-
Notifications
You must be signed in to change notification settings - Fork 1
/
pretrain.py
67 lines (47 loc) · 2.69 KB
/
pretrain.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
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--ID', default='0', help='run ID')
parser.add_argument('--gpu', default='1', type=str, help='gpu device numbers')
parser.add_argument('--load_trained', type=bool, default=True, help='load trained model or not')
parser.add_argument('--epoch', type=int, default=100,help='random seed')
parser.add_argument('--batch_size', type=int, default=10,help='batch_size')
parser.add_argument('--num_workers', type=int, default=4,help='num_of_workers')
parser.add_argument('--lr', type=float, default=1e-4,help='learning rate')
parser.add_argument('--dir_path', type=str, default="EMNLP/", help='file path')
parser.add_argument('--alm_path', type=str, default="./ckpt/saved_ckpt/ALM.pt", help='pretrained ALM path')
parser.add_argument('--print_every', type=int, default=100,help='how many iter for averaging results')
parser.add_argument('--seed', type=int, default=666,help='random seed')
# parser.add_argument('--sent_max_length', type=int, default=60,help='sent_max_length')
args = parser.parse_args()
import os
# main_gpu = int(args.gpu.split(",")[0])
# torch.cuda.set_device(main_gpu)
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
print("CUDA_VISIBLE_DEVICES", os.environ["CUDA_VISIBLE_DEVICES"])
from tqdm import tqdm
from transformers import RobertaModel, RobertaTokenizerFast
from trainer import Subst_Trainer
from tester import Subst_Tester
from subst_model import Subst_model
from lexsub_data_loader import get_lexsub_loader
CAND_MAX_LEN = 10
def main(args):
tokenizer = RobertaTokenizerFast.from_pretrained("roberta-large", cache_dir="./ckpt/roberta")
sentence_encoder = RobertaModel.from_pretrained("roberta-large", cache_dir="./ckpt/roberta")
lex_loader, test_loader = get_lexsub_loader(["coinco", "semeval_all"], "data", batch_size=args.batch_size,
tokenizer=tokenizer, candidate_max_length=CAND_MAX_LEN, num_workers=args.num_workers)
if torch.cuda.is_available():
device = "cuda"
else:
device = "cpu"
subst_model = Subst_model(sentence_encoder, tokenizer, device, candidate_max_length=CAND_MAX_LEN).to(device)
subst_model.load_state_dict(torch.load(args.alm_path))
optimizer = torch.optim.Adam(subst_model.parameters(), lr=args.lr, betas=(0.9, 0.99))
pretrainer = Subst_Trainer(subst_model, lex_loader, tokenizer, optimizer, test_loader,
device, num_of_epo=args.epoch, dir_path=args.dir_path+"pretrain/output", print_every=args.print_every)
pretrainer.train()
tester = Subst_Tester(subst_model, test_loader, args.dir_path+"pretrain/output", device)
tester.test()
if __name__ == "__main__":
main(args)