Hugging Face's logo

InternRobotics
/
InternW0-Libero

Robotics
PyTorch
English
internw0
world-model
robot-learning
manipulation
flow-matching
video-action-model
libero
InternW0
Model card Files
xet
 Community
InternW0-Libero
File size: 7,372 Bytes
f751b92
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
# R13 step-000660 LIBERO evaluation reference.
# This records the complete 50-trial/task protocol used for the selected checkpoint.
# Paths are portable environment-variable overrides; execution examples are in PROVENANCE.md.

output_dir: ${oc.env:W0_LIBERO_EVAL_OUTPUT,outputs/libero/evaluation/step_000600}
batch_size: 48
num_workers: 4
gc_collect_every: 1
lr_scheduler_type: cosine
learning_rate: 5.0e-05
soft_prompt_lr_scale: 1.0
num_epochs: 10
max_steps: null
log_every: 10
save_every: 100
keep_training_states: 1
checkpoint_min_free_gib: 128
eval_every: 50
eval_num_inference_steps: 10
gradient_accumulation_steps: 1
mixed_precision: bf16
seed: 42
max_grad_norm: 1.0
weight_decay: 0.0
freeze_video_dit: false
resume: null
init_checkpoint: ${oc.env:W0_LIBERO_INIT_CHECKPOINT,checkpoints/libero_init_step_2640.pt}
init_soft_prompts: false
resume_global_batch_size: null
wandb:
  enabled: false
  workspace: null
  project: w0
  name: w0-lib-gl-rf-c16-e10-32g-r13
  group: null
  mode: online
data:
  train:
    _target_: w0.datasets.libero_posttrain.LiberoPosttrainDataset
    dataset_root: ${oc.env:W0_LIBERO_DATA,data/libero}
    is_training_set: true
    samples_per_epoch: 100000
    epoch_sampling: random_frame
    seed: 42
    val_set_proportion: 0.01
    pretrained_norm_stats: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json}
    max_action_offset: 15
    action_horizon: 64
    action_chunk_size: 16
    video_plan_horizon: 96
    action_video_freq_ratio: 8
    text_embedding_cache_dir: ${oc.env:W0_LIBERO_TEXT_CACHE,assets/libero/text_embeds_cache}
    include_suites:
    - libero_goal
    - libero_10
    color_jitter:
      enabled: false
      brightness: 0.2
      contrast: 0.2
      saturation: 0.2
      hue: 0.0
    processor:
      _target_: w0.datasets.libero_posttrain.LiberoPosttrainProcessor
      stats_path: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json}
  val:
    _target_: w0.datasets.libero_posttrain.LiberoPosttrainDataset
    dataset_root: ${oc.env:W0_LIBERO_DATA,data/libero}
    is_training_set: false
    samples_per_epoch: 128
    epoch_sampling: random_frame
    seed: 42
    val_set_proportion: 0.01
    pretrained_norm_stats: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json}
    max_action_offset: 15
    action_horizon: 64
    action_chunk_size: 16
    video_plan_horizon: 96
    action_video_freq_ratio: 8
    text_embedding_cache_dir: ${oc.env:W0_LIBERO_TEXT_CACHE,assets/libero/text_embeds_cache}
    include_suites:
    - libero_goal
    - libero_10
    processor:
      _target_: w0.datasets.libero_posttrain.LiberoPosttrainProcessor
      stats_path: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json}
    color_jitter:
      enabled: false
      brightness: 0.2
      contrast: 0.2
      saturation: 0.2
      hue: 0.0
model:
  _target_: w0.runtime.create_w0
  model_id: ${oc.env:W0_WAN_MODEL_DIR,assets/wan22}
  tokenizer_model_id: Wan-AI/Wan2.1-T2V-1.3B
  tokenizer_max_len: 128
  load_text_encoder: true
  proprio_dim: 37
  redirect_common_files: true
  mot_checkpoint_mixed_attn: true
  action_dit_pretrained_path: null
  skip_dit_load_from_pretrain: true
  legacy_checkpoint_config: null
  observation_encoder: dino
  dino_config:
