forked from microsoft/Swin-Transformer
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main_simmim_pt.py
234 lines (194 loc) · 9.47 KB
/
main_simmim_pt.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
# --------------------------------------------------------
# SimMIM
# Copyright (c) 2021 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Ze Liu
# Modified by Zhenda Xie
# --------------------------------------------------------
import os
import time
import argparse
import datetime
import numpy as np
import torch
import torch.backends.cudnn as cudnn
import torch.distributed as dist
import torch.cuda.amp as amp
from timm.utils import AverageMeter
from config import get_config
from models import build_model
from data import build_loader
from lr_scheduler import build_scheduler
from optimizer import build_optimizer
from logger import create_logger
from utils_simmim import load_checkpoint, save_checkpoint, get_grad_norm, auto_resume_helper
# pytorch major version (1.x or 2.x)
PYTORCH_MAJOR_VERSION = int(torch.__version__.split('.')[0])
def parse_option():
parser = argparse.ArgumentParser('SimMIM pre-training script', add_help=False)
parser.add_argument('--cfg', type=str, required=True, metavar="FILE", help='path to config file', )
parser.add_argument(
"--opts",
help="Modify config options by adding 'KEY VALUE' pairs. ",
default=None,
nargs='+',
)
# easy config modification
parser.add_argument('--batch-size', type=int, help="batch size for single GPU")
parser.add_argument('--data-path', type=str, help='path to dataset')
parser.add_argument('--resume', help='resume from checkpoint')
parser.add_argument('--accumulation-steps', type=int, help="gradient accumulation steps")
parser.add_argument('--use-checkpoint', action='store_true',
help="whether to use gradient checkpointing to save memory")
parser.add_argument('--enable-amp', action='store_true')
parser.add_argument('--disable-amp', action='store_false', dest='enable_amp')
parser.set_defaults(enable_amp=True)
parser.add_argument('--output', default='output', type=str, metavar='PATH',
help='root of output folder, the full path is <output>/<model_name>/<tag> (default: output)')
parser.add_argument('--tag', help='tag of experiment')
# distributed training
# for pytorch >= 2.0, use `os.environ['LOCAL_RANK']` instead
# (see https://pytorch.org/docs/stable/distributed.html#launch-utility)
if PYTORCH_MAJOR_VERSION == 1:
parser.add_argument("--local_rank", type=int, required=True, help='local rank for DistributedDataParallel')
args = parser.parse_args()
config = get_config(args)
return args, config
def main(config):
data_loader_train = build_loader(config, simmim=True, is_pretrain=True)
logger.info(f"Creating model:{config.MODEL.TYPE}/{config.MODEL.NAME}")
model = build_model(config, is_pretrain=True)
model.cuda()
logger.info(str(model))
optimizer = build_optimizer(config, model, simmim=True, is_pretrain=True)
model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[config.LOCAL_RANK], broadcast_buffers=False)
model_without_ddp = model.module
n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
logger.info(f"number of params: {n_parameters}")
if hasattr(model_without_ddp, 'flops'):
flops = model_without_ddp.flops()
logger.info(f"number of GFLOPs: {flops / 1e9}")
lr_scheduler = build_scheduler(config, optimizer, len(data_loader_train))
scaler = amp.GradScaler()
if config.TRAIN.AUTO_RESUME:
resume_file = auto_resume_helper(config.OUTPUT, logger)
if resume_file:
if config.MODEL.RESUME:
logger.warning(f"auto-resume changing resume file from {config.MODEL.RESUME} to {resume_file}")
config.defrost()
config.MODEL.RESUME = resume_file
config.freeze()
logger.info(f'auto resuming from {resume_file}')
else:
logger.info(f'no checkpoint found in {config.OUTPUT}, ignoring auto resume')
if config.MODEL.RESUME:
load_checkpoint(config, model_without_ddp, optimizer, lr_scheduler, scaler, logger)
logger.info("Start training")
start_time = time.time()
for epoch in range(config.TRAIN.START_EPOCH, config.TRAIN.EPOCHS):
data_loader_train.sampler.set_epoch(epoch)
train_one_epoch(config, model, data_loader_train, optimizer, epoch, lr_scheduler, scaler)
if dist.get_rank() == 0 and (epoch % config.SAVE_FREQ == 0 or epoch == (config.TRAIN.EPOCHS - 1)):
