-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_c2i.py
286 lines (257 loc) · 12.8 KB
/
train_c2i.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
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
import os
import math
import tqdm
import argparse
from omegaconf import OmegaConf
from contextlib import nullcontext
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader
from torch.utils.data.distributed import DistributedSampler
from torch.nn.parallel import DistributedDataParallel as DDP
from torchvision.utils import save_image
from datasets import CachedFolder
from models import make_vqmodel, EMA, MaskGITSampler
from utils.data import load_data
from utils.logger import get_logger
from utils.tracker import StatusTracker
from utils.misc import get_time_str, check_freq, set_seed
from utils.experiment import create_exp_dir, find_resume_checkpoint, instantiate_from_config
from utils.distributed import init_distributed_mode, is_main_process, on_main_process, is_dist_avail_and_initialized
from utils.distributed import wait_for_everyone, cleanup, gather_tensor, get_rank, get_world_size, get_local_rank, main_process_first
def get_parser():
parser = argparse.ArgumentParser()
parser.add_argument('-c', '--config', type=str, required=True, help='Path to configuration file')
parser.add_argument('-e', '--exp_dir', type=str, help='Path to the experiment directory. Default to be ./runs/exp-{current time}/')
parser.add_argument('-r', '--resume', type=str, help='Resume from a checkpoint. Could be a path or `best` or `latest`')
parser.add_argument('-mp', '--mixed_precision', type=str, default=None, choices=['fp16', 'bf16'], help='Mixed precision training')
parser.add_argument('-cd', '--cover_dir', action='store_true', default=False, help='Cover the experiment directory if it exists')
return parser
def main():
# PARSE ARGS AND CONFIGS
args, unknown_args = get_parser().parse_known_args()
args.time_str = get_time_str()
if args.exp_dir is None:
args.exp_dir = os.path.join('runs', f'exp-{args.time_str}')
unknown_args = [(a[2:] if a.startswith('--') else a) for a in unknown_args]
unknown_args = [f'{k}={v}' for k, v in zip(unknown_args[::2], unknown_args[1::2])]
conf = OmegaConf.load(args.config)
conf = OmegaConf.merge(conf, OmegaConf.from_dotlist(unknown_args))
# INITIALIZE DISTRIBUTED MODE
device = init_distributed_mode()
print(f'Process {get_rank()} using device: {device}', flush=True)
wait_for_everyone()
# CREATE EXPERIMENT DIRECTORY
exp_dir = args.exp_dir
if is_main_process():
create_exp_dir(
exp_dir=exp_dir, conf_yaml=OmegaConf.to_yaml(conf), subdirs=['ckpt', 'samples'],
time_str=args.time_str, exist_ok=args.resume is not None, cover_dir=args.cover_dir,
)
# INITIALIZE LOGGER
logger = get_logger(
log_file=os.path.join(exp_dir, f'output-{args.time_str}.log'),
use_tqdm_handler=True, is_main_process=is_main_process(),
)
# INITIALIZE STATUS TRACKER
status_tracker = StatusTracker(
logger=logger, print_freq=conf.train.print_freq,
tensorboard_dir=os.path.join(exp_dir, 'tensorboard'),
is_main_process=is_main_process(),
)
# SET MIXED PRECISION
if args.mixed_precision == 'fp16':
mp_dtype = torch.float16
elif args.mixed_precision == 'bf16':
mp_dtype = torch.bfloat16
else:
mp_dtype = torch.float32
# SET SEED
set_seed(conf.seed + get_rank())
logger.info('=' * 19 + ' System Info ' + '=' * 18)
logger.info(f'Experiment directory: {exp_dir}')
logger.info(f'Number of processes: {get_world_size()}')
logger.info(f'Distributed mode: {is_dist_avail_and_initialized()}')
logger.info(f'Mixed precision: {args.mixed_precision}')
wait_for_everyone()
# BUILD DATASET AND DATALOADER
assert conf.train.batch_size % get_world_size() == 0
bspp = conf.train.batch_size // get_world_size() # batch size per process