    model_name: dinov3-vitb16-hf
    weights_path: ${oc.env:W0_DINO_WEIGHTS,assets/dinov3-vitb16-pretrain-lvd1689m}
    input_resolution:
    - 384
    - 320
    feature_dim: 768
    patch_size: 16
    use_cls_token: false
    normalize_features: false
    encode_microbatch_size: 16
  checkpoint_observation_encoder_adapt: false
  action_horizon: 64
  action_chunk_size: 16
  max_action_offset: 15
  action_train_timestep_mode: per_chunk
  future_training_mode: idm
  idm_video_cond_noise_prob: 0.5
  idm_checkpoint_full_layer: true
  idm_checkpoint_layer_stride: 1
  action_obs_downsample_factor: 1
  chunk_kv_editor_num_queries: 32
  num_history_frames: 0
  action_video_read_mode: future
  video_rope_frame_stride: 1
  video_dit_config:
    compact_timestep_embedding: true
    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: true
    video_attention_mask_mode: first_frame_causal
    action_conditioned: false
    action_dim: 37
    action_group_causal_mask_mode: group_diagonal
  action_dit_config:
    num_soft_prompts: 8
    soft_prompt_domains:
    - libero/franka
    soft_prompt_init_std: 0.02
    domain_configs:
    - name: libero/franka
      dataset: libero
      embodiment: franka
      action_dim: 37
      proprio_dim: 37
      action_valid_indices:
      - 7
      - 8
      - 9
      - 10
      - 11
      - 12
      - 13
      - 14
      - 15
      - 16
      proprio_valid_indices:
      - 7
      - 8
      - 9
      - 10
      - 11
      - 12
      - 13
      - 14
      - 15
      - 16
      raw_eef_names:
      - eef_x
      - eef_y
      - eef_z
      - rot6d_col0_x
      - rot6d_col0_y
      - rot6d_col0_z
      - rot6d_col1_x
      - rot6d_col1_y
      - rot6d_col1_z
      - gripper_open_scale
      action_semantics: native_delta_eef10_row_aligned
    action_dim: 37
    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
    autoregressive_teacher_forcing: true
    action_chunk_size: 16
    action_teacher_forcing_mask_mode: stage1_chunkwise
    use_gradient_checkpointing: true
  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
    lambda_action_prior: 1.0
  video_plan_horizon: 96
  video_action_stride: 8
  video_plan_steps: 10
  domain_registry_path: null
posttrain_required: true
posttrain_stats: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json}
history_aware_batching: false
optimizer_group_max_numel: 200000000
ckpt: ${oc.env:W0_LIBERO_CHECKPOINT,step_000600.pt}
gpu_id: 0
EVALUATION:
  protocol: w0-stochastic-50-trial-v1
  task_suite_name: libero_goal
  task_suite_names:
  - libero_spatial
  - libero_object
  - libero_goal
  - libero_10
  task_id: 0
  num_trials: 50
  trial_start: 0
  seed: 42
  policy_seed: null
  stochastic: true
  execute_n_actions: 16
  settle_steps: 30
  settle_action:
  - 0
  - 0
  - 0
  - 0
  - 0
  - 0
  - -1
  camera_height: 256
  camera_width: 256
  image_transform: rotate_180
  image_codec: PNG
  binarize_gripper: true
  max_steps_by_suite:
    libero_spatial: 400
    libero_object: 400
    libero_goal: 400
    libero_10: 700
  output_dir: ${oc.env:W0_LIBERO_EVAL_OUTPUT,outputs/libero/evaluation/step_000600}
  action_horizon: 64
  chunks_per_video_prefill: 1
  num_inference_steps: ${eval_num_inference_steps}
  dataset_stats_path: ${oc.env:W0_LIBERO_STATS,assets/libero/normalization.json}
  text_cache_dir: ${oc.env:W0_LIBERO_TEXT_CACHE,assets/libero/text_embeds_cache}
  device: cuda
  training_config: training_config.yaml
  training_config_sha256: a9209138c249fd07bed47d6d95096d9b0ec14606ff771ad24d52b1539713835b