def __init__(self,
vocab_file: Optional[str] = None,
merges_file: Optional[str] = None,
unk_token: Optional[str] = "<unk>",
suffix: Optional[str] = "</w>",
dropout: Optional[float] = None):
if vocab_file is not None and merges_file is not None:
tokenizer = Tokenizer(
BPE.from_files(vocab_file,
merges_file,
dropout=dropout,
unk_token=unk_token,
end_of_word_suffix=suffix))
else:
tokenizer = Tokenizer(BPE.empty())
tokenizer.normalizer = Sequence.new([NFKC.new(), Lowercase.new()])
tokenizer.pre_tokenizer = pre_tokenizers.Whitespace.new()
tokenizer.decoder = decoders.BPEDecoder.new(suffix=suffix)
parameters = {
"model": "BPE",
"unk_token": unk_token,
"suffix": suffix,
"dropout": dropout,
}
super().__init__(tokenizer, parameters)
def get_tokenizer(args):
tokenizer = Tokenizer(models.BPE())
tokenizer.normalizer = Sequence(
[NFKC(), Replace('\r', ''),
Replace('\n', ' ')])
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel()
tokenizer.decoder = decoders.ByteLevel()
if os.path.isdir(args.tokenizer_dir):
vocab_fn = os.path.join(args.tokenizer_dir, 'vocab.json')
merge_fn = os.path.join(args.tokenizer_dir, 'merges.txt')
tokenizer.model = models.BPE.from_file(vocab_fn, merge_fn)
else:
os.makedirs(args.tokenizer_dir)
trainer = trainers.BpeTrainer(
vocab_size=args.vocab_size,
special_tokens=["[UNK]", "[PAD]", "[BOS]", "[EOS]"])
files = [
os.path.join(args.data_dir, split)
for split in ['train.json', 'val.json', 'test.json']
]
tokenizer.train(files=files, trainer=trainer)
tokenizer.model.save(args.tokenizer_dir)
return tokenizer
def test_get_set_components(self):
toki = Tokenizer(models.BPE())
toki.normalizer = normalizers.NFC()
toki.pre_tokenizer = pre_tokenizers.ByteLevel()
toki.post_processor = processors.BertProcessing(("A", 0), ("B", 1))
toki.decoder = decoders.ByteLevel()
tokenizer = BaseTokenizer(toki)
assert isinstance(tokenizer.model, models.BPE)
assert isinstance(tokenizer.normalizer, normalizers.NFC)
assert isinstance(tokenizer.pre_tokenizer, pre_tokenizers.ByteLevel)
assert isinstance(tokenizer.post_processor, processors.BertProcessing)
assert isinstance(tokenizer.decoder, decoders.ByteLevel)
tokenizer.model = models.Unigram()
assert isinstance(tokenizer.model, models.Unigram)
tokenizer.normalizer = normalizers.NFD()
assert isinstance(tokenizer.normalizer, normalizers.NFD)
tokenizer.pre_tokenizer = pre_tokenizers.Whitespace()
assert isinstance(tokenizer.pre_tokenizer, pre_tokenizers.Whitespace)
tokenizer.post_processor = processors.ByteLevel()
assert isinstance(tokenizer.post_processor, processors.ByteLevel)
tokenizer.decoder = decoders.WordPiece()
assert isinstance(tokenizer.decoder, decoders.WordPiece)
def converted(self) -> Tokenizer:
ot = self.original_tokenizer
vocab = ot.encoder
merges = list(ot.bpe_ranks.keys())
tokenizer = Tokenizer(
BPE(
vocab=vocab,
merges=merges,
dropout=None,
continuing_subword_prefix="",
end_of_word_suffix="",
fuse_unk=False,
))
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(
add_prefix_space=ot.add_prefix_space)
tokenizer.decoder = decoders.ByteLevel()
tokenizer.post_processor = processors.TemplateProcessing(
single="[CLS]:0 $A:0 [SEP]:0",
pair="[CLS]:0 $A:0 [SEP]:0 $B:0 [SEP]:0",
special_tokens=[
("[CLS]",
self.original_tokenizer.convert_tokens_to_ids("[CLS]")),
("[SEP]",
self.original_tokenizer.convert_tokens_to_ids("[SEP]")),
],
)
return tokenizer
def train_tokenizer(input_dir: str,
save_path: str,
tokenizer_type: str = "BPE",
vocab_size: int = 52000):
"""
Trains a tokenizer on all the json files in `input_dir` and saves it to `save_path`
:param input_dir: input directory containing jsonl files
:param save_path: path to save tokenizer to
:param tokenizer_type: type of tokenizer to train.
