def train_tokenizer(args):
"""[summary]
Arguments:
args {[dictionary]} -- [arguments객체]
"""
# Tokenizer train
morpheme_func = None
if args.tokenizer.pretokenizer_type == "khaiii":
api = KhaiiiApi()
morpheme_func = api.analyze
elif args.tokenizer.pretokenizer_type == "mecab":
mecab = Mecab()
morpheme_func = mecab.morphs
# tokenizer-type", type=str, choices=["bbpe", "cbpe", "wp"], default="bbpe"
if args.tokenizer.tokenizer_type == "bbpe":
# tokenizer = BytelevelBPETokenizer()
tokenizer = Tokenizer(BPE())
# tokenizer.pre_tokenizer = BertPreTokenizer()
trainer = BpeTrainer(
special_tokens=omegalist_to_list(args.tokenizer.special_tokens),
vocab_size=args.tokenizer.vocab_size,
min_frequency=args.tokenizer.min_frequency,
)
elif args.tokenizer.tokenizer_type == "cbpe":
tokenizer = Tokenizer(BPE())
tokenizer.pre_tokenizer = CharDelimiterSplit
trainer = BpeTrainer(
special_tokens=omegalist_to_list(args.tokenizer.special_tokens),
vocab_size=args.tokenizer.vocab_size,
min_frequency=args.tokenizer.min_frequency,
)
elif args.tokenizer.tokenizer_type == "wp":
tokenizer = Tokenizer(WordPiece())
# tokenizer.pre_tokenizer = Whitespace
trainer = WordPieceTrainer(
special_tokens=omegalist_to_list(args.tokenizer.special_tokens),
vocab_size=args.tokenizer.vocab_size,
min_frequency=args.tokenizer.min_frequency,
)
tokenizer.train_from_iterator(get_pretokenize_generator(morpheme_func))
tokenizer.save(f"../vocab/{args.tokenizer.tokenizer_type}.vocab")
test_string = "안녕하세요 이것은 테스트입니다. 구름은 하늘에 떠 있고 우리는 여기있어"
output = tokenizer.encode(test_string)
print(f"output:{output}")
print(f"tokens:{output.tokens}")
print(f"ids :{output.ids}")
print(f"offset:{output.offsets}")
print(f"decode:{tokenizer.decode(output.ids)}")
datasets = get_datasets(args.tokenizer.data_path)
for line in datasets:
print(line)
break
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_size=25000,
min_freq=5,
lang="en",
files=[None, None]) -> None:
"""
Args:
vocab_size: (int)
min_freq: minimum frequency
lang:
files: (List[str]) ["vocab.json", "merge.txt"]
"""
super(BPETokenizer, self).__init__()
self.tokenizer = Tokenizer(BPE(files[0], files[1]))
self.lang = lang
self.trainer = BpeTrainer(vocab_size=vocab_size,
min_frequency=min_freq,
special_tokens=["[PAD]", "[SEP]"],
initial_alphabet=ByteLevel.alphabet())
# https://huggingface.co/docs/tokenizers/python/latest/components.html#normalizers
self.tokenizer.normalizer = Sequence([NFKC(), Lowercase()])
# https://huggingface.co/docs/tokenizers/python/latest/components.html#pre-tokenizers
self.tokenizer.pre_tokenizer = ByteLevel()
self.tokenizer.decoder = ByteLevelDecoder()
def bpe_train(self, paths):
trainer = BpeTrainer(
vocab_size=50000,
show_progress=True,
inital_alphabet=ByteLevel.alphabet(),
special_tokens=["<s>", "<pad>", "</s>", "<unk>", "<mask>"])
self.tokenizer.train(paths, trainer)
def train(
self,
files: Union[str, List[str]],
vocab_size: int = 30000,
min_frequency: int = 2,
special_tokens: List[str] = ["<unk>"],
limit_alphabet: int = 1000,
initial_alphabet: List[str] = [],
suffix: Optional[str] = "</w>",
show_progress: bool = True,
):
""" Train the model using the given files """
trainer = BpeTrainer(
vocab_size=vocab_size,
min_frequency=min_frequency,
special_tokens=special_tokens,
limit_alphabet=limit_alphabet,
initial_alphabet=initial_alphabet,
end_of_word_suffix=suffix,
show_progress=show_progress,
)
if isinstance(files, str):
files = [files]
self._tokenizer.train(trainer, files)
def __init__(
self,
load_from: str = None,
vocab_size: int = 10000,
max_example_len: int = 128,
batch_size: int = 16,
num_stopwords: int = 250,
