File size: 7,481 Bytes
5e21358
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
datamodule:
  transforms:
    combine_goal_obs: false
    move_axis: false
    bytes_to_string: true
    adjust_type: null
    add_robot_information: false
  language_encoders:
    _target_: medit.agents.input_encoders.goal_encoders.language_encoders.clip_tokens.TokenLangClip
    _recursive_: false
    model_name: ${clip_lang_model_name}
  _target_: oxe_torch_dataloader.uha.uha_datamodule.UhaDataModule
  _recursive_: false
  num_workers: ${num_workers}
  batch_size: ${batch_size}
  pin_memory: ${pin_memory}
  drop_last: ${drop_last}
  datasets:
    DATA_NAME: ${DATA_NAME}
    DATA_PATH: gs://gresearch/robotics
    load_camera_views: ${load_camera_views}
    dataset_size_limit: ${dataset_size_limit}
    action_proprio_normalization_type: bounds
    interleaved_dataset_cfg:
      shuffle_buffer_size: ${shuffle_buffer_size}
      balance_weights: true
      traj_transform_kwargs:
        goal_relabeling_strategy: ${goal_relabeling_strategy}
        goal_relabeling_kwargs: ${goal_relabeling_kwargs}
        window_size: ${window_size}
        action_horizon: ${act_seq_len}
        subsample_length: ${subsample_length}
        skip_unlabeled: ${skip_unlabeled}
      frame_transform_kwargs:
        image_augment_kwargs:
          primary:
            random_resized_crop:
              scale:
              - 0.8
              - 1.0
              ratio:
              - 0.9
              - 1.1
            random_brightness:
            - 0.1
            random_contrast:
            - 0.9
            - 1.1
            random_saturation:
            - 0.9
            - 1.1
            random_hue:
            - 0.05
            augment_order:
            - random_resized_crop
            - random_brightness
            - random_contrast
            - random_saturation
            - random_hue
          secondary:
            random_resized_crop:
              scale:
              - 0.8
              - 1.0
              ratio:
              - 0.9
              - 1.1
            random_brightness:
            - 0.1
            random_contrast:
            - 0.9
            - 1.1
            random_saturation:
            - 0.9
            - 1.1
            random_hue:
            - 0.05
            augment_order:
            - random_resized_crop
            - random_brightness
            - random_contrast
            - random_saturation
            - random_hue
          wrist:
            random_brightness:
            - 0.1
            random_contrast:
            - 0.9
            - 1.1
            random_saturation:
            - 0.9
            - 1.1
            random_hue:
            - 0.05
            augment_order:
            - random_brightness
            - random_contrast
            - random_saturation
            - random_hue
        resize_size:
          primary:
          - 224
          - 224
          secondary:
          - 224
          - 224
          wrist:
          - 224
          - 224
        resize_size_future_obs:
          primary:
          - 112
          - 112
          secondary:
          - 112
          - 112
          wrist:
          - 112
          - 112
        num_parallel_calls: 128
      traj_transform_threads: 64
      traj_read_threads: 32
trainer:
  agent:
    agent:
      language_goal:
        _target_: medit.agents.input_encoders.goal_encoders.language_encoders.clip_tokens.LangClip
        _recursive_: false
        freeze_backbone: true
        model_name: ${clip_lang_model_name}
      model:
        _target_: medit.agents.inner_models.edm_diffusion_policy.score_wrappers.GCDenoiser
        _recursive_: true
        sigma_data: 0.5
        inner_model:
          _target_: medit.agents.inner_models.modedit.MoDeDiT
