Aluode/PerceptionLabPortable

1# coding=utf-82# Copyright 2023 The Facebook Inc. and The HuggingFace Inc. team. All rights reserved.3#4# Licensed under the Apache License, Version 2.0 (the "License");5# you may not use this file except in compliance with the License.6# You may obtain a copy of the License at7#8#     http://www.apache.org/licenses/LICENSE-2.09#10# Unless required by applicable law or agreed to in writing, software11# distributed under the License is distributed on an "AS IS" BASIS,12# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.13# See the License for the specific language governing permissions and14# limitations under the License.15"""Tokenization class for SpeechT5."""16 17import os18from shutil import copyfile19from typing import Any, Optional20 21import sentencepiece as spm22 23from ...tokenization_utils import PreTrainedTokenizer24from ...utils import logging25from ...utils.import_utils import requires26from .number_normalizer import EnglishNumberNormalizer27 28 29logger = logging.get_logger(__name__)30 31VOCAB_FILES_NAMES = {"vocab_file": "spm_char.model"}32 33 34@requires(backends=("sentencepiece",))35class SpeechT5Tokenizer(PreTrainedTokenizer):36    """37    Construct a SpeechT5 tokenizer. Based on [SentencePiece](https://github.com/google/sentencepiece).38 39    This tokenizer inherits from [`PreTrainedTokenizer`] which contains most of the main methods. Users should refer to40    this superclass for more information regarding those methods.41 42    Args:43        vocab_file (`str`):44            [SentencePiece](https://github.com/google/sentencepiece) file (generally has a *.spm* extension) that45            contains the vocabulary necessary to instantiate a tokenizer.46        bos_token (`str`, *optional*, defaults to `"<s>"`):47            The begin of sequence token.48        eos_token (`str`, *optional*, defaults to `"</s>"`):49            The end of sequence token.50        unk_token (`str`, *optional*, defaults to `"<unk>"`):51            The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this52            token instead.53        pad_token (`str`, *optional*, defaults to `"<pad>"`):54            The token used for padding, for example when batching sequences of different lengths.55        normalize (`bool`, *optional*, defaults to `False`):56            Whether to convert numeric quantities in the text to their spelt-out english counterparts.57        sp_model_kwargs (`dict`, *optional*):58            Will be passed to the `SentencePieceProcessor.__init__()` method. The [Python wrapper for59            SentencePiece](https://github.com/google/sentencepiece/tree/master/python) can be used, among other things,60            to set:61 62            - `enable_sampling`: Enable subword regularization.63            - `nbest_size`: Sampling parameters for unigram. Invalid for BPE-Dropout.64 65              - `nbest_size = {0,1}`: No sampling is performed.66              - `nbest_size > 1`: samples from the nbest_size results.67              - `nbest_size < 0`: assuming that nbest_size is infinite and samples from the all hypothesis (lattice)68                using forward-filtering-and-backward-sampling algorithm.69 70            - `alpha`: Smoothing parameter for unigram sampling, and dropout probability of merge operations for71              BPE-dropout.72 73    Attributes:74        sp_model (`SentencePieceProcessor`):75            The *SentencePiece* processor that is used for every conversion (string, tokens and IDs).76    """77 78    vocab_files_names = VOCAB_FILES_NAMES79    model_input_names = ["input_ids", "attention_mask"]80 81    def __init__(82        self,83        vocab_file,84        bos_token="<s>",85        eos_token="</s>",86        unk_token="<unk>",87        pad_token="<pad>",88        normalize=False,89        sp_model_kwargs: Optional[dict[str, Any]] = None,90        **kwargs,91    ) -> None:92        self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs93        self.vocab_file = vocab_file94        self.normalize = normalize95        self._normalizer = None96 97        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)98        self.sp_model.Load(vocab_file)99 100        super().