mask_output_len: int = 4,
):
self.char_dict: Dict[str, int] = {}
self.char_rev: Dict[int, str] = {}
self.token_dict: Dict[str, int] = {}
self.token_rev: Dict[str, int] = {}
self.vocab_size = vocab_size
self.max_example_len = max_example_len
self.batch_size = batch_size
self.num_stopwords = num_stopwords
self.mask_output_len = mask_output_len
self.tokenizer_fit = False
self.tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
self.tokenizer.pre_tokenizer = Whitespace()
self.tokenizer.normalizer = Sequence(
[NFD(), Lowercase(), StripAccents()])
self.tok_trainer = BpeTrainer(special_tokens=["[UNK]", "[MASK]"],
vocab_size=self.vocab_size)
if load_from:
self._load(load_from)
def train_tokenizer(lang, dataset, vocab_size):
# Byte-pair encoding
tokenizer = Tokenizer(BPE(unk_token='[UNK]'))
# trainer
trainer = BpeTrainer(
special_tokens=['[MASK]', '[CLS]', '[SEP]', '[PAD]', '[UNK]'],
vocab_size=vocab_size)
# pre tokenizer with whitespace
tokenizer.pre_tokenizer = Whitespace()
# train
tokenizer.train_from_iterator(dataset[lang], 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 prepare_trainer(self):
return BpeTrainer(vocab_size=30000, show_progress=True, min_frequency=2, special_tokens=[
"<s>",
"<pad>",
"</s>",
"<unk>",
"<mask>",
])
def generate_tokenizer(equations, output, vocab_size):
from tokenizers import Tokenizer, pre_tokenizers
from tokenizers.models import BPE
from tokenizers.trainers import BpeTrainer
tokenizer = Tokenizer(BPE())
tokenizer.pre_tokenizer = pre_tokenizers.ByteLevel(add_prefix_space=False)
trainer = BpeTrainer(special_tokens=["[PAD]", "[BOS]", "[EOS]"], vocab_size=vocab_size, show_progress=True)
tokenizer.train(trainer, equations)
tokenizer.save(path=output, pretty=False)
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 bpe_train(self, paths):
trainer = BpeTrainer(vocab_size=50000,
show_progress=True,
inital_alphabet=ByteLevel.alphabet(),
special_tokens=[
"<s>",
"<pad>",
"</s>",
"<unk>",
"<mask>",
"<company>",
"<label>",
"<category>",
"<review>",
])
self.tokenizer.train(trainer, paths)
def create_train_bpe_tokenizer(
bpe_vocab_size,
asr_text_filepath='asr.txt',
ttx_text_filepath='ttx.txt',
save_tokenizer=True,
tokenizer_filename=".\\data\\tokenizer-test.json"):
tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
trainer = BpeTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"],
vocab_size=bpe_vocab_size)
tokenizer.pre_tokenizer = Whitespace()
files = [asr_text_filepath, ttx_text_filepath]
files = [file for file in files if file] # Get rid of None's
tokenizer.train(files, trainer)
if save_tokenizer:
tokenizer.save(tokenizer_filename)
return tokenizer
def get_tokenizer_trainer():
# START init_tokenizer
from tokenizers import Tokenizer
from tokenizers.models import BPE
tokenizer = Tokenizer(BPE())
# END init_tokenizer
# START init_trainer
from tokenizers.trainers import BpeTrainer
trainer = BpeTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"])
# END init_trainer
# START init_pretok
from tokenizers.pre_tokenizers import Whitespace
tokenizer.pre_tokenizer = Whitespace()
# END init_pretok
return tokenizer, trainer
def build_new_vocab():
tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
trainer = BpeTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"])
tokenizer.pre_tokenizer = Whitespace()
# files = [f"/daintlab/home/moo/NLU/biobert-pytorch/datasets/QA/BioASQ/BioASQ-{split}-factoid-7b.json" for split in ["train_split", "dev"]]
files = "/daintlab/home/moo/NLU/biobert-pytorch/datasets/QA/BioASQ/BioASQ-train-factoid-7b.json"
with open(files) as f:
file = json.load(f)
contexts = []
