forked from mit-han-lab/once-for-all
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_ofa_net.py
223 lines (194 loc) · 8.24 KB
/
train_ofa_net.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
# Once for All: Train One Network and Specialize it for Efficient Deployment
# Han Cai, Chuang Gan, Tianzhe Wang, Zhekai Zhang, Song Han
# International Conference on Learning Representations (ICLR), 2020.
import argparse
import numpy as np
import os
import random
import horovod.torch as hvd
import torch
from ofa.elastic_nn.modules.dynamic_op import DynamicSeparableConv2d
from ofa.elastic_nn.networks import OFAMobileNetV3
from ofa.imagenet_codebase.run_manager import DistributedImageNetRunConfig
from ofa.imagenet_codebase.run_manager.distributed_run_manager import DistributedRunManager
from ofa.imagenet_codebase.data_providers.base_provider import MyRandomResizedCrop
from ofa.utils import download_url
from ofa.elastic_nn.training.progressive_shrinking import load_models
parser = argparse.ArgumentParser()
parser.add_argument('--task', type=str, default='depth', choices=[
'kernel', 'depth', 'expand',
])
parser.add_argument('--phase', type=int, default=1, choices=[1, 2])
args = parser.parse_args()
if args.task == 'kernel':
args.path = 'exp/normal2kernel'
args.dynamic_batch_size = 1
args.n_epochs = 120
args.base_lr = 3e-2
args.warmup_epochs = 5
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '6'
args.depth_list = '4'
elif args.task == 'depth':
args.path = 'exp/kernel2kernel_depth/phase%d' % args.phase
args.dynamic_batch_size = 2
if args.phase == 1:
args.n_epochs = 25
args.base_lr = 2.5e-3
args.warmup_epochs = 0
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '6'
args.depth_list = '3,4'
else:
args.n_epochs = 120
args.base_lr = 7.5e-3
args.warmup_epochs = 5
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '6'
args.depth_list = '2,3,4'
elif args.task == 'expand':
args.path = 'exp/kernel_depth2kernel_depth_width/phase%d' % args.phase
args.dynamic_batch_size = 4
if args.phase == 1:
args.n_epochs = 25
args.base_lr = 2.5e-3
args.warmup_epochs = 0
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '4,6'
args.depth_list = '2,3,4'
else:
args.n_epochs = 120
args.base_lr = 7.5e-3
args.warmup_epochs = 5
args.warmup_lr = -1
args.ks_list = '3,5,7'
args.expand_list = '3,4,6'
args.depth_list = '2,3,4'
else:
raise NotImplementedError
args.manual_seed = 0
args.lr_schedule_type = 'cosine'
args.base_batch_size = 64
args.valid_size = 10000
args.opt_type = 'sgd'
args.momentum = 0.9
args.no_nesterov = False
args.weight_decay = 3e-5
args.label_smoothing = 0.1
args.no_decay_keys = 'bn#bias'
args.fp16_allreduce = False
args.model_init = 'he_fout'
args.validation_frequency = 1
args.print_frequency = 10
args.n_worker = 8
args.resize_scale = 0.08
args.distort_color = 'tf'
args.image_size = '128,160,192,224'
args.continuous_size = True
args.not_sync_distributed_image_size = False
args.bn_momentum = 0.1
args.bn_eps = 1e-5
args.dropout = 0.1
args.base_stage_width = 'proxyless'
args.width_mult_list = '1.0'
args.dy_conv_scaling_mode = 1
args.independent_distributed_sampling = False
args.kd_ratio = 1.0
args.kd_type = 'ce'
if __name__ == '__main__':
os.makedirs(args.path, exist_ok=True)
# Initialize Horovod
hvd.init()
# Pin GPU to be used to process local rank (one GPU per process)
torch.cuda.set_device(hvd.local_rank())
args.teacher_path = download_url(
'https://hanlab.mit.edu/files/OnceForAll/ofa_checkpoints/ofa_D4_E6_K7',
model_dir='.torch/ofa_checkpoints/%d' % hvd.rank()
)
num_gpus = hvd.size()
torch.manual_seed(args.manual_seed)
torch.cuda.manual_seed_all(args.manual_seed)
np.random.seed(args.manual_seed)
random.seed(args.manual_seed)
# image size
args.image_size = [int(img_size) for img_size in args.image_size.split(',')]
if len(args.image_size) == 1:
args.image_size = args.image_size[0]
MyRandomResizedCrop.CONTINUOUS = args.continuous_size
MyRandomResizedCrop.SYNC_DISTRIBUTED = not args.not_sync_distributed_image_size
# build run config from args
args.lr_schedule_param = None
args.opt_param = {
'momentum': args.momentum,
'nesterov': not args.no_nesterov,
}
args.init_lr = args.base_lr * num_gpus # linearly rescale the learning rate
if args.warmup_lr < 0:
args.warmup_lr = args.base_lr
args.train_batch_size = args.base_batch_size
args.test_batch_size = args.base_batch_size * 4
run_config = DistributedImageNetRunConfig(**args.__dict__, num_replicas=num_gpus, rank=hvd.rank())
# print run config information
if hvd.rank() == 0:
print('Run config:')
for k, v in run_config.config.items():
print('\t%s: %s' % (k, v))
if args.dy_conv_scaling_mode == -1:
args.dy_conv_scaling_mode = None
DynamicSeparableConv2d.KERNEL_TRANSFORM_MODE = args.dy_conv_scaling_mode
# build net from args
args.width_mult_list = [float(width_mult) for width_mult in args.width_mult_list.split(',')]
args.ks_list = [int(ks) for ks in args.ks_list.split(',')]
args.expand_list = [int(e) for e in args.expand_list.split(',')]
args.depth_list = [int(d) for d in args.depth_list.split(',')]
net = OFAMobileNetV3(
n_classes=run_config.data_provider.n_classes, bn_param=(args.bn_momentum, args.bn_eps),
dropout_rate=args.dropout, base_stage_width=args.base_stage_width, width_mult_list=args.width_mult_list,
ks_list=args.ks_list, expand_ratio_list=args.expand_list, depth_list=args.depth_list
)
# teacher model
if args.kd_ratio > 0:
args.teacher_model = OFAMobileNetV3(
n_classes=run_config.data_provider.n_classes, bn_param=(args.bn_momentum, args.bn_eps),
dropout_rate=0, width_mult_list=1.0, ks_list=7, expand_ratio_list=6, depth_list=4,
)
args.teacher_model.cuda()
""" Distributed RunManager """
# Horovod: (optional) compression algorithm.
compression = hvd.Compression.fp16 if args.fp16_allreduce else hvd.Compression.none
distributed_run_manager = DistributedRunManager(
args.path, net, run_config, compression, backward_steps=args.dynamic_batch_size, is_root=(hvd.rank() == 0)
)
distributed_run_manager.save_config()
# hvd broadcast
distributed_run_manager.broadcast()
# load teacher net weights
if args.kd_ratio > 0:
load_models(distributed_run_manager, args.teacher_model, model_path=args.teacher_path)
# training
from ofa.elastic_nn.training.progressive_shrinking import validate, train
validate_func_dict = {'image_size_list': {224} if isinstance(args.image_size, int) else sorted({160, 224}),
'width_mult_list': sorted({0, len(args.width_mult_list) - 1}),
'ks_list': sorted({min(args.ks_list), max(args.ks_list)}),
'expand_ratio_list': sorted({min(args.expand_list), max(args.expand_list)}),
'depth_list': sorted({min(net.depth_list), max(net.depth_list)})}
if args.task == 'kernel':
validate_func_dict['ks_list'] = sorted(args.ks_list)
if distributed_run_manager.start_epoch == 0:
model_path = download_url('https://hanlab.mit.edu/files/OnceForAll/ofa_checkpoints/ofa_D4_E6_K7',
model_dir='.torch/ofa_checkpoints/%d' % hvd.rank())
load_models(distributed_run_manager, distributed_run_manager.net, model_path=model_path)
distributed_run_manager.write_log('%.3f\t%.3f\t%.3f\t%s' %
validate(distributed_run_manager, **validate_func_dict), 'valid')
train(distributed_run_manager, args,
lambda _run_manager, epoch, is_test: validate(_run_manager, epoch, is_test, **validate_func_dict))
elif args.task == 'depth':
from ofa.elastic_nn.training.progressive_shrinking import supporting_elastic_depth
supporting_elastic_depth(train, distributed_run_manager, args, validate_func_dict)
else:
from ofa.elastic_nn.training.progressive_shrinking import supporting_elastic_expand
supporting_elastic_expand(train, distributed_run_manager, args, validate_func_dict)