:param vocab_size: int, size of tokenizer's vocab
:return:
"""
if tokenizer_type == "BPE":
model = models.BPE()
else:
raise NotImplementedError(
f'Tokenizer type {tokenizer_type} not implemented')
tokenizer = Tokenizer(model)
# Customize pre-tokenization and decoding
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(add_prefix_space=True)
tokenizer.decoder = decoders.ByteLevel()
tokenizer.post_processor = processors.ByteLevel(trim_offsets=True)
tokenizer.normalizer = NFKC()
# And then train
trainer = trainers.BpeTrainer(
vocab_size=vocab_size, special_tokens=["<|endoftext|>", "<|padding|>"])
tokenizer.train_from_iterator(json_iterator(input_dir), trainer)
# And Save it
tokenizer.save(save_path, pretty=True)
print(f'Tokenizer saved at {save_path}')
def __init__(
self,
vocab: Optional[Union[str, Dict[str, int]]] = None,
merges: Optional[Union[str, Dict[Tuple[int, int], Tuple[int,
int]]]] = None,
unk_token: Union[str, AddedToken] = "<unk>",
replacement: str = "▁",
add_prefix_space: bool = True,
dropout: Optional[float] = None,
):
if vocab is not None and merges is not None:
tokenizer = Tokenizer(
BPE(vocab, merges, dropout=dropout, unk_token=unk_token))
else:
tokenizer = Tokenizer(BPE())
if tokenizer.token_to_id(str(unk_token)) is not None:
tokenizer.add_special_tokens([str(unk_token)])
tokenizer.normalizer = NFKC()
tokenizer.pre_tokenizer = pre_tokenizers.Metaspace(
replacement=replacement, add_prefix_space=add_prefix_space)
tokenizer.decoder = decoders.Metaspace(
replacement=replacement, add_prefix_space=add_prefix_space)
parameters = {
"model": "SentencePieceBPE",
"unk_token": unk_token,
"replacement": replacement,
"add_prefix_space": add_prefix_space,
"dropout": dropout,
}
super().__init__(tokenizer, parameters)
def __init__(
self,
vocab: Optional[str] = None,
replacement: str = "▁",
add_prefix_space: bool = True,
):
if vocab is not None:
# Let Unigram(..) fail if only one of them is None
tokenizer = Tokenizer(Unigram(vocab))
else:
tokenizer = Tokenizer(Unigram())
tokenizer.normalizer = normalizers.Sequence([
normalizers.Nmt(),
normalizers.NFKC(),
])
tokenizer.pre_tokenizer = pre_tokenizers.Sequence([
pre_tokenizers.WhitespaceSplit(),
pre_tokenizers.Metaspace(replacement=replacement,
add_prefix_space=add_prefix_space),
])
tokenizer.decoder = decoders.Metaspace(
replacement=replacement, add_prefix_space=add_prefix_space)
parameters = {
"model": "SentencePieceUnigram",
"replacement": replacement,
"add_prefix_space": add_prefix_space,
}
super().__init__(tokenizer, parameters)
def converted(self) -> Tokenizer:
tokenizer_info_str = "#version:"
token_suffix = "</w>"
vocab = self.original_tokenizer.encoder
merges = list(self.original_tokenizer.bpe_ranks.keys())
if tokenizer_info_str in merges[0][0]:
merges = merges[1:]
tokenizer = Tokenizer(
BPE(
vocab,
merges,
dropout=None,
unk_token=self.original_tokenizer.unk_token,
end_of_word_suffix=token_suffix,
))
tokenizer.normalizer = normalizers.BertNormalizer(lowercase=False,
strip_accents=False)
tokenizer.pre_tokenizer = pre_tokenizers.BertPreTokenizer()
tokenizer.decoder = decoders.BPEDecoder(suffix=token_suffix)
tokenizer.post_processor = processors.BertProcessing(
sep=(self.original_tokenizer.sep_token,
self.original_tokenizer.sep_token_id),
cls=(self.original_tokenizer.cls_token,
self.original_tokenizer.cls_token_id),
)
return tokenizer
def converted(self) -> Tokenizer:
ot = self.original_tokenizer
vocab = ot.encoder
merges = list(ot.bpe_ranks.keys())
tokenizer = Tokenizer(
BPE(
vocab=vocab,
merges=merges,
dropout=None,
continuing_subword_prefix="",
end_of_word_suffix="",
fuse_unk=False,
))
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(
add_prefix_space=ot.add_prefix_space)
tokenizer.decoder = decoders.ByteLevel()
tokenizer.post_processor = processors.RobertaProcessing(
sep=(ot.sep_token, ot.sep_token_id),
cls=(ot.cls_token, ot.cls_token_id),
add_prefix_space=ot.add_prefix_space,
trim_offsets=True, # True by default on Roberta (historical)
)
return tokenizer
def train_tokenizer(langs, dataset, vocab_size):
"""Train a tokenizer on given list of languages.
