-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathget_trie_dict_4structured_id.py
213 lines (192 loc) · 8.04 KB
/
get_trie_dict_4structured_id.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
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
from tqdm import tqdm
import re
import json
from utils import Trie
import pickle
from open_flamingo import create_model_and_transforms
import argparse
arg_parser = argparse.ArgumentParser()
arg_parser.add_argument(
"--output_dir",
type=str,
help="Pass in the directory where the output shards (as tar files) will be written to.",
)
arg_parser.add_argument(
"--json_file",
type=str,
help="image-caption pairs json_file",
)
arg_parser.add_argument(
"--image_name2id_dict",
type=str,
)
arg_parser.add_argument(
"--id2image_name_dict",
type=str,
)
arg_parser.add_argument(
"--identifier_type",
choices=["semantic_identifier", "structured_identifier", "automatic_identifier", "string_identifier", "numeric_identifier"]
)
args = arg_parser.parse_args()
if args.identifier_type == "structured_identifier":
with open(args.image_name2id_dict, 'rb') as f:
image_name2id_dict = pickle.load(f)
model, image_processor, tokenizer = create_model_and_transforms(
clip_vision_encoder_path="ViT-L-14",
clip_vision_encoder_pretrained="openai",
lang_encoder_path="anas-awadalla/mpt-1b-redpajama-200b",
tokenizer_path="anas-awadalla/mpt-1b-redpajama-200b",
cross_attn_every_n_layers=1,
add_extra_id_tokens = args.id2image_name_dict
)
id_dict_test_set = {}
with open(args.json_file, 'r') as f:
data = json.load(f)['images']
for original_sample_data in data:
if original_sample_data['split'] =='test':
id_dict_test_set[str(image_name2id_dict[original_sample_data['filename']])] = 1
print(str(image_name2id_dict[original_sample_data['filename']]))
end_of_chunk_id = tokenizer.encode("<|endofchunk|>")[0]
print('end_of_chunk_id',end_of_chunk_id)
caption_sequence = []
for caption in tqdm(id_dict_test_set):
input_ids = tokenizer.encode(
"<image>id "+ caption,
add_special_tokens=True,
max_length=20,
truncation=True)
caption_sequence.append(input_ids+[end_of_chunk_id])
print(input_ids)
decoder_trie = Trie(caption_sequence)
print("decoder_trie len %s", decoder_trie.len)
with open(args.output_dir, 'wb') as f:
pickle.dump(decoder_trie.trie_dict, f)
if args.identifier_type == "semantic_identifier":
with open(args.image_name2id_dict, 'rb') as f:
image_name2id_dict = pickle.load(f)
model, image_processor, tokenizer = create_model_and_transforms(
clip_vision_encoder_path="ViT-L-14",
clip_vision_encoder_pretrained="openai",
lang_encoder_path="anas-awadalla/mpt-1b-redpajama-200b",
tokenizer_path="anas-awadalla/mpt-1b-redpajama-200b",
cross_attn_every_n_layers=1,
)
id_dict_test_set = {}
with open(args.json_file, 'r') as f:
data = json.load(f)['images']
for original_sample_data in data:
if original_sample_data['split'] =='test':
for identifier in image_name2id_dict[original_sample_data['filename']]:
id_dict_test_set[str(identifier)] = 1
print(str(identifier))
end_of_chunk_id = tokenizer.encode("<|endofchunk|>")[0]
print('end_of_chunk_id',end_of_chunk_id)
caption_sequence = []
for caption in tqdm(id_dict_test_set):
input_ids = tokenizer.encode(
"<image>id "+ caption,
add_special_tokens=True,
max_length=30,
truncation=True)
caption_sequence.append(input_ids+[end_of_chunk_id])
print(input_ids)
decoder_trie = Trie(caption_sequence)
print("decoder_trie len %s", decoder_trie.len)
with open(args.output_dir, 'wb') as f:
pickle.dump(decoder_trie.trie_dict, f)
if args.identifier_type == "automatic_identifier":