for question in file['data']:
for paragraph in question['paragraphs']:
contexts.append(paragraph['context'])
tokenizer.train_from_iterator(contexts, trainer)
additional_vocab = [k for k, v in tokenizer.get_vocab().items()]
tokenizer.save("tokenizer/tokenizer-bioasq.json")
return additional_vocab
from tokenizers import Tokenizer
from tokenizers.decoders import ByteLevel as ByteLevelDecoder
from tokenizers.models import BPE
from tokenizers.normalizers import Lowercase, NFKC, Sequence
from tokenizers.pre_tokenizers import ByteLevel
from tokenizers.trainers import BpeTrainer
path_data = "../../ml-datasets/wmt14/tokenizer/"
path_train_src = "../../ml-datasets/wmt14/train.en"
path_train_tgt = "../../ml-datasets/wmt14/train.de"
tokenizer = Tokenizer(BPE())
tokenizer.normalizer = Sequence([
NFKC(),
Lowercase()
])
tokenizer.pre_tokenizer = ByteLevel()
tokenizer.decoder = ByteLevelDecoder()
trainer = BpeTrainer(vocab_size=25000, show_progress=True, initial_alphabet=ByteLevel.alphabet(),
min_frequency=2, special_tokens=["<pad>", "<s>", "</s>", "<unk>", "<mask>", ])
tokenizer.train(trainer, [path_train_src, path_train_tgt])
print("Trained vocab size: {}".format(tokenizer.get_vocab_size()))
tokenizer.model.save(path_data)
tokenizer = Tokenizer(BPE())
# string normalization
tokenizer.normalizer = Sequence([NFD(), StripAccents(), Lowercase()])
tokenizer.pre_tokenizer = ByteLevel()
tokenizer.decoder = ByteLevelDecoder()
return tokenizer
if __name__ == "__main__":
# preparing corpus for wiki
en_vocab_size = 50257
wiki_txt = load_text_file_json('text/AA/wiki_00.json', 'text')
write_text_file(wiki_txt, 'wiki-corpus.txt')
corpus_files = {
'wiki-corpus': 'wiki-corpus.txt',
'oscar-corpus': 'shuff-dedup/ceb/ceb_dedup.txt'
}
# define a trainer for the tokenizer
trainer = BpeTrainer(vocab_size=en_vocab_size,
show_progress=True,
initial_alphabet=ByteLevel.alphabet(),
special_tokens=['<|endoftext|>', '<pad>'])
for corpus, path in corpus_files.items():
tokenizer = tokenizer_pipeline()
tokenizer.train([path], trainer)
tokenizer.save(f'model/{corpus}-tokenizer.json')
from tokenizers import Tokenizer
from tokenizers.models import BPE
from tokenizers.trainers import BpeTrainer
#from tokenizers.pre_tokenizers import Whitespace
tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
trainer = BpeTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"])
#tokenizer.pre_tokenizer = Whitespace()
files = ['./processed/processed_wiki_ko.txt']
tokenizer.train(files, trainer)
tokenizer.save("wiki_tokenizer.json")
tokenizer.pre_tokenizer = Sequence([Whitespace(), Digits()])
tokenizer.normalizer = Lowercase()
trainer = WordPieceTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"],
vocab_size=70000)
tokenizer.train_from_iterator(item_names, trainer)
tokenizer.save(os.path.join(OUTPUT_PATH, 'wordpiece_70k.json'))
# BPE tokenizer
tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
tokenizer.pre_tokenizer = Sequence([Whitespace(), Digits()])
tokenizer.normalizer = Lowercase()
trainer = BpeTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"],
vocab_size=60000)
tokenizer.train_from_iterator(item_names, trainer)
tokenizer.save(os.path.join(OUTPUT_PATH, 'bpe_60k.json'))
# Unigram tokenizer
tokenizer = Tokenizer(Unigram())
tokenizer.pre_tokenizer = Sequence([Whitespace(), Digits()])
tokenizer.normalizer = Lowercase()
trainer = UnigramTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"],
vocab_size=50000)
tokenizer.train_from_iterator(item_names, trainer)
tokenizer.save(os.path.join(OUTPUT_PATH, 'unigram_50k.json'))
tokenizer.normalizer = Sequence([NFKC(), Lowercase()])
