Yi Zhang commited on
Commit
3d8adcb
·
1 Parent(s): a5d74a9

Initial commit of pretrained policy

Browse files
Files changed (3) hide show
  1. config.json +104 -0
  2. model.safetensors +3 -0
  3. train_config.json +263 -0
config.json ADDED
@@ -0,0 +1,104 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "dit_moe",
3
+ "n_obs_steps": 5,
4
+ "normalization_mapping": {
5
+ "VISUAL": "MEAN_STD",
6
+ "STATE": "MIN_MAX",
7
+ "ACTION": "MIN_MAX"
8
+ },
9
+ "input_features": {
10
+ "observation.images.image": {
11
+ "type": "VISUAL",
12
+ "shape": [
13
+ 3,
14
+ 256,
15
+ 256
16
+ ]
17
+ },
18
+ "observation.images.wrist_image": {
19
+ "type": "VISUAL",
20
+ "shape": [
21
+ 3,
22
+ 256,
23
+ 256
24
+ ]
25
+ },
26
+ "observation.state": {
27
+ "type": "STATE",
28
+ "shape": [
29
+ 8
30
+ ]
31
+ },
32
+ "observation.state.joint": {
33
+ "type": "STATE",
34
+ "shape": [
35
+ 7
36
+ ]
37
+ }
38
+ },
39
+ "output_features": {
40
+ "action": {
41
+ "type": "ACTION",
42
+ "shape": [
43
+ 7
44
+ ]
45
+ }
46
+ },
47
+ "device": "cuda",
48
+ "use_amp": false,
49
+ "horizon": 16,
50
+ "n_action_steps": 8,
51
+ "vision_backbone_size": 18,
52
+ "pretrained_backbone_weights": null,
53
+ "avg_pool": true,
54
+ "vision_backbone_norm_name": "group_norm",
55
+ "vision_backbone_norm_num_groups": 16,
56
+ "use_spatial_softmax": false,
57
+ "vision_backbone": "resnet18",
58
+ "crop_shape": null,
59
+ "crop_is_random": true,
60
+ "use_group_norm": true,
61
+ "spatial_softmax_num_keypoints": 32,
62
+ "dim_model": 512,
63
+ "n_heads": 8,
64
+ "dropout": 0.1,
65
+ "dim_feedforward": 2048,
66
+ "feedforward_activation": "gelu",
67
+ "n_encoder_layers": 6,
68
+ "n_decoder_layers": 6,
69
+ "only_expand_on_transformer_encoder": true,
70
+ "num_expandable_network_each_layer": [
71
+ 1,
72
+ 1,
73
+ 1,
74
+ 1,
75
+ 1,
76
+ 1
77
+ ],
78
+ "threshold": 0.0,
79
+ "autoencoder_hidden_dim": 256,
80
+ "autoencoder_latent_dim": 128,
81
+ "moe_only_on_transformer_encoder": true,
82
+ "noise_scheduler_type": "DDIM",
83
+ "num_train_timesteps": 100,
84
+ "beta_schedule": "squaredcos_cap_v2",
85
+ "beta_start": 0.0001,
86
+ "beta_end": 0.02,
87
+ "prediction_type": "epsilon",
88
+ "clip_sample": true,
89
+ "set_alpha_to_one": true,
90
+ "steps_offset": 0,
91
+ "num_inference_steps": 10,
92
+ "do_mask_loss_for_padding": false,
93
+ "optimizer_lr": 0.0001,
94
+ "optimizer_betas": [
95
+ 0.95,
96
+ 0.999
97
+ ],
98
+ "optimizer_eps": 1e-08,
99
+ "optimizer_weight_decay": 1e-06,
100
+ "optimizer_grad_clip_norm": 10.0,
101
+ "scheduler_name": "cosine",
102
+ "scheduler_warmup_steps": 500,
103
+ "num_learned_tasks": 0
104
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c27dba8c702ecf571b57188f29add7fcf1edd456884fcdd5121397c9cc46a182
3
+ size 292608156
