DoruC/Grounded-Segment-Anything

1# coding=utf-82# Copyright 2018 Salesforce and The HuggingFace Inc. team.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 Salesforce CTRL."""16 17 18import json19import os20from typing import Optional, Tuple21 22import regex as re23 24from ...tokenization_utils import PreTrainedTokenizer25from ...utils import logging26 27 28logger = logging.get_logger(__name__)29 30VOCAB_FILES_NAMES = {31    "vocab_file": "vocab.json",32    "merges_file": "merges.txt",33}34 35PRETRAINED_VOCAB_FILES_MAP = {36    "vocab_file": {"ctrl": "https://raw.githubusercontent.com/salesforce/ctrl/master/ctrl-vocab.json"},37    "merges_file": {"ctrl": "https://raw.githubusercontent.com/salesforce/ctrl/master/ctrl-merges.txt"},38}39 40PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {41    "ctrl": 256,42}43 44CONTROL_CODES = {45    "Pregnancy": 168629,46    "Christianity": 7675,47    "Explain": 106423,48    "Fitness": 63440,49    "Saving": 63163,50    "Ask": 27171,51    "Ass": 95985,52    "Joke": 163509,53    "Questions": 45622,54    "Thoughts": 49605,55    "Retail": 52342,56    "Feminism": 164338,57    "Writing": 11992,58    "Atheism": 192263,59    "Netflix": 48616,60    "Computing": 39639,61    "Opinion": 43213,62    "Alone": 44967,63    "Funny": 58917,64    "Gaming": 40358,65    "Human": 4088,66    "India": 1331,67    "Joker": 77138,68    "Diet": 36206,69    "Legal": 11859,70    "Norman": 4939,71    "Tip": 72689,72    "Weight": 52343,73    "Movies": 46273,74    "Running": 23425,75    "Science": 2090,76    "Horror": 37793,77    "Confession": 60572,78    "Finance": 12250,79    "Politics": 16360,80    "Scary": 191985,81    "Support": 12654,82    "Technologies": 32516,83    "Teenage": 66160,84    "Event": 32769,85    "Learned": 67460,86    "Notion": 182770,87    "Wikipedia": 37583,88    "Books": 6665,89    "Extract": 76050,90    "Confessions": 102701,91    "Conspiracy": 75932,92    "Links": 63674,93    "Narcissus": 150425,94    "Relationship": 54766,95    "Relationships": 134796,96    "Reviews": 41671,97    "News": 4256,98    "Translation": 26820,99    "multilingual": 128406,100}101 102 103def get_pairs(word):104    """105    Return set of symbol pairs in a word.106 107    Word is represented as tuple of symbols (symbols being variable-length strings).108    """109    pairs = set()110    prev_char = word[0]111    for char in word[1:]:112        pairs.add((prev_char, char))113        prev_char = char114 115    pairs = set(pairs)116    return pairs117 118 119class CTRLTokenizer(PreTrainedTokenizer):120    """121    Construct a CTRL tokenizer. Based on Byte-Pair-Encoding.122 123    This tokenizer inherits from [`PreTrainedTokenizer`] which contains most of the main methods. Users should refer to124    this superclass for more information regarding those methods.125 126    Args:127        vocab_file (`str`):128            Path to the vocabulary file.129        merges_file (`str`):130            Path to the merges file.131        unk_token (`str`, *optional*, defaults to `"<unk>"`):132            The unknown token. A token that is not in the vocabulary cannot be converted to an ID and is set to be this133            token instead.134    """135 136    vocab_files_names = VOCAB_FILES_NAMES137    pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP138    max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES139    control_codes = CONTROL_CODES140 141    def __init__(self, vocab_file, merges_file, unk_token="<unk>", **kwargs):142        with open(vocab_file, encoding="utf-8") as vocab_handle:143            self.encoder = json.load(vocab_handle)144        self.decoder = {v: k for k, v in self.encoder.items()}145        with open(merges_file, encoding="utf-8") as merges_handle:146            merges = merges_handle.read().split("\n")[1:-1]147        merges = [tuple(merge.split()) for merge in merges]148        self.bpe_ranks = dict(zip(merges, range(len(merges))))149        self.cache = {}150        super().