Hugging Face's logo

deformable-bench
/
fastwam-flatten-silk

Robotics
fastwam
deformable-manipulation
Model card Files
xet
 Community
fastwam-flatten-silk
File size: 5,056 Bytes
bd36ac3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
output_dir: /horizon-bucket/robot_lab/users/wenkang.hu-labs/projects/deformable_bench/runs/flatten-silk/fastwam/production-20260718T232050Z
batch_size: 16
num_workers: 0
lr_scheduler_type: cosine
learning_rate: 0.0001
num_epochs: 5
max_steps: 3145
log_every: 10
save_every: 0
eval_every: 0
eval_num_inference_steps: 10
gradient_accumulation_steps: 1
mixed_precision: bf16
seed: 42
max_grad_norm: 1.0
weight_decay: 0.01
resume: null
wandb:
  enabled: false
  workspace: null
  project: fast-wam
  name: silk_grasp_uncond_3cam_384_1e-4
  group: null
  mode: online
data:
  train:
    _target_: fastwam.datasets.lerobot.robot_video_dataset.RobotVideoDataset
    dataset_dirs:
    - /horizon-bucket/robot_lab/users/wenkang.hu-labs/projects/deformable_bench/datasets/huggingface/deformable-bench/flatten-silk/e465c61e9c03674cd46f215c3c1fa86e4071a6e2
    shape_meta:
      images:
      - key: static_cam
        raw_shape:
        - 3
        - 720
        - 1280
        shape:
        - 3
        - 240
        - 320
      - key: left_hand_cam
        raw_shape:
        - 3
        - 720
        - 1280
        shape:
        - 3
        - 240
        - 320
      - key: right_hand_cam
        raw_shape:
        - 3
        - 720
        - 1280
        shape:
        - 3
        - 240
        - 320
      action:
      - key: default
        raw_shape: 14
        shape: 14
      state:
      - key: default
        raw_shape: 14
        shape: 14
    num_frames: 33
    global_sample_stride: 1
    action_video_freq_ratio: 4
    video_size:
    - 384
    - 320
    camera_key: null
    val_set_proportion: 0.0
    is_training_set: true
    pretrained_norm_stats: null
    skip_padding_as_possible: false
    concat_multi_camera: robotwin
    processor:
      _target_: fastwam.datasets.lerobot.processors.fastwam_processor.FastWAMProcessor
      shape_meta:
        images:
        - key: static_cam
          raw_shape:
          - 3
          - 720
          - 1280
          shape:
          - 3
          - 240
          - 320
        - key: left_hand_cam
          raw_shape:
          - 3
          - 720
          - 1280
          shape:
          - 3
          - 240
          - 320
        - key: right_hand_cam
          raw_shape:
          - 3
          - 720
          - 1280
          shape:
          - 3
          - 240
          - 320
        action:
        - key: default
          raw_shape: 14
          shape: 14
        state:
        - key: default
          raw_shape: 14
          shape: 14
      num_obs_steps: 33
      num_output_cameras: 3
      action_output_dim: 14
      proprio_output_dim: 14
      action_state_transforms: null
      use_stepwise_action_norm: false
      norm_default_mode: z-score
      norm_exception_mode: null
      action_state_merger:
        _target_: fastwam.datasets.lerobot.transforms.action_state_merger.ConcatLeftAlign
      train_transforms:
      - _target_: fastwam.datasets.lerobot.transforms.image.ToTensor
      - _target_: torchvision.transforms.Resize
        size:
        - 240
        - 320
      val_transforms:
      - _target_: fastwam.datasets.lerobot.transforms.image.ToTensor
      - _target_: torchvision.transforms.Resize
        size:
        - 240
        - 320
    text_embedding_cache_dir: /horizon-bucket/robot_lab/users/wenkang.hu-labs/projects/deformable_bench/derived/fastwam/text-embeddings/deformable-bench-four-datasets/v3-wan22ti2v5b-t5len128
    context_len: 128
model:
  _target_: fastwam.runtime.create_fastwam
  model_id: Wan-AI/Wan2.2-TI2V-5B
  tokenizer_model_id: Wan-AI/Wan2.1-T2V-1.3B
  tokenizer_max_len: 128
  load_text_encoder: false
  proprio_dim: 14
  redirect_common_files: false
  mot_checkpoint_mixed_attn: false
  action_dit_pretrained_path: /horizon-bucket/robot_lab/users/wenkang.hu-labs/projects/deformable_bench/derived/fastwam/actiondit/sha256-7bd65e1986accaaf2e8dd5e59b5e98ef348d7c2f1d88ab2df010eb954a441445/ActionDiT_linear_interp_Wan22_alphascale_1024hdim.pt
  skip_dit_load_from_pretrain: false
  video_dit_config:
    has_image_input: false
    patch_size:
    - 1
    - 2
    - 2
    in_dim: 48
    hidden_dim: 3072
    ffn_dim: 14336
    freq_dim: 256
    text_dim: 4096
    out_dim: 48
    num_heads: 24
    attn_head_dim: 128
    num_layers: 30
    eps: 1.0e-06
    seperated_timestep: true
    require_clip_embedding: false
    require_vae_embedding: false
    fuse_vae_embedding_in_latents: true
    use_gradient_checkpointing: false
    video_attention_mask_mode: first_frame_causal
    action_conditioned: false
    action_dim: 14
    action_group_causal_mask_mode: group_diagonal
  action_dit_config:
    action_dim: 14
    hidden_dim: 1024
    ffn_dim: 4096
    num_heads: 24
    attn_head_dim: 128
    num_layers: 30
    text_dim: 4096
    freq_dim: 256
    eps: 1.0e-06
    use_gradient_checkpointing: false
  video_scheduler:
    train_shift: 5.0
    infer_shift: 5.0
    num_train_timesteps: 1000
  action_scheduler:
    train_shift: 5.0
    infer_shift: 5.0
    num_train_timesteps: 1000
  loss:
    lambda_action: 1.0