Big Bird Fast Tokenizer implementation (#11075)

* Added Big Bird Fast Tokenizer initial file

* style fixes

* flake fixes

* Added big bird fast tokenizer to init files

* Added big bird fast to Auto tokenization

* fix styles

* minor quality fixes

* Added initial test code

* Fix SpmConverter when precompiled_charsmap doesn't exist

* fixed post processor

* minor style fix

* minor fix input names

* Actually fix identity normalization

* style

* Added token type ids to fast tokenizer

* style

* flake fix

* fix copies

Co-authored-by: Anthony MOI <m.anthony.moi@gmail.com>

docs/source/index.rst

@@ -276,7 +276,7 @@ Flax), PyTorch, and/or TensorFlow.
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       Bert Generation       |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
-|           BigBird           |       ✅       |       ❌       |       ✅        |         ❌         |      ❌      |
+|           BigBird           |       ✅       |       ✅       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+
 |       BigBirdPegasus        |       ❌       |       ❌       |       ✅        |         ❌         |      ❌      |
 +-----------------------------+----------------+----------------+-----------------+--------------------+--------------+

docs/source/model_doc/bigbird.rst

@@ -67,6 +67,11 @@ BigBirdTokenizer
     :members: build_inputs_with_special_tokens, get_special_tokens_mask,
         create_token_type_ids_from_sequences, save_vocabulary
 
+BigBirdTokenizerFast
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+.. autoclass:: transformers.BigBirdTokenizerFast
+    :members:
 
 BigBird specific outputs
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

src/transformers/__init__.py

@@ -320,6 +320,7 @@ if is_tokenizers_available():
     _import_structure["models.bart"].append("BartTokenizerFast")
     _import_structure["models.barthez"].append("BarthezTokenizerFast")
     _import_structure["models.bert"].append("BertTokenizerFast")
+    _import_structure["models.big_bird"].append("BigBirdTokenizerFast")
     _import_structure["models.camembert"].append("CamembertTokenizerFast")
     _import_structure["models.deberta"].append("DebertaTokenizerFast")
     _import_structure["models.distilbert"].append("DistilBertTokenizerFast")
@@ -1712,6 +1713,7 @@ if TYPE_CHECKING:
         from .models.bart import BartTokenizerFast
         from .models.barthez import BarthezTokenizerFast
         from .models.bert import BertTokenizerFast
+        from .models.big_bird import BigBirdTokenizerFast
         from .models.camembert import CamembertTokenizerFast
         from .models.convbert import ConvBertTokenizerFast
         from .models.deberta import DebertaTokenizerFast

src/transformers/convert_slow_tokenizer.py

@@ -373,9 +373,12 @@ class SpmConverter(Converter):
 
     def normalizer(self, proto):
         precompiled_charsmap = proto.normalizer_spec.precompiled_charsmap
-        return normalizers.Sequence(
+        if not precompiled_charsmap:
-            [normalizers.Precompiled(precompiled_charsmap), normalizers.Replace(Regex(" {2,}"), " ")]
+            return normalizers.Sequence([normalizers.Replace(Regex(" {2,}"), " ")])
-        )
+        else:
+            return normalizers.Sequence(
+                [normalizers.Precompiled(precompiled_charsmap), normalizers.Replace(Regex(" {2,}"), " ")]
+            )
 
     def pre_tokenizer(self, replacement, add_prefix_space):
         return pre_tokenizers.Metaspace(replacement=replacement, add_prefix_space=add_prefix_space)
@@ -686,11 +689,24 @@ class T5Converter(SpmConverter):
         )
 
 
+class BigBirdConverter(SpmConverter):
+    def post_processor(self):
+        return processors.TemplateProcessing(
+            single="[CLS]:0 $A:0 [SEP]:0",
+            pair="[CLS]:0 $A:0 [SEP]:0 $B:1 [SEP]:1",
+            special_tokens=[
+                ("[CLS]", self.original_tokenizer.convert_tokens_to_ids("[CLS]")),
+                ("[SEP]", self.original_tokenizer.convert_tokens_to_ids("[SEP]")),
+            ],
+        )
+
+
 SLOW_TO_FAST_CONVERTERS = {
     "AlbertTokenizer": AlbertConverter,
     "BartTokenizer": RobertaConverter,
     "BarthezTokenizer": BarthezConverter,
     "BertTokenizer": BertConverter,
+    "BigBirdTokenizer": BigBirdConverter,
     "CamembertTokenizer": CamembertConverter,
     "ConvBertTokenizer": BertConverter,
     "DebertaTokenizer": DebertaConverter,

