargs):
         super(CTRLTokenizer, 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
 
-        self.encoder = json.load(open(vocab_file, encoding="utf-8"))
+        with open(vocab_file, encoding="utf-8") as vocab_handle:
+            self.encoder = json.load(vocab_handle)
         self.decoder = {v:k for k,v in self.encoder.items()}
-        merges = open(merges_file, encoding='utf-8').read().split('\n')[1:-1]
+        with open(merges_file, encoding='utf-8') as merges_handle:
+            merges = merges_handle.read().split('\n')[1:-1]
         merges = [tuple(merge.split()) for merge in merges]
         self.bpe_ranks = dict(zip(merges, range(len(merges))))
         self.cache = {}
@@ -137,9 +194,9 @@ class CTRLTokenizer(PreTrainedTokenizer):
         """
         split_tokens = []
 
-        text = text.split(' ')
+        words = re.findall(r'\S+\n?', text)
 
-        for token in text:
+        for token in words:
             split_tokens.extend([t for t in self.bpe(token).split(' ')])
         return split_tokens
 

transformers/tokenization_distilbert.py

@@ -33,12 +33,16 @@ PRETRAINED_VOCAB_FILES_MAP = {
     {
         '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",
+        'distilbert-base-german-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/distilbert-base-german-cased-vocab.txt",
+        'distilbert-base-multilingual-cased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-multilingual-cased-vocab.txt",
     }
 }
 
 PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     'distilbert-base-uncased': 512,
     'distilbert-base-uncased-distilled-squad': 512,
+    'distilbert-base-german-cased': 512,
+    'distilbert-base-multilingual-cased': 512,
 }
 
 

transformers/tokenization_gpt2.py

@@ -46,6 +46,7 @@ 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",
+        'gpt2-xl': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-xl-vocab.json",
         'distilgpt2': "https://s3.amazonaws.com/models.huggingface.co/bert/distilgpt2-vocab.json",
     },
     'merges_file':
@@ -53,6 +54,7 @@ PRETRAINED_VOCAB_FILES_MAP = {
         '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",
+        'gpt2-xl': "https://s3.amazonaws.com/models.huggingface.co/bert/gpt2-xl-merges.txt",
         'distilgpt2': "https://s3.amazonaws.com/models.huggingface.co/bert/distilgpt2-merges.txt",
     },
 }
@@ -61,6 +63,7 @@ PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     'gpt2': 1024,
     'gpt2-medium': 1024,
     'gpt2-large': 1024,
+    'gpt2-xl': 1024,
     'distilgpt2': 1024,
 }
 
@@ -69,7 +72,7 @@ def bytes_to_unicode():
     """
     Returns list of utf-8 byte and a mapping to unicode strings.
     We specifically avoids mapping to whitespace/control characters the bpe code barfs on.
-    
+
     The reversible bpe codes work on unicode strings.
     This means you need a large # of unicode characters in your vocab if you want to avoid UNKs.
     When you're at something like a 10B token dataset you end up needing around 5K for decent coverage.
@@ -104,10 +107,10 @@ class GPT2Tokenizer(PreTrainedTokenizer):
     """
     GPT-2 BPE tokenizer. Peculiarities:
         - Byte-level Byte-Pair-Encoding
-        - Requires a space to start the input string => the encoding methods should be called with the
+        - Requires a space to start the input string => the encoding and tokenize methods should be called with the
           ``add_prefix_space`` flag set to ``True``.
-          Otherwise, this tokenizer ``encode`` and ``decode`` method will not conserve
-          the absence of a space at the beginning of a string: `tokenizer.decode(tokenizer.encode("Hello")) = " Hello"`
+          Otherwise, this tokenizer's ``encode``, ``decode``, and ``tokenize`` methods will not conserve
+          the spaces at the beginning of a string: `tokenizer.decode(tokenizer.encode(" Hello")) = "Hello"`
     """
     vocab_files_names = VOCAB_FILES_NAMES
     pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
@@ -119,13 +122,15 @@ class GPT2Tokenizer(PreTrainedTokenizer):
         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, encoding="utf-8"))
+        with open(vocab_file, encoding="utf-8") as vocab_handle:
+            self.encoder = json.load(vocab_handle)
         self.decoder = {v: k for k, v in self.encoder.items()}
         self.errors = errors  # how to handle errors in decoding
         self.byte_encoder = bytes_to_unicode()
         self.byte_decoder = {v: k for k, v in self.byte_encoder.items()}
-        bpe_data = open(merges_file, encoding='utf-8').read().split('\n')[1:-1]
-        bpe_merges = [tuple(merge.split()) for merge in bpe_data]
+        with open(merges_file, encoding='utf-8') as merges_handle:
+            bpe_merges = merges_handle.read().split('\n')[1:-1]
+        bpe_merges = [tuple(merge.split()) for merge in bpe_merges]
         self.bpe_ranks = dict(zip(bpe_merges, range(len(bpe_merges))))
         self.cache = {}
 