with open(args.image_name2id_dict, 'rb') as f:
image_name2id_dict = pickle.load(f)
model, image_processor, tokenizer = create_model_and_transforms(
clip_vision_encoder_path="ViT-L-14",
clip_vision_encoder_pretrained="openai",
lang_encoder_path="anas-awadalla/mpt-1b-redpajama-200b",
tokenizer_path="anas-awadalla/mpt-1b-redpajama-200b",
cross_attn_every_n_layers=1,
add_extra_id_tokens = args.id2image_name_dict
)
id_dict_test_set = {}
with open(args.json_file, 'r') as f:
data = json.load(f)['images']
for original_sample_data in data:
if original_sample_data['split'] =='test':
id_dict_test_set[str(image_name2id_dict[original_sample_data['filename']])] = 1
print(str(image_name2id_dict[original_sample_data['filename']]))
end_of_chunk_id = tokenizer.encode("<|endofchunk|>")[0]
print('end_of_chunk_id',end_of_chunk_id)
caption_sequence = []
for caption in tqdm(id_dict_test_set):
input_ids = tokenizer.encode(
"<image>id "+ caption,
add_special_tokens=True,
max_length=20,
truncation=True)
caption_sequence.append(input_ids+[end_of_chunk_id])
print(input_ids)
decoder_trie = Trie(caption_sequence)
print("decoder_trie len %s", decoder_trie.len)
with open(args.output_dir, 'wb') as f:
pickle.dump(decoder_trie.trie_dict, f)
if args.identifier_type == "string_identifier":
with open(args.image_name2id_dict, 'rb') as f:
image_name2id_dict = pickle.load(f)
model, image_processor, tokenizer = create_model_and_transforms(
clip_vision_encoder_path="ViT-L-14",
clip_vision_encoder_pretrained="openai",
lang_encoder_path="anas-awadalla/mpt-1b-redpajama-200b",
tokenizer_path="anas-awadalla/mpt-1b-redpajama-200b",
cross_attn_every_n_layers=1
)
id_dict_test_set = {}
with open(args.json_file, 'r') as f:
data = json.load(f)['images']
for original_sample_data in data:
if original_sample_data['split'] =='test':
id_dict_test_set[str(image_name2id_dict[original_sample_data['filename']])] = 1
end_of_chunk_id = tokenizer.encode("<|endofchunk|>")[0]
print('end_of_chunk_id',end_of_chunk_id)
caption_sequence = []
for caption in tqdm(id_dict_test_set):
input_ids = tokenizer.encode(
"<image>id "+ caption,
add_special_tokens=True,
max_length=10,
truncation=True)
caption_sequence.append(input_ids+[end_of_chunk_id])
print(input_ids)
decoder_trie = Trie(caption_sequence)
print("decoder_trie len %s", decoder_trie.len)
with open(args.output_dir, 'wb') as f:
pickle.dump(decoder_trie.trie_dict, f)
if args.identifier_type == "numeric_identifier":
with open(args.image_name2id_dict, 'rb') as f:
image_name2id_dict = pickle.load(f)
model, image_processor, tokenizer = create_model_and_transforms(
clip_vision_encoder_path="ViT-L-14",
clip_vision_encoder_pretrained="openai",
lang_encoder_path="anas-awadalla/mpt-1b-redpajama-200b",
tokenizer_path="anas-awadalla/mpt-1b-redpajama-200b",
cross_attn_every_n_layers=1
)
id_dict_test_set = {}
with open(args.json_file, 'r') as f:
data = json.load(f)['images']
for original_sample_data in data:
if original_sample_data['split'] =='test':
id_dict_test_set[str(image_name2id_dict[original_sample_data['filename']])] = 1
end_of_chunk_id = tokenizer.encode("<|endofchunk|>")[0]
print('end_of_chunk_id',end_of_chunk_id)
caption_sequence = []
for caption in tqdm(id_dict_test_set):
input_ids = tokenizer.encode(
"<image>id "+ caption,
add_special_tokens=True,
max_length=10,
truncation=True)
caption_sequence.append(input_ids+[end_of_chunk_id])
print(input_ids)
decoder_trie = Trie(caption_sequence)
print("decoder_trie len %s", decoder_trie.len)
with open(args.output_dir, 'wb') as f:
pickle.dump(decoder_trie.trie_dict, f)