src/transformers/models/auto/tokenization_auto.py

@@ -157,6 +157,7 @@ if is_tokenizers_available():
     from ..bart.tokenization_bart_fast import BartTokenizerFast
     from ..barthez.tokenization_barthez_fast import BarthezTokenizerFast
     from ..bert.tokenization_bert_fast import BertTokenizerFast
+    from ..big_bird.tokenization_big_bird_fast import BigBirdTokenizerFast
     from ..camembert.tokenization_camembert_fast import CamembertTokenizerFast
     from ..convbert.tokenization_convbert_fast import ConvBertTokenizerFast
     from ..deberta.tokenization_deberta_fast import DebertaTokenizerFast
@@ -190,6 +191,7 @@ else:
     BartTokenizerFast = None
     BarthezTokenizerFast = None
     BertTokenizerFast = None
+    BigBirdTokenizerFast = None
     CamembertTokenizerFast = None
     ConvBertTokenizerFast = None
     DebertaTokenizerFast = None
@@ -268,7 +270,7 @@ TOKENIZER_MAPPING = OrderedDict(
         (TapasConfig, (TapasTokenizer, None)),
         (LEDConfig, (LEDTokenizer, LEDTokenizerFast)),
         (ConvBertConfig, (ConvBertTokenizer, ConvBertTokenizerFast)),
-        (BigBirdConfig, (BigBirdTokenizer, None)),
+        (BigBirdConfig, (BigBirdTokenizer, BigBirdTokenizerFast)),
         (IBertConfig, (RobertaTokenizer, RobertaTokenizerFast)),
         (Wav2Vec2Config, (Wav2Vec2CTCTokenizer, None)),
         (GPTNeoConfig, (GPT2Tokenizer, GPT2TokenizerFast)),

src/transformers/models/big_bird/__init__.py

@@ -17,14 +17,25 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...file_utils import _BaseLazyModule, is_torch_available
+from ...file_utils import (
+    _BaseLazyModule,
+    is_sentencepiece_available,
+    is_tf_available,
+    is_tokenizers_available,
+    is_torch_available,
+)
 
 
 _import_structure = {
     "configuration_big_bird": ["BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP", "BigBirdConfig"],
-    "tokenization_big_bird": ["BigBirdTokenizer"],
 }
 
+if is_sentencepiece_available():
+    _import_structure["tokenization_big_bird"] = ["BigBirdTokenizer"]
+
+if is_tokenizers_available():
+    _import_structure["tokenization_big_bird_fast"] = ["BigBirdTokenizerFast"]
+
 if is_torch_available():
     _import_structure["modeling_big_bird"] = [
         "BIG_BIRD_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -44,7 +55,12 @@ if is_torch_available():
 
 if TYPE_CHECKING:
     from .configuration_big_bird import BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP, BigBirdConfig
-    from .tokenization_big_bird import BigBirdTokenizer
+
+    if is_sentencepiece_available():
+        from .tokenization_big_bird import BigBirdTokenizer
+
+    if is_tokenizers_available():
+        from .tokenization_big_bird_fast import BigBirdTokenizerFast
 