__init__(101            bos_token=bos_token,102            eos_token=eos_token,103            unk_token=unk_token,104            pad_token=pad_token,105            normalize=normalize,106            sp_model_kwargs=self.sp_model_kwargs,107            **kwargs,108        )109 110    def prepare_for_tokenization(self, text, is_split_into_words=False, **kwargs):111        normalize = kwargs.pop("normalize", self.normalize)112        if is_split_into_words:113            text = " " + text114        if normalize:115            text = self.normalizer(text)116        return (text, kwargs)117 118    @property119    def vocab_size(self):120        return self.sp_model.get_piece_size()121 122    @property123    def normalizer(self):124        if self._normalizer is None:125            self._normalizer = EnglishNumberNormalizer()126        return self._normalizer127 128    @normalizer.setter129    def normalizer(self, value):130        self._normalizer = value131 132    def get_vocab(self):133        vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}134        vocab.update(self.added_tokens_encoder)135        return vocab136 137    def __getstate__(self):138        state = self.__dict__.copy()139        state["sp_model"] = None140        return state141 142    def __setstate__(self, d):143        self.__dict__ = d144 145        # for backward compatibility146        if not hasattr(self, "sp_model_kwargs"):147            self.sp_model_kwargs = {}148 149        self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)150        self.sp_model.Load(self.vocab_file)151 152    def _tokenize(self, text: str) -> list[str]:153        """Take as input a string and return a list of strings (tokens) for words/sub-words"""154        return self.sp_model.encode(text, out_type=str)155 156    def _convert_token_to_id(self, token):157        """Converts a token (str) in an id using the vocab."""158        return self.sp_model.piece_to_id(token)159 160    def _convert_id_to_token(self, index):161        """Converts an index (integer) in a token (str) using the vocab."""162        token = self.sp_model.IdToPiece(index)163        return token164 165    # Copied from transformers.models.albert.tokenization_albert.AlbertTokenizer.convert_tokens_to_string166    def convert_tokens_to_string(self, tokens):167        """Converts a sequence of tokens (string) in a single string."""168        current_sub_tokens = []169        out_string = ""170        prev_is_special = False171        for token in tokens:172            # make sure that special tokens are not decoded using sentencepiece model173            if token in self.all_special_tokens:174                if not prev_is_special:175                    out_string += " "176                out_string += self.sp_model.decode(current_sub_tokens) + token177                prev_is_special = True178                current_sub_tokens = []179            else:180                current_sub_tokens.append(token)181                prev_is_special = False182        out_string += self.sp_model.decode(current_sub_tokens)183        return out_string.strip()184 185    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None) -> list[int]:186        """Build model inputs from a sequence by appending eos_token_id."""187        if token_ids_1 is None:188            return token_ids_0 + [self.eos_token_id]189        # We don't expect to process pairs, but leave the pair logic for API consistency190        return token_ids_0 + token_ids_1 + [self.eos_token_id]191 192    def get_special_tokens_mask(193        self, token_ids_0: list[int], token_ids_1: Optional[list[int]] = None, already_has_special_tokens: bool = False194    ) -> list[int]:195        if already_has_special_tokens:196            return super().get_special_tokens_mask(197                token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True198            )199 200        suffix_ones = [1]201        if token_ids_1 is None:202            return ([0] * len(token_ids_0)) + suffix_ones203        return ([0] * len(token_ids_0)) + ([0] * len(token_ids_1)) + suffix_ones204 205    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> tuple[str]:206        if not os.path.isdir(save_directory):207            logger.error(f"Vocabulary path ({save_directory}) should be a directory")208            return209        out_vocab_file = os.path.join(210            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]211        )212 213        if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):214            copyfile(self.vocab_file, out_vocab_file)215        elif not os.path.isfile(self.vocab_file):216            with open(out_vocab_file, "wb") as fi:217                content_spiece_model = self.sp_model.serialized_model_proto()218                fi.write(content_spiece_model)219 220        return (out_vocab_file,)221 222 223__all__ = ["SpeechT5Tokenizer"]224