timm
/

Image Classification
timm
PyTorch
Safetensors
rwightman HF staff commited on
Commit
4b8f42b
1 Parent(s): 7f00ccb

Upload pretrain_hparams.yaml with huggingface_hub

Browse files
Files changed (1) hide show
  1. pretrain_hparams.yaml +146 -0
pretrain_hparams.yaml ADDED
@@ -0,0 +1,146 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ aa: rand-m7-inc1-mstd1.0-n3
2
+ amp: true
3
+ amp_dtype: float16
4
+ amp_impl: native
5
+ aug_repeats: 0.0
6
+ aug_splits: 0
7
+ batch_size: 256
8
+ bce_loss: true
9
+ bce_pos_weight: null
10
+ bce_sum: true
11
+ bce_target_thresh: 0.2
12
+ bn_eps: null
13
+ bn_momentum: null
14
+ channels_last: false
15
+ checkpoint_hist: 10
16
+ class_map: ''
17
+ clip_grad: 5.0
18
+ clip_mode: norm
19
+ color_jitter: null
20
+ color_jitter_prob: 0.2
21
+ cooldown_epochs: 10
22
+ crop_pct: 0.95
23
+ cutmix: 1.0
24
+ cutmix_minmax: null
25
+ data: null
26
+ data_dir: /data/tfds/
27
+ dataset: tfds/imagenet12k
28
+ dataset_download: false
29
+ decay_epochs: 100
30
+ decay_milestones:
31
+ - 90
32
+ - 180
33
+ - 270
34
+ decay_rate: 0.1
35
+ dist_bn: reduce
36
+ drop: 0.0
37
+ drop_block: null
38
+ drop_connect: null
39
+ drop_path: 0.2
40
+ epoch_repeats: 0.0
41
+ epochs: 150
42
+ eval_metric: top1
43
+ experiment: ''
44
+ fast_norm: false
45
+ force_cpu: false
46
+ fuser: ''
47
+ gaussian_blur_prob: 0.05
48
+ gp: null
49
+ grad_accum_steps: 1
50
+ grad_checkpointing: false
51
+ grayscale_prob: 0.1
52
+ head_init_bias: null
53
+ head_init_scale: 0.0
54
+ hflip: 0.5
55
+ img_size: 256
56
+ in_chans: null
57
+ initial_checkpoint: ''
58
+ input_img_mode: null
59
+ input_key: null
60
+ input_size: null
61
+ interpolation: ''
62
+ jsd_loss: false
63
+ layer_decay: null
64
+ local_rank: 0
65
+ log_interval: 50
66
+ log_wandb: false
67
+ lr: null
68
+ lr_base: 0.0001
69
+ lr_base_scale: ''
70
+ lr_base_size: 512
71
+ lr_cycle_decay: 0.5
72
+ lr_cycle_limit: 1
73
+ lr_cycle_mul: 1.0
74
+ lr_k_decay: 1.0
75
+ lr_noise: null
76
+ lr_noise_pct: 0.67
77
+ lr_noise_std: 1.0
78
+ mean: null
79
+ min_lr: 5.0e-07
80
+ mixup: 0.2
81
+ mixup_mode: batch
82
+ mixup_off_epoch: 0
83
+ mixup_prob: 1.0
84
+ mixup_switch_prob: 0.5
85
+ model: vit_betwixt_patch16_reg4_gap_256
86
+ model_ema: true
87
+ model_ema_decay: 0.99993
88
+ model_ema_force_cpu: false
89
+ model_kwargs:
90
+ fix_init: true
91
+ momentum: 0.9
92
+ no_aug: false
93
+ no_ddp_bb: false
94
+ no_prefetcher: false
95
+ no_resume_opt: false
96
+ num_classes: 11821
97
+ opt: nadamw
98
+ opt_betas: null
99
+ opt_eps: 1.0e-08
100
+ opt_kwargs: {}
101
+ output: ''
102
+ patience_epochs: 10
103
+ pin_mem: true
104
+ pretrained: false
105
+ pretrained_path: null
106
+ ratio:
107
+ - 0.75
108
+ - 1.3333333333333333
109
+ recount: 1
110
+ recovery_interval: 0
111
+ remode: pixel
112
+ reprob: 0.2
113
+ resplit: false
114
+ resume: ''
115
+ save_images: false
116
+ scale:
117
+ - 0.2
118
+ - 1.0
119
+ sched: cosine
120
+ sched_on_updates: true
121
+ seed: 0
122
+ smoothing: 0.0
123
+ split_bn: false
124
+ start_epoch: null
125
+ std: null
126
+ sync_bn: false
127
+ synchronize_step: false
128
+ target_key: null
129
+ torchcompile: inductor
130
+ torchscript: false
131
+ train_crop_mode: rkrr
132
+ train_interpolation: random
133
+ train_num_samples: null
134
+ train_split: train
135
+ tta: 0
136
+ use_multi_epochs_loader: false
137
+ val_num_samples: null
138
+ val_split: validation
139
+ validation_batch_size: null
140
+ vflip: 0.0
141
+ warmup_epochs: 3
142
+ warmup_lr: 5.0e-07
143
+ warmup_prefix: false
144
+ weight_decay: 0.2
145
+ worker_seeding: all
146
+ workers: 8