# Our tokenizer also needs a pre-tokenizer responsible for converting the input
# to a ByteLevel representation.
tokenizer.pre_tokenizer = ByteLevel()
# And finally, let's plug a decoder so we can recover from a tokenized input
# to the original one
tokenizer.decoder = ByteLevelDecoder()
from tokenizers.trainers import BpeTrainer
# We initialize our trainer, giving him the details about the vocabulary we want
# to generate
trainer = BpeTrainer(vocab_size=25000,
show_progress=True,
initial_alphabet=ByteLevel.alphabet())
tokenizer.train(trainer,
["/Volumes/750GB-HDD/root/Question-Answering/pyData/big.txt"])
print("Trained vocab size: {}".format(tokenizer.get_vocab_size()))
# Et voilà ! You trained your very first tokenizer from scratch using tokenizers.
# Of course, this covers only the basics, and you may want to have a look at the
# add_special_tokens or special_tokens parameters on the Trainer class, but the
# overall process should be very similar.
# You will see the generated files in the output.
tokenizer.model.save('/Volumes/750GB-HDD/root/Question-Answering/pyData')
def train_tokenizer_vocab(dataset, style='BPE', force_retrain=True):
"""
if force_retrain: overwrite the stored tokenizer from tokenizers dir (by retraining)
else: load the tokenizer if it exists
"""
assert dataset in VALID_DATASETS
assert style in VALID_TOKENIZATIONS
tpath_expected = default_tpath(dataset, style)
train = True
if not force_retrain and os.path.isfile(tpath_expected):
tokenizer = Tokenizer.from_file(tpath_expected)
train = False
else:
print('%s tokenizer file does not exist; training new tokenizer' %
tpath_expected)
if train:
# load data associated with one of the valid datasets (from /data/ directory)
datafiles = load_dataset(dataset)
# Steps for each algo (e.g. BPE):
# - init Tokenizer using algo
# - specify algo specific trainer
# - specify any pre-processing of text (will affect decoding)
# see: https://huggingface.co/docs/tokenizers/python/latest/components.html#decoders
# - different training calls if its the arxiv dataset or wikitext
# see https://blog.einstein.ai/the-wikitext-long-term-dependency-language-modeling-dataset/
if style == 'BPE':
tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
trainer = BpeTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"])
tokenizer.pre_tokenizer = ByteLevel()
if dataset == 'arxiv':
tokenizer.train_from_iterator(datafiles, trainer=trainer)
else:
tokenizer.train(datafiles, trainer=trainer)
tokenizer.decoder = decoders.ByteLevel()
else:
assert style == 'WordLevel'
tokenizer = Tokenizer(WordLevel(unk_token="[UNK]"))
trainer = WordLevelTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"])
tokenizer.pre_tokenizer = Whitespace()
if dataset == 'arxiv':
tokenizer.train_from_iterator(datafiles, trainer=trainer)
else:
tokenizer.train(datafiles, trainer=trainer)
tokenizer.decoder = decoders.WordPiece(
) # WordPiece seems to work (adds back spaces)
# Save to tokenizers directory
tokenizer.save(tpath_expected)
# Generate vocab object based on tokenizer.decoder() method
# ... TODO implement the same vocabulary functionality, or ensure it is present in Tokenizer and then code it elsewhere...
