tokenizer-arena / utils /zh_util.py
xu-song's picture
add more tokenizers
f4973d4
raw
history blame
4.45 kB
"""
TODO: 繁体、简体、语种、
"""
import os
import json
from collections import Counter
from utils.text_util import is_chinese, get_zh_count, get_digit_count
from zhon.hanzi import punctuation as zh_punc
CURRENT_DIR = os.path.dirname(os.path.abspath(__file__))
zh_tokens = [line.strip() for line in open(os.path.join(CURRENT_DIR, "vocab.jd.txt.v2"), "r", encoding="utf-8") if
is_chinese(line.strip())]
def zh_iterator():
for idx in range(ord(u'\u4e00'), ord(u'\u9fa5')):
yield (chr(idx))
def get_coding_length(tokenizer, vocab, filter=None):
"""
计算编码长度。(有些中文汉字被解码成多个token)
"""
all_length = []
for word in vocab:
if len(word) > 1:
continue
if filter is not None and filter(word):
continue
tokens = tokenizer.encode(word)
all_length.append(len(tokens))
# if len(tokens.ids) > 1:
# if len(tokens) > 3:
# print(word, tokens)
dist_length = Counter(all_length)
mean_length = round(sum(all_length) / len(all_length), 2)
return dist_length, mean_length
def has_zh_punc(text):
"""
是否包含中文标点
"""
return any(ch in zh_punc for ch in text)
def get_space_count(text):
space_count = 0
for char in text:
if len(char.strip()) == 0:
space_count += 1
return space_count
def remove_special_char():
"""
:return:
"""
# bert词典有 ##开头的
# byteBPE词典有带空格的
# decode_str = decode_str.strip().replace("#", "") # TODO, 按类型
pass
cache = {}
def iter_vocab(tokenizer, name="", from_cache=True):
if from_cache and name in cache:
return cache[name]
f_out = open(name + "_vocab.zh.jsonl", "w", encoding="utf-8")
zh_token_count = {"total": 0, "中文单字": 0, "中文多字": 0}
# zh_token_count = {"total": 0, "包含1个中文单字": 0, "中文多字": 0}
# symbol_count = 0
all_single_zh_tokens = set()
zh_symbol_count = 0
for token_id in range(tokenizer.vocab_size):
decode_str = tokenizer.decode([token_id], skip_special_tokens=False)
token = tokenizer.convert_ids_to_tokens([token_id], skip_special_tokens=False)[0]
# tokenizer.convert_tokens_to_string(tokens)
if token is None: # 有些词典有空的id(不连续)
continue
if isinstance(token, bytes):
token = token.decode("utf-8", errors="ignore")
digit_count = get_digit_count(token)
zh_count = get_zh_count(decode_str)
space_count = get_space_count(decode_str)
f_out.write(json.dumps(
{"id": token_id,
"token": token,
"token_decode": decode_str,
"token_len": len(token),
"zh_count": zh_count,
"space_count": space_count,
"digit_count": digit_count,
"zh_symbol_count": zh_symbol_count,
},
ensure_ascii=False) + "\n"
)
if zh_count >= 1:
zh_token_count["total"] += 1
if zh_count > 1:
zh_token_count["中文多字"] += 1
else:
zh_token_count["中文单字"] += 1
all_single_zh_tokens.add(decode_str.strip().replace("#", ""))
#
dist_length, mean_length = get_coding_length(tokenizer, zh_tokens, filter=lambda k: not is_chinese(k))
# TODO: 繁体字,简体字
zh_token_count["中文单字-去重后"] = len(all_single_zh_tokens)
result = {
"name": name,
"impl": str(tokenizer.__class__),
"vocab_size": tokenizer.vocab_size,
"中文汉字数": zh_token_count,
"中文标点数": zh_symbol_count,
"中文汉字编码长度均值": mean_length,
"中文汉字编码长度分布": json.dumps(dist_length),
}
cache[name] = result
return result
if __name__ == "__main__":
# test_coding_length(jd_vocab_tokens, filter=lambda k: not is_chinese(k))
# test_coding_length(zh_punc)
# test_coding_length(zh_iterator())
from vocab.chatglm2_6b import tokenizer; name = "chatglm2_6b"
# from vocab.chatglm_6b import tokenizer; name="chatglm_6b"
# from vocab.baichuan2 import tokenizer; name="baichuan2"
# from vocab.gpt_4 import tokenizer; name="gpt4"
print(iter_vocab(tokenizer, name=name))