-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_util.py
50 lines (41 loc) · 1.52 KB
/
train_util.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
import torch
import numpy as np
import pickle
from tqdm import tqdm
def create_train_data(config, tokenizer, use_pickle=False) -> list:
r"""Create train data
If the train data is saved as pickle object load that.
else create, save and return the data.
"""
data = []
if use_pickle:
with open(config['train_data'], 'rb') as file:
data = pickle.load(file)
else:
with open(config['text_data'], 'r', encoding='utf-8') as file:
lines = file.readlines()
for i in tqdm(range(0, len(lines), 3)):
li = []
for line in lines[i: i+2]:
li.append(line[:config['max_len']])
data.append(tuple(map(tokenizer.encode, li)))
with open(config['train_data'], 'wb') as file:
pickle.dump(data, file)
return data
def subsequent_mask(seq):
r""" For masking out the subsequent info. """
_, size = seq.size()
subsequent_mask = (1 - torch.triu(
torch.ones((1, size, size), device=seq.device), diagonal=1)).bool()
return subsequent_mask
def create_masks(source, target, pad=0):
r"""Create source and target mask"""
source_mask = (source != pad).unsqueeze(-2).to(source.device)
target_mask = (target != pad).unsqueeze(-2).to(target.device)
target_mask = target_mask & subsequent_mask(target)
return source_mask, target_mask
def seed_everything(seed):
np.random.seed(seed)
torch.random.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.backends.cudnn.benchmark = True