-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy patheval_ofa_net.py
166 lines (140 loc) · 5.36 KB
/
eval_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
"""
Randomly samples --n_arch subnets from a given OFAKWSNet.
Evaluates them and saves their performance results to json files.
"""
import argparse
import os
import random
import time
import numpy as np
import torch
from once_for_all.elastic_nn.networks.ofa_kws_net import OFAKWSNet
from once_for_all.elastic_nn.training.progressive_shrinking import load_models
from once_for_all.evaluation.perf_dataset import PerformanceDataset
from once_for_all.run_manager import RunManager, KWSRunConfig
from utils.config_utils import set_ft_extr_params_to_args
parser = argparse.ArgumentParser()
parser.add_argument("--ft_extr_type",
type=str,
default="mfcc",
choices=[
"mfcc",
"mel_spectrogram",
"log_mel_spectrogram",
"spectrogram",
"log_spectrogram",
"linear_stft",
"raw"
])
parser.add_argument("--params_id", type=int, default=1)
parser.add_argument("--n_arch", type=int, default=5)
parser.add_argument('--use_csv', action='store_true')
parser.add_argument('--use_json', dest='use_csv', action='store_false')
parser.set_defaults(use_csv=False)
parser.add_argument("--load_from",
type=str,
default="expand",
choices=[
"normal",
"kernel",
"depth",
"expand",
])
parser.add_argument('--measure_latency', action='store_true')
parser.set_defaults(measure_latency=True)
args = parser.parse_args()
# Path parameters
args.path = "eval/" + args.ft_extr_type + str(args.params_id) + "_v4" + "/"
args.ofa_checkpoint_path = "exp/" + args.ft_extr_type + str(args.params_id)
"""Set which model step to evaluate, width_mult_list, ks_list, expand_list and depth_list"""
if args.load_from == "normal":
args.width_mult_list = "1.0"
args.ks_list = "7"
args.depth_list = "4"
args.expand_list = "3"
args.ofa_checkpoint_path += "/normal/checkpoint/model_best.pth.tar"
elif args.load_from == "kernel":
args.width_mult_list = "1.0"
args.ks_list = "3,5,7"
args.depth_list = "4"
args.expand_list = "3"
args.ofa_checkpoint_path += "/normal2kernel/checkpoint/model_best.pth.tar"
elif args.load_from == "depth":
args.width_mult_list = "1.0"
args.ks_list = "3,5,7"
args.depth_list = "1,2,3,4"
args.expand_list = "3"
args.ofa_checkpoint_path += "/kernel2kernel_depth/phase2/checkpoint/model_best.pth.tar"
elif args.load_from == "expand":
args.width_mult_list = "1.0"
args.ks_list = "3,5,7"
args.depth_list = "1,2,3,4"
args.expand_list = "1,2,3"
args.ofa_checkpoint_path += "/kernel_depth2kernel_depth_expand/phase2/checkpoint/model_best.pth.tar"
args.measure_latency = "gpu4#cpu" if args.measure_latency else None
args.measure_latency = None
"""Set ft_extr_params_list depending on the ft_extr_type"""
args = set_ft_extr_params_to_args(args)
# Other parameters
args.manual_seed = 0
args.n_worker = 8
args.bn_momentum = 0.1
args.bn_eps = 1e-5
args.dropout = 0.1
if __name__ == "__main__":
os.makedirs(args.path, exist_ok=True)
start = time.time()
num_gpus = torch.cuda.device_count()
print("Using %f cuda devices" % num_gpus)
# Set random seed
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)
# Cuda Setup
if torch.cuda.is_available():
# Pin GPU to be used to process local rank (one GPU per process)
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
torch.cuda.manual_seed(args.manual_seed)
print('Using GPU.')
else:
print('Using CPU.')
args.base_batch_size = 128
args.test_batch_size = args.base_batch_size * 4
run_config = KWSRunConfig(**args.__dict__, num_replicas=num_gpus)
# Print run config information
print("Run config:")
for k, v in run_config.config.items():
print("\t%s: %s" % (k, v))
args.ks_list = [int(ks) for ks in args.ks_list.split(",")]
args.depth_list = [int(d) for d in args.depth_list.split(",")]
args.expand_list = [int(e) for e in args.expand_list.split(",")]
args.width_mult_list = [float(w) for w in args.width_mult_list.split(",")]
"""Instantiate OFAKWSNet and load trained model"""
ofa_net = OFAKWSNet(
n_classes=12,
bn_param=(args.bn_momentum, args.bn_eps),
dropout_rate=args.dropout,
width_mult_list=args.width_mult_list,
ks_list=args.ks_list,
expand_ratio_list=args.expand_list,
depth_list=args.depth_list,
)
""" RunManager """
run_manager = RunManager(
args.path,
ofa_net,
run_config,
)
load_models(
run_manager,
run_manager.net,
args.ofa_checkpoint_path,
)
"""
Create & build the performance dataset
build_dataset randomly samples n_arch subnets and saves their config, accuracy, n_params, flops, latency
"""
performance_dataset = PerformanceDataset(args.path, use_csv=args.use_csv)
performance_dataset.build_dataset(run_manager, ofa_net, n_arch=args.n_arch, ft_extr_params_list=args.ft_extr_params_list, measure_latency=args.measure_latency)