@@ -181,7 +186,7 @@ class GPT2Tokenizer(PreTrainedTokenizer):
         """ Tokenize a string.
             Args:
                 - add_prefix_space (boolean, default False):
-                    Begin the sentence with at least one space toto get invariance to word order in GPT-2 (and RoBERTa) tokenizers.
+                    Begin the sentence with at least one space to get invariance to word order in GPT-2 (and RoBERTa) tokenizers.
         """
         if add_prefix_space:
             text = ' ' + text
@@ -231,4 +236,4 @@ class GPT2Tokenizer(PreTrainedTokenizer):
                 writer.write(' '.join(bpe_tokens) + u'\n')
                 index += 1
 
-        return vocab_file, merge_file
+        return vocab_file, merge_file
\ No newline at end of file

transformers/tokenization_openai.py

@@ -101,9 +101,11 @@ class OpenAIGPTTokenizer(PreTrainedTokenizer):
             self.nlp = BasicTokenizer(do_lower_case=True)
             self.fix_text = None
 
-        self.encoder = json.load(open(vocab_file, encoding="utf-8"))
+        with open(vocab_file, encoding="utf-8") as vocab_handle:
+            self.encoder = json.load(vocab_handle)
         self.decoder = {v:k for k,v in self.encoder.items()}
-        merges = open(merges_file, encoding='utf-8').read().split('\n')[1:-1]
+        with open(merges_file, encoding='utf-8') as merges_handle:
+            merges = merges_handle.read().split('\n')[1:-1]
         merges = [tuple(merge.split()) for merge in merges]
         self.bpe_ranks = dict(zip(merges, range(len(merges))))
         self.cache = {}

transformers/tokenization_roberta.py

@@ -47,6 +47,8 @@ PRETRAINED_VOCAB_FILES_MAP = {
         'roberta-large': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-vocab.json",
         'roberta-large-mnli': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-mnli-vocab.json",
         'distilroberta-base': "https://s3.amazonaws.com/models.huggingface.co/bert/distilroberta-base-vocab.json",
+        'roberta-base-openai-detector': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-base-vocab.json",
+        'roberta-large-openai-detector': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-vocab.json",
     },
     'merges_file':
     {
@@ -54,6 +56,8 @@ PRETRAINED_VOCAB_FILES_MAP = {
         'roberta-large': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-merges.txt",
         'roberta-large-mnli': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-mnli-merges.txt",
         'distilroberta-base': "https://s3.amazonaws.com/models.huggingface.co/bert/distilroberta-base-merges.txt",
+        'roberta-base-openai-detector': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-base-merges.txt",
+        'roberta-large-openai-detector': "https://s3.amazonaws.com/models.huggingface.co/bert/roberta-large-merges.txt",
     },
 }
 
@@ -62,6 +66,8 @@ PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
     'roberta-large': 512,
     'roberta-large-mnli': 512,
     'distilroberta-base': 512,
+    'roberta-base-openai-detector': 512,
+    'roberta-large-openai-detector': 512,
 }
 
 
@@ -114,7 +120,7 @@ class RobertaTokenizer(GPT2Tokenizer):
                 special tokens for the model
 
         Returns:
-            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
+            A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
         """
         if already_has_special_tokens:
             if token_ids_1 is not None:

transformers/tokenization_utils.py

@@ -21,9 +21,11 @@ import os
 import json
 import six
 import copy
+import itertools
+import re
 from io import open
 
-from .file_utils import cached_path, is_tf_available, is_torch_available
+from .file_utils import cached_path, is_remote_url, hf_bucket_url, is_tf_available, is_torch_available
 
 if is_tf_available():
     import tensorflow as tf
@@ -76,6 +78,8 @@ class PreTrainedTokenizer(object):
                                  "pad_token", "cls_token", "mask_token",
                                  "additional_special_tokens"]
 
+    padding_side = "right"
+
     @property
     def bos_token(self):
         """ Beginning of sentence token (string). Log an error if used while not having been set. """
@@ -189,6 +193,11 @@ class PreTrainedTokenizer(object):
         """ Id of the padding token in the vocabulary. Log an error if used while not having been set. """
         return self.convert_tokens_to_ids(self.pad_token)
 
+    @property
+    def pad_token_type_id(self):
+        """ Id of the padding token type in the vocabulary."""
+        return self._pad_token_type_id
+
     @property
     def cls_token_id(self):
         """ Id of the classification token in the vocabulary. E.g. to extract a summary of an input sequence leveraging self-attention along the full depth of the model. Log an error if used while not having been set. """
@@ -212,10 +221,14 @@ class PreTrainedTokenizer(object):
         self._pad_token = None
         self._cls_token = None
         self._mask_token = None
+        self._pad_token_type_id = 0
         self._additional_special_tokens = []
 
         self.max_len = max_len if max_len is not None else int(1e12)
 
