Hugging Face's logo

JayLee131
/
vqbet_pusht2

Robotics
Transformers
Safetensors
Inference Endpoints
Model card Files Community
JayLee131 commited on
Commit
115b29c
•
1 Parent(s): 32c05a6

Upload 3 files

Browse files
Files changed (3) hide show
  1. config.json +53 -0
  2. config.yaml +184 -0
  3. model.safetensors +3 -0
config.json ADDED
@@ -0,0 +1,53 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_chunk_size": 5,
3
+ "bet_softmax_temperature": 0.1,
4
+ "crop_is_random": true,
5
+ "crop_shape": [
6
+ 84,
7
+ 84
8
+ ],
9
+ "dropout": 0.1,
10
+ "gpt_block_size": 500,
11
+ "gpt_hidden_dim": 512,
12
+ "gpt_input_dim": 512,
13
+ "gpt_n_head": 8,
14
+ "gpt_n_layer": 8,
15
+ "gpt_output_dim": 512,
16
+ "input_normalization_modes": {
17
+ "observation.image": "mean_std",
18
+ "observation.state": "min_max"
19
+ },
20
+ "input_shapes": {
21
+ "observation.image": [
22
+ 3,
23
+ 96,
24
+ 96
25
+ ],
26
+ "observation.state": [
27
+ 2
28
+ ]
29
+ },
30
+ "mlp_hidden_dim": 1024,
31
+ "n_action_pred_token": 7,
32
+ "n_obs_steps": 5,
33
+ "n_vqvae_training_steps": 20000,
34
+ "offset_loss_weight": 10000.0,
35
+ "output_normalization_modes": {
36
+ "action": "min_max"
37
+ },
38
+ "output_shapes": {
39
+ "action": [
40
+ 2
41
+ ]
42
+ },
43
+ "pretrained_backbone_weights": null,
44
+ "primary_code_loss_weight": 5.0,
45
+ "secondary_code_loss_weight": 0.5,
46
+ "sequentially_select": false,
47
+ "spatial_softmax_num_keypoints": 32,
48
+ "use_group_norm": true,
49
+ "vision_backbone": "resnet18",
50
+ "vqvae_embedding_dim": 256,
51
+ "vqvae_enc_hidden_dim": 128,
52
+ "vqvae_n_embed": 16
53
+ }
config.yaml ADDED
@@ -0,0 +1,184 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ resume: false
2
+ device: cuda
3
+ use_amp: false
4
+ seed: 100000
5
+ dataset_repo_id: lerobot/pusht
6
+ video_backend: pyav
7
+ training:
8
+ offline_steps: 250000
9
+ online_steps: 0
10
+ online_steps_between_rollouts: 1
11
+ online_sampling_ratio: 0.5
12
+ online_env_seed: ???
13
+ eval_freq: 20000
14
+ log_freq: 250
15
+ save_checkpoint: true
16
+ save_freq: 20000
17
+ num_workers: 4
18
+ batch_size: 64
19
+ image_transforms:
20
+ enable: false
21
+ max_num_transforms: 3
22
+ random_order: false
23
+ brightness:
24
+ weight: 1
25
+ min_max:
26
+ - 0.8
27
+ - 1.2
28
+ contrast:
29
+ weight: 1
30
+ min_max:
31
+ - 0.8
32
+ - 1.2
33
+ saturation:
34
+ weight: 1
35
+ min_max:
36
+ - 0.5
37
+ - 1.5
38
+ hue:
39
+ weight: 1
40
+ min_max:
41
+ - -0.05
42
+ - 0.05
43
+ sharpness:
44
+ weight: 1
45
+ min_max:
46
+ - 0.8
47
+ - 1.2
48
+ grad_clip_norm: 10
49
+ lr: 0.0001
50
+ lr_scheduler: cosine
51
+ lr_warmup_steps: 500
52
+ adam_betas:
53
+ - 0.95
54
+ - 0.999
55
+ adam_eps: 1.0e-08
56
+ adam_weight_decay: 1.0e-06
