-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain.py
175 lines (160 loc) · 6.48 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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
from __future__ import division
# Standard Library
import argparse
import sys
import os
os.environ['DEFAULT_TASKS']="cls"
# Import from third library
import torch.multiprocessing as mp
from up.utils.env.dist_helper import setup_distributed, finalize, gpu_check
from up.utils.general.yaml_loader import load_yaml # IncludeLoader
from up.utils.env.launch import launch
from up.utils.general.user_analysis_helper import send_info
# Import from local
from up.commands.subcommand import Subcommand
from up.utils.general.registry_factory import SUBCOMMAND_REGISTRY, RUNNER_REGISTRY
from up.utils.general.global_flag import DIST_BACKEND
__all__ = ['Train']
def add_arguments(parser):
parser.add_argument('-e',
'--evaluate',
dest='evaluate',
action='store_true',
help='evaluate model on validation set')
parser.add_argument('--debug',
dest='debug',
default=False,
help='debug')
parser.add_argument(
'--fork-method',
dest='fork_method',
type=str,
default='fork',
choices=['spawn', 'fork'],
help='method to fork subprocess, especially for dataloader')
parser.add_argument('--backend',
dest='backend',
type=str,
default='dist',
help='model backend')
parser.add_argument(
'--nocudnn',
dest='nocudnn',
action='store_true',
help='Whether to use cudnn backend or not. Please disable cudnn when running on V100'
)
parser.add_argument(
'--allow_dead_parameter',
action='store_true',
help='dead parameter (defined in model but not used in forward pass) is allowed'
)
parser.add_argument('--config',
dest='config',
required=True,
help='settings of detection in yaml format')
parser.add_argument('--display',
dest='display',
type=int,
default=20,
help='display intervel')
parser.add_argument('--async',
dest='asynchronize',
action='store_true',
help='whether to use asynchronize mode(linklink)')
parser.add_argument('--ng', '--num_gpus_per_machine',
dest='num_gpus_per_machine',
type=int,
default=1,
help='num_gpus_per_machine')
parser.add_argument('--nm', '--num_machines',
dest='num_machines',
type=int,
default=1,
help='num_machines')
parser.add_argument('--launch',
dest='launch',
type=str,
default='pytorch',
help='launch backend')
parser.add_argument('--port',
dest='port',
type=int,
default=13333,
help='dist port')
parser.add_argument('--no_running_config',
action='store_true',
help='disable display running config')
parser.add_argument('--phase', default='train', help="train phase")
parser.add_argument('--cfg_type',
dest='cfg_type',
type=str,
default='up',
help='config type (up or pod)')
parser.add_argument('--opts',
help='options to replace yaml config',
default=None,
nargs=argparse.REMAINDER)
parser.add_argument('--test_gpu',
dest='test_gpu',
action='store_true',
help='test if gpus work properly before training')
parser.set_defaults(run=_main)
# return parser
def main(args):
if args.test_gpu:
gpu_check()
cfg = load_yaml(args.config, args.cfg_type)
cfg['args'] = {
'ddp': args.backend == 'dist',
'config_path': args.config,
'asynchronize': args.asynchronize,
'nocudnn': args.nocudnn,
'display': args.display,
'no_running_config': args.no_running_config,
'allow_dead_parameter': args.allow_dead_parameter,
'opts': args.opts,
'debug': args.debug
}
train_phase = args.phase
cfg['runtime'] = cfg.setdefault('runtime', {})
runner_cfg = cfg['runtime'].get('runner', {})
runner_cfg['type'] = runner_cfg.get('type', 'base')
runner_cfg['kwargs'] = runner_cfg.get('kwargs', {})
cfg['runtime']['runner'] = runner_cfg
training = True
if args.evaluate:
training = False
train_phase = "eval"
send_info(cfg, train_phase)
runner_cfg['kwargs']['training'] = training
runner = RUNNER_REGISTRY.get(runner_cfg['type'])(cfg, **runner_cfg['kwargs'])
train_func = {"train": runner.train, "eval": runner.evaluate}
if runner_cfg['type'] == 'bignas':
train_func = {
"train_supnet": runner.train,
"sample_flops": runner.sample_multiple_subnet_flops,
"sample_accuracy":
runner.sample_multiple_subnet_accuracy,
"evaluate_subnet": runner.evaluate_subnet,
"finetune_subnet": runner.finetune_subnet,
"sample_subnet": runner.sample_subnet_weight
}
assert train_phase in train_func, f"{train_phase} is not supported"
train_func[train_phase]()
finalize()
def _main(args):
DIST_BACKEND.backend = args.backend
if args.launch == 'pytorch':
launch(main, args.num_gpus_per_machine, args.num_machines, args=args, start_method=args.fork_method)
else:
mp.set_start_method(args.fork_method, force=True)
fork_method = mp.get_start_method(allow_none=True)
assert fork_method == args.fork_method
sys.stdout.flush()
setup_distributed(args.port, args.launch, args.backend)
main(args)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Run Training")
add_arguments(parser)
args = parser.parse_args()
args.run(args)