tokenizer-arena / util.py
xu-song's picture
update
d10ecd7
raw
history blame
3.21 kB
import gradio as gr
import json
import pandas as pd
from vocab import load_tokener
from utils.zh_util import iter_vocab
def tokenize(text, tokenizer_type, color_num=5):
"""
TODO: cache tokenizer
"""
print(f"入参:tokenize, {text}, {tokenizer_type}")
pos_tokens = []
tokenizer = load_tokener(tokenizer_type)
encoding = tokenizer.encode(text)
table = []
for idx, token_id in enumerate(encoding):
decode_text = tokenizer.decode([token_id]) # 特殊字符解码后会统一变成 �,对应 "\ufffd"
pos_tokens.extend([(decode_text, str(idx % color_num))])
# token "Byte": # 这是 utf-8编码吧?
token = tokenizer.convert_ids_to_tokens([token_id])[0]
if isinstance(token, bytes):
try:
token_str = token.decode("utf-8")
except:
token_str = token.decode("utf-8", errors="ignore")
print("decode_error", tokenizer_type, token, token_str)
token_bytes = token
json_dumps = json.dumps(token_str)
elif isinstance(token, str):
token_str = token
token_bytes = bytes(token_str, "utf-8")
json_dumps = json.dumps(token_str)
else:
return
# ⭐
table.append(
{"TokenID": token_id,
"Token": token_str, # utf-8解码后的字符串,为什么有些是 <0xE7>,表示什么?比如llama
"Text": decode_text, #
# "Bytes": token_bytes, # bytes类型在gradio前端页面被解码成字符串,比如 b'\xe4\xb8\xad' 仍然显示成 "中"。因此 str(token_bytes)
"Bytes": str(token_bytes),
# "Unicode": json_dumps # unicode, 如果是ascii码,就直接显示。如果不是ascii码,就显示unicode
}
)
table_df = pd.DataFrame(table)
print(f"Tokenization[{tokenizer_type}]: {table}")
# print(table_df)
return gr.update(value=pos_tokens, label=f"Tokens: {len(encoding)}"), table_df
def tokenize_pair(text, tokenizer_type_1, tokenizer_type_2):
pos_tokens_1, table_df_1 = tokenize(text, tokenizer_type_1)
pos_tokens_2, table_df_2 = tokenize(text, tokenizer_type_2)
return pos_tokens_1, table_df_1, pos_tokens_2, table_df_2
def basic_count(tokenizer_type):
tokenizer = load_tokener(tokenizer_type)
stats = iter_vocab(tokenizer, tokenizer_type)
return tokenizer.vocab_size, f'{stats["中文汉字数"]["中文单字"]}/{stats["中文汉字数"]["中文多字"]}'
def get_overlap_token_size(tokenizer_type_1, tokenizer_type_2):
tokenizer1 = load_tokener(tokenizer_type_1)
tokenizer2 = load_tokener(tokenizer_type_2)
vocab1 = tokenizer1.get_vocab()
vocab2 = tokenizer2.get_vocab()
overlap_tokens = vocab1.keys() & vocab2.keys()
overlap_token_size = len(overlap_tokens)
print(f"OverlapTokens: {tokenizer_type_1}, {tokenizer_type_2} {list(overlap_tokens)[:10]}")
return overlap_token_size, overlap_token_size
def test_coding():
bytes1 = b'\xe4\xb8\xad'
print(bytes1) # b'\xe4\xb8\xad'
if __name__ == "__main__":
print(basic_count("internlm_chat_7b"))