Ross Wightman commited on
Commit
3a7d5cc
1 Parent(s): 18f6c60

Add weights and args

Browse files
Files changed (2) hide show
  1. pytorch_model.bin +3 -0
  2. train_args.yaml +130 -0
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:049e8080eedb7246de64c48d8b1a229560892edfa46895bc5258f742ac95f773
3
+ size 346296033
train_args.yaml ADDED
@@ -0,0 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ aa: rand-m8-inc1-mstd101-n3
2
+ amp: true
3
+ amp_dtype: float16
4
+ amp_impl: native
5
+ aot_autograd: false
6
+ apex_amp: false
7
+ aug_repeats: 0
8
+ aug_splits: 0
9
+ batch_size: 512
10
+ bce_loss: false
11
+ bce_target_thresh: null
12
+ bn_eps: null
13
+ bn_momentum: null
14
+ channels_last: false
15
+ checkpoint_hist: 10
16
+ class_map: ''
17
+ clip_grad: 3.0
18
+ clip_mode: norm
19
+ color_jitter: 0.4
20
+ cooldown_epochs: 10
21
+ crop_pct: 1.0
22
+ cutmix: 0.0
23
+ cutmix_minmax: null
24
+ data_dir: /data/imagenet/
25
+ dataset: ''
26
+ dataset_download: false
27
+ decay_epochs: 100
28
+ decay_milestones:
29
+ - 30
30
+ - 60
31
+ decay_rate: 0.1
32
+ dist_bn: reduce
33
+ drop: 0.0
34
+ drop_block: null
35
+ drop_connect: null
36
+ drop_path: 0.1
37
+ dynamo: false
38
+ dynamo_backend: null
39
+ epoch_repeats: 0.0
40
+ epochs: 50
41
+ eval_metric: top1
42
+ experiment: ''
43
+ fast_norm: false
44
+ fuser: ''
45
+ gp: null
46
+ grad_checkpointing: true
47
+ hflip: 0.5
48
+ img_size: null
49
+ in_chans: null
50
+ initial_checkpoint: ''
51
+ input_size: null
52
+ interpolation: ''
53
+ jsd_loss: false
54
+ layer_decay: 0.75
55
+ local_rank: 0
56
+ log_interval: 50
57
+ log_wandb: false
58
+ lr: 0.0001
59
+ lr_base: 0.1
60
+ lr_base_scale: ''
61
+ lr_base_size: 256
62
+ lr_cycle_decay: 0.5
63
+ lr_cycle_limit: 1
64
+ lr_cycle_mul: 1.0
65
+ lr_k_decay: 1.0
66
+ lr_noise:
67
+ - 0.1
68
+ - 1.0
69
+ lr_noise_pct: 0.67
70
+ lr_noise_std: 1.0
71
+ mean: null
72
+ min_lr: 5.0e-07
73
+ mixup: 0.0
74
+ mixup_mode: batch
75
+ mixup_off_epoch: 0
76
+ mixup_prob: 1.0
77
+ mixup_switch_prob: 0.5
78
+ model: vit_base_patch16_clip_224.laion2b_ft_in12k
79
+ model_ema: true
80
+ model_ema_decay: 0.9998
81
+ model_ema_force_cpu: false
82
+ momentum: 0.9
83
+ native_amp: false
84
+ no_aug: false
85
+ no_ddp_bb: false
86
+ no_prefetcher: false
87
+ no_resume_opt: false
88
+ num_classes: 1000
89
+ opt: adamw
90
+ opt_betas: null
91
+ opt_eps: null
92
+ output: ''
93
+ patience_epochs: 10
94
+ pin_mem: false
95
+ pretrained: true
96
+ ratio:
97
+ - 0.75
98
+ - 1.3333333333333333
99
+ recount: 1
100
+ recovery_interval: 0
101
+ remode: pixel
102
+ reprob: 0.3
103
+ resplit: false
104
+ resume: ''
105
+ save_images: false
106
+ scale:
107
+ - 0.08
108
+ - 1.0
109
+ sched: cosine
110
+ sched_on_updates: false
111
+ seed: 42
112
+ smoothing: 0.1
113
+ split_bn: false
114
+ start_epoch: null
115
+ std: null
116
+ sync_bn: false
117
+ torchscript: false
118
+ train_interpolation: random
119
+ train_split: train
120
+ tta: 0
121
+ use_multi_epochs_loader: false
122
+ val_split: validation
123
+ validation_batch_size: null
124
+ vflip: 0.0
125
+ warmup_epochs: 10
126
+ warmup_lr: 1.0e-06
127
+ warmup_prefix: false
128
+ weight_decay: 0.05
129
+ worker_seeding: all
130
+ workers: 8