micro_batch_size = conf.train.micro_batch_size or bspp # actual batch size in each iteration
train_set = load_data(conf.data, split='all' if conf.data.name.lower() == 'ffhq' else 'train')
train_sampler = DistributedSampler(train_set, num_replicas=get_world_size(), rank=get_rank(), shuffle=True)
train_loader = DataLoader(train_set, batch_size=bspp, sampler=train_sampler, drop_last=True, **conf.dataloader)
logger.info('=' * 19 + ' Data Info ' + '=' * 20)
logger.info(f'Size of training set: {len(train_set)}')
logger.info(f'Batch size per process: {bspp}')
logger.info(f'Micro batch size: {micro_batch_size}')
logger.info(f'Gradient accumulation steps: {math.ceil(bspp / micro_batch_size)}')
logger.info(f'Total batch size: {conf.train.batch_size}')
# LOAD PRETRAINED VQMODEL
with main_process_first():
vqmodel = make_vqmodel(conf.vqmodel.model_name)
vqmodel = vqmodel.requires_grad_(False).eval().to(device)
logger.info('=' * 19 + ' Model Info ' + '=' * 19)
logger.info(f'Successfully load pretrained vqmodel: {conf.vqmodel.model_name}')
logger.info(f'Number of parameters of vqmodel: {sum(p.numel() for p in vqmodel.parameters()):,}')
# BUILD MODEL AND OPTIMIZERS
model = instantiate_from_config(conf.transformer).to(device)
ema = EMA(model.parameters(), **getattr(conf.train, 'ema', dict()))
optimizer = instantiate_from_config(conf.train.optim, params=model.parameters())
scheduler = instantiate_from_config(conf.train.sched, optimizer=optimizer)
scaler = torch.amp.GradScaler('cuda', enabled=args.mixed_precision == 'fp16')
logger.info(f'Number of parameters of transformer: {sum(p.numel() for p in model.parameters()):,}')
logger.info('=' * 50)
# BUILD SAMPLER
fm_size = conf.data.img_size // vqmodel.downsample_factor # feature map size
sampler = MaskGITSampler(model=model, sequence_length=fm_size ** 2, sampling_steps=8, device=device)
# RESUME TRAINING
step, epoch = 0, 0
if args.resume is not None:
resume_path = find_resume_checkpoint(exp_dir, args.resume)
logger.info(f'Resume from {resume_path}')
# load model
ckpt = torch.load(os.path.join(resume_path, 'model.pt'), map_location='cpu', weights_only=True)
model.load_state_dict(ckpt['model'])
logger.info(f'Successfully load model from {resume_path}')
# load training states (ema, optimizer, scheduler, scaler, step, epoch)
ckpt = torch.load(os.path.join(resume_path, 'training_states.pt'), map_location='cpu', weights_only=True)
ema.load_state_dict(ckpt['ema'])
optimizer.load_state_dict(ckpt['optimizer'])
scheduler.load_state_dict(ckpt['scheduler'])
scaler.load_state_dict(ckpt['scaler'])
step = ckpt['step'] + 1
epoch = ckpt['epoch']
logger.info(f'Successfully load training states from {resume_path}')
logger.info(f'Restart training at step {step}')
del ckpt
# PREPARE FOR DISTRIBUTED TRAINING
if is_dist_avail_and_initialized():
model = DDP(model, device_ids=[get_local_rank()], output_device=get_local_rank())
model_wo_ddp = model.module if is_dist_avail_and_initialized() else model
ema.to(device)
wait_for_everyone()
# TRAINING FUNCTIONS
@on_main_process
def save_ckpt(save_path: str):
os.makedirs(save_path, exist_ok=True)
# save model and ema model
torch.save(dict(model=model_wo_ddp.state_dict()), os.path.join(save_path, 'model.pt'))
with ema.scope(model.parameters()):
torch.save(dict(model=model_wo_ddp.state_dict()), os.path.join(save_path, 'model_ema.pt'))
# save training states (ema, optimizer, scheduler, scaler, step, epoch)
torch.save(dict(
ema=ema.state_dict(),
optimizer=optimizer.state_dict(),
scheduler=scheduler.state_dict(),
scaler=scaler.state_dict(),
step=step,
epoch=epoch,
), os.path.join(save_path, 'training_states.pt'))
def train_micro_batch(micro_batch, loss_scale, no_sync):
idx, y = micro_batch