train_config.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "yzembodied/libero_10_image_task_3",
4
+ "root": null,
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": false,
8
+ "max_num_transforms": 3,
9
+ "random_order": false,
10
+ "tfs": {
11
+ "brightness": {
12
+ "weight": 1.0,
13
+ "type": "ColorJitter",
14
+ "kwargs": {
15
+ "brightness": [
16
+ 0.8,
17
+ 1.2
18
+ ]
19
+ }
20
+ },
21
+ "contrast": {
22
+ "weight": 1.0,
23
+ "type": "ColorJitter",
24
+ "kwargs": {
25
+ "contrast": [
26
+ 0.8,
27
+ 1.2
28
+ ]
29
+ }
30
+ },
31
+ "saturation": {
32
+ "weight": 1.0,
33
+ "type": "ColorJitter",
34
+ "kwargs": {
35
+ "saturation": [
36
+ 0.5,
37
+ 1.5
38
+ ]
39
+ }
40
+ },
41
+ "hue": {
42
+ "weight": 1.0,
43
+ "type": "ColorJitter",
44
+ "kwargs": {
45
+ "hue": [
46
+ -0.05,
47
+ 0.05
48
+ ]
49
+ }
50
+ },
51
+ "sharpness": {
52
+ "weight": 1.0,
53
+ "type": "SharpnessJitter",
54
+ "kwargs": {
55
+ "sharpness": [
56
+ 0.5,
57
+ 1.5
58
+ ]
59
+ }
60
+ }
61
+ }
62
+ },
63
+ "revision": null,
64
+ "use_imagenet_stats": true,
65
+ "video_backend": "pyav"
66
+ },
67
+ "env": {
68
+ "type": "libero",
69
+ "task": "Libero_10_Task_3",
70
+ "fps": 20,
71
+ "features": {
72
+ "action": {
73
+ "type": "ACTION",
74
+ "shape": [
75
+ 7
76
+ ]
77
+ },
78
+ "agent_pos": {
79
+ "type": "STATE",
80
+ "shape": [
81
+ 8
82
+ ]
83
+ },
84
+ "joint_state": {
85
+ "type": "STATE",
86
+ "shape": [
87
+ 7
88
+ ]
89
+ },
90
+ "pixels/image": {
91
+ "type": "VISUAL",
92
+ "shape": [
93
+ 256,
94
+ 256,
95
+ 3
96
+ ]
97
+ },
98
+ "pixels/wrist_image": {
99
+ "type": "VISUAL",
100
+ "shape": [
101
+ 256,
102
+ 256,
103
+ 3
104
+ ]
105
+ }
106
+ },
107
+ "features_map": {
108
+ "action": "action",
109
+ "agent_pos": "observation.state",
110
+ "joint_state": "observation.state.joint",
111
+ "pixels/image": "observation.images.image",
112
+ "pixels/wrist_image": "observation.images.wrist_image"
113
+ },
114
+ "benchmark": "libero_10",
115
+ "task_id": "task_0",
116
+ "episode_length": 500
117
+ },
118
+ "policy": {
119
+ "type": "dit_moe",
120
+ "n_obs_steps": 5,
121
+ "normalization_mapping": {
122
+ "VISUAL": "MEAN_STD",
123
+ "STATE": "MIN_MAX",
124
+ "ACTION": "MIN_MAX"
125
+ },
126
+ "input_features": {
127
+ "observation.images.image": {
128
+ "type": "VISUAL",
129
+ "shape": [
130
+ 3,
131
+ 256,
132
+ 256
133
+ ]
134
+ },
135
+ "observation.images.wrist_image": {
136
+ "type": "VISUAL",
137
+ "shape": [
138
+ 3,
139
+ 256,
140
+ 256
141
+ ]
142
+ },
143
+ "observation.state": {
144
+ "type": "STATE",
145
+ "shape": [
146
+ 8
147
+ ]
148
+ },
149
+ "observation.state.joint": {
150
+ "type": "STATE",
151
+ "shape": [
152
+ 7
153
+ ]
154
+ }
155
+ },
156
+ "output_features": {
157
+ "action": {
158
+ "type": "ACTION",