Reserves a special token for each language which is
[LANG] where LANG is the language tag. These are assigned
to tokens 5, 6, ..., len(langs) + 4.
"""
# Byte-pair encoding
tokenizer = Tokenizer(BPE(unk_token='[UNK]'))
# trainer
lang_tokens = ['[' + lang + ']' for lang in langs]
special_tokens = ['[MASK]', '[CLS]', '[SEP]', '[PAD]', '[UNK]'] + lang_tokens
trainer = BpeTrainer(
special_tokens=special_tokens,
vocab_size=vocab_size)
# normalise and pre tokenize
tokenizer.normalizer = Sequence([NFD(), Lowercase(), StripAccents()])
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel()
tokenizer.decoder = decoders.ByteLevel()
# create iterator and train
iterator = _MultilingualIterator(dataset, langs)
tokenizer.train_from_iterator(iterator, trainer)
# post process start/end tokens
tokenizer.post_processor = TemplateProcessing(
single="[CLS] $A [SEP]",
pair="[CLS] $A [SEP] $B:1 [SEP]:1",
special_tokens=[
("[CLS]", tokenizer.token_to_id("[CLS]")),
("[SEP]", tokenizer.token_to_id("[SEP]")),
], )
return tokenizer
def __init__(self,
vocab_file: Optional[str] = None,
merges_file: Optional[str] = None,
unk_token: str = "<unk>",
replacement: str = "▁",
add_prefix_space: bool = True,
dropout: Optional[float] = None):
if vocab_file is not None and merges_file is not None:
tokenizer = Tokenizer(
BPE.from_files(vocab_file,
merges_file,
dropout=dropout,
unk_token=unk_token))
else:
tokenizer = Tokenizer(BPE.empty())
tokenizer.add_special_tokens([unk_token])
tokenizer.normalizer = NFKC.new()
tokenizer.pre_tokenizer = pre_tokenizers.Metaspace.new(
replacement=replacement, add_prefix_space=add_prefix_space)
tokenizer.decoder = decoders.Metaspace.new(
replacement=replacement, add_prefix_space=add_prefix_space)
parameters = {
"model": "SentencePieceBPE",
"unk_token": unk_token,
"replacement": replacement,
"add_prefix_space": add_prefix_space,
"dropout": dropout,
}
super().__init__(tokenizer, parameters)
def __init__(
self,
vocab: Optional[Union[str, Dict[str, int]]] = None,
merges: Optional[Union[str, Dict[Tuple[int, int], Tuple[int,
int]]]] = None,
add_prefix_space: bool = False,
lowercase: bool = False,
dropout: Optional[float] = None,
unicode_normalizer: Optional[str] = None,
continuing_subword_prefix: Optional[str] = None,
end_of_word_suffix: Optional[str] = None,
trim_offsets: bool = False,
):
if vocab is not None and merges is not None:
tokenizer = Tokenizer(
BPE(
vocab,
merges,
dropout=dropout,
continuing_subword_prefix=continuing_subword_prefix or "",
end_of_word_suffix=end_of_word_suffix or "",
))
else:
tokenizer = Tokenizer(BPE())
# Check for Unicode normalization first (before everything else)
normalizers = []
if unicode_normalizer:
normalizers += [unicode_normalizer_from_str(unicode_normalizer)]
if lowercase:
normalizers += [Lowercase()]
# Create the normalizer structure
if len(normalizers) > 0:
if len(normalizers) > 1:
tokenizer.normalizer = Sequence(normalizers)
else:
tokenizer.normalizer = normalizers[0]
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(
add_prefix_space=add_prefix_space)
tokenizer.decoder = decoders.ByteLevel()
tokenizer.post_processor = processors.ByteLevel(
trim_offsets=trim_offsets)
parameters = {
"model": "ByteLevelBPE",
"add_prefix_space": add_prefix_space,
"lowercase": lowercase,
"dropout": dropout,
"unicode_normalizer": unicode_normalizer,
"continuing_subword_prefix": continuing_subword_prefix,
"end_of_word_suffix": end_of_word_suffix,
"trim_offsets": trim_offsets,
}
super().__init__(tokenizer, parameters)
def __init__(
self,
vocab_file: Optional[str] = None,
merges_file: Optional[str] = None,
unk_token: Union[str, AddedToken] = "<unk>",
replacement: str = "▁",
add_prefix_space: bool = True,
no_consecutive_space: bool = True,
dropout: Optional[float] = None,
clean_text: bool = True,
handle_chinese_chars: bool = True,
separate_numbers: bool = True,
strip_accents: bool = True,
lowercase: bool = True,
wordpieces_prefix: str = "##",
special_chars: str = SPECIAL_CHARS,
zh_norm: bool = True,
):
if vocab_file is not None and merges_file is not None:
tokenizer = Tokenizer(
BPE(vocab_file,
merges_file,
dropout=dropout,
unk_token=unk_token))
else:
tokenizer = Tokenizer(BPE())
if tokenizer.token_to_id(str(unk_token)) is not None:
tokenizer.add_special_tokens([str(unk_token)])
tokenizer.normalizer = Sequence([
NFKC(),
BertNormalizer(clean_text=clean_text,
handle_chinese_chars=handle_chinese_chars,
separate_numbers=separate_numbers,
strip_accents=strip_accents,
lowercase=lowercase,
special_chars=special_chars,
zh_norm=zh_norm)
])
tokenizer.pre_tokenizer = pre_tokenizers.Metaspace(
replacement=replacement,
add_prefix_space=add_prefix_space,
no_consecutive_space=no_consecutive_space)
tokenizer.decoder = decoders.Metaspace(
replacement=replacement,
add_prefix_space=add_prefix_space,
no_consecutive_space=no_consecutive_space)
parameters = {
"model": "SentencePieceBPE",
"unk_token": unk_token,
"replacement": replacement,
"add_prefix_space": add_prefix_space,
"no_consecutive_space": no_consecutive_space,
"dropout": dropout,
}
super().__init__(tokenizer, parameters)
def from_spm(filename: str):
try:
import sys
sys.path.append(".")