     if is_torch_available():
         from .modeling_big_bird import (

src/transformers/models/big_bird/tokenization_big_bird_fast.py

@@ -0,0 +1,240 @@
+# 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 Big Bird model."""
+
+
+import os
+from shutil import copyfile
+from typing import List, Optional, Tuple
+
+from ...file_utils import is_sentencepiece_available
+from ...tokenization_utils import AddedToken
+from ...tokenization_utils_fast import PreTrainedTokenizerFast
+from ...utils import logging
+
+
+if is_sentencepiece_available():
+    from .tokenization_big_bird import BigBirdTokenizer
+else:
+    BigBirdTokenizer = None
+
+logger = logging.get_logger(__name__)
+VOCAB_FILES_NAMES = {"vocab_file": "spiece.model", "tokenizer_file": "tokenizer.json"}
+
+PRETRAINED_VOCAB_FILES_MAP = {
+    "vocab_file": {
+        "google/bigbird-roberta-base": "https://huggingface.co/google/bigbird-roberta-base/resolve/main/spiece.model",
+        "google/bigbird-roberta-large": "https://huggingface.co/google/bigbird-roberta-large/resolve/main/spiece.model",
+        "google/bigbird-base-trivia-itc": "https://huggingface.co/google/bigbird-base-trivia-itc/resolve/main/spiece.model",
+    },
+    "tokenizer_file": {
+        "google/bigbird-roberta-base": "https://huggingface.co/google/bigbird-roberta-base/resolve/main/tokenizer.json",
+        "google/bigbird-roberta-large": "https://huggingface.co/google/bigbird-roberta-large/resolve/main/tokenizer.json",
+        "google/bigbird-base-trivia-itc": "https://huggingface.co/google/bigbird-base-trivia-itc/resolve/main/tokenizer.json",
+    },
+}
+
+PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
+    "google/bigbird-roberta-base": 4096,
+    "google/bigbird-roberta-large": 4096,
+    "google/bigbird-base-trivia-itc": 4096,
+}
+
+
+SPIECE_UNDERLINE = "▁"
+
+
+class BigBirdTokenizerFast(PreTrainedTokenizerFast):
+    """
+    Construct a "fast" BigBird tokenizer (backed by HuggingFace's `tokenizers` library). Based on `Unigram
+    <https://huggingface.co/docs/tokenizers/python/latest/components.html?highlight=unigram#models>`__. This tokenizer
+    inherits from :class:`~transformers.PreTrainedTokenizerFast` which contains most of the main methods. Users should
+    refer to this superclass for more information regarding those methods
+
+    Args:
+        vocab_file (:obj:`str`):
+            `SentencePiece <https://github.com/google/sentencepiece>`__ file (generally has a `.spm` extension) that
+            contains the vocabulary necessary to instantiate a tokenizer.
+        bos_token (:obj:`str`, `optional`, defaults to :obj:`"[CLS]"`):
+            The beginning of sequence token that was used during pretraining. Can be used a sequence classifier token.
+
+            .. note::
+
+               When building a sequence using special tokens, this is not the token that is used for the beginning of
+               sequence. The token used is the :obj:`cls_token`.
+        eos_token (:obj:`str`, `optional`, defaults to :obj:`"[SEP]"`):
+            The end of sequence token. .. note:: When building a sequence using special tokens, this is not the token
+            that is used for the end of sequence. The token used is the :obj:`sep_token`.
+        unk_token (:obj:`str`, `optional`, defaults to :obj:`"<unk>"`):
+            The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this
+            token instead.
+        sep_token (:obj:`str`, `optional`, defaults to :obj:`"[SEP]"`):
+            The separator token, which is used when building a sequence from multiple sequences, e.g. two sequences for
+            sequence classification or for a text and a question for question answering. It is also used as the last
+            token of a sequence built with special tokens.
+        pad_token (:obj:`str`, `optional`, defaults to :obj:`"<pad>"`):
+            The token used for padding, for example when batching sequences of different lengths.
+        cls_token (:obj:`str`, `optional`, defaults to :obj:`"[CLS]"`):
+            The classifier token which is used when doing sequence classification (classification of the whole sequence
+            instead of per-token classification). It is the first token of the sequence when built with special tokens.
+        mask_token (:obj:`str`, `optional`, defaults to :obj:`"[MASK]"`):
+            The token used for masking values. This is the token used when training this model with masked language
+            modeling. This is the token which the model will try to predict.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
+    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
+    slow_tokenizer_class = BigBirdTokenizer
+    model_input_names = ["input_ids", "attention_mask"]
+    prefix_tokens: List[int] = []
+
+    def __init__(
+        self,
+        vocab_file,
+        tokenizer_file=None,
+        unk_token="<unk>",
+        bos_token="<s>",
+        eos_token="</s>",
+        pad_token="<pad>",
+        sep_token="[SEP]",
+        mask_token="[MASK]",
+        cls_token="[CLS]",
+        **kwargs
+    ):
+        bos_token = AddedToken(bos_token, lstrip=False, rstrip=False) if isinstance(bos_token, str) else bos_token
+        eos_token = AddedToken(eos_token, lstrip=False, rstrip=False) if isinstance(eos_token, str) else eos_token
+        unk_token = AddedToken(unk_token, lstrip=False, rstrip=False) if isinstance(unk_token, str) else unk_token
+        pad_token = AddedToken(pad_token, lstrip=False, rstrip=False) if isinstance(pad_token, str) else pad_token
+        cls_token = AddedToken(cls_token, lstrip=False, rstrip=False) if isinstance(cls_token, str) else cls_token
+        sep_token = AddedToken(sep_token, lstrip=False, rstrip=False) if isinstance(sep_token, str) else sep_token
+
+        # Mask token behave like a normal word, i.e. include the space before it
+        mask_token = AddedToken(mask_token, lstrip=True, rstrip=False) if isinstance(mask_token, str) else mask_token
+
+        super().__init__(
+            vocab_file,
+            tokenizer_file=tokenizer_file,
+            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.vocab_file = vocab_file
+
+    def build_inputs_with_special_tokens(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
+        adding special tokens. An BigBird sequence has the following format:
+
+        - single sequence: ``[CLS] X [SEP]``
+        - pair of sequences: ``[CLS] A [SEP] B [SEP]``
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of IDs to which the special tokens will be added
+            token_ids_1 (:obj:`List[int]`, `optional`, defaults to :obj:`None`):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            :obj:`List[int]`: list of `input IDs <../glossary.html#input-ids>`__ with the appropriate special tokens.
+        """
+        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: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
+    ) -> List[int]:
+        """
+        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`` method.
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of ids.
+            token_ids_1 (:obj:`List[int]`, `optional`, defaults to :obj:`None`):
+                Optional second list of IDs for sequence pairs.
+            already_has_special_tokens (:obj:`bool`, `optional`, defaults to :obj:`False`):
+                Set to True if the token list is already formatted with special tokens for the model
+
+        Returns:
+            :obj:`List[int]`: 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 formatted 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] + ([0] * len(token_ids_1)) + [1]
+
+    def create_token_type_ids_from_sequences(
+        self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        """
+        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
+            | first sequence    | second sequence |
+
+        if token_ids_1 is None, only returns the first portion of the mask (0s).
+
+        Args:
+            token_ids_0 (:obj:`List[int]`):
+                List of ids.
+            token_ids_1 (:obj:`List[int]`, `optional`, defaults to :obj:`None`):
+                Optional second list of IDs for sequence pairs.
+
+        Returns:
+            :obj:`List[int]`: List of `token type IDs <../glossary.html#token-type-ids>`_ according to the given
+            sequence(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: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
+        if not os.path.isdir(save_directory):
+            logger.error(f"Vocabulary path ({save_directory}) should be a directory")
+            return
+        out_vocab_file = os.path.join(
+            save_directory, (filename_prefix + "-" if filename_prefix else "") + 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,)

