villekuosmanen commited on
Commit
4cda834
·
verified ·
1 Parent(s): 4e03287

Upload policy weights, train config and readme

Browse files
Files changed (4) hide show
  1. README.md +68 -0
  2. config.json +122 -0
  3. model.safetensors +3 -0
  4. train_config.json +263 -0
README.md ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ datasets:
3
+ - villekuosmanen/dAgger_build_block_tower_1.0.0
4
+ - villekuosmanen/build_block_tower
5
+ - villekuosmanen/dAgger_build_block_tower_1.1.0
6
+ - villekuosmanen/dAgger_build_block_tower_1.2.0
7
+ - villekuosmanen/dAgger_build_block_tower_1.3.0
8
+ - villekuosmanen/dAgger_build_block_tower_1.4.0
9
+ library_name: lerobot
10
+ license: apache-2.0
11
+ model_name: diffusion_mp
12
+ pipeline_tag: robotics
13
+ tags:
14
+ - lerobot
15
+ - robotics
16
+ - diffusion_mp
17
+ ---
18
+
19
+ # Model Card for diffusion_mp
20
+
21
+ <!-- Provide a quick summary of what the model is/does. -->
22
+
23
+
24
+ _Model type not recognized — please update this template._
25
+
26
+
27
+ This policy has been trained and pushed to the Hub using [LeRobot](https://github.com/huggingface/lerobot).
28
+ See the full documentation at [LeRobot Docs](https://huggingface.co/docs/lerobot/index).
29
+
30
+ ---
31
+
32
+ ## How to Get Started with the Model
33
+
34
+ For a complete walkthrough, see the [training guide](https://huggingface.co/docs/lerobot/il_robots#train-a-policy).
35
+ Below is the short version on how to train and run inference/eval:
36
+
37
+ ### Train from scratch
38
+
39
+ ```bash
40
+ lerobot-train \
41
+ --dataset.repo_id=${HF_USER}/<dataset> \
42
+ --policy.type=act \
43
+ --output_dir=outputs/train/<desired_policy_repo_id> \
44
+ --job_name=lerobot_training \
45
+ --policy.device=cuda \
46
+ --policy.repo_id=${HF_USER}/<desired_policy_repo_id>
47
+ --wandb.enable=true
48
+ ```
49
+
50
+ _Writes checkpoints to `outputs/train/<desired_policy_repo_id>/checkpoints/`._
51
+
52
+ ### Evaluate the policy/run inference
53
+
54
+ ```bash
55
+ lerobot-record \
56
+ --robot.type=so100_follower \
57
+ --dataset.repo_id=<hf_user>/eval_<dataset> \
58
+ --policy.path=<hf_user>/<desired_policy_repo_id> \
59
+ --episodes=10
60
+ ```
61
+
62
+ Prefix the dataset repo with **eval\_** and supply `--policy.path` pointing to a local or hub checkpoint.
63
+
64
+ ---
65
+
66
+ ## Model Details
67
+
68
+ - **License:** apache-2.0
config.json ADDED
@@ -0,0 +1,122 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "diffusion_mp",
3
+ "n_obs_steps": 2,
4
+ "input_features": {
5
+ "observation.velocity": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 7
9
+ ]
10
+ },
11
+ "observation.state": {
12
+ "type": "STATE",
13
+ "shape": [
14
+ 7
15
+ ]
16
+ },
17
+ "observation.images.wrist": {
18
+ "type": "VISUAL",
19
+ "shape": [
20
+ 3,
21
+ 480,
22
+ 640
23
+ ]
24
+ },
25
+ "observation.effort": {
26
+ "type": "STATE",
27
+ "shape": [
28
+ 7
29
+ ]
30
+ },
31
+ "observation.images.front": {
32
+ "type": "VISUAL",
33
+ "shape": [
34
+ 3,
35
+ 480,
36
+ 640
37
+ ]
38
+ },
39
+ "observation.eef_6d_pose": {
40
+ "type": "STATE",
41
+ "shape": [
42
+ 6
43
+ ]
44
+ }
45
+ },
46
+ "output_features": {
47
+ "action.eef_pose": {
48
+ "type": "ACTION",
49
+ "shape": [
50
+ 7
51
+ ]
52
+ },
53
+ "action": {
54
+ "type": "ACTION",
55
+ "shape": [
56
+ 7
57
+ ]
58
+ }
59
+ },
60
+ "device": "cuda",
61
+ "use_amp": false,
62
+ "use_peft": false,
63
+ "push_to_hub": true,
64
+ "repo_id": "villekuosmanen/build_block_tower_diffusion_motion_primitives",