import sentencepiece_model_pb2 as model
except Exception:
raise Exception(
"You don't seem to have the required protobuf file, in order to use this function you need to run `pip install protobuf` and `wget https://raw.githubusercontent.com/google/sentencepiece/master/python/sentencepiece_model_pb2.py` for us to be able to read the intrinsics of your spm_file. `pip install sentencepiece` is not required."
)
m = model.ModelProto()
m.ParseFromString(open(filename, "rb").read())
precompiled_charsmap = m.normalizer_spec.precompiled_charsmap
vocab = [(piece.piece, piece.score) for piece in m.pieces]
unk_id = m.trainer_spec.unk_id
model_type = m.trainer_spec.model_type
if model_type != 1:
raise Exception(
"You're trying to run a `Unigram` model but you're file was trained with a different algorithm"
)
data = {"unk_id": unk_id, "vocab": vocab}
replacement = "▁"
add_prefix_space = True
out_vocab_filename = f"{filename}.json"
try:
with open(out_vocab_filename, "w") as f:
json.dump(data, f, indent=4)
tokenizer = Tokenizer(Unigram(out_vocab_filename))
finally:
os.remove(out_vocab_filename)
tokenizer.normalizer = normalizers.Precompiled(precompiled_charsmap)
tokenizer.pre_tokenizer = pre_tokenizers.Sequence(
[
pre_tokenizers.WhitespaceSplit(),
pre_tokenizers.Metaspace(
replacement=replacement, add_prefix_space=add_prefix_space
),
]
)
tokenizer.decoder = decoders.Metaspace(
replacement=replacement, add_prefix_space=add_prefix_space
)
parameters = {
"model": "SentencePieceUnigram",
}
obj = BaseTokenizer.__new__(SentencePieceUnigramTokenizer, tokenizer, parameters)
BaseTokenizer.__init__(obj, tokenizer, parameters)
return obj
def __init__(
self,
vocab_file: Optional[str] = None,
add_special_tokens: bool = True,
unk_token: str = "[UNK]",
sep_token: str = "[SEP]",
cls_token: str = "[CLS]",
clean_text: bool = True,
handle_chinese_chars: bool = True,
strip_accents: bool = True,
lowercase: bool = True,
wordpieces_prefix: str = "##",
):
if vocab_file is not None:
tokenizer = Tokenizer(
WordPiece.from_files(vocab_file, unk_token=unk_token))
else:
tokenizer = Tokenizer(WordPiece.empty())
tokenizer.add_special_tokens([unk_token, sep_token, cls_token])
tokenizer.normalizer = BertNormalizer(
clean_text=clean_text,
handle_chinese_chars=handle_chinese_chars,
strip_accents=strip_accents,
lowercase=lowercase,
)
tokenizer.pre_tokenizer = BertPreTokenizer()
if add_special_tokens and vocab_file is not None:
sep_token_id = tokenizer.token_to_id(sep_token)
if sep_token_id is None:
raise TypeError("sep_token not found in the vocabulary")
cls_token_id = tokenizer.token_to_id(cls_token)
if cls_token_id is None:
raise TypeError("cls_token not found in the vocabulary")
tokenizer.post_processor = BertProcessing(
(sep_token, sep_token_id), (cls_token, cls_token_id))
tokenizer.decoder = decoders.WordPiece(prefix=wordpieces_prefix)
parameters = {
"model": "BertWordPiece",
"add_special_tokens": add_special_tokens,
"unk_token": unk_token,
"sep_token": sep_token,
"cls_token": cls_token,
"clean_text": clean_text,
"handle_chinese_chars": handle_chinese_chars,
"strip_accents": strip_accents,
"lowercase": lowercase,
"wordpieces_prefix": wordpieces_prefix,
}
super().__init__(tokenizer, parameters)
def __init__(
self,
vocab_file: Optional[str] = None,
merges_file: Optional[str] = None,
add_prefix_space: bool = False,
lowercase: bool = False,
dropout: Optional[float] = None,
unicode_normalizer: Optional[str] = None,