159
+ "shape": [
160
+ 7
161
+ ]
162
+ }
163
+ },
164
+ "device": "cuda",
165
+ "use_amp": false,
166
+ "horizon": 16,
167
+ "n_action_steps": 8,
168
+ "vision_backbone_size": 18,
169
+ "pretrained_backbone_weights": null,
170
+ "avg_pool": true,
171
+ "vision_backbone_norm_name": "group_norm",
172
+ "vision_backbone_norm_num_groups": 16,
173
+ "use_spatial_softmax": false,
174
+ "vision_backbone": "resnet18",
175
+ "crop_shape": null,
176
+ "crop_is_random": true,
177
+ "use_group_norm": true,
178
+ "spatial_softmax_num_keypoints": 32,
179
+ "dim_model": 512,
180
+ "n_heads": 8,
181
+ "dropout": 0.1,
182
+ "dim_feedforward": 2048,
183
+ "feedforward_activation": "gelu",
184
+ "n_encoder_layers": 6,
185
+ "n_decoder_layers": 6,
186
+ "only_expand_on_transformer_encoder": true,
187
+ "num_expandable_network_each_layer": [
188
+ 1,
189
+ 1,
190
+ 1,
191
+ 1,
192
+ 1,
193
+ 1
194
+ ],
195
+ "threshold": 0.0,
196
+ "autoencoder_hidden_dim": 256,
197
+ "autoencoder_latent_dim": 128,
198
+ "moe_only_on_transformer_encoder": true,
199
+ "noise_scheduler_type": "DDIM",
200
+ "num_train_timesteps": 100,
201
+ "beta_schedule": "squaredcos_cap_v2",
202
+ "beta_start": 0.0001,
203
+ "beta_end": 0.02,
204
+ "prediction_type": "epsilon",
205
+ "clip_sample": true,
206
+ "set_alpha_to_one": true,
207
+ "steps_offset": 0,
208
+ "num_inference_steps": 10,
209
+ "do_mask_loss_for_padding": false,
210
+ "optimizer_lr": 0.0001,
211
+ "optimizer_betas": [
212
+ 0.95,
213
+ 0.999
214
+ ],
215
+ "optimizer_eps": 1e-08,
216
+ "optimizer_weight_decay": 1e-06,
217
+ "optimizer_grad_clip_norm": 10.0,
218
+ "scheduler_name": "cosine",
219
+ "scheduler_warmup_steps": 500,
220
+ "num_learned_tasks": 0
221
+ },
222
+ "output_dir": "outputs/train/2025-04-02/04-26-20_libero10_task3_dit_moe",
223
+ "job_name": "libero10_task3_dit_moe",
224
+ "resume": false,
225
+ "seed": 1000,
226
+ "num_workers": 16,
227
+ "batch_size": 32,
228
+ "steps": 80000,
229
+ "eval_freq": 20000,
230
+ "log_freq": 500,
231
+ "save_checkpoint": true,
232
+ "save_freq": 20000,
233
+ "use_policy_training_preset": true,
234
+ "optimizer": {
235
+ "type": "adamw",
236
+ "lr": 0.0001,
237
+ "weight_decay": 1e-06,
238
+ "grad_clip_norm": 10.0,
239
+ "betas": [
240
+ 0.95,
241
+ 0.999
242
+ ],
243
+ "eps": 1e-08
244
+ },
245
+ "scheduler": {
246
+ "type": "diffuser",
247
+ "num_warmup_steps": 500,
248
+ "name": "cosine"
249
+ },
250
+ "eval": {
251
+ "n_episodes": 50,
252
+ "batch_size": 1,
253
+ "use_async_envs": false,
254
+ "max_episodes_rendered": 4
255
+ },
256
+ "wandb": {
257
+ "enable": true,
258
+ "disable_artifact": true,
259
+ "project": "lifelong_learning",
260
+ "entity": "470620104-technical-university-of-munich",
261
+ "notes": null
262
+ }
263
+ }