-
Notifications
You must be signed in to change notification settings - Fork 6
/
resnet18_linear_eval_imagenet.yaml
124 lines (124 loc) · 5.67 KB
/
resnet18_linear_eval_imagenet.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
SEED: 100
MODEL:
ARCH: resnet18
INPUTSHAPE: [224, 224]
#PRETRAINED: checkpoints/resnet50.pth
#PRETRAINED: /data/train_log/imagenet/simsiam_imagenet_r18_imagenet_baseline_rerun/checkpoint.pth.tar
#PRETRAINED: train_log/imagenet/simsiam_imagenet_r18_imagenet_baseline/checkpoint.pth.tar
#PRETRAINED: train_log/imagenet/simsiam_200ep_imagenet_r18_imagenet_baseline/checkpoint.pth.tar
#PRETRAINED: train_log/imagenet/RQbyol_randombit_4_16_add_float_r18_imagenet_baseline/checkpoint.pth.tar
#PRETRAINED: train_log/imagenet/RQbyol_randombit_w_2_8_f_4_8_add_float_r18_imagenet_baseline/checkpoint.pth.tar
#PRETRAINED: /data/train_log/imagenet/RQbyol_randombit_w_2_8_f_4_8+floatloss_add_float_r18_imagenet_baseline/checkpoint.pth.tar
#PRETRAINED: /data/train_log/imagenet/RQbyol_200ep_randombit_w_2_8_f_4_8+floatloss_add_float_r18_imagenet_baseline/checkpoint.pth.tar
#PRETRAINED: /data/train_log/imagenet/byol_imagenet_r18_imagenet_baseline/checkpoint.pth.tar
#PRETRAINED: /data/train_log/imagenet/byol_imagenet_r18_imagenet_baseline_rerun/checkpoint.pth.tar
#CHECKPOINT: train_log_NEW/resnet18_float_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_NEW/resnet18_4w4f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_NEW/resnet18_RQbyol_4_16_float_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_NEW/resnet18_RQbyol_4_16_8w8f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_NEW/resnet18_RQbyol_4_16_4w4f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_NEW/resnet18_RQbyol_randombit_w_2_8_f_4_8_float_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_NEW/resnet18_RQbyol_randombit_w_2_8_f_4_8_4w4f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_lars/resnet18_simsiam_200ep_4w4f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_lars/resnet18_simsiam_200ep_float_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_lars/resnet18_RQbyol_randombit_w_2_8_f_4_8_float_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_lars/resnet18_RQbyol_randombit_w_2_8_f_4_8+floatloss_float_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_lars/resnet18_RQbyol_randombit_w_2_8_f_4_8+floatloss_4w4f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_lars/resnet18_RQbyol_200ep_randombit_w_2_8_f_4_8+floatloss_float_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: train_log_lars/resnet18_RQbyol_200ep_randombit_w_2_8_f_4_8+floatloss_4w4f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: /data/train_log_lars/resnet18_byol_float_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: /data/train_log_lars/resnet18_byol_4w4f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: /data/train_log_lars/resnet18_simsiam_rerun_float_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: /data/train_log_lars/resnet18_RQbyol_200ep_randombit_w_2_8_f_4_8+floatloss_2w4f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: /data/train_log_lars/resnet18_simsiam_rerun_3w3f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: /data/train_log_lars/resnet18_simsiam_5w5f_imagenet_linear_eval/checkpoint.pth.tar
#CHECKPOINT: /data/train_log_lars/resnet18_RQCL_5w5f_imagenet_linear_eval/checkpoint.pth.tar
#PRETRAINED: /data/train_log_rebuttal/imagenet/RQbyol_200ep_randombit_w_2_8_f_4_8+floatloss_add_float_r18_imagenet_baseline/checkpoint_0050.pth.tar
PRETRAINED: /data/train_log_rebuttal/imagenet/QQbyol_200ep_randombit_w_2_8_f_4_8_add_float_r18_imagenet_baseline/checkpoint_0050.pth.tar
NUM_CLASSES: 1000
TRAIN:
EPOCHS: 60
USE_DDP: True
LINEAR_EVAL: True
DATASET: imagenet
BATCH_SIZE: 256 # per-gpu
OPTIMIZER:
NAME: lars
MOMENTUM: 0.9
WEIGHT_DECAY: 0.000 # 1e-5
LR_SCHEDULER:
WARMUP_EPOCHS: 0
WARMUP_LR: 0.0002 # 1e-4
BASE_LR: 0.8 # 1e-2
MIN_LR: 0.
TYPE: cosine
DECAY_RATE: 0.1
DECAY_MILESTONES : [30, 40]
LOSS:
CRITERION:
NAME: CrossEntropy
#REGULARIZER:
# NAME: PACT
LAMBDA: 0.0001
METER:
NAME: ACC
ACC:
TOPK: [1, 5]
RUNNER:
NAME: default
AUG:
TRAIN:
HORIZONTAL_FLIP:
PROB: 0.5
RANDOMRESIZEDCROP:
ENABLE: True
SCALE: (0.08, 1.0)
INTERPOLATION: bilinear
NORMLIZATION:
MEAN: [0.485, 0.456, 0.406]
STD: [0.229, 0.224, 0.225]
EVALUATION:
RESIZE:
ENABLE: True
SIZE: [256, 256]
CENTERCROP:
ENABLE: True
NORMLIZATION:
MEAN: [0.485, 0.456, 0.406]
STD: [0.229, 0.224, 0.225]
QUANT:
TYPE: ptq
CALIBRATION:
TYPE: tar
PATH: calibrations/imagenet_100.tar
SIZE: 100
BATCHSIZE: 25
W:
BIT: 0
SYMMETRY: True
QUANTIZER: uniform
GRANULARITY : channelwise
OBSERVER_METHOD:
NAME: MINMAX
A:
BIT: 0
SYMMETRY: False
QUANTIZER: uniform
GRANULARITY : layerwise
OBSERVER_METHOD:
NAME: MINMAX
BIT_CONFIG: [{
"conv1": {"w": 8, "a": 8},
"layer1.0.conv2": {"a": 0},
"layer1.1.conv2": {"a": 0},
"layer2.0.conv2": {"a": 0},
"layer2.0.downsample": {"a": 0},
"layer2.1.conv2": {"a": 0},
"layer3.0.downsample": {"a": 0},
"layer3.0.conv2": {"a": 0},
"layer3.1.conv2": {"a": 0},
"layer4.0.downsample": {"a": 0},
"layer4.0.conv2": {"a": 0},
"layer4.1.conv2": {"a": 0},
"fc": {"w": 8, "a": 0},
}]