Aluode/PerceptionLabPortable

1# coding=utf-82# Copyright 2021 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 classes for RoFormer."""16 17import json18from typing import Optional19 20from tokenizers import normalizers21from tokenizers.pre_tokenizers import BertPreTokenizer, PreTokenizer22 23from ...tokenization_utils_fast import PreTrainedTokenizerFast24from ...utils import logging25from .tokenization_roformer import RoFormerTokenizer26from .tokenization_utils import JiebaPreTokenizer27 28 29logger = logging.get_logger(__name__)30 31VOCAB_FILES_NAMES = {"vocab_file": "vocab.txt", "tokenizer_file": "tokenizer.json"}32 33 34class RoFormerTokenizerFast(PreTrainedTokenizerFast):35    r"""36    Construct a "fast" RoFormer tokenizer (backed by HuggingFace's *tokenizers* library).37 38    [`RoFormerTokenizerFast`] is almost identical to [`BertTokenizerFast`] and runs end-to-end tokenization:39    punctuation splitting and wordpiece. There are some difference between them when tokenizing Chinese.40 41    This tokenizer inherits from [`PreTrainedTokenizerFast`] which contains most of the main methods. Users should42    refer to this superclass for more information regarding those methods.43 44    Example:45 46    ```python47    >>> from transformers import RoFormerTokenizerFast48 49    >>> tokenizer = RoFormerTokenizerFast.from_pretrained("junnyu/roformer_chinese_base")50    >>> tokenizer.tokenize("今天天气非常好。")51    ['今', '天', '天', '气', '非常', '好', '。']52    ```"""53 54    vocab_files_names = VOCAB_FILES_NAMES55    slow_tokenizer_class = RoFormerTokenizer56 57    def __init__(58        self,59        vocab_file=None,60        tokenizer_file=None,61        do_lower_case=True,62        unk_token="[UNK]",63        sep_token="[SEP]",64        pad_token="[PAD]",65        cls_token="[CLS]",66        mask_token="[MASK]",67        tokenize_chinese_chars=True,68        strip_accents=None,69        **kwargs,70    ):71        super().__init__(72            vocab_file,73            tokenizer_file=tokenizer_file,74            do_lower_case=do_lower_case,75            unk_token=unk_token,76            sep_token=sep_token,77            pad_token=pad_token,78            cls_token=cls_token,79            mask_token=mask_token,80            tokenize_chinese_chars=tokenize_chinese_chars,81            strip_accents=strip_accents,82            **kwargs,83        )84 85        normalizer_state = json.loads(self.backend_tokenizer.normalizer.__getstate__())86        if (87            normalizer_state.get("lowercase", do_lower_case) != do_lower_case88            or normalizer_state.get("strip_accents", strip_accents) != strip_accents89        ):90            normalizer_class = getattr(normalizers, normalizer_state.pop("type"))91            normalizer_state["lowercase"] = do_lower_case92            normalizer_state["strip_accents"] = strip_accents93            self.backend_tokenizer.normalizer = normalizer_class(**normalizer_state)94 95        # Make sure we correctly set the custom PreTokenizer96        vocab = self.backend_tokenizer.get_vocab()97        self.backend_tokenizer.pre_tokenizer = PreTokenizer.custom(JiebaPreTokenizer(vocab))98 99        self.do_lower_case = do_lower_case100 101    def __getstate__(self):102        state = self.__dict__.copy()103        state["_tokenizer"].pre_tokenizer = BertPreTokenizer()104        return state105 106    def __setstate__(self, d):107        self.__dict__ = d108        vocab = self.__dict__["_tokenizer"].get_vocab()109        self.__dict__["_tokenizer"].pre_tokenizer = PreTokenizer.custom(JiebaPreTokenizer(vocab))110 111    def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):112        """113        Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and114        adding special tokens. A RoFormer sequence has the following format:115 116        - single sequence: `[CLS] X [SEP]`117        - pair of sequences: `[CLS] A [SEP] B [SEP]`118 119        Args:120            token_ids_0 (`List[int]`):121                List of IDs to which the special tokens will be added.122            token_ids_1 (`List[int]`, *optional*):123                Optional second list of IDs for sequence pairs.124 125        Returns:126            `List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.127        """128        output = [self.cls_token_id] + token_ids_0 + [self.sep_token_id]129 130        if token_ids_1 is not None:131            output += token_ids_1 + [self.sep_token_id]132 133        return output134 135    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> tuple[str]:136        files = self._tokenizer.model.save(save_directory, name=filename_prefix)137        return tuple(files)138 139    def save_pretrained(140        self,141        save_directory,142        legacy_format=None,143        filename_prefix=None,144        push_to_hub=False,145        **kwargs,146    ):147        self.backend_tokenizer.pre_tokenizer = BertPreTokenizer()148        return super().save_pretrained(save_directory, legacy_format, filename_prefix, push_to_hub, **kwargs)149 150 151__all__ = ["RoFormerTokenizerFast"]152