-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
155 lines (110 loc) · 5.38 KB
/
train.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
import os
import torch
from torch.utils.data import DataLoader
from torch import nn
from torch.optim import Adam
from pathlib import Path
from tqdm.auto import tqdm
from torch.nn.utils import clip_grad_norm_
def exists(x):
return x is not None
def cycle(dl):
while True:
for data in dl:
yield data
class Trainer(object):
def __init__(self, preprocess_model, diffusion_model, data_tensor, results_folder='./Model', train_lr=1e-4,
train_num_steps=10000, adam_betas=(0.9, 0.99), train_batch_size=32, shuffle=True, pre_epoch=10000,
gradient_accumulate_every=5):
super().__init__()
self.preprocess_model = preprocess_model
self.pre_epoch = pre_epoch
self.model = diffusion_model
self.device = diffusion_model.alphas_cumprod.device
self.batch_size = train_batch_size
self.train_num_steps = train_num_steps
self.seq_length = diffusion_model.seq_length
self.gradient_accumulate_every = gradient_accumulate_every
self.save_cycle = 10000
# assert train_num_steps % 10 == 0, 'number of train steps must be n*10'
dl = DataLoader(data_tensor, batch_size=train_batch_size, shuffle=shuffle, num_workers=0)
self.dl = cycle(dl)
self.step = 0
self.results_folder = Path(results_folder)
self.results_folder.mkdir(exist_ok=True)
self.opt = Adam(diffusion_model.parameters(), lr=train_lr, betas=adam_betas)
self.opt_ae = Adam(preprocess_model.parameters(), lr=train_lr, betas=adam_betas)
self.loss = nn.L1Loss().to(self.device)
def save(self, milestone):
data = {
'step': self.step,
'model': self.model.state_dict(),
'opt': self.opt.state_dict(),
}
torch.save(data, str(self.results_folder / f'model-{milestone}.pt'))
def load(self, milestone):
device = self.device
data = torch.load(str(self.results_folder / f'model-{milestone}.pt'), map_location=device)
self.model.load_state_dict(data['model'])
self.step = data['step']
self.opt.load_state_dict(data['opt'])
self.preprocess_model.load_state_dict(torch.load(os.path.join(self.results_folder, "er.pt")))
def train(self):
device = self.device
"""First, train embedding and recovery network."""
print("Start Embedding Network Training.")
with tqdm(initial=self.step, total=self.pre_epoch) as pbar:
while self.step < self.pre_epoch:
total_loss = 0.
for _ in range(self.gradient_accumulate_every):
data = next(self.dl).to(device)
data_hat = self.preprocess_model(data)
loss = self.loss(data_hat, data)
loss = loss / self.gradient_accumulate_every
loss.backward()
total_loss += loss.item()
pbar.set_description(f'loss: {total_loss:.6f}')
clip_grad_norm_(self.preprocess_model.parameters(), 1.0)
self.opt_ae.step()
self.opt_ae.zero_grad()
self.step += 1
pbar.update(1)
print("Finish Embedding Network Training. Now Start Joint Training.")
self.step = 0
milestone = 0
with tqdm(initial=self.step, total=self.train_num_steps) as pbar:
while self.step < self.train_num_steps:
total_loss = 0.
for _ in range(2):
loss_ae = 0.
for _ in range(self.gradient_accumulate_every):
data = next(self.dl).to(device)
data_hat = self.preprocess_model(data)
er_loss = self.loss(data_hat, data)
er_loss = er_loss / self.gradient_accumulate_every
er_loss.backward()
loss_ae += er_loss.item()
clip_grad_norm_(self.preprocess_model.parameters(), 1.0)
self.opt_ae.step()
self.opt_ae.zero_grad()
self.preprocess_model.requires_grad_(False)
for _ in range(self.gradient_accumulate_every):
data = next(self.dl).to(device)
data = self.preprocess_model.embedding(data)
loss = self.model(data.reshape(-1, 8, 8))
loss = loss / self.gradient_accumulate_every
loss.backward()
total_loss += loss.item()
pbar.set_description(f'loss: {total_loss:.6f} and AE_loss: {loss_ae:.6f}')
clip_grad_norm_(self.model.parameters(), 1.0)
self.opt.step()
self.opt.zero_grad()
self.preprocess_model.requires_grad_(True)
self.step += 1
with torch.no_grad():
if self.step != 0 and self.step % self.save_cycle == 0:
milestone = self.step // self.save_cycle
self.save(milestone)
pbar.update(1)
torch.save(self.preprocess_model.state_dict(), os.path.join(self.results_folder, "er.pt"))
print('training complete')