65
+ "private": null,
66
+ "tags": null,
67
+ "license": null,
68
+ "pretrained_path": "outputs/train/build_block_tower_diffusion_motion_primitives/checkpoints/135000/pretrained_model",
69
+ "horizon": 96,
70
+ "n_action_steps": 48,
71
+ "normalization_mapping": {
72
+ "VISUAL": "MEAN_STD",
73
+ "STATE": "MIN_MAX",
74
+ "ACTION": "MIN_MAX"
75
+ },
76
+ "drop_n_last_frames": 0,
77
+ "vision_backbone": "resnet18",
78
+ "resize_shape": null,
79
+ "crop_ratio": 1.0,
80
+ "crop_shape": null,
81
+ "crop_is_random": true,
82
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
83
+ "use_group_norm": false,
84
+ "spatial_softmax_num_keypoints": 32,
85
+ "use_separate_rgb_encoder_per_camera": true,
86
+ "down_dims": [
87
+ 512,
88
+ 1024,
89
+ 2048
90
+ ],
91
+ "kernel_size": 5,
92
+ "n_groups": 8,
93
+ "diffusion_step_embed_dim": 128,
94
+ "use_film_scale_modulation": true,
95
+ "noise_scheduler_type": "DDPM",
96
+ "num_train_timesteps": 100,
97
+ "beta_schedule": "squaredcos_cap_v2",
98
+ "beta_start": 0.0001,
99
+ "beta_end": 0.02,
100
+ "prediction_type": "epsilon",
101
+ "clip_sample": false,
102
+ "clip_sample_range": 1.0,
103
+ "num_inference_steps": 20,
104
+ "compile_model": false,
105
+ "compile_mode": "reduce-overhead",
106
+ "do_mask_loss_for_padding": true,
107
+ "optimizer_lr": 0.0001,
108
+ "optimizer_betas": [
109
+ 0.95,
110
+ 0.999
111
+ ],
112
+ "optimizer_eps": 1e-08,
113
+ "optimizer_weight_decay": 1e-06,
114
+ "scheduler_name": "cosine",
115
+ "scheduler_warmup_steps": 500,
116
+ "motion_primitive_dim": 7,
117
+ "motion_primitive_dropout": 0.3,
118
+ "all_motion_prompting_dropout": 0.2,
119
+ "observation_state_dropout": 0.2,
120
+ "use_delta_actions": false,
121
+ "delta_action_state_key": "observation.state"
122
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b75f90cc1cea7edb4e262cf7b4ea657d05ac818e9d40abfceee1a46f2ab84a9
3
+ size 1113257956
train_config.json ADDED
@@ -0,0 +1,263 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": [
4
+ "villekuosmanen/dAgger_build_block_tower_1.0.0",
5
+ "villekuosmanen/build_block_tower",
6
+ "villekuosmanen/dAgger_build_block_tower_1.1.0",
7
+ "villekuosmanen/dAgger_build_block_tower_1.2.0",
8
+ "villekuosmanen/dAgger_build_block_tower_1.3.0",
9
+ "villekuosmanen/dAgger_build_block_tower_1.4.0"
10
+ ],
11
+ "root": null,
12
+ "episodes": null,
13
+ "image_transforms": {
14
+ "enable": true,
15
+ "max_num_transforms": 3,
16
+ "random_order": false,
17
+ "tfs": {
18
+ "brightness": {
19
+ "weight": 1.0,
20
+ "type": "ColorJitter",
21
+ "kwargs": {
22
+ "brightness": [
23
+ 0.8,
24
+ 1.2
25
+ ]
26
+ }
27
+ },
28
+ "contrast": {
29
+ "weight": 1.0,
30
+ "type": "ColorJitter",
31
+ "kwargs": {
32
+ "contrast": [
33
+ 0.8,
34
+ 1.2
35
+ ]
36
+ }
37
+ },
38
+ "saturation": {
39
+ "weight": 1.0,
40
+ "type": "ColorJitter",
41
+ "kwargs": {
42
+ "saturation": [
43
+ 0.5,
44
+ 1.5
45
+ ]
46
+ }
47
+ },
48
+ "hue": {
49
+ "weight": 1.0,
50
+ "type": "ColorJitter",
51
+ "kwargs": {
52
+ "hue": [
53
+ -0.05,
54
+ 0.05
55
+ ]
56
+ }
57
+ },
58
+ "sharpness": {
59
+ "weight": 1.0,
60
+ "type": "SharpnessJitter",
61
+ "kwargs": {
62
+ "sharpness": [
63
+ 0.5,
64
+ 1.5
65
+ ]
66
+ }
67
+ },
68
+ "affine": {
69
+ "weight": 1.0,
70
+ "type": "RandomAffine",
71
+ "kwargs": {
72
+ "degrees": [
73
+ -5.0,
74
+ 5.0
75
+ ],
76
+ "translate": [
77
+ 0.05,
78
+ 0.05
79
+ ]
80
+ }
81
+ }
82
+ }
83
+ },
84
+ "revision": null,
85
+ "use_imagenet_stats": true,
86
+ "video_backend": "torchcodec",