__init__(unk_token=unk_token, **kwargs)151 152    @property153    def vocab_size(self):154        return len(self.encoder)155 156    def get_vocab(self):157        return dict(self.encoder, **self.added_tokens_encoder)158 159    def bpe(self, token):160        if token in self.cache:161            return self.cache[token]162        word = tuple(token)163        word = tuple(list(word[:-1]) + [word[-1] + "</w>"])164        pairs = get_pairs(word)165 166        if not pairs:167            return token168 169        while True:170            bigram = min(pairs, key=lambda pair: self.bpe_ranks.get(pair, float("inf")))171            if bigram not in self.bpe_ranks:172                break173            first, second = bigram174            new_word = []175            i = 0176            while i < len(word):177                try:178                    j = word.index(first, i)179                except ValueError:180                    new_word.extend(word[i:])181                    break182                else:183                    new_word.extend(word[i:j])184                    i = j185 186                if word[i] == first and i < len(word) - 1 and word[i + 1] == second:187                    new_word.append(first + second)188                    i += 2189                else:190                    new_word.append(word[i])191                    i += 1192            new_word = tuple(new_word)193            word = new_word194            if len(word) == 1:195                break196            else:197                pairs = get_pairs(word)198        word = "@@ ".join(word)199        word = word[:-4]200        self.cache[token] = word201        return word202 203    def _tokenize(self, text):204        """Tokenize a string."""205        split_tokens = []206 207        words = re.findall(r"\S+\n?", text)208 209        for token in words:210            split_tokens.extend(list(self.bpe(token).split(" ")))211        return split_tokens212 213    def _convert_token_to_id(self, token):214        """Converts a token (str) in an id using the vocab."""215        return self.encoder.get(token, self.encoder.get(self.unk_token))216 217    def _convert_id_to_token(self, index):218        """Converts an index (integer) in a token (str) using the vocab."""219        return self.decoder.get(index, self.unk_token)220 221    def convert_tokens_to_string(self, tokens):222        """Converts a sequence of tokens (string) in a single string."""223        out_string = " ".join(tokens).replace("@@ ", "").strip()224        return out_string225 226    def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:227        if not os.path.isdir(save_directory):228            logger.error(f"Vocabulary path ({save_directory}) should be a directory")229            return230        vocab_file = os.path.join(231            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]232        )233        merge_file = os.path.join(234            save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["merges_file"]235        )236 237        with open(vocab_file, "w", encoding="utf-8") as f:238            f.write(json.dumps(self.encoder, indent=2, sort_keys=True, ensure_ascii=False) + "\n")239 240        index = 0241        with open(merge_file, "w", encoding="utf-8") as writer:242            writer.write("#version: 0.2\n")243            for bpe_tokens, token_index in sorted(self.bpe_ranks.items(), key=lambda kv: kv[1]):244                if index != token_index:245                    logger.warning(246                        f"Saving vocabulary to {merge_file}: BPE merge indices are not consecutive."247                        " Please check that the tokenizer is not corrupted!"248                    )249                    index = token_index250                writer.write(" ".join(bpe_tokens) + "\n")251                index += 1252 253        return vocab_file, merge_file254 255    # def decode(self, token_ids, skip_special_tokens=False, clean_up_tokenization_spaces=True):256    #     filtered_tokens = ' '.join(self.convert_ids_to_tokens(token_ids, skip_special_tokens=skip_special_tokens))257    #     tokens_generated_so_far = re.sub('(@@ )', '', string=filtered_tokens)258    #     tokens_generated_so_far = re.sub('(@@ ?$)', '', string=tokens_generated_so_far)259    #     return ''.join(tokens_generated_so_far)260