forked from DLLXW/baby-llama2-chinese
-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_process.py
160 lines (151 loc) · 4.54 KB
/
data_process.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
import json
import glob
import numpy as np
from tqdm import tqdm
from chatglm_tokenizer.tokenization_chatglm import ChatGLMTokenizer
import pandas as pd
#from zhconv import convert
def process_wiki_clean():
with open('./data/wikipedia-cn-20230720-filtered.json','r') as f:
data=json.load(f)
doc_ids=[]
for line in tqdm(data):
text=line['completion']
text_id=tokenizer.encode(text,add_special_tokens=False)
text_id.append(tokenizer.special_tokens['<eos>'])
if len(text_id)>5:
doc_ids+=text_id
arr = np.array(doc_ids,dtype=np.uint16)
with open('./data/wiki.bin','wb') as f:
f.write(arr.tobytes())
def process_medical(data_path,name):
f=open(data_path,'r')
doc_ids=[]
while True:
line=f.readline()
if not line:
break
line=json.loads(line)
text=line['text']
text_id=tokenizer.encode(text,add_special_tokens=False)
text_id.append(tokenizer.special_tokens['<eos>'])
if len(text_id)>5:
doc_ids+=text_id
arr = np.array(doc_ids,dtype=np.uint16)
with open('./data/medical_{}.bin'.format(name),'wb') as f:
f.write(arr.tobytes())
def sft_to_pretrain():
df=pd.read_csv('./data/medical_qa_144w.csv')
doc_ids=[]
for _,q,a in tqdm(df.itertuples()):
q_id = tokenizer.encode(q,add_special_tokens=False)
a_id = tokenizer.encode(a,add_special_tokens=False)
#
print(q)
print(a)
print('-----')
text_id=q_id+a_id+[tokenizer.special_tokens['<eos>']]
if len(text_id)>5:
doc_ids+=text_id
arr = np.array(doc_ids,dtype=np.uint16)
print(arr.shape)
with open('./data/medical_qa.bin','wb') as f:
f.write(arr.tobytes())
def sft_process():
with open('./data/alpaca_gpt4_data_zh.json','r') as f:
data=json.load(f)
#
q_lst=[]
a_lst=[]
for per in data:
q=per['instruction']
i=per['input']
a=per['output']
q=q+i
if len(q)<10 or len(a)<5:
continue
if len(q)>256 or len(a)>256:
continue
q_lst.append(q)
a_lst.append(a)
#
# with open('../track1/train_valid.json','r') as f:
# data=json.load(f)
# #
# for l in data:
# q_lst.append(l['question'])
# a_lst.append(l['answer'])
#
f = open('./data/Belle_open_source_1M.json','r')
#s
while True:
line = f.readline()
if not line:
break
per=json.loads(line)
q=per['instruction']
i=per['input']
a=per['output']
q=q+i
if len(q)<10 or len(a)<5:
continue
if len(q)>256 or len(a)>256:
continue
q_lst.append(q)
a_lst.append(a)
df=pd.DataFrame(columns=['prompt','answer'])
df['prompt']=q_lst
df['answer']=a_lst
df.to_csv('data/sft_data.csv',index=False)
print(df)
def process_baidu():
f=open('./data/563w_baidubaike.json','r')
cnt=0
token=0
doc_ids=[]
while True:
line = f.readline()
if not line:
break
line=json.loads(line)
text=''
try:
text+=line['title']+':'+line['summary']
except:
pass
for per in line['sections']:
text+=per['title']+':'+per['content']+'。'
text_id=tokenizer.encode(text,add_special_tokens=False)
text_id.append(tokenizer.special_tokens['<eos>'])
if len(text_id)>5:
doc_ids+=text_id
cnt+=1
if cnt%10000==0:
print(cnt)
arr = np.array(doc_ids,dtype=np.uint16)
print(arr.shape)
with open('./data/baidubaike_563w.bin','wb') as f:
f.write(arr.tobytes())
if __name__=="__main__":
tokenizer=ChatGLMTokenizer(vocab_file='./chatglm_tokenizer/tokenizer.model')
# process_wiki_clean()
# process_medical('./data/medical_book_zh.json','book')
# process_medical('./data/train_encyclopedia.json','encyclopedia')
# sft_to_pretrain()
# sft_process()
#process_baidu()
data_path_list=[
'./data/baidubaike_563w.bin',
'./data/medical_book.bin',
'./data/medical_encyclopedia.bin',
'./data/wiki.bin'
]
data_lst=[]
for data_path in data_path_list:
with open(data_path,'rb') as f:
data=np.fromfile(f,dtype=np.uint16)
data_lst.append(data)
arr = np.concatenate(data_lst)
print(arr.shape)
with open('./data/pretrain_data.bin','wb') as f:
f.write(arr.tobytes())