tokenizer-arena / vocab /gpt_nexo_20b /test_tokenizer_HF.py
xu-song's picture
update
751936e
raw
history blame
1.3 kB
"""
最简单的tokenizer
"""
import json
from vocab.gpt_nexo_20b.tokenizer.tokenizer import HFTokenizer
tokenizer = HFTokenizer("20B_tokenizer.json")
print("vocab_size with added_tokens:", tokenizer.vocab_size)
vocab = tokenizer.vocab
def test_single_token():
"""
单个字符的编码(一个字符可能会编码成多个id)
"""
for word in "中国解决方法黑白侗鸩,。!?;":
encoding = tokenizer.tokenize(word)
for token_id in encoding:
decode_str = tokenizer.detokenize([token_id]) # 特殊字符解码后会统一变成 �,对应 "\ufffd"
# token = tokenizer.tokenizer.id_to_token(token_id)
print(word, token_id, decode_str, json.dumps(decode_str), )
# print(word, token_id, decode_str, json.dumps(decode_str), token, json.dumps(token))
def test_encode():
text = "中国解决方法黑白侗鸩,。!?;一个人去哪里 一 个"
encoding = tokenizer.tokenize(text)
for token_id in encoding:
decode_str = tokenizer.detokenize([token_id]) # 特殊字符解码后会统一变成 �,对应 "\ufffd"
token = tokenizer.tokenizer.id_to_token(token_id)
print(token_id, decode_str, json.dumps(decode_str), token, json.dumps(token))
test_encode()