          action_dim: ${act_dim}
          goal_dim: ${goal_dim}
          obs_dim: 2048
          goal_conditioned: true
          causal: true
          use_custom_attn_mask: false
          use_proprio: false
          state_dim: 8
          embed_dim: 1024
          n_layers: 12
          goal_seq_len: 1
          obs_seq_len: ${obs_seq_len}
          action_seq_len: ${act_seq_len}
          embed_pdrob: 0
          goal_drop: 0.1
          attn_pdrop: 0.3
          mlp_pdrop: 0.1
          n_heads: 8
          linear_output: true
          cond_router: true
          num_experts: 4
          top_k: 2
          router_normalize: true
          use_goal_in_routing: false
          use_argmax: false
          use_shared_expert: false
          use_noise_token_as_input: true
          init_style: olmoe
      _target_: medit.agents.mode_agent.MoDEAgent
      _recursive_: false
      latent_dim: 1024
      multistep: 5
      sampler_type: ddim
      num_sampling_steps: 5
      sigma_data: 0.5
      sigma_min: 0.001
      sigma_max: 80
      noise_scheduler: exponential
      sigma_sample_density_type: loglogistic
      act_window_size: ${act_seq_len}
      act_dim: ${act_dim}
      seed: ${seed}
      obs_modalities: ${obs_modalities}
      goal_modalities: ${goal_modalities}
      img_modalities: ${img_modalities}
      lang_modalities: ${lang_modalities}
      target_modality: ${target_modality}
      entropy_gamma: 0.01
      router_z_delta: 0.0
      resnet_type: '50'
    _target_: medit.agents.ddp_wrapper.DDPAgentWrapper
    _recursive_: false
    obs_modalities: ${obs_modalities}
    goal_modalities: ${goal_modalities}
    img_modalities: ${img_modalities}
    lang_modalities: ${lang_modalities}
    target_modality: ${target_modality}
  _target_: medit.trainers.accelerate_trainer.AccelerateTrainer
  _recursive_: false
  weight_decay:
    transformer_weight_decay: 0.1
    obs_encoder_weight_decay: 0.1
    perceptual_encoder_lr: 0.0001
  lr_scheduler: ${lr_scheduler}
  eval_every_n_steps: ${eval_every_n_steps}
  save_every_n_steps: ${save_every_n_steps}
  max_train_steps: ${max_train_steps}
  max_eval_steps: ${max_eval_steps}
  use_ema: true
  decay: ${decay}
  rampup_ratio: ${rampup_ratio}
  update_ema_every_n_steps: ${update_ema_every_n_steps}
  batch_size: ${batch_size}
  obs_modalities: ${obs_modalities}
  goal_modalities: ${goal_modalities}
  img_modalities: ${img_modalities}
  lang_modalities: ${lang_modalities}
  target_modality: ${target_modality}
vis_clip_model_name: ViT-B/16
clip_lang_model_name: ViT-B/32
DATA_NAME: MO
wandb:
  name: uha_${now:%H-%M-%S}
  group: ${now:%Y-%m-%d}
  project: simulation_eval
  entity: irl-masterthesis
  mode: null
lr_scheduler:
  _target_: medit.agents.utils.lr_schedulers.InverseSquareRootLRSchedule
  num_warmup_steps: 1000
  timescale: ${max_train_steps}
log_dir: logs/
window_size: 1
obs_seq_len: 1
goal_window_size: 1
seed: 42
obs_dim: 512
goal_dim: 512
act_seq_len: 10
update_ema_every_n_steps: 1
decay: 0.999
rampup_ratio: 0.001
gen_img_res: 112
num_tokens_voltron: 10
img_gen_frame_diff: 3
use_modality_encoder: false
goal_relabeling_strategy: null
goal_relabeling_kwargs:
  min_bound: 20
  max_bound: 50
  frame_diff: ${img_gen_frame_diff}
subsample_length: null
skip_unlabeled: true
load_camera_views:
- primary
- secondary
- wrist
obs_modalities: observation
goal_modalities: task
img_modalities:
- image_primary
- image_secondary
- image_wrist
lang_modalities:
- language_instruction
target_modality: action
drop_last: true
pin_memory: true
num_workers: 0
gradient_accumulation_steps: 1
act_dim: 7
max_train_steps: 300000
max_eval_steps: 200
eval_every_n_steps: 5000
save_every_n_steps: 5000
shuffle_buffer_size: 400000
batch_size: 512
dataset_size_limit: null