B, L = idx.shape
with model.no_sync() if no_sync else nullcontext():
with torch.autocast(device_type='cuda', dtype=mp_dtype):
# transformer forward
mask = model_wo_ddp.get_random_mask(B, L) # (B, L)
masked_idx = torch.where(mask, model_wo_ddp.mask_token_id, idx) # (B, L)
preds = model(masked_idx, y=y, cond_drop_prob=conf.train.cond_drop_prob) # (B, L, C)
preds = preds.reshape(B * L, -1) # (B * L, C)
mask = mask.reshape(B * L) # (B * L)
# cross-entropy loss
target = idx.reshape(-1) # (B * L)
target = torch.where(mask, target, -100)
loss = F.cross_entropy(
input=preds, target=target, ignore_index=-100,
label_smoothing=conf.train.label_smoothing,
)
loss = loss * loss_scale
# backward
scaler.scale(loss).backward()
return loss
def train_step(batch):
# get data
if isinstance(train_set, CachedFolder):
idx = batch['idx'].long().to(device)
y = batch['y'].long().to(device)
B, L = idx.shape
else:
x = batch[0].float().to(device)
y = batch[1].long().to(device)
B, N = x.shape[0], conf.data.img_size // vqmodel.downsample_factor
L = N * N
# vqmodel encode
with torch.no_grad():
idx = vqmodel.encode(x)['indices'].reshape(B, L)
# zero the gradients
optimizer.zero_grad()
# forward and backward with gradient accumulation
loss = torch.tensor(0., device=device)
for i in range(0, B, micro_batch_size):
idx_micro_batch = idx[i:i+micro_batch_size]
y_micro_batch = y[i:i+micro_batch_size]
loss_scale = idx_micro_batch.shape[0] / B
no_sync = i + micro_batch_size < B and is_dist_avail_and_initialized()
loss_micro_batch = train_micro_batch((idx_micro_batch, y_micro_batch), loss_scale, no_sync)
loss = loss + loss_micro_batch
# optimize
if conf.train.get('clip_grad_norm', None):
scaler.unscale_(optimizer)
nn.utils.clip_grad_norm_(model.parameters(), max_norm=conf.train.clip_grad_norm)
scaler.step(optimizer)
scaler.update()
ema.update(model.parameters())
scheduler.step()
return dict(loss=loss.item(), lr=optimizer.param_groups[0]['lr'])
@torch.no_grad()
def sample(savepath):
samples_list = []
for c in conf.train.sample_class_ids:
n_samples = math.ceil(conf.train.n_samples_per_class / get_world_size())
y = torch.full((n_samples, ), c, dtype=torch.long, device=device)
with ema.scope(model.parameters()):
idx = sampler.sample(n_samples=n_samples, y=y)
samples = vqmodel.decode_indices(idx, shape=(n_samples, fm_size, fm_size, -1)).clamp(-1, 1)
samples = torch.cat(gather_tensor(samples), dim=0)[:conf.train.n_samples_per_class]
samples_list.append(samples.cpu())
samples = torch.cat(samples_list, dim=0)
if is_main_process():
nrow = conf.train.n_samples_per_class
save_image(samples, savepath, nrow=nrow, normalize=True, value_range=(-1, 1))
# START TRAINING
logger.info('Start training...')
while step < conf.train.n_steps:
if hasattr(train_loader.sampler, 'set_epoch'):
train_loader.sampler.set_epoch(epoch)
for _batch in tqdm.tqdm(train_loader, desc='Epoch', leave=False, disable=not is_main_process()):
# train a step
model.train()
train_status = train_step(_batch)
status_tracker.track_status('Train', train_status, step)
wait_for_everyone()
# validate
model.eval()
# save checkpoint
if check_freq(conf.train.save_freq, step):
save_ckpt(os.path.join(exp_dir, 'ckpt', f'step{step:0>7d}'))
wait_for_everyone()
# sample from current model
if check_freq(conf.train.sample_freq, step):
sample(os.path.join(exp_dir, 'samples', f'step{step:0>7d}.png'))
wait_for_everyone()
step += 1
if step >= conf.train.n_steps:
break
epoch += 1
# save the last checkpoint if not saved
if not check_freq(conf.train.save_freq, step - 1):
save_ckpt(os.path.join(exp_dir, 'ckpt', f'step{step-1:0>7d}'))
wait_for_everyone()
# END OF TRAINING
status_tracker.close()
cleanup()
logger.info('End of training')
if __name__ == '__main__':
main()