continuing_subword_prefix: Optional[str] = None,
end_of_word_suffix: Optional[str] = None,
):
if vocab_file is not None and merges_file is not None:
tokenizer = Tokenizer(
BPE.from_files(
vocab_file,
merges_file,
dropout=dropout,
continuing_subword_prefix=continuing_subword_prefix or "",
end_of_word_suffix=end_of_word_suffix or "",
))
else:
tokenizer = Tokenizer(BPE.empty())
# Check for Unicode normalization first (before everything else)
normalizers = []
if unicode_normalizer:
normalizers += [unicode_normalizer_from_str(unicode_normalizer)]
if lowercase:
normalizers += [Lowercase()]
# Create the normalizer structure
if len(normalizers) > 0:
if len(normalizers) > 1:
tokenizer.normalizer = Sequence(normalizers)
else:
tokenizer.normalizer = normalizers[0]
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(
add_prefix_space=add_prefix_space)
tokenizer.decoder = decoders.ByteLevel()
parameters = {
"model": "ByteLevelBPE",
"add_prefix_space": add_prefix_space,
"lowercase": lowercase,
"dropout": dropout,
"unicode_normalizer": unicode_normalizer,
"continuing_subword_prefix": continuing_subword_prefix,
"end_of_word_suffix": end_of_word_suffix,
}
super().__init__(tokenizer, parameters)
def setup_tokenizer(_):
# Initialize a tokenizer
tokenizer = Tokenizer(models.BPE())
# Customize pre-tokenization and decoding
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(add_prefix_space=True)
tokenizer.decoder = decoders.ByteLevel()
tokenizer.post_processor = processors.ByteLevel(trim_offsets=True)
normalizers = [NFKC()]
tokenizer.normalizer = Sequence(normalizers)
return tokenizer
def tokenizer_pipeline():
"""
specific pipeline for Cebuano Corpus tokenization
- Uses a Byte pair encoding (BPE) tokenizer
"""
tokenizer = Tokenizer(BPE())
# string normalization
tokenizer.normalizer = Sequence([NFD(), StripAccents(), Lowercase()])
tokenizer.pre_tokenizer = ByteLevel()
tokenizer.decoder = ByteLevelDecoder()
return tokenizer
def load_or_train_tokenizer(file_paths, tokenizer_mode_path):
'''
Tries to load saved text tokenizer
If there is none, trains the new tokenizer and saves is
'''
if not os.path.exists(tokenizer_mode_path):
print('Tokenizer model not found, training one')
from tokenizers.models import BPE
from tokenizers import Tokenizer
from tokenizers.decoders import ByteLevel as ByteLevelDecoder
from tokenizers.normalizers import NFKC, Sequence
from tokenizers.pre_tokenizers import ByteLevel
from tokenizers.trainers import BpeTrainer
tokenizer = Tokenizer(BPE())
tokenizer.normalizer = Sequence([
NFKC()
])
tokenizer.pre_tokenizer = ByteLevel()
tokenizer.decoder = ByteLevelDecoder()
trainer = BpeTrainer(
vocab_size=50000,
show_progress=True,
inital_alphabet=ByteLevel.alphabet(),
special_tokens=[
"<s>",
"<pad>",
"</s>",
"<unk>",
"<mask>"
]
)
tokenizer.train(file_paths, trainer)
if not os.path.exists(tokenizer_mode_path):
os.makedirs(tokenizer_mode_path)
tokenizer.model.save(tokenizer_mode_path, None)
print('Loading trained tokenizer model')
tokenizer = GPT2Tokenizer.from_pretrained(tokenizer_mode_path)
tokenizer.add_special_tokens({
'eos_token': '</s>',
'bos_token': '<s>',
'unk_token': '<unk>',
'pad_token': '<pad>',
'mask_token': '<mask>'
})
return tokenizer
def tokenize_corpus(
input_file: str,
output_file: str,
vocab_file: str,
unk_token: str = '<unk>',
control_tokens: List[str] = []):
r"""Tokenize corpus sentences through trained **WordPiece** model.
