/
large_gpu.py
39 lines (31 loc) · 1.35 KB
/
large_gpu.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
from src.models import model_factory
from src.dataloaders import dataloader_factory
from src.datasets import dataset_factory
from src.trainers import trainer_factory
from src.utils.options import parser
from src.utils.utils import *
if __name__ == '__main__':
# hyper-parameter config
args = parser.parse_args()
ckpt_root = setup_train(args)
# dataset and data loader
dataset = dataset_factory(args)
train_loader, val_loader, test_loader, dataset = dataloader_factory(args, dataset)
# pretrained products vectors
pretrained_item_vectors = None
if args.use_pretrained_vectors:
pretrained_item_vectors = dataset.meta['p2v']
# model setup
model = model_factory(args, pretrained_item_vectors)
if args.load_pretrained_weights is not None:
print("weights loading from %s ..." % args.load_pretrained_weights)
model = load_pretrained_weights(model, args.load_pretrained_weights)
print("Model size:", sum(p.numel() for p in model.parameters() if p.requires_grad))
# trainer setup
trainer = trainer_factory(args, model, train_loader, val_loader, test_loader, ckpt_root, dataset.data)
# model training
trainer.train()
# model testing and saving
trainer.test()
trainer.logger_service.complete({'state_dict': (trainer._create_state_dict())})
trainer.writer.close()