57
+ vqvae_lr: 0.001
58
+ n_vqvae_training_steps: 20000
59
+ bet_weight_decay: 0.0002
60
+ bet_learning_rate: 5.5e-05
61
+ bet_betas:
62
+ - 0.9
63
+ - 0.999
64
+ delta_timestamps:
65
+ observation.image:
66
+ - -0.4
67
+ - -0.3
68
+ - -0.2
69
+ - -0.1
70
+ - 0.0
71
+ observation.state:
72
+ - -0.4
73
+ - -0.3
74
+ - -0.2
75
+ - -0.1
76
+ - 0.0
77
+ action:
78
+ - -0.4
79
+ - -0.3
80
+ - -0.2
81
+ - -0.1
82
+ - 0.0
83
+ - 0.1
84
+ - 0.2
85
+ - 0.3
86
+ - 0.4
87
+ - 0.5
88
+ - 0.6
89
+ - 0.7
90
+ - 0.8
91
+ - 0.9
92
+ - 1.0
93
+ eval:
94
+ n_episodes: 500
95
+ batch_size: 50
96
+ use_async_envs: false
97
+ wandb:
98
+ enable: true
99
+ disable_artifact: false
100
+ project: lerobot
101
+ notes: ''
102
+ fps: 10
103
+ env:
104
+ name: pusht
105
+ task: PushT-v0
106
+ image_size: 96
107
+ state_dim: 2
108
+ action_dim: 2
109
+ fps: ${fps}
110
+ episode_length: 300
111
+ gym:
112
+ obs_type: pixels_agent_pos
113
+ render_mode: rgb_array
114
+ visualization_width: 384
115
+ visualization_height: 384
116
+ override_dataset_stats:
117
+ observation.image:
118
+ mean:
119
+ - - - 0.5
120
+ - - - 0.5
121
+ - - - 0.5
122
+ std:
123
+ - - - 0.5
124
+ - - - 0.5
125
+ - - - 0.5
126
+ observation.state:
127
+ min:
128
+ - 13.456424
129
+ - 32.938293
130
+ max:
131
+ - 496.14618
132
+ - 510.9579
133
+ action:
134
+ min:
135
+ - 12.0
136
+ - 25.0
137
+ max:
138
+ - 511.0
139
+ - 511.0
140
+ policy:
141
+ name: vqbet
142
+ n_obs_steps: 5
143
+ n_action_pred_token: 7
144
+ action_chunk_size: 5
145
+ input_shapes:
146
+ observation.image:
147
+ - 3
148
+ - 96
149
+ - 96
150
+ observation.state:
151
+ - ${env.state_dim}
152
+ output_shapes:
153
+ action:
154
+ - ${env.action_dim}
155
+ input_normalization_modes:
156
+ observation.image: mean_std
157
+ observation.state: min_max
158
+ output_normalization_modes:
159
+ action: min_max
160
+ vision_backbone: resnet18
161
+ crop_shape:
162
+ - 84
163
+ - 84
164
+ crop_is_random: true
165
+ pretrained_backbone_weights: null
166
+ use_group_norm: true
167
+ spatial_softmax_num_keypoints: 32
168
+ n_vqvae_training_steps: ${training.n_vqvae_training_steps}
169
+ vqvae_n_embed: 16
170
+ vqvae_embedding_dim: 256
171
+ vqvae_enc_hidden_dim: 128
172
+ gpt_block_size: 500
173
+ gpt_input_dim: 512
174
+ gpt_output_dim: 512
175
+ gpt_n_layer: 8
176
+ gpt_n_head: 8
177
+ gpt_hidden_dim: 512
178
+ dropout: 0.1
179
+ mlp_hidden_dim: 1024
180
+ offset_loss_weight: 10000.0
181
+ primary_code_loss_weight: 5.0
182
+ secondary_code_loss_weight: 0.5
183
+ bet_softmax_temperature: 0.1
184
+ sequentially_select: false
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:21a3edc7aae94b8f12aba85b309c0f8311533a09fd773982b610ebe61f951b59
3
+ size 158152442