Arguments:
input_file (str): Input corpus file path.
output_file (str): Output file path.
vocab_file (str): Trained vocabulary file path.
unk_token (str): Unknown token in the vocabulary.
control_tokens (list): Control tokens in the vocabulary.
"""
# Create `WordPiece` model and add special tokens. Note that `unk_token`
# is also a special token.normalizer and pre-tokenizer.
tokenizer = Tokenizer(models.WordPiece(vocab_file, unk_token=unk_token))
tokenizer.add_special_tokens([unk_token] + control_tokens)
# Use BERT-specific normalizer, pre-tokenizer and **WordPiece** decoder.
tokenizer.normalizer = BertNormalizer(strip_accents=False)
tokenizer.pre_tokenizer = BertPreTokenizer()
tokenizer.decoder = decoders.WordPiece(prefix='##')
with open(input_file, 'r', encoding='utf-8') as src, \
open(output_file, 'w', encoding='utf-8') as dst:
# Count total lines in corpus.
total_lines = 0
for _ in src:
total_lines += 1
# Move the corpus file to first.
src.seek(0)
buffer = []
for line in tqdm.tqdm(src,
desc='[*] tokenize corpus',
total=total_lines):
buffer.append(line)
# Tokenize buffered sentences and write to `output_file`.
if len(buffer) > 10000:
for t in tokenizer.encode_batch(buffer):
dst.write(' '.join(t.tokens) + '\n')
buffer.clear()
# Process the remained buffer.
if buffer:
for t in tokenizer.encode_batch(buffer):
dst.write(' '.join(t.tokens) + '\n')
def get_tokenizer(self, tokenizer_dir):
tokenizer = Tokenizer(models.BPE())
tokenizer.normalizer = Sequence(
[NFKC(), Replace('\r', ''),
Replace('\n', ' ')])
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel()
tokenizer.decoder = decoders.ByteLevel()
vocab_fn = os.path.join(tokenizer_dir, 'vocab.json')
merge_fn = os.path.join(tokenizer_dir, 'merges.txt')
tokenizer.model = models.BPE.from_file(vocab_fn, merge_fn)
tokenizer.add_special_tokens(['[UNK]', '[PAD]', '[BOS]', '[EOS]'])
return tokenizer
def converted(self) -> Tokenizer:
vocab = self.original_tokenizer.vocab
tokenizer = Tokenizer(
WordPiece(vocab, unk_token=str(self.original_tokenizer.unk_token)))
# # Let the tokenizer know about special tokens if they are part of the vocab
# if tokenizer.token_to_id(str(self.original_tokenizer.unk_token)) is not None:
# tokenizer.add_special_tokens([str(self.original_tokenizer.unk_token)])
# if tokenizer.token_to_id(str(self.original_tokenizer.sep_token)) is not None:
# tokenizer.add_special_tokens([str(self.original_tokenizer.sep_token)])
# if tokenizer.token_to_id(str(self.original_tokenizer.cls_token)) is not None:
# tokenizer.add_special_tokens([str(self.original_tokenizer.cls_token)])
# if tokenizer.token_to_id(str(self.original_tokenizer.pad_token)) is not None:
# tokenizer.add_special_tokens([str(self.original_tokenizer.pad_token)])
# if tokenizer.token_to_id(str(self.original_tokenizer.mask_token)) is not None:
# tokenizer.add_special_tokens([str(self.original_tokenizer.mask_token)])
tokenize_chinese_chars = False
strip_accents = False
do_lower_case = False
if hasattr(self.original_tokenizer, "basic_tokenizer"):
tokenize_chinese_chars = self.original_tokenizer.basic_tokenizer.tokenize_chinese_chars
strip_accents = self.original_tokenizer.basic_tokenizer.strip_accents
do_lower_case = self.original_tokenizer.basic_tokenizer.do_lower_case
tokenizer.normalizer = normalizers.BertNormalizer(
clean_text=True,
handle_chinese_chars=tokenize_chinese_chars,
strip_accents=strip_accents,
lowercase=do_lower_case,
)
tokenizer.pre_tokenizer = pre_tokenizers.BertPreTokenizer()
cls = str(self.original_tokenizer.cls_token)
sep = str(self.original_tokenizer.sep_token)
cls_token_id = self.original_tokenizer.cls_token_id
sep_token_id = self.original_tokenizer.sep_token_id
tokenizer.post_processor = processors.TemplateProcessing(
single=
f"{cls}:2 $A:0 {sep}:0", # token_type_id is 2 for Funnel transformer
pair=f"{cls}:2 $A:0 {sep}:0 $B:1 {sep}:1",
special_tokens=[
(cls, cls_token_id),
(sep, sep_token_id),
