-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain_main.py
179 lines (160 loc) · 8.82 KB
/
train_main.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
import pytorch_lightning as pl
import torch
import clip
from PIL import Image
from src.lightning_classes import SciMMIR_FT_DataMoudle , SciMMIR_FT
from pytorch_lightning.callbacks import ModelCheckpoint
from src.MyBlip.MyBlip import Blip2Model
import os
import argparse
from transformers import BertTokenizer, BertModel, BlipModel, BlipProcessor, Blip2Processor
#from lavis.models import load_model_and_preprocess
from math import floor
def run(args):
#fixed random seed
pl.seed_everything(1234)
if args.Use_BERT == True:
logs_path = os.path.join("./checkpoints/" , f'MMIR_BERT_{args.model_name}_{args.image_type}')
else:
logs_path = os.path.join("./checkpoints/" , f'MMIR_{args.model_name}_{args.image_type}')
if os.path.exists(logs_path) == False:
os.makedirs(logs_path)
if os.path.exists(args.result_save_path) == False:
os.makedirs(args.result_save_path)
device = "cuda" if torch.cuda.is_available() else "cpu"
if args.model_name == 'CLIP':
model, preprocess = clip.load("ViT-B/32", device=device)
elif args.model_name == 'BLIP':
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base", size = args.image_size)
model = BlipModel.from_pretrained("Salesforce/blip-image-captioning-base").cuda().float()
preprocess = processor
elif args.model_name == 'BLIP-large':
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large", size = args.image_size)
model = BlipModel.from_pretrained("Salesforce/blip-image-captioning-large").cuda().float()
preprocess = processor
elif args.model_name == 'BLIP-FLAN-T5-XL':
model = Blip2Model.from_pretrained("Salesforce/blip2-flan-t5-xl").cuda().float()
processor = Blip2Processor.from_pretrained("Salesforce/blip2-flan-t5-xl")
preprocess = processor
elif args.model_name == 'BLIP-FLAN-T5-XXL':
model = Blip2Model.from_pretrained("Salesforce/blip2-flan-t5-xxl").cuda().float()
processor = Blip2Processor.from_pretrained("Salesforce/blip2-flan-t5-xxl")
preprocess = processor
if args.Use_BERT == True:
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
BERT_model = BertModel.from_pretrained("bert-base-uncased")
else:
tokenizer = clip.tokenize
# lightningmodule checkpoint
checkpoint_callback = ModelCheckpoint(
monitor="Validation/MRR_forward",
dirpath=logs_path,
filename="checkpoints--{Validation/MRR_forward:.4f}",
save_top_k=args.save_top_k,
mode="max",
)
callbacks = [checkpoint_callback]
if args.multi_gpus > 0:
trainer = pl.Trainer(
callbacks = callbacks,
accelerator="gpu",
devices = list(range(int(args.multi_gpus))),
max_epochs = args.max_epochs,
fast_dev_run=False,
gradient_clip_val=args.gradient_clip_val,
num_sanity_val_steps=0,
check_val_every_n_epoch = 1,
strategy="ddp_find_unused_parameters_true",
)
else:
trainer = pl.Trainer(
callbacks = callbacks,
accelerator="gpu",
devices = args.gpus,
max_epochs = args.max_epochs,
fast_dev_run=False,
gradient_clip_val=args.gradient_clip_val,
num_sanity_val_steps=0,
check_val_every_n_epoch = 1,
)
DM = SciMMIR_FT_DataMoudle(
preprocess = preprocess,
tokenizer = tokenizer,
config = args,
)
if args.Use_BERT == True:
Model = SciMMIR_FT(
config = args,
model = model,
BERT_model = BERT_model,
)
else:
Model = SciMMIR_FT(
config = args,
model = model,
)
DM.setup( )
# Training Model
trainer.fit(model=Model, datamodule=DM)
# test
DM.setup('test')
trainer.test(model=Model, datamodule=DM, ckpt_path = 'best') #, ckpt_path = 'best' ; , ckpt_path = './checkpoints_384/MMIR_BLIP-FLAN-T5-XL_overall/checkpoints--Validation/MRR_forward=0.1153.ckpt'
if __name__ == '__main__':
parser = argparse.ArgumentParser(description="config for SciMMIR model")
parser.add_argument("--Use_BERT" , type = int , default = False)
parser.add_argument("--image_type" , type = str , default = 'overall')
