forked from OpenGVLab/OmniQuant
-
Notifications
You must be signed in to change notification settings - Fork 0
/
datautils.py
195 lines (159 loc) · 6.53 KB
/
datautils.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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
import pdb
from transformers import AutoTokenizer
from datasets import load_dataset
import numpy as np
import torch
import random
def set_seed(seed):
np.random.seed(seed)
torch.random.manual_seed(seed)
def get_pile(nsamples, seed, seqlen, model):
print("get_pile")
traindata = load_dataset("json", data_files='/cpfs01/user/chenmengzhao/prompt_quantization/val.jsonl.zst', split="train")
tokenizer = AutoTokenizer.from_pretrained(model, use_fast=False)
trainenc = tokenizer("\n\n".join(traindata['text'][:1000]), return_tensors='pt')
random.seed(seed)
trainloader = []
for _ in range(nsamples):
i = random.randint(0, trainenc.input_ids.shape[1] - seqlen - 1)
j = i + seqlen
inp = trainenc.input_ids[:, i:j]
tar = inp.clone()
tar[:, :-1] = -100
trainloader.append((inp, tar))
return trainloader, None
def get_wikitext2(nsamples, seed, seqlen, model):
print("get_wikitext2")
traindata = load_dataset('wikitext', 'wikitext-2-raw-v1', split='train')
testdata = load_dataset('wikitext', 'wikitext-2-raw-v1', split='test')
tokenizer = AutoTokenizer.from_pretrained(model, use_fast=False)
trainenc = tokenizer("\n\n".join(traindata['text']), return_tensors='pt')
testenc = tokenizer("\n\n".join(testdata['text']), return_tensors='pt')
random.seed(seed)
trainloader = []
for _ in range(nsamples):
i = random.randint(0, trainenc.input_ids.shape[1] - seqlen - 1)
j = i + seqlen
inp = trainenc.input_ids[:, i:j]
tar = inp.clone()
tar[:, :-1] = -100
trainloader.append((inp, tar))
return trainloader, testenc
def get_ptb(nsamples, seed, seqlen, model):
print("get_ptb")
traindata = load_dataset('ptb_text_only', 'penn_treebank', split='train')
valdata = load_dataset('ptb_text_only', 'penn_treebank', split='validation')
tokenizer = AutoTokenizer.from_pretrained(model, use_fast=False)
trainenc = tokenizer("\n\n".join(traindata['sentence']), return_tensors='pt')
testenc = tokenizer("\n\n".join(valdata['sentence']), return_tensors='pt')
random.seed(seed)
trainloader = []
for _ in range(nsamples):
i = random.randint(0, trainenc.input_ids.shape[1] - seqlen - 1)
j = i + seqlen
inp = trainenc.input_ids[:, i:j]
tar = inp.clone()
tar[:, :-1] = -100
trainloader.append((inp, tar))
return trainloader, testenc
def get_c4(nsamples, seed, seqlen, model):
print("get_c4")
traindata = load_dataset(
'allenai/c4', data_files={'train': 'en/c4-train.00000-of-01024.json.gz'}, split='train'
)
valdata = load_dataset(
'allenai/c4', data_files={'validation': 'en/c4-validation.00000-of-00008.json.gz'}, split='validation'
)
tokenizer = AutoTokenizer.from_pretrained(model, use_fast=False)
random.seed(seed)
trainloader = []
for _ in range(nsamples):
while True:
i = random.randint(0, len(traindata) - 1)
trainenc = tokenizer(traindata[i]['text'], return_tensors='pt')
if trainenc.input_ids.shape[1] >= seqlen:
break
i = random.randint(0, trainenc.input_ids.shape[1] - seqlen - 1)
j = i + seqlen
inp = trainenc.input_ids[:, i:j]
tar = inp.clone()
tar[:, :-1] = -100
trainloader.append((inp, tar))
random.seed(0)
valenc = []
for _ in range(256):
while True:
i = random.randint(0, len(valdata) - 1)
tmp = tokenizer(valdata[i]['text'], return_tensors='pt')
if tmp.input_ids.shape[1] >= seqlen:
break
i = random.randint(0, tmp.input_ids.shape[1] - seqlen - 1)
j = i + seqlen
valenc.append(tmp.input_ids[:, i:j])
valenc = torch.hstack(valenc)
return trainloader, valenc
def get_ptb_new(nsamples, seed, seqlen, model):
print("get_ptb_new")
traindata = load_dataset('ptb_text_only', 'penn_treebank', split='train')
testdata = load_dataset('ptb_text_only', 'penn_treebank', split='test')
tokenizer = AutoTokenizer.from_pretrained(model, use_fast=False)
trainenc = tokenizer(" ".join(traindata["sentence"]), return_tensors="pt")
testenc = tokenizer(" ".join(testdata ["sentence"]), return_tensors="pt")
random.seed(seed)
trainloader = []
for _ in range(nsamples):
i = random.randint(0, trainenc.input_ids.shape[1] - seqlen - 1)
j = i + seqlen
inp = trainenc.input_ids[:, i:j]
tar = inp.clone()
tar[:, :-1] = -100
trainloader.append((inp, tar))
return trainloader, testenc
def get_c4_new(nsamples, seed, seqlen, model):
print("get_c4_new")
traindata = load_dataset(
'allenai/c4', data_files={'train': 'en/c4-train.00000-of-01024.json.gz'}, split='train'
)
valdata = load_dataset(
'allenai/c4', data_files={'validation': 'en/c4-validation.00000-of-00008.json.gz'}, split='validation'
)
tokenizer = AutoTokenizer.from_pretrained(model, use_fast=False)
random.seed(seed)
trainloader = []
for _ in range(nsamples):
while True:
i = random.randint(0, len(traindata) - 1)
trainenc = tokenizer(traindata[i]["text"], return_tensors="pt")
if trainenc.input_ids.shape[1] >= seqlen:
break
i = random.randint(0, trainenc.input_ids.shape[1] - seqlen - 1)
j = i + seqlen
inp = trainenc.input_ids[:, i:j]
tar = inp.clone()
tar[:, :-1] = -100
trainloader.append((inp, tar))
valenc = tokenizer(" ".join(valdata[:1100]["text"]), return_tensors="pt")
valenc = valenc.input_ids[:, : (256 * seqlen)]
return trainloader, valenc
def get_loaders(
name, nsamples=128, seed=0, seqlen=2048, model='',
):
if 'wikitext2' in name:
return get_wikitext2(nsamples, seed, seqlen, model)
if 'pile' in name:
return get_pile(nsamples, seed, seqlen, model)
if 'ptb' in name:
if 'new' in name:
return get_ptb_new(nsamples, seed, seqlen, model)
return get_ptb(nsamples, seed, seqlen, model)
if 'c4' in name:
if 'new' in name:
return get_c4_new(nsamples, seed, seqlen, model)
return get_c4(nsamples, seed, seqlen, model)
if 'mix' in name:
wiki_train,wiki_val=get_wikitext2(nsamples//3, seed, seqlen, model)
ptb_train,ptb_val=get_ptb(nsamples//3, seed, seqlen, model)
c4_train,c4_val=get_c4(nsamples//3, seed, seqlen, model)
train=wiki_train+ptb_train+c4_train
val=None
return train,val