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 normalizer(self, proto):
normalizers = [Replace("``", '"'), Replace("''", '"')]
if not self.original_tokenizer.keep_accents:
normalizers.append(NFKD())
normalizers.append(StripAccents())
if self.original_tokenizer.do_lower_case:
normalizers.append(Lowercase())
precompiled_charsmap = proto.normalizer_spec.precompiled_charsmap
normalizers.append(Precompiled(precompiled_charsmap))
normalizers.append(Replace(Regex(" {2,}"), " "))
return Sequence(normalizers)
def normalizer(self, proto):
list_normalizers = []
if self.original_tokenizer.do_lower_case:
list_normalizers.append(normalizers.Lowercase())
list_normalizers.append(normalizers.Strip())
precompiled_charsmap = proto.normalizer_spec.precompiled_charsmap
if precompiled_charsmap:
list_normalizers.append(
normalizers.Precompiled(precompiled_charsmap))
list_normalizers.append(normalizers.Replace(Regex(" {2,}"), " "))
return normalizers.Sequence(list_normalizers)
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 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/src/sentencepiece/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"
)
replacement = "▁"
add_prefix_space = True
tokenizer = Tokenizer(Unigram(vocab, unk_id))
tokenizer.normalizer = normalizers.Sequence(
[
normalizers.Precompiled(precompiled_charsmap),
normalizers.Replace(Regex(" {2,}"), " "),
]
)
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": "SentencePieceUnigram",
}
obj = BaseTokenizer.__new__(SentencePieceUnigramTokenizer, tokenizer, parameters)
BaseTokenizer.__init__(obj, tokenizer, parameters)
return obj
def normalizer(self, proto):
list_normalizers = [
normalizers.Replace("``", '"'),
normalizers.Replace("''", '"'),
]
if not self.original_tokenizer.keep_accents:
list_normalizers.append(normalizers.NFKD())
list_normalizers.append(normalizers.StripAccents())
if self.original_tokenizer.do_lower_case:
list_normalizers.append(normalizers.Lowercase())
precompiled_charsmap = proto.normalizer_spec.precompiled_charsmap
list_normalizers.append(normalizers.Precompiled(precompiled_charsmap))
list_normalizers.append(normalizers.Replace(Regex(" {2,}"), " "))
return normalizers.Sequence(list_normalizers)
def normalizer(self, proto):
precompiled_charsmap = proto.normalizer_spec.precompiled_charsmap
return Sequence(
[Precompiled(precompiled_charsmap),
Replace(Regex(" {2,}"), " ")])
from itertools import product
from tokenizers import Tokenizer, Regex
from tokenizers.models import WordLevel
from tokenizers.normalizers import NFD
from tokenizers.trainers import WordLevelTrainer
from tokenizers.pre_tokenizers import Split, WhitespaceSplit
## Loop which creates and loads the tokenizer
def stackTraceTokenizer(tokens:tuple, events:tuple, vocab_size=2_000, min_freq=3):
for norm, event in product(tokens, events):
print(norm, event)
tokenizer = Tokenizer(WordLevel(unk_token="[UNK]"))
tokenizer.normalizer = NFD()
if norm == 'white':
tokenizer.pre_tokenizer = WhitespaceSplit()
else:
tokenizer.pre_tokenizer = Split(pattern=Regex("[A-Z]+[a-z0-9]+|[.A-Z]+|[a-z0-9]+"),
behavior='isolated')
trainer = WordLevelTrainer(vocab_size=vocab_size, min_frequency=min_freq,
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"])
tokenizer.train(["vocab-{}.txt".format(event)], trainer)
print(f"Trained tokenizer for {norm}:{event}")
yield tokenizer, event, norm
