mirror of
https://github.com/mii443/tokenizers.git
synced 2025-12-03 11:18:29 +00:00
52 lines
1.3 KiB
Python
52 lines
1.3 KiB
Python
import argparse
|
|
import glob
|
|
|
|
from tokenizers import BertWordPieceTokenizer
|
|
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument("--files",
|
|
default=None,
|
|
metavar="path",
|
|
type=str,
|
|
required=True,
|
|
help="The files to use as training; accept '**/*.txt' type of patterns \
|
|
if enclosed in quotes")
|
|
parser.add_argument("--out",
|
|
default="./",
|
|
type=str,
|
|
help="Path to the output directory, where the files will be saved")
|
|
parser.add_argument("--name",
|
|
default="bert-wordpiece",
|
|
type=str,
|
|
help="The name of the output vocab files")
|
|
args = parser.parse_args()
|
|
|
|
files = glob.glob(args.files)
|
|
if not files:
|
|
print(f"File does not exist: {args.files}")
|
|
exit(1)
|
|
|
|
|
|
# Initialize an empty tokenizer
|
|
tokenizer = BertWordPieceTokenizer(
|
|
clean_text=True,
|
|
handle_chinese_chars=True,
|
|
strip_accents=True,
|
|
lowercase=True,
|
|
|
|
)
|
|
|
|
# And then train
|
|
trainer = tokenizer.train(
|
|
files,
|
|
vocab_size=10000,
|
|
min_frequency=2,
|
|
show_progress=True,
|
|
special_tokens=["[PAD]", "[UNK]", "[CLS]", "[SEP]", "[MASK]"],
|
|
limit_alphabet=1000,
|
|
wordpieces_prefix="##"
|
|
)
|
|
|
|
# Save the files
|
|
tokenizer.save(args.out, args.name)
|