save_checkpoint(config, epoch, model_without_ddp, 0., optimizer, lr_scheduler, scaler, logger)
total_time = time.time() - start_time
total_time_str = str(datetime.timedelta(seconds=int(total_time)))
logger.info('Training time {}'.format(total_time_str))
def train_one_epoch(config, model, data_loader, optimizer, epoch, lr_scheduler, scaler):
model.train()
optimizer.zero_grad()
num_steps = len(data_loader)
batch_time = AverageMeter()
loss_meter = AverageMeter()
norm_meter = AverageMeter()
loss_scale_meter = AverageMeter()
start = time.time()
end = time.time()
for idx, (img, mask, _) in enumerate(data_loader):
img = img.cuda(non_blocking=True)
mask = mask.cuda(non_blocking=True)
with amp.autocast(enabled=config.ENABLE_AMP):
loss = model(img, mask)
if config.TRAIN.ACCUMULATION_STEPS > 1:
loss = loss / config.TRAIN.ACCUMULATION_STEPS
scaler.scale(loss).backward()
if config.TRAIN.CLIP_GRAD:
scaler.unscale_(optimizer)
grad_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), config.TRAIN.CLIP_GRAD)
else:
grad_norm = get_grad_norm(model.parameters())
if (idx + 1) % config.TRAIN.ACCUMULATION_STEPS == 0:
scaler.step(optimizer)
optimizer.zero_grad()
scaler.update()
lr_scheduler.step_update(epoch * num_steps + idx)
else:
optimizer.zero_grad()
scaler.scale(loss).backward()
if config.TRAIN.CLIP_GRAD:
scaler.unscale_(optimizer)
grad_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), config.TRAIN.CLIP_GRAD)
else:
grad_norm = get_grad_norm(model.parameters())
scaler.step(optimizer)
scaler.update()
lr_scheduler.step_update(epoch * num_steps + idx)
torch.cuda.synchronize()
loss_meter.update(loss.item(), img.size(0))
norm_meter.update(grad_norm)
loss_scale_meter.update(scaler.get_scale())
batch_time.update(time.time() - end)
end = time.time()
if idx % config.PRINT_FREQ == 0:
lr = optimizer.param_groups[0]['lr']
memory_used = torch.cuda.max_memory_allocated() / (1024.0 * 1024.0)
etas = batch_time.avg * (num_steps - idx)
logger.info(
f'Train: [{epoch}/{config.TRAIN.EPOCHS}][{idx}/{num_steps}]\t'
f'eta {datetime.timedelta(seconds=int(etas))} lr {lr:.6f}\t'
f'time {batch_time.val:.4f} ({batch_time.avg:.4f})\t'
f'loss {loss_meter.val:.4f} ({loss_meter.avg:.4f})\t'
f'grad_norm {norm_meter.val:.4f} ({norm_meter.avg:.4f})\t'
f'loss_scale {loss_scale_meter.val:.4f} ({loss_scale_meter.avg:.4f})\t'
f'mem {memory_used:.0f}MB')
epoch_time = time.time() - start
logger.info(f"EPOCH {epoch} training takes {datetime.timedelta(seconds=int(epoch_time))}")
if __name__ == '__main__':
_, config = parse_option()
if 'RANK' in os.environ and 'WORLD_SIZE' in os.environ:
rank = int(os.environ["RANK"])
world_size = int(os.environ['WORLD_SIZE'])
print(f"RANK and WORLD_SIZE in environ: {rank}/{world_size}")
else:
rank = -1
world_size = -1
torch.cuda.set_device(config.LOCAL_RANK)
torch.distributed.init_process_group(backend='nccl', init_method='env:https://', world_size=world_size, rank=rank)
torch.distributed.barrier()
seed = config.SEED + dist.get_rank()
torch.manual_seed(seed)
np.random.seed(seed)
cudnn.benchmark = True
# linear scale the learning rate according to total batch size, may not be optimal
linear_scaled_lr = config.TRAIN.BASE_LR * config.DATA.BATCH_SIZE * dist.get_world_size() / 512.0
linear_scaled_warmup_lr = config.TRAIN.WARMUP_LR * config.DATA.BATCH_SIZE * dist.get_world_size() / 512.0
linear_scaled_min_lr = config.TRAIN.MIN_LR * config.DATA.BATCH_SIZE * dist.get_world_size() / 512.0
# gradient accumulation also need to scale the learning rate
if config.TRAIN.ACCUMULATION_STEPS > 1:
linear_scaled_lr = linear_scaled_lr * config.TRAIN.ACCUMULATION_STEPS
linear_scaled_warmup_lr = linear_scaled_warmup_lr * config.TRAIN.ACCUMULATION_STEPS
linear_scaled_min_lr = linear_scaled_min_lr * config.TRAIN.ACCUMULATION_STEPS
config.defrost()
config.TRAIN.BASE_LR = linear_scaled_lr
config.TRAIN.WARMUP_LR = linear_scaled_warmup_lr
config.TRAIN.MIN_LR = linear_scaled_min_lr
config.freeze()
os.makedirs(config.OUTPUT, exist_ok=True)
logger = create_logger(output_dir=config.OUTPUT, dist_rank=dist.get_rank(), name=f"{config.MODEL.NAME}")
if dist.get_rank() == 0:
path = os.path.join(config.OUTPUT, "config.json")
with open(path, "w") as f:
f.write(config.dump())
logger.info(f"Full config saved to {path}")
# print config
logger.info(config.dump())
main(config)