-
Notifications
You must be signed in to change notification settings - Fork 1
/
muisc_model.py
203 lines (148 loc) · 6.73 KB
/
muisc_model.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
196
197
198
import logging
import math
import os
import sys
from typing import Optional
from collections import OrderedDict
import numpy as np
from dataclasses import dataclass, field
import torch
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader
import datasets
from datasets import load_dataset
import transformers
from transformers import (
CONFIG_MAPPING,
MODEL_FOR_MASKED_LM_MAPPING,
AutoConfig,
AutoModelForMaskedLM,
AutoTokenizer,
DataCollatorForLanguageModeling,
HfArgumentParser,
BertLayer,
GPT2DoubleHeadsModel,
ViTModel,
ViTFeatureExtractor,
)
from transformers.utils import check_min_version
from transformers.utils.versions import require_version
import argparse
from PIL import Image
import os
os.environ["TOKENIZERS_PARALLELISM"] = "false"
# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
check_min_version("4.9.0.dev0")
require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/language-modeling/requirements.txt")
logger = logging.getLogger(__name__)
MODEL_CONFIG_CLASSES = list(MODEL_FOR_MASKED_LM_MAPPING.keys())
MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)
class DS(Dataset):
def _read_image_data_from_file(self, filenames, shape=(224, 224, 3)):
try:
image = [Image.open(filename).convert('RGB') for filename in filenames]
image_data = self.feature_extractor(images=image, return_tensors="pt") #['pixel_values']
#print('_read_image_data_from_file: Got one. {}'.format(image_data.shape))
except:
image_data = self.feature_extractor(images=[np.zeros(shape)]*3, return_tensors="pt") # hard-encoded
print("An exception occurred in _read_image_data_from_file {}.".format(filenames))
return image_data
def __init__(self, data, tokenizer, max_length=1024):
self.data = data
self.tok = tokenizer
self.max_length = max_length
self.feature_extractor = ViTFeatureExtractor.from_pretrained('google/vit-base-patch16-224-in21k')
def __len__(self):
return len(self.data)
def __getitem__(self, index):
assert 'img_0' in self.data[index] and 'img_1' in self.data[index] and 'img_2' in self.data[index]
img_files = [self.data[index]['img_0'], self.data[index]['img_1'], self.data[index]['img_2']]
#
if 'cid1' in self.data[index]:
cid1 = self.data[index]['cid1']
else: cid1 = 0
if 'cid2' in self.data[index]:
cid2 = self.data[index]['cid2']
else: cid2 = 0
cid_text = "_".join([str(cid1), str(cid2)])
if 'title' in self.data[index]:
title = self.data[index]['title']
else:
title = ''
#
mc_labels = 0
offreason = self.data[index]['offreason']
#text = cid_text + "_" + title + "[MASK]" + offreason
text = title + "[MASK]" + offreason
print("text = {}".format(text))
line = self.tok.encode_plus(
text,
max_length=self.max_length,
truncation=True,
padding="max_length",
return_tensors="pt",
)
line = {kk: vv.squeeze(0) for kk, vv in line.items()}
imgs = self._read_image_data_from_file(img_files)
line['image_data'] = imgs
line['image_files'] = img_files
line['mc_token_ids'] = line['input_ids'].tolist().index(self.tok.mask_token_id)
line['labels'] = line['input_ids'].clone()
for ii in range(line['mc_token_ids']+0): # including mc_toke itself
line['labels'][ii] = -100
line['offreason'] = offreason
line['mc_labels'] = mc_labels
line['wid'] = 0
return line
class ImageEncoderModel(nn.Module):
def __init__(self, out_embed, pos_num=197*3):
super(ImageEncoderModel, self).__init__()
single_image_model_name = 'google/vit-base-patch16-224-in21k'
single_image_config = AutoConfig.from_pretrained(single_image_model_name)
#single_image_config.num_hidden_layers = 3
self.image_tower = ViTModel.from_pretrained(single_image_model_name, config=single_image_config)
print('xc====== single_image_config = {}'.format(single_image_config))
self.use_image_cross = True
self.pos_num = pos_num
if self.use_image_cross is True:
print("xc====== Use BertLayer")
cross_image_config = AutoConfig.from_pretrained('bert-base-chinese')
cross_image_config.num_hidden_layers = 1
# cross_image_config.num_attention_heads = 12
cross_image_config.add_cross_attention = False
cross_image_config.is_decoder = False
self.self = BertLayer(cross_image_config)
self.wpe = nn.Embedding(pos_num, out_embed)
self.position_ids = torch.tensor(list(range(pos_num)), dtype=torch.int32)
#"""Initialize weights"""
#self.wpe.apply(self._init_weights)
#if self.use_image_cross == True:
#self.self.apply(self._init_weights)
print('xc====== use_image_cross is {}'.format(self.use_image_cross))
#self.encoder_attention_masks = self._build_encoder_attention_mask(image_seq_len=pos_num, head_num=12) # pos_num
self.encoder_attention_masks = None
self.dropout_prob = 0.0
print("xc====== ImageEncoderModel: dropout_prob = {}".format(self.dropout_prob))
def forward(self, image_data=None, device=None, eval=True):
if image_data is None:
return None
else:
input = image_data
if device is not None:
input['pixel_values'] = input['pixel_values'].to(device)
assert len(input['pixel_values'].size()) == 4 or len(input['pixel_values'].size()) == 5
if len(input['pixel_values'].size()) == 4:
batch_size = 1
else:
batch_size = input['pixel_values'].shape[0]
input['pixel_values'] = input['pixel_values'].view((-1, input['pixel_values'].shape[-3],
input['pixel_values'].shape[-2],
input['pixel_values'].shape[-1]))
outputs = self.image_tower(**input)
x = outputs.last_hidden_state
if self.use_image_cross == True:
position_embeds = self.wpe(self.position_ids.to(input['pixel_values'].device))
xx = x.view((batch_size, -1, x.shape[-1]))
x = self.self(xx + position_embeds, attention_mask=None)[0]
x = x.view((batch_size, 1, -1, x.shape[-1]))
return x