+        # Padding side is right by default and over-riden in subclasses. If specified in the kwargs, it is changed.
+        self.padding_side = kwargs.pop('padding_side', self.padding_side)
+        
         # Added tokens
         self.added_tokens_encoder = {}
         self.added_tokens_decoder = {}
@@ -242,6 +255,7 @@ class PreTrainedTokenizer(object):
             pretrained_model_name_or_path: either:
 
                 - a string with the `shortcut name` of a predefined tokenizer to load from cache or download, e.g.: ``bert-base-uncased``.
+                - a string with the `identifier name` of a predefined tokenizer that was user-uploaded to our S3, e.g.: ``dbmdz/bert-base-german-cased``.
                 - a path to a `directory` containing vocabulary files required by the tokenizer, for instance saved using the :func:`~transformers.PreTrainedTokenizer.save_pretrained` method, e.g.: ``./my_model_directory/``.
                 - (not applicable to all derived classes) a path or url to a single saved vocabulary file if and only if the tokenizer only requires a single vocabulary file (e.g. Bert, XLNet), e.g.: ``./my_model_directory/vocab.txt``.
 
@@ -251,6 +265,9 @@ class PreTrainedTokenizer(object):
             force_download: (`optional`) boolean, default False:
                 Force to (re-)download the vocabulary files and override the cached versions if they exists.
 
+            resume_download: (`optional`) boolean, default False:
+                Do not delete incompletely recieved file. Attempt to resume the download if such a file 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.
@@ -266,6 +283,9 @@ class PreTrainedTokenizer(object):
             # Download vocabulary from S3 and cache.
             tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
 
+            # Download vocabulary from S3 (user-uploaded) and cache.
+            tokenizer = BertTokenizer.from_pretrained('dbmdz/bert-base-german-cased')
+
             # If vocabulary files are in a directory (e.g. tokenizer was saved using `save_pretrained('./test/saved_model/')`)
             tokenizer = BertTokenizer.from_pretrained('./test/saved_model/')
 
@@ -286,6 +306,7 @@ class PreTrainedTokenizer(object):
     def _from_pretrained(cls, pretrained_model_name_or_path, *init_inputs, **kwargs):
         cache_dir = kwargs.pop('cache_dir', None)
         force_download = kwargs.pop('force_download', False)
+        resume_download = kwargs.pop('resume_download', False)
         proxies = kwargs.pop('proxies', None)
 
         s3_models = list(cls.max_model_input_sizes.keys())
@@ -310,12 +331,15 @@ class PreTrainedTokenizer(object):
                 if os.path.isdir(pretrained_model_name_or_path):
                     # If a directory is provided we look for the standard filenames
                     full_file_name = os.path.join(pretrained_model_name_or_path, file_name)
-                else:
+                    if not os.path.exists(full_file_name):
+                        logger.info("Didn't find file {}. We won't load it.".format(full_file_name))
+                        full_file_name = None
+                elif os.path.isfile(pretrained_model_name_or_path) or is_remote_url(pretrained_model_name_or_path):
                     # If a path to a file is provided we use it (will only work for non-BPE tokenizer using a single vocabulary file)
                     full_file_name = pretrained_model_name_or_path
-                if not os.path.exists(full_file_name):
-                    logger.info("Didn't find file {}. We won't load it.".format(full_file_name))
-                    full_file_name = None
+                else:
+                    full_file_name = hf_bucket_url(pretrained_model_name_or_path, postfix=file_name)
+                
                 vocab_files[file_id] = full_file_name
 
             # Look for the additional tokens files
@@ -342,7 +366,7 @@ class PreTrainedTokenizer(object):
                     "We assumed '{}' was a path or url to a directory containing vocabulary files "
                     "named {} but couldn't find such vocabulary files at this path or url.".format(
                         pretrained_model_name_or_path, ', '.join(s3_models),
-                        pretrained_model_name_or_path, 
+                        pretrained_model_name_or_path,
                         list(cls.vocab_files_names.values())))
 
         # Get files from url, cache, or disk depending on the case
@@ -352,7 +376,7 @@ 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, force_download=force_download, proxies=proxies)
+                    resolved_vocab_files[file_id] = cached_path(file_path, cache_dir=cache_dir, force_download=force_download, proxies=proxies, resume_download=resume_download)
         except EnvironmentError:
             if pretrained_model_name_or_path in s3_models:
                 msg = "Couldn't reach server at '{}' to download vocabulary files."
@@ -377,7 +401,8 @@ class PreTrainedTokenizer(object):
         # Did we saved some inputs and kwargs to reload ?
         tokenizer_config_file = resolved_vocab_files.pop('tokenizer_config_file', None)
         if tokenizer_config_file is not None:
-            init_kwargs = json.load(open(tokenizer_config_file, encoding="utf-8"))
+            with open(tokenizer_config_file, encoding="utf-8") as tokenizer_config_handle:
+                init_kwargs = json.load(tokenizer_config_handle)
             saved_init_inputs = init_kwargs.pop('init_inputs', ())
             if not init_inputs:
                 init_inputs = saved_init_inputs
@@ -402,7 +427,8 @@ class PreTrainedTokenizer(object):
             if args_name not in init_kwargs:
                 init_kwargs[args_name] = file_path
         if special_tokens_map_file is not None:
-            special_tokens_map = json.load(open(special_tokens_map_file, encoding="utf-8"))
+            with open(special_tokens_map_file, encoding="utf-8") as special_tokens_map_handle:
+                special_tokens_map = json.load(special_tokens_map_handle)
             for key, value in special_tokens_map.items():
                 if key not in init_kwargs:
                     init_kwargs[key] = value
@@ -416,7 +442,8 @@ class PreTrainedTokenizer(object):
 
