-
Notifications
You must be signed in to change notification settings - Fork 9
/
config.yaml
238 lines (238 loc) · 3.86 KB
/
config.yaml
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
ANYNET:
BLOCK_TYPE: res_bottleneck_block
BOT_MULS: []
DEPTHS: []
GROUP_WS: []
SE_ON: false
SE_R: 0.25
STEM_TYPE: simple_stem_in
STEM_W: 32
STRIDES: []
WIDTHS: []
BN:
CUSTOM_WEIGHT_DECAY: 0.0
EPS: 1.0e-05
MOM: 0.1
NUM_SAMPLES_PRECISE: 8192
USE_CUSTOM_WEIGHT_DECAY: false
USE_PRECISE_STATS: true
ZERO_INIT_FINAL_GAMMA: true
CFG_DEST: config.yaml
CUDNN:
BENCHMARK: true
DATA_LOADER:
NUM_WORKERS: 4
PIN_MEMORY: true
DIST_BACKEND: nccl
DOWNLOAD_CACHE: /tmp/pycls-download-cache
EN:
DC_RATIO: 0.0
DEPTHS: []
DROPOUT_RATIO: 0.0
EXP_RATIOS: []
HEAD_W: 1280
KERNELS: []
SE_R: 0.25
STEM_W: 32
STRIDES: []
WIDTHS: []
FINETUNE: false
HOST: localhost
LOG_DEST: stdout
LOG_PERIOD: 50
MEM:
RELU_INPLACE: true
MODEL:
DEPTH: 50
LOSS_FUN: cross_entropy
NUM_CLASSES: 1000
TYPE: resnet
NUM_GPUS: 8
OPTIM:
BASE_LR: 0.1
DAMPENING: 0.0
GAMMA: 0.1
LR_MULT: 0.1
LR_POLICY: steps
MAX_EPOCH: 100
METHOD: SGD
MOMENTUM: 0.9
NESTEROV: true
STEPS:
- 0
- 30
- 60
- 90
WARMUP_EPOCHS: 0
WARMUP_FACTOR: 0.1
WEIGHT_DECAY: 0.0001
OUT_DIR: .
PORT_RANGE:
- 10000
- 65000
PREC_TIME:
NUM_ITER: 30
WARMUP_ITER: 3
PRUNE:
FACTOR: null
GET_MASK: false
MIN_CHANNELS: 1
PRUNE_RESIDUAL: false
SPUER_NORM_LAYERS:
- - s1.b1.f.c_bn
- s1.b2.f.c_bn
- s1.b3.f.c_bn
- s1.b1.bn
- - s2.b1.f.c_bn
- s2.b2.f.c_bn
- s2.b3.f.c_bn
- s2.b4.f.c_bn
- s2.b1.bn
- - s3.b1.f.c_bn
- s3.b2.f.c_bn
- s3.b3.f.c_bn
- s3.b4.f.c_bn
- s3.b5.f.c_bn
- s3.b6.f.c_bn
- s3.b1.bn
- - s4.b1.f.c_bn
- s4.b2.f.c_bn
- s4.b3.f.c_bn
- s4.b1.bn
SPUER_PREC_LAYERS:
- - s1.b1.f.c
- s1.b2.f.c
- s1.b3.f.c
- s1.b1.proj
- - s2.b1.f.c
- s2.b2.f.c
- s2.b3.f.c
- s2.b4.f.c
- s2.b1.proj
- - s3.b1.f.c
- s3.b2.f.c
- s3.b3.f.c
- s3.b4.f.c
- s3.b5.f.c
- s3.b6.f.c
- s3.b1.proj
- - s4.b1.f.c
- s4.b2.f.c
- s4.b3.f.c
- s4.b1.proj
SPUER_SUCC_LAYERS:
- - s1.b2.f.a
- s1.b3.f.a
- s2.b1.f.a
- s2.b1.proj
- - s2.b2.f.a
- s2.b3.f.a
- s2.b4.f.a
- s3.b1.f.a
- s3.b1.proj
- - s3.b2.f.a
- s3.b3.f.a
- s3.b4.f.a
- s3.b5.f.a
- s3.b6.f.a
- s4.b1.f.a
- s4.b1.proj
- - s4.b2.f.a
- s4.b3.f.a
- head.fc
THRESHOLD_METHOD: th
USE_MASK: null
REBUILD: false
REGNET:
BLOCK_TYPE: res_bottleneck_block
BOT_MUL: 1.0
DEPTH: 10
GROUP_W: 16
SE_ON: false
SE_R: 0.25
STEM_TYPE: simple_stem_in
STEM_W: 32
STRIDE: 2
W0: 32
WA: 5.0
WM: 2.5
RESNET:
NUM_GROUPS: 1
STRIDE_1X1: false
TRANS_FUN: bottleneck_transform
WIDTH_PER_GROUP: 64
RNG_SEED: 1
SCRATCH: false
SPARSE:
MASK_DICT: null
NORM_LAYER_NAMES:
- stem.bn
- s1.b1.f.a_bn
- s1.b1.f.b_bn
- s1.b2.f.a_bn
- s1.b2.f.b_bn
- s1.b3.f.a_bn
- s1.b3.f.b_bn
- s1.b1.f.c_bn
- s1.b2.f.c_bn
- s1.b3.f.c_bn
- s1.b1.bn
- s2.b1.f.a_bn
- s2.b1.f.b_bn
- s2.b2.f.a_bn
- s2.b2.f.b_bn
- s2.b3.f.a_bn
- s2.b3.f.b_bn
- s2.b4.f.a_bn
- s2.b4.f.b_bn
- s2.b1.f.c_bn
- s2.b2.f.c_bn
- s2.b3.f.c_bn
- s2.b4.f.c_bn
- s2.b1.bn
- s3.b1.f.a_bn
- s3.b1.f.b_bn
- s3.b2.f.a_bn
- s3.b2.f.b_bn
- s3.b3.f.a_bn
- s3.b3.f.b_bn
- s3.b4.f.a_bn
- s3.b4.f.b_bn
- s3.b5.f.a_bn
- s3.b5.f.b_bn
- s3.b6.f.a_bn
- s3.b6.f.b_bn
- s3.b1.f.c_bn
- s3.b2.f.c_bn
- s3.b3.f.c_bn
- s3.b4.f.c_bn
- s3.b5.f.c_bn
- s3.b6.f.c_bn
- s3.b1.bn
- s4.b1.f.a_bn
- s4.b1.f.b_bn
- s4.b2.f.a_bn
- s4.b2.f.b_bn
- s4.b3.f.a_bn
- s4.b3.f.b_bn
- s4.b1.f.c_bn
- s4.b2.f.c_bn
- s4.b3.f.c_bn
- s4.b1.bn
REGULAR_METHOD: L1
SPARSITY: null
TEST:
BATCH_SIZE: 200
DATASET: imagenet
IM_SIZE: 256
SPLIT: val
WEIGHTS: ''
TRAIN:
AUTO_RESUME: true
BATCH_SIZE: 256
CHECKPOINT_PERIOD: 1
DATASET: imagenet
EVAL_PERIOD: 1
IM_SIZE: 224
SPLIT: train
WEIGHTS: paper/imagenet_uniform/prune_ratio_0.5/prune/th-p0.01_model.pt