87
+ "streaming": false
88
+ },
89
+ "env": null,
90
+ "policy": {
91
+ "type": "diffusion_mp",
92
+ "n_obs_steps": 2,
93
+ "input_features": {
94
+ "observation.velocity": {
95
+ "type": "STATE",
96
+ "shape": [
97
+ 7
98
+ ]
99
+ },
100
+ "observation.state": {
101
+ "type": "STATE",
102
+ "shape": [
103
+ 7
104
+ ]
105
+ },
106
+ "observation.images.wrist": {
107
+ "type": "VISUAL",
108
+ "shape": [
109
+ 3,
110
+ 480,
111
+ 640
112
+ ]
113
+ },
114
+ "observation.effort": {
115
+ "type": "STATE",
116
+ "shape": [
117
+ 7
118
+ ]
119
+ },
120
+ "observation.images.front": {
121
+ "type": "VISUAL",
122
+ "shape": [
123
+ 3,
124
+ 480,
125
+ 640
126
+ ]
127
+ },
128
+ "observation.eef_6d_pose": {
129
+ "type": "STATE",
130
+ "shape": [
131
+ 6
132
+ ]
133
+ }
134
+ },
135
+ "output_features": {
136
+ "action.eef_pose": {
137
+ "type": "ACTION",
138
+ "shape": [
139
+ 7
140
+ ]
141
+ },
142
+ "action": {
143
+ "type": "ACTION",
144
+ "shape": [
145
+ 7
146
+ ]
147
+ }
148
+ },
149
+ "device": "cuda",
150
+ "use_amp": false,
151
+ "use_peft": false,
152
+ "push_to_hub": true,
153
+ "repo_id": "villekuosmanen/build_block_tower_diffusion_motion_primitives",
154
+ "private": null,
155
+ "tags": null,
156
+ "license": null,
157
+ "pretrained_path": "outputs/train/build_block_tower_diffusion_motion_primitives/checkpoints/135000/pretrained_model",
158
+ "horizon": 96,
159
+ "n_action_steps": 48,
160
+ "normalization_mapping": {
161
+ "VISUAL": "MEAN_STD",
162
+ "STATE": "MIN_MAX",
163
+ "ACTION": "MIN_MAX"
164
+ },
165
+ "drop_n_last_frames": 0,
166
+ "vision_backbone": "resnet18",
167
+ "resize_shape": null,
168
+ "crop_ratio": 1.0,
169
+ "crop_shape": null,
170
+ "crop_is_random": true,
171
+ "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
172
+ "use_group_norm": false,
173
+ "spatial_softmax_num_keypoints": 32,
174
+ "use_separate_rgb_encoder_per_camera": true,
175
+ "down_dims": [
176
+ 512,
177
+ 1024,
178
+ 2048
179
+ ],
180
+ "kernel_size": 5,
181
+ "n_groups": 8,
182
+ "diffusion_step_embed_dim": 128,
183
+ "use_film_scale_modulation": true,
184
+ "noise_scheduler_type": "DDPM",
185
+ "num_train_timesteps": 100,
186
+ "beta_schedule": "squaredcos_cap_v2",
187
+ "beta_start": 0.0001,
188
+ "beta_end": 0.02,
189
+ "prediction_type": "epsilon",
190
+ "clip_sample": false,
191
+ "clip_sample_range": 1.0,
192
+ "num_inference_steps": 20,
193
+ "compile_model": false,
194
+ "compile_mode": "reduce-overhead",
195
+ "do_mask_loss_for_padding": true,
196
+ "optimizer_lr": 0.0001,
197
+ "optimizer_betas": [
198
+ 0.95,
199
+ 0.999
200
+ ],
201
+ "optimizer_eps": 1e-08,
202
+ "optimizer_weight_decay": 1e-06,
203
+ "scheduler_name": "cosine",
204
+ "scheduler_warmup_steps": 500,
205
+ "motion_primitive_dim": 7,
206
+ "motion_primitive_dropout": 0.3,
207
+ "all_motion_prompting_dropout": 0.2,
208
+ "observation_state_dropout": 0.2,
209
+ "use_delta_actions": false,
210
+ "delta_action_state_key": "observation.state"
211
+ },
212
+ "output_dir": "outputs/train/build_block_tower_diffusion_motion_primitives",
213
+ "job_name": "build_block_tower_diffusion_motion_primitives",
214
+ "resume": true,
215
+ "seed": 1000,
216
+ "num_workers": 18,
217
+ "batch_size": 24,
218
+ "steps": 200000,
219
+ "eval_freq": -1,
220
+ "log_freq": 20,
221
+ "tolerance_s": 0.0001,
222
+ "save_checkpoint": true,
223
+ "save_freq": 5000,
224
+ "use_policy_training_preset": true,
225
+ "optimizer": {
226
+ "type": "adam",
227
+ "lr": 0.0001,
228
+ "weight_decay": 