         # Add supplementary tokens.
         if added_tokens_file is not None:
-            added_tok_encoder = json.load(open(added_tokens_file, encoding="utf-8"))
+            with open(added_tokens_file, encoding="utf-8") as added_tokens_handle:
+                added_tok_encoder = json.load(added_tokens_handle)
             added_tok_decoder = {v:k for k, v in added_tok_encoder.items()}
             tokenizer.added_tokens_encoder.update(added_tok_encoder)
             tokenizer.added_tokens_decoder.update(added_tok_decoder)
@@ -512,6 +539,8 @@ class PreTrainedTokenizer(object):
         to_add_tokens = []
         for token in new_tokens:
             assert isinstance(token, str) or (six.PY2 and isinstance(token, unicode))
+            if self.init_kwargs.get('do_lower_case', False) and token not in self.all_special_tokens:
+                token = token.lower()
             if token != self.unk_token and \
                     self.convert_tokens_to_ids(token) == self.convert_tokens_to_ids(self.unk_token) and \
                     token not in to_add_tokens:
@@ -604,7 +633,23 @@ class PreTrainedTokenizer(object):
             vocabularies (BPE/SentencePieces/WordPieces).
 
             Take care of added tokens.
+
+            text: The sequence to be encoded.
+            **kwargs: passed to the child `self.tokenize()` method
         """
+        def lowercase_text(t):
+            # convert non-special tokens to lowercase
+            escaped_special_toks = [re.escape(s_tok) for s_tok in self.all_special_tokens]
+            pattern = r'(^' + r'|'.join(escaped_special_toks) + r')|' + \
+                      r'(.+?)'
+            return re.sub(
+                pattern,
+                lambda m: m.groups()[0] or m.groups()[1].lower(),
+                t)
+
+        if self.init_kwargs.get('do_lower_case', False):
+            text = lowercase_text(text)
+
         def split_on_token(tok, text):
             result = []
             split_text = text.split(tok)
@@ -624,7 +669,7 @@ class PreTrainedTokenizer(object):
             return result
 
         def split_on_tokens(tok_list, text):
-            if not text:
+            if not text.strip():
                 return []
             if not tok_list:
                 return self._tokenize(text, **kwargs)
@@ -641,9 +686,9 @@ class PreTrainedTokenizer(object):
                         tokenized_text += [sub_text]
                 text_list = tokenized_text
 
-            return sum((self._tokenize(token, **kwargs) if token not \
+            return list(itertools.chain.from_iterable((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), [])
+                    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)
@@ -671,10 +716,6 @@ class PreTrainedTokenizer(object):
         ids = []
         for token in tokens:
             ids.append(self._convert_token_to_id_with_added_voc(token))
-        if len(ids) > self.max_len:
-            logger.warning("Token indices sequence length is longer than the specified maximum sequence length "
-                           "for this model ({} > {}). Running this sequence through the model will result in "
-                           "indexing errors".format(len(ids), self.max_len))
         return ids
 
     def _convert_token_to_id_with_added_voc(self, token):
@@ -689,14 +730,15 @@ 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',
+               pad_to_max_length=False,
+               return_tensors=None,
+               **kwargs):
         """
         Converts a string in a sequence of ids (integer), using the tokenizer and vocabulary.
 
@@ -721,6 +763,12 @@ class PreTrainedTokenizer(object):
                 - 'only_first': Only truncate the first sequence
                 - 'only_second': Only truncate the second sequence
                 - 'do_not_truncate': Does not truncate (raise an error if the input sequence is longer than max_length)
+            pad_to_max_length: if set to True, the returned sequences will be padded according to the model's padding side and
+                padding index, up to their max length. If no max length is specified, the padding is done up to the model's max length.
+                The tokenizer padding sides are handled by the following strings:
+                - 'left': pads on the left of the sequences
+                - 'right': pads on the right of the sequences   
+                Defaults to False: no padding.
             return_tensors: (optional) can be set to 'tf' or 'pt' to return respectively TensorFlow tf.constant
                 or PyTorch torch.Tensor instead of a list of python integers.
             **kwargs: passed to the `self.tokenize()` method
@@ -731,6 +779,7 @@ class PreTrainedTokenizer(object):
                                           add_special_tokens=add_special_tokens,
                                           stride=stride,
                                           truncation_strategy=truncation_strategy,
+                                          pad_to_max_length=pad_to_max_length,
                                           return_tensors=return_tensors,
                                           **kwargs)
 