# parser.add_argument("--image_type" , type = str , default = 'fig_architecture')
# parser.add_argument("--image_type" , type = str , default = 'fig_illustration')
# parser.add_argument("--image_type" , type = str , default = 'fig_result')
# parser.add_argument("--image_type" , type = str , default = 'table_result')
# parser.add_argument("--image_type" , type = str , default = 'table_parameter')
parser.add_argument("--multi_gpus" , type = int , default = False)
parser.add_argument("--use_ocr" , type = int , default = False)
#parser.add_argument("--ckpt_path" , type = str , default = '')
parser.add_argument("--select_print" , type = str , default = 'print_all_setting')
parser.add_argument("--score_method" , type = str , default = 'Matrix Dot Product')
parser.add_argument("--model_name" , type = str , default = 'CLIP')
parser.add_argument("--training_data_len" , type = int , default = 504731)
parser.add_argument("--figure_process_mat_num_valid", type=int, default=4)
parser.add_argument("--figure_process_mat_num_test", type=int, default=107)#107
parser.add_argument("--top_k" , type = int , default = 100)
parser.add_argument("--embedding_size" , type = int , default = 512)
parser.add_argument("--text_feature_len", type=int, default=768)
parser.add_argument("--image_feature_len", type=int, default=512)
parser.add_argument("--val_batch_size", type=int, default=128)
parser.add_argument("--test_batch_size", type=int, default=32)
parser.add_argument("--train_batch_size" , type = int , default = 210)
parser.add_argument("--save_top_k", type = int , default = 1)
parser.add_argument("--max_epochs" , type = int ,default = 5)
parser.add_argument("--text_process_num" , type = int , default = 2000)
parser.add_argument("--figure_process_num" , type = int , default = 800)
parser.add_argument("--gradient_clip_val", type = float , default= 5.0 )
parser.add_argument("--weight_decay", type=float, default= 0.01)
parser.add_argument("--alpha", type=float, default= 0.05)
parser.add_argument("--lr_CLIP", type=float, default= 2e-5)
parser.add_argument("--lr_BERT", type=float, default= 2e-5)
parser.add_argument("--lr_Linner", type=float, default=1e-4)
parser.add_argument("--epsilon", type=float, default= 1e-8)
parser.add_argument("--gpus", type=list, default=[0])
parser.add_argument("--context_length", type=int, default=77)
parser.add_argument("--image_size", type=int, default=224) #384
parser.add_argument("--datasets_saved_path", type=str, default="./data/")
parser.add_argument("--result_save_path", type=str, default="./data/result/")
parser.add_argument("--num_warmup_steps", type=int, default= floor(floor(parser.parse_args().training_data_len / parser.parse_args().train_batch_size) * parser.parse_args().max_epochs / 10))
parser.add_argument("--num_training_steps", type=int, default=floor(parser.parse_args().training_data_len / parser.parse_args().train_batch_size) * parser.parse_args().max_epochs)#12020
parser.add_argument("--fign2ocr", type=str, default= './data/fign2ocr_dic.json')
parser.add_argument("--text_2_image_index_valid", type=str, default=f"{parser.parse_args().model_name}_torch_data_valid/text_2_image_index.json")
parser.add_argument("--text_2_image_index_test", type=str, default=f"{parser.parse_args().model_name}_torch_data_test/text_2_image_index.json")
parser.add_argument("--figure_process_mat_path_valid", type=str, default=f"{parser.parse_args().model_name}_torch_data_valid/")
parser.add_argument("--figure_process_mat_path_test", type=str, default=f"{parser.parse_args().model_name}_torch_data_test/")
if parser.parse_args().Use_BERT == True:
model_name = "BERT"
else:
model_name = parser.parse_args().model_name
parser.add_argument("--text_2_index_valid", type=str, default=f"{model_name}_torch_data_valid/text_index.json")
parser.add_argument("--text_2_index_test", type=str, default=f"{model_name}_torch_data_test/text_index.json")
parser.add_argument("--text_process_mat_path_valid", type=str, default=f"{model_name}_torch_data_valid/")
parser.add_argument("--text_process_mat_path_test", type=str, default=f"{model_name}_torch_data_test/")
args = parser.parse_args()
print(args)
run(args)