File size: 1,524 Bytes
207c208
 
b341bc3
207c208
b341bc3
207c208
b341bc3
 
 
 
207c208
b341bc3
 
207c208
 
 
 
 
b341bc3
 
 
 
 
 
 
3fe00bc
b341bc3
 
 
207c208
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b341bc3
 
 
207c208
 
 
b341bc3
 
 
207c208
 
b341bc3
 
 
207c208
 
 
b341bc3
 
 
 
 
 
 
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
project:
  name: hyper3-clip
seed: 31
model:
  objective: uncha
  vision_backbone: vit_base_patch16_224
  vision_pretrained: false
  vision_global_pool: token
  vision_use_sincos2d_pos: true
  vision_timm_norm_layer: layer_norm
  text_model_name: openai/clip-vit-base-patch32
  text_pretrained: false
  text_pooling: auto
  embed_dim: 512
  curv_init: 1.0
  learn_curv: true
  inter_aperture_scale: 0.7
  intra_aperture_scale: 1.2
  uncha_entailment_geometry: lorentz
  uncha_contrastive_loss: ce
  beta_clip_beta: 0.5
  beta_clip_similarity: dot
  beta_clip_num_heads: 8
  beta_clip_mlp_ratio: 4.0
  beta_clip_drop_cls_token: true
  beta_clip_variant: ce
  fuse_beta_query_encoder_forwards: true
  group_beta_query_pooling: true
  phyclip_product_metric: l1
training:
  total_steps: 500000
  global_batch_size: 768
  grad_accum_steps: 1
  lr: 0.0005
  weight_decay: 0.2
  betas:
  - 0.9
  - 0.98
  warmup_steps: 4000
  amp: true
  max_grad_norm: 1.0
optimizer:
  no_decay_params:
  - logit_scale
  - global_logit_scale
  - local_logit_scale
  - global_local_logit_scale
  - visual_alpha
  - textual_alpha
  - log_curv
  - global_logit_bias
  - local_logit_bias
  - global_local_logit_bias
data:
  type: processed_grit
  part_sampling: all
  max_parts: 5
  train_transform: tight_crop_color_jitter_gray
  shuffle_buffer: 4000
  image_size: 224
  max_text_length: 77
  image_normalization: imagenet
  beta_clip:
    enabled: true
    max_sentences: 5
    max_phrases: 30
    max_queries_per_image: 6
    use_part_texts: true