@@ -739,11 +788,16 @@ 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',
+                    pad_to_max_length=False,
                     return_tensors=None,
+                    return_token_type_ids=True,
+                    return_attention_mask=True,
+                    return_overflowing_tokens=False,
+                    return_special_tokens_mask=False,
                     **kwargs):
         """
         Returns a dictionary containing the encoded sequence or sequence pair and additional informations:
@@ -768,9 +822,40 @@ class PreTrainedTokenizer(object):
                 - 'only_first': Only truncate the first sequence
                 - 'only_second': Only truncate the second sequence
                 - 'do_not_truncate': Does not truncate (raise an error if the input sequence is longer than max_length)
+            pad_to_max_length: if set to True, the returned sequences will be padded according to the model's padding side and
+                padding index, up to their max length. If no max length is specified, the padding is done up to the model's max length.
+                The tokenizer padding sides are handled by the following strings:
+                - 'left': pads on the left of the sequences
+                - 'right': pads on the right of the sequences   
+                Defaults to False: no padding.
             return_tensors: (optional) can be set to 'tf' or 'pt' to return respectively TensorFlow tf.constant
                 or PyTorch torch.Tensor instead of a list of python integers.
+            return_token_type_ids: (optional) Set to False to avoid returning token_type_ids (default True).
+            return_attention_mask: (optional) Set to False to avoir returning attention mask (default True)
+            return_overflowing_tokens: (optional) Set to True to return overflowing token information (default False).
+            return_special_tokens_mask: (optional) Set to True to return special tokens mask information (default False).
             **kwargs: passed to the `self.tokenize()` method
+
+        Return:
+            A Dictionary of shape::
+
+                {
+                    input_ids: list[int],
+                    token_type_ids: list[int] if return_token_type_ids is True (default)
+                    attention_mask: list[int] if return_attention_mask is True (default)
+                    overflowing_tokens: list[int] if a ``max_length`` is specified and return_overflowing_tokens is True
+                    num_truncated_tokens: int if a ``max_length`` is specified and return_overflowing_tokens is True
+                    special_tokens_mask: list[int] if ``add_special_tokens`` if set to ``True`` and return_special_tokens_mask is True
+                }
+
+            With the fields:
+                ``input_ids``: list of token ids to be fed to a model
+                ``token_type_ids``: list of token type ids to be fed to a model
+                ``attention_mask``: list of indices specifying which tokens should be attended to by the model
+                ``overflowing_tokens``: list of overflowing tokens if a max length is specified.
+                ``num_truncated_tokens``: number of overflowing tokens a ``max_length`` is specified
+                ``special_tokens_mask``: if adding special tokens, this is a list of [0, 1], with 0 specifying special added
+                tokens and 1 specifying sequence tokens.
         """
 
         def get_input_ids(text):
@@ -789,13 +874,24 @@ class PreTrainedTokenizer(object):
         return self.prepare_for_model(first_ids,
                                       pair_ids=second_ids,
                                       max_length=max_length,
+                                      pad_to_max_length=pad_to_max_length,
                                       add_special_tokens=add_special_tokens,
                                       stride=stride,
                                       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,
-                          truncation_strategy='longest_first', return_tensors=None):
+                                      return_tensors=return_tensors,
+                                      return_attention_mask=return_attention_mask,
+                                      return_token_type_ids=return_token_type_ids,
+                                      return_overflowing_tokens=return_overflowing_tokens,
+                                      return_special_tokens_mask=return_special_tokens_mask)
+
+    def prepare_for_model(self, ids, pair_ids=None, max_length=None, add_special_tokens=True, stride=0,
+                          truncation_strategy='longest_first',
+                          pad_to_max_length=False,
+                          return_tensors=None,
+                          return_token_type_ids=True,
+                          return_attention_mask=True,
+                          return_overflowing_tokens=False,
+                          return_special_tokens_mask=False):
         """
         Prepares a sequence of input id, or a pair of sequences of inputs ids so that it can be used by the model.
         It adds special tokens, truncates
@@ -818,23 +914,36 @@ class PreTrainedTokenizer(object):
                 - 'only_first': Only truncate the first sequence
                 - 'only_second': Only truncate the second sequence
                 - 'do_not_truncate': Does not truncate (raise an error if the input sequence is longer than max_length)
+            pad_to_max_length: if set to True, the returned sequences will be padded according to the model's padding side and
+                padding index, up to their max length. If no max length is specified, the padding is done up to the model's max length.
+                The tokenizer padding sides are handled by the following strings:
+                - 'left': pads on the left of the sequences
+                - 'right': pads on the right of the sequences   
+                Defaults to False: no padding.
             return_tensors: (optional) can be set to 'tf' or 'pt' to return respectively TensorFlow tf.constant
                 or PyTorch torch.Tensor instead of a list of python integers.
+            return_token_type_ids: (optional) Set to False to avoid returning token_type_ids (default True).
+            return_attention_mask: (optional) Set to False to avoid returning attention mask (default True)
+            return_overflowing_tokens: (optional) Set to True to return overflowing token information (default False).
+            return_special_tokens_mask: (optional) Set to True to return special tokens mask information (default False).
 