],
)
tokenizer.decoder = decoders.WordPiece(prefix="##")
return tokenizer
def __init__(
self,
replacement: str = "▁",
add_prefix_space: bool = True,
unk_token: Union[str, AddedToken] = "<unk>",
eos_token: Union[str, AddedToken] = "</s>",
pad_token: Union[str, AddedToken] = "<pad>",
):
self.special_tokens = {
"pad": {"id": 0, "token": pad_token},
"eos": {"id": 1, "token": eos_token},
"unk": {"id": 2, "token": unk_token},
}
self.special_tokens_list = [None] * len(self.special_tokens)
for token_dict in self.special_tokens.values():
self.special_tokens_list[token_dict["id"]] = token_dict["token"]
tokenizer = Tokenizer(Unigram())
tokenizer.normalizer = normalizers.Sequence(
[
normalizers.Nmt(),
normalizers.NFKC(),
normalizers.Replace(Regex(" {2,}"), " "),
normalizers.Lowercase(),
]
)
tokenizer.pre_tokenizer = pre_tokenizers.Sequence(
[
pre_tokenizers.Metaspace(replacement=replacement, add_prefix_space=add_prefix_space),
pre_tokenizers.Digits(individual_digits=True),
pre_tokenizers.Punctuation(),
]
)
tokenizer.decoder = decoders.Metaspace(replacement=replacement, add_prefix_space=add_prefix_space)
tokenizer.post_processor = TemplateProcessing(
single=f"$A {self.special_tokens['eos']['token']}",
special_tokens=[(self.special_tokens["eos"]["token"], self.special_tokens["eos"]["id"])],
)
parameters = {
"model": "SentencePieceUnigram",
"replacement": replacement,
"add_prefix_space": add_prefix_space,
}
super().__init__(tokenizer, parameters)
def converted(self) -> Tokenizer:
vocab = self.original_tokenizer.vocab
tokenizer = Tokenizer(
WordPiece(vocab, unk_token=str(self.original_tokenizer.unk_token)))
tokenize_chinese_chars = False
strip_accents = False
do_lower_case = False
if hasattr(self.original_tokenizer, "basic_tokenizer"):
tokenize_chinese_chars = self.original_tokenizer.basic_tokenizer.tokenize_chinese_chars
strip_accents = self.original_tokenizer.basic_tokenizer.strip_accents
do_lower_case = self.original_tokenizer.basic_tokenizer.do_lower_case
tokenizer.normalizer = normalizers.BertNormalizer(
clean_text=True,
handle_chinese_chars=tokenize_chinese_chars,
strip_accents=strip_accents,
lowercase=do_lower_case,
)
tokenizer.pre_tokenizer = pre_tokenizers.BertPreTokenizer()
cls = str(self.original_tokenizer.cls_token)
sep = str(self.original_tokenizer.sep_token)
question = str(self.original_tokenizer.question_token)
dot = "."
cls_token_id = self.original_tokenizer.cls_token_id
sep_token_id = self.original_tokenizer.sep_token_id
question_token_id = self.original_tokenizer.question_token_id
dot_token_id = self.original_tokenizer.convert_tokens_to_ids(".")
if self.original_tokenizer.padding_side == "right":
pair = f"{cls}:0 $A:0 {question} {dot} {sep}:0 $B:1 {sep}:1"
else:
pair = f"{cls}:0 $A:0 {sep}:0 $B:1 {question} {dot} {sep}:1"
tokenizer.post_processor = processors.TemplateProcessing(
single=f"{cls}:0 $A:0 {sep}:0",
pair=pair,
special_tokens=[
(cls, cls_token_id),
(sep, sep_token_id),
(question, question_token_id),
(dot, dot_token_id),
],
)
tokenizer.decoder = decoders.WordPiece(prefix="##")
return tokenizer
def __init__(
self,
vocab_file: Optional[str] = None,
merges_file: Optional[str] = None,
unk_token: Optional[str] = "<unk>",
suffix: Optional[str] = "</w>",
dropout: Optional[float] = None,
unicode_normalizer: Optional[str] = None,
):
if vocab_file is not None and merges_file is not None:
tokenizer = Tokenizer(
BPE.from_files(vocab_file,
merges_file,
dropout=dropout,
unk_token=unk_token,
end_of_word_suffix=suffix))
else:
tokenizer = Tokenizer(BPE.empty())
# Check for Unicode normalization first (before everything else)
normalizers = []
if unicode_normalizer:
normalizers += [unicode_normalizer_from_str(unicode_normalizer)]
# OpenAI normalization is the same as Bert
normalizers += [BertNormalizer()]
# Create the normalizer structure
if len(normalizers) > 0:
if len(normalizers) > 1:
tokenizer.normalizer = Sequence(normalizers)
