speedinghzl commited on
Commit
2d68eec
·
verified ·
1 Parent(s): 600f5a8

Upload folder using huggingface_hub

Browse files
exp_rope_clipcls_vit_b16_s512m_bs8k/checkpoints/epoch_4.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5711fc667c8378c10b4ecfcedce976907b2dad6acf6a1af366387b890d3365c4
3
+ size 2337851367
exp_rope_clipcls_vit_b16_s512m_bs8k/checkpoints/results.jsonl CHANGED
@@ -1 +1,12 @@
1
  {"imagenet-zeroshot-val-top1": 0.0015595918367346938, "imagenet-zeroshot-val-top5": 0.0029268367346938777}
 
 
 
 
 
 
 
 
 
 
 
 
1
  {"imagenet-zeroshot-val-top1": 0.0015595918367346938, "imagenet-zeroshot-val-top5": 0.0029268367346938777}
2
+ {"imagenet-zeroshot-val-top1": 0.0019320408163265306, "imagenet-zeroshot-val-top5": 0.0033858163265306123}
3
+ {"imagenet-zeroshot-val-top1": 0.0021146938775510205, "imagenet-zeroshot-val-top5": 0.0035695918367346937}
4
+ {"imagenet-zeroshot-val-top1": 0.0022825510204081633, "imagenet-zeroshot-val-top5": 0.0037456122448979593}
5
+ {"imagenet-zeroshot-val-top1": 0.002414795918367347, "imagenet-zeroshot-val-top5": 0.0038751020408163265}
6
+ {"imagenet-zeroshot-val-top1": 0.002423469387755102, "imagenet-zeroshot-val-top5": 0.0038877551020408165}
7
+ {"imagenet-zeroshot-val-top1": 0.0025968367346938777, "imagenet-zeroshot-val-top5": 0.004040102040816326}
8
+ {"imagenet-zeroshot-val-top1": 0.002754183673469388, "imagenet-zeroshot-val-top5": 0.004176122448979592}
9
+ {"imagenet-zeroshot-val-top1": 0.0029424489795918368, "imagenet-zeroshot-val-top5": 0.004326020408163265}
10
+ {"imagenet-zeroshot-val-top1": 0.0031240816326530614, "imagenet-zeroshot-val-top5": 0.004441938775510204}
11
+ {"imagenet-zeroshot-val-top1": 0.0031896938775510205, "imagenet-zeroshot-val-top5": 0.004487142857142857}
12
+ {"imagenet-zeroshot-val-top1": 0.0031844897959183672, "imagenet-zeroshot-val-top5": 0.004485714285714286}
exp_rope_clipcls_vit_b16_s512m_bs8k/out.log CHANGED
The diff for this file is too large to render. See raw diff
 
exp_rope_clipcls_vit_b16_s512m_bs8k/params.txt CHANGED
@@ -24,7 +24,7 @@ distill: False
24
  distill_model: None
25
  distill_pretrained: None
26
  distributed: True
27
- epochs: 4
28
  epochs_cooldown: None
29
  eps: 1e-06
30
  force_custom_text: False
@@ -42,6 +42,7 @@ image_resize_mode: None
42
  image_std: None
43
  imagenet_v2: None
44
  imagenet_val: /mnt/bn/zilongdata-hl/dataset/imagenet/val
 
45
  local_loss: True
46
  local_rank: 0
47
  lock_image: False
@@ -55,7 +56,7 @@ log_level: 20
55
  log_local: False
56
  log_path: ./logs-lr1e-3-datacomp/exp_rope_clipcls_vit_b16_s512m_bs8k/out.log
57
  logs: ./logs-lr1e-3-datacomp
58
- lr: 0.001
59
  lr_cooldown_end: 0.0
60
  lr_cooldown_power: 1.0
61
  lr_scheduler: cosine
@@ -66,7 +67,7 @@ native_dynamic_resolution: True
66
  no_set_device_rank: False
67
  only_packing: True
68
  precision: amp_bfloat16
69
- pretrained:
70
  pretrained_image:
71
  pretrained_text:
72
  rank: 0
@@ -74,7 +75,7 @@ remote_sync: None
74
  remote_sync_frequency: 300
75
  remote_sync_protocol: s3
76
  report_to: wandb
77
- resume: None
78
  rope_attn_num_heads: 12
79
  rope_model_width: 768
80
  save_every_n_steps: 6104
@@ -88,9 +89,9 @@ tensorboard_path:
88
  torchcompile: False
89
  torchscript: False
90
  trace: False
91
- train_data: /mnt/bn/bytenas-weixian/data/Recap-DataComp-1B-Dataset/{000000..140146}.tar
92
  train_data_upsampling_factors: None
93
- train_num_samples: 128000000
94
  use_bn_sync: False
95
  use_bnb_linear: None
96
  val_data: None
 
24
  distill_model: None
25
  distill_pretrained: None
26
  distributed: True
27
+ epochs: 1
28
  epochs_cooldown: None
29
  eps: 1e-06
30
  force_custom_text: False
 
42
  image_std: None
43
  imagenet_v2: None
44
  imagenet_val: /mnt/bn/zilongdata-hl/dataset/imagenet/val
45
+ is_cls_token: True
46
  local_loss: True
47
  local_rank: 0
48
  lock_image: False
 
56
  log_local: False
57
  log_path: ./logs-lr1e-3-datacomp/exp_rope_clipcls_vit_b16_s512m_bs8k/out.log
58
  logs: ./logs-lr1e-3-datacomp
59
+ lr: 0.0005
60
  lr_cooldown_end: 0.0
61
  lr_cooldown_power: 1.0
62
  lr_scheduler: cosine
 
67
  no_set_device_rank: False
68
  only_packing: True
69
  precision: amp_bfloat16
70
+ pretrained: /mnt/bn/zilongdata-hl/code/SuperCls/opencls/logs-lr1e-3-datacomp/exp_rope_clipcls_vit_b16_s512m_bs8k/checkpoints/epoch_4.pt
71
  pretrained_image:
72
  pretrained_text:
73
  rank: 0
 
75
  remote_sync_frequency: 300
76
  remote_sync_protocol: s3
77
  report_to: wandb
78
+ resume: ./logs-lr1e-3-datacomp/exp_rope_clipcls_vit_b16_s512m_bs8k/checkpoints/step_61040.pt
79
  rope_attn_num_heads: 12
80
  rope_model_width: 768
81
  save_every_n_steps: 6104
 
89
  torchcompile: False
90
  torchscript: False
91
  trace: False
92
+ train_data:
93
  train_data_upsampling_factors: None
94
+ train_num_samples: 512000000
95
  use_bn_sync: False
96
  use_bnb_linear: None
97
  val_data: None