         Return:
             A Dictionary of shape::
 
                 {
                     input_ids: list[int],
-                    overflowing_tokens: list[int] if a ``max_length`` is specified, else None
-                    special_tokens_mask: list[int] if ``add_special_tokens`` if set to ``True``
+                    token_type_ids: list[int] if return_token_type_ids is True (default)
+                    overflowing_tokens: list[int] if a ``max_length`` is specified and return_overflowing_tokens is True
+                    num_truncated_tokens: int if a ``max_length`` is specified and return_overflowing_tokens is True
+                    special_tokens_mask: list[int] if ``add_special_tokens`` if set to ``True`` and return_special_tokens_mask is True
                 }
 
             With the fields:
-                ``input_ids``: list of tokens to be fed to a model
+                ``input_ids``: list of token ids to be fed to a model
+                ``token_type_ids``: list of token type ids to be fed to a model
 
                 ``overflowing_tokens``: list of overflowing tokens if a max length is specified.
-
+                ``num_truncated_tokens``: number of overflowing tokens a ``max_length`` is specified
                 ``special_tokens_mask``: if adding special tokens, this is a list of [0, 1], with 0 specifying special added
                 tokens and 1 specifying sequence tokens.
         """
@@ -843,39 +952,98 @@ class PreTrainedTokenizer(object):
         len_pair_ids = len(pair_ids) if pair else 0
 
         encoded_inputs = {}
+
+        # Handle max sequence length
         total_len = len_ids + len_pair_ids + (self.num_added_tokens(pair=pair) if add_special_tokens else 0)
         if max_length and total_len > max_length:
             ids, pair_ids, overflowing_tokens = self.truncate_sequences(ids, pair_ids=pair_ids,
                                                                         num_tokens_to_remove=total_len-max_length,
                                                                         truncation_strategy=truncation_strategy,
                                                                         stride=stride)
-            encoded_inputs["overflowing_tokens"] = overflowing_tokens
-            encoded_inputs["num_truncated_tokens"] = total_len - max_length
+            if return_overflowing_tokens:
+                encoded_inputs["overflowing_tokens"] = overflowing_tokens
+                encoded_inputs["num_truncated_tokens"] = total_len - max_length
 
+        # Handle special_tokens
         if add_special_tokens:
             sequence = self.build_inputs_with_special_tokens(ids, pair_ids)
             token_type_ids = self.create_token_type_ids_from_sequences(ids, pair_ids)
-            encoded_inputs["special_tokens_mask"] = self.get_special_tokens_mask(ids, pair_ids)
         else:
             sequence = ids + pair_ids if pair else ids
             token_type_ids = [0] * len(ids) + ([1] * len(pair_ids) if pair else [])
 
-        if return_tensors == 'tf' and is_tf_available():
-            sequence = tf.constant([sequence])
-            token_type_ids = tf.constant([token_type_ids])
-        elif return_tensors == 'pt' and is_torch_available():
-            sequence = torch.tensor([sequence])
-            token_type_ids = torch.tensor([token_type_ids])
-        elif return_tensors is not None:
-            logger.warning("Unable to convert output to tensors format {}, PyTorch or TensorFlow is not available.".format(return_tensors))
+        if return_special_tokens_mask:
+            encoded_inputs["special_tokens_mask"] = self.get_special_tokens_mask(ids, pair_ids)
 
         encoded_inputs["input_ids"] = sequence
-        encoded_inputs["token_type_ids"] = token_type_ids
+        if return_token_type_ids:
+            encoded_inputs["token_type_ids"] = token_type_ids
 