src/transformers/utils/dummy_tokenizers_objects.py

@@ -38,6 +38,15 @@ class BertTokenizerFast:
         requires_backends(self, ["tokenizers"])
 
 
+class BigBirdTokenizerFast:
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["tokenizers"])
+
+    @classmethod
+    def from_pretrained(self, *args, **kwargs):
+        requires_backends(self, ["tokenizers"])
+
+
 class CamembertTokenizerFast:
     def __init__(self, *args, **kwargs):
         requires_backends(self, ["tokenizers"])

tests/test_tokenization_big_bird.py

@@ -17,9 +17,9 @@
 import os
 import unittest
 
-from transformers import BigBirdTokenizer
+from transformers import BigBirdTokenizer, BigBirdTokenizerFast
 from transformers.file_utils import cached_property
-from transformers.testing_utils import require_sentencepiece, require_torch, slow
+from transformers.testing_utils import require_sentencepiece, require_tokenizers, require_torch, slow
 
 from .test_tokenization_common import TokenizerTesterMixin
 
@@ -30,9 +30,12 @@ SAMPLE_VOCAB = os.path.join(os.path.dirname(os.path.abspath(__file__)), "fixture
 
 
 @require_sentencepiece
+@require_tokenizers
 class BigBirdTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
 
     tokenizer_class = BigBirdTokenizer
+    rust_tokenizer_class = BigBirdTokenizerFast
+    test_rust_tokenizer = True
 
     def setUp(self):
         super().setUp()
@@ -40,6 +43,28 @@ class BigBirdTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
         tokenizer = BigBirdTokenizer(SAMPLE_VOCAB, keep_accents=True)
         tokenizer.save_pretrained(self.tmpdirname)
 
+    def test_rust_and_python_full_tokenizers(self):
+        if not self.test_rust_tokenizer:
+            return
+
+        tokenizer = self.get_tokenizer()
+        rust_tokenizer = self.get_rust_tokenizer()
+
+        sequence = "I was born in 92000, and this is falsé."
+
+        tokens = tokenizer.tokenize(sequence)
+        rust_tokens = rust_tokenizer.tokenize(sequence)
+        self.assertListEqual(tokens, rust_tokens)
+
+        ids = tokenizer.encode(sequence, add_special_tokens=False)
+        rust_ids = rust_tokenizer.encode(sequence, add_special_tokens=False)
+        self.assertListEqual(ids, rust_ids)
+
+        rust_tokenizer = self.get_rust_tokenizer()
+        ids = tokenizer.encode(sequence)
+        rust_ids = rust_tokenizer.encode(sequence)
+        self.assertListEqual(ids, rust_ids)
+
     def test_full_tokenizer(self):
         tokenizer = BigBirdTokenizer(SAMPLE_VOCAB, keep_accents=True)