else:
tokenizer.normalizer = normalizers[0]
tokenizer.pre_tokenizer = BertPreTokenizer()
tokenizer.decoder = BPEDecoder(suffix=suffix)
parameters = {
"model": "BPE",
"unk_token": unk_token,
"suffix": suffix,
"dropout": dropout,
}
super().__init__(tokenizer, parameters)
def get_tokenizer_trainer():
# START init_tokenizer_trainer
from tokenizers import Tokenizer, models, normalizers, pre_tokenizers, decoders, trainers
tokenizer = Tokenizer(models.Unigram())
tokenizer.normalizer = normalizers.NFKC()
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel()
tokenizer.decoder = decoders.ByteLevel()
trainer = trainers.UnigramTrainer(
vocab_size=20000,
initial_alphabet=pre_tokenizers.ByteLevel.alphabet(),
special_tokens=["<PAD>", "<BOS>", "<EOS>"],
)
# END init_tokenizer_trainer
trainer.show_progress = False
return tokenizer, trainer
def create_tokenizer(sentence_list):
filename = f'temp_{time.strftime("%Y%m%d-%H%M%S")}.txt'
with open(filename, 'w') as f:
for s in sentence_list:
f.write(f'{s}\n')
tokenizer = Tokenizer(WordPiece())
tokenizer.pre_tokenizer = Whitespace()
tokenizer.decoder = decoders.WordPiece()
tokenizer.enable_padding(pad_token='[PAD]', pad_id=0)
trainer = WordPieceTrainer(
vocab_size=3000, special_tokens=['[PAD]', '[S]', '[/S]', '[UNK]'])
tokenizer.train(trainer, [filename])
os.remove(filename)
return tokenizer
def __init__(
self,
vocab: Union[str, List],
merges: Union[str, None],
unk_token: str = "<unk>",
replacement: str = "▁",
add_prefix_space: bool = True,
dropout: Optional[float] = None,
normalize: bool = True,
):
if merges:
n_model = "BPE"
tokenizer = Tokenizer(
BPE(
vocab, # type: ignore
merges,
unk_token=unk_token,
fuse_unk=True,
))
else:
n_model = "Unigram"
tokenizer = Tokenizer(Unigram(vocab, 1)) # type: ignore
if normalize:
tokenizer.normalizer = NFKC()
tokenizer.pre_tokenizer = pre_tokenizers.Metaspace(
replacement=replacement,
add_prefix_space=add_prefix_space,
)
tokenizer.decoder = decoders.Metaspace(
replacement=replacement,
add_prefix_space=add_prefix_space,
)
parameters = {
"model": f"SentencePiece{n_model}",
"unk_token": unk_token,
"replacement": replacement,
"add_prefix_space": add_prefix_space,
"dropout": dropout,
}
super().__init__(tokenizer, parameters)
def converted(self) -> Tokenizer:
vocab = self.original_tokenizer.encoder
merges = list(self.original_tokenizer.bpe_ranks.keys())
unk_token = self.original_tokenizer.unk_token
tokenizer = Tokenizer(
BPE(
vocab=vocab,
merges=merges,
dropout=None,
continuing_subword_prefix="",
end_of_word_suffix="</w>",
fuse_unk=False,
unk_token=str(unk_token),
))
tokenizer.normalizer = normalizers.Sequence([
normalizers.NFC(),
normalizers.Replace(Regex(r"\s+"), " "),
normalizers.Lowercase()
])
tokenizer.pre_tokenizer = pre_tokenizers.Sequence([
pre_tokenizers.Split(
Regex(
r"""'s|'t|'re|'ve|'m|'ll|'d|[\p{L}]+|[\p{N}]|[^\s\p{L}\p{N}]+"""
),
behavior="removed",
invert=True,
),
pre_tokenizers.ByteLevel(add_prefix_space=False),
])
tokenizer.decoder = decoders.ByteLevel()
# Hack to have a ByteLevel and TemplaceProcessor
tokenizer.post_processor = processors.RobertaProcessing(
sep=(self.original_tokenizer.eos_token,
self.original_tokenizer.eos_token_id),
cls=(self.original_tokenizer.bos_token,
self.original_tokenizer.bos_token_id),
add_prefix_space=False,
trim_offsets=False,
)
return tokenizer
def __init__(self,
vocab_file: Optional[str]=None,
merges_file: Optional[str]=None,
add_prefix_space: bool=False):
if vocab_file is not None and merges_file is not None:
tokenizer = Tokenizer(BPE.from_files(vocab_file, merges_file))
else:
tokenizer = Tokenizer(BPE.empty())
tokenizer.normalizer = NFKC.new()
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel.new(add_prefix_space=add_prefix_space)
tokenizer.decoder = decoders.ByteLevel.new()
parameters = {
"model": "ByteLevelBPE",
"add_prefix_space": add_prefix_space,
}
super().__init__(tokenizer, parameters)