         if max_length and len(encoded_inputs["input_ids"]) > max_length:
             encoded_inputs["input_ids"] = encoded_inputs["input_ids"][:max_length]
-            encoded_inputs["token_type_ids"] = encoded_inputs["token_type_ids"][:max_length]
-            encoded_inputs["special_tokens_mask"] = encoded_inputs["special_tokens_mask"][:max_length]
+            if return_token_type_ids:
+                encoded_inputs["token_type_ids"] = encoded_inputs["token_type_ids"][:max_length]
+            if return_special_tokens_mask:
+                encoded_inputs["special_tokens_mask"] = encoded_inputs["special_tokens_mask"][:max_length]
+
+        if max_length is None and len(encoded_inputs["input_ids"]) > self.max_len:
+            logger.warning("Token indices sequence length is longer than the specified maximum sequence length "
+                           "for this model ({} > {}). Running this sequence through the model will result in "
+                           "indexing errors".format(len(ids), self.max_len))
+                           
+        needs_to_be_padded = pad_to_max_length and (
+            max_length and len(encoded_inputs["input_ids"]) < max_length
+            or 
+            max_length is None and len(encoded_inputs["input_ids"]) < self.max_len and self.max_len <= 10000
+        )
+
+        if pad_to_max_length and max_length is None and self.max_len > 10000:
+            logger.warning("Sequence can't be padded as no maximum length is specified and the model maximum length is too high.")
+
+        if needs_to_be_padded:
+            difference = (max_length if max_length is not None else self.max_len) - len(encoded_inputs["input_ids"])
+
+            if self.padding_side == 'right':
+                if return_attention_mask:
+                    encoded_inputs["attention_mask"] = [1] * len(encoded_inputs["input_ids"]) + [0] * difference
+                if return_token_type_ids:
+                    encoded_inputs["token_type_ids"] = encoded_inputs["token_type_ids"] + [self.pad_token_type_id] * difference
+                if return_special_tokens_mask:
+                    encoded_inputs["special_tokens_mask"] = encoded_inputs["special_tokens_mask"] + [1] * difference
+                encoded_inputs["input_ids"] = encoded_inputs["input_ids"] + [self.pad_token_id] * difference
+            elif self.padding_side == 'left':
+                if return_attention_mask:
+                    encoded_inputs["attention_mask"] = [0] * difference + [1] * len(encoded_inputs["input_ids"])
+                if return_token_type_ids:
+                    encoded_inputs["token_type_ids"] = [self.pad_token_type_id] * difference + encoded_inputs["token_type_ids"]
+                if return_special_tokens_mask:
+                    encoded_inputs["special_tokens_mask"] = [1] * difference + encoded_inputs["special_tokens_mask"]
+                encoded_inputs["input_ids"] = [self.pad_token_id] * difference + encoded_inputs["input_ids"]
+
+            else:
+                raise ValueError("Invalid padding strategy:" + str(self.padding_side))
+            
+        elif return_attention_mask:
+            encoded_inputs["attention_mask"] = [1] * len(encoded_inputs["input_ids"])
+
+        # Prepare inputs as tensors if asked
+        if return_tensors == 'tf' and is_tf_available():
+            encoded_inputs["input_ids"] = tf.constant([encoded_inputs["input_ids"]])
+            encoded_inputs["token_type_ids"] = tf.constant([encoded_inputs["token_type_ids"]])
+
+            if "attention_mask" in encoded_inputs:
+                encoded_inputs["attention_mask"] = tf.constant([encoded_inputs["attention_mask"]])
+
+        elif return_tensors == 'pt' and is_torch_available():
+            encoded_inputs["input_ids"] = torch.tensor([encoded_inputs["input_ids"]])
+            encoded_inputs["token_type_ids"] = torch.tensor([encoded_inputs["token_type_ids"]])
+
+            if "attention_mask" in encoded_inputs:
+                encoded_inputs["attention_mask"] = torch.tensor([encoded_inputs["attention_mask"]])
+        elif return_tensors is not None:
+            logger.warning(
+                "Unable to convert output to tensors format {}, PyTorch or TensorFlow is not available.".format(
+                    return_tensors))
 
         return encoded_inputs
 
@@ -920,7 +1088,6 @@ class PreTrainedTokenizer(object):
         return (ids, pair_ids, overflowing_tokens)
 
     def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
-        logger.warning("This tokenizer does not make use of special tokens.")
         if token_ids_1 is None:
             return len(token_ids_0) * [0]
         return [0] * len(token_ids_0) + [1] * len(token_ids_1)
@@ -933,7 +1100,6 @@ class PreTrainedTokenizer(object):
             single sequence: <s> X </s>
             pair of sequences: <s> A </s></s> B </s>
         """
-        logger.warning("This tokenizer does not make use of special tokens. Input is returned with no modification.")
         if token_ids_1 is None:
             return token_ids_0
         return token_ids_0 + token_ids_1
@@ -951,7 +1117,7 @@ class PreTrainedTokenizer(object):
                 special tokens for the model
 
         Returns:
-            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
+            A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
         """
         return [0] * ((len(token_ids_1) if token_ids_1 else 0) + len(token_ids_0))
 
@@ -1055,7 +1221,7 @@ class PreTrainedTokenizer(object):
             class attributes (cls_token, unk_token...).
         """
         all_toks = self.all_special_tokens
-        all_ids = list(self._convert_token_to_id(t) for t in all_toks)
+        all_ids = self.convert_tokens_to_ids(all_toks)
         return all_ids
 
     @staticmethod

transformers/tokenization_xlm.py

@@ -12,7 +12,7 @@
 # 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 OpenAI GPT."""
+"""Tokenization classes for XLM."""
 from __future__ import (absolute_import, division, print_function,
                         unicode_literals)
 