# Features we need to match:
# from torchtext.legacy.vocab import Vocab as RetiredVocab
# ntokens = len(vocab.stoi) ---> ntokens = tokenizer.(...)
# data = [torch.tensor([vocab[token] for token in tokenizer(item)],
# dtype=torch.long) for item in raw_text_iter]
# tokenized_text_ints = torch.tensor([vocab[token] for token in tokenized_text], dtype=torch.long)
# running_context_string = ' '.join([vocab.itos[src[k]] for k in range(src.shape[0])])
# unk_index = vocab.unk_index
vocab = None
return tokenizer, vocab
parser.add_argument('--languages',
help='dataset languages to tokenize',
type=str,
required=True)
parser.add_argument('--tokenizer-out',
help='tokenizer output file',
type=str,
required=True)
parser.add_argument('--special-tokens',
type=str,
default="[UNK],[SEP],[PAD],[MASK],[ECHO],[TRANSLATE]")
args = parser.parse_args()
# translation_dataset = load_dataset(args.dataset, args.languages)
# translation_datase t.set_format(columns='translation')
translation_dataset = NewsCommentaryTranslationDataset()
tokenizer_file = args.tokenizer_out
special_tokens = args.special_tokens.split(",")
tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
tokenizer.pre_tokenizer = Whitespace()
trainer = BpeTrainer(special_tokens=special_tokens)
all_translation_sentences = map(
lambda x: [x['translation'][lang] for lang in x['translation'].keys()],
translation_dataset)
tokenizer.train_from_iterator(all_translation_sentences, trainer=trainer)
tokenizer.save(tokenizer_file)
print(f"Unique names: {len(names)}\n")
name_words = {n: " ".join(split_to_words(n)) for n in names}
with open(f"{proc_path}/names.txt", "w") as f:
f.write("\n".join(list(name_words.values())))
# f.write("\n".join(words))
tokenizer = Tokenizer(BPE())
tokenizer.normalizer = Sequence([
# NFKC(),
Lowercase()
])
tokenizer.pre_tokenizer = ByteLevel()
tokenizer.decoder = ByteLevelDecoder()
trainer = BpeTrainer(vocab_size=int(vocab_size), show_progress=True)
tokenizer.train(trainer, [f"{proc_path}/names.txt"])
print("Trained vocab size: {}".format(tokenizer.get_vocab_size()))
tokenizer.model.save(proc_path)
tokenizer.model = BPE.from_file(f'{proc_path}/vocab.json',
f'{proc_path}/merges.txt')
with open(f"{proc_path}/vocab.json", "r") as f:
bpe_vocab = json.load(f)
bpe_vocab_idx = {v: k for k, v in bpe_vocab.items()}
char_map = {k: v + 1 for k, v in bpe_vocab.items() if len(k) == 1}
if __name__ == "__main__":
args = parser.parse_args()
for f in ['ewe-fon', "ewe", "fon"]:
# instantiate tokenizer
tokenizer = Tokenizer(BPE(unk_token="[UNK]"))
# splitting our inputs into words
tokenizer.pre_tokenizer = Whitespace()
# instantiate trainer
trainer = BpeTrainer(
special_tokens=["[UNK]", "[CLS]", "[SEP]", "[PAD]", "[MASK]"],
min_frequency=2)
# get files
files = [os.path.join(args.data_dir, f"{f}-sentences.txt")]
# train tokenizer
tokenizer.train(files=files, trainer=trainer)
# save tokenizer config file
tokenizer.save(os.path.join(args.save_dir, f"tokenizer-{f}.json"))
# load trained tokenizers
for f in ['ewe-fon', "ewe", "fon"]:
print(f'Using {f} tokenizer : \n')
try:
import fire
from tokenizers import Tokenizer
from tokenizers.models import BPE
from tokenizers.trainers import BpeTrainer
from tokenizers.pre_tokenizers import Whitespace
from tokenizers.normalizers import Sequence, Lowercase, Strip
def train(dataset_path,
output_dir='data/tokenizer/',
vocab_size=30_000,
min_frequency=3):
trainer = BpeTrainer(
vocab_size=vocab_size,
min_frequency=min_frequency,
special_tokens=['[UNK]', '[CLS]', '[SEP]', '[PAD]', '[MASK]'])
tokenizer = Tokenizer(BPE())
tokenizer.pre_tokenizer = Whitespace()
tokenizer.normalizer = Sequence([Lowercase(), Strip()])
files = [dataset_path]
tokenizer.train(trainer, files)
files = tokenizer.model.save(output_dir)
tokenizer.model = BPE.from_file(*files, unk_token='[UNK]')
tokenizer.save(f'{output_dir}/tokenizer.json')
if __name__ == '__main__':