@@ -524,7 +524,7 @@ class XLMTokenizer(PreTrainedTokenizer):
 
         - argument ``special_tokens`` and function ``set_special_tokens``, can be used to add additional symbols \
         (ex: "__classify__") to a vocabulary
-        
+
         - `lang2id` attribute maps the languages supported by the model with their ids if provided (automatically set for pretrained vocabularies)
 
         - `id2lang` attributes does reverse mapping if provided (automatically set for pretrained vocabularies)
@@ -549,6 +549,10 @@ class XLMTokenizer(PreTrainedTokenizer):
                                            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 
+
         # cache of sm.MosesPunctNormalizer instance
         self.cache_moses_punct_normalizer = dict()
         # cache of sm.MosesTokenizer instance
@@ -564,9 +568,11 @@ class XLMTokenizer(PreTrainedTokenizer):
         self.ja_word_tokenizer = None
         self.zh_word_tokenizer = None
 
-        self.encoder = json.load(open(vocab_file, encoding="utf-8"))
+        with open(vocab_file, encoding="utf-8") as vocab_handle:
+            self.encoder = json.load(vocab_handle)
         self.decoder = {v:k for k,v in self.encoder.items()}
-        merges = open(merges_file, encoding='utf-8').read().split('\n')[:-1]
+        with open(merges_file, encoding='utf-8') as merges_handle:
+            merges = merges_handle.read().split('\n')[:-1]
         merges = [tuple(merge.split()[:2]) for merge in merges]
         self.bpe_ranks = dict(zip(merges, range(len(merges))))
         self.cache = {}
@@ -758,9 +764,9 @@ class XLMTokenizer(PreTrainedTokenizer):
         """
         Build model inputs from a sequence or a pair of sequence for sequence classification tasks
         by concatenating and adding special tokens.
-        A RoBERTa sequence has the following format:
+        A XLM sequence has the following format:
             single sequence: <s> X </s>
-            pair of sequences: <s> A </s></s> B </s>
+            pair of sequences: <s> A </s> B </s>
         """
         if token_ids_1 is None:
             return [self.cls_token_id] + token_ids_0 + [self.sep_token_id]
@@ -781,7 +787,7 @@ class XLMTokenizer(PreTrainedTokenizer):
                 special tokens for the model
 
         Returns:
-            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
+            A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
         """
 
         if already_has_special_tokens:

transformers/tokenization_xlnet.py

@@ -60,6 +60,7 @@ class XLNetTokenizer(PreTrainedTokenizer):
     vocab_files_names = VOCAB_FILES_NAMES
     pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
     max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    padding_side = "left"
 
     def __init__(self, vocab_file,
                  do_lower_case=False, remove_space=True, keep_accents=False,
@@ -74,6 +75,7 @@ class XLNetTokenizer(PreTrainedTokenizer):
 
         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
+        self._pad_token_type_id = 3
 
         try:
             import sentencepiece as spm
@@ -141,7 +143,7 @@ class XLNetTokenizer(PreTrainedTokenizer):
             pieces = self.sp_model.SampleEncodeAsPieces(text, 64, 0.1)
         new_pieces = []
         for piece in pieces:
-            if len(piece) > 1 and piece[-1] == ',' and piece[-2].isdigit():
+            if len(piece) > 1 and piece[-1] == str(',') and piece[-2].isdigit():
                 cur_pieces = self.sp_model.EncodeAsPieces(
                     piece[:-1].replace(SPIECE_UNDERLINE, ''))
                 if piece[0] != SPIECE_UNDERLINE and cur_pieces[0][0] == SPIECE_UNDERLINE:
@@ -185,9 +187,9 @@ class XLNetTokenizer(PreTrainedTokenizer):
         """
         Build model inputs from a sequence or a pair of sequence for sequence classification tasks
         by concatenating and adding special tokens.
-        A RoBERTa sequence has the following format:
-            single sequence: <s> X </s>
-            pair of sequences: <s> A </s></s> B </s>
+        An XLNet sequence has the following format:
+            single sequence: X <sep> <cls>
+            pair of sequences: A <sep> B <sep> <cls>
         """
         sep = [self.sep_token_id]
         cls = [self.cls_token_id]
@@ -208,7 +210,7 @@ class XLNetTokenizer(PreTrainedTokenizer):
                 special tokens for the model
 
         Returns:
-            A list of integers in the range [0, 1]: 0 for a special token, 1 for a sequence token.
+            A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
         """
 
         if already_has_special_tokens:
@@ -224,10 +226,10 @@ class XLNetTokenizer(PreTrainedTokenizer):
     def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
         """
         Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
-        A BERT sequence pair mask has the following format:
+        An XLNet sequence pair mask has the following format:
         0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 2
         | first sequence    | second sequence     | CLS segment ID
-        
+
         if token_ids_1 is None, only returns the first portion of the mask (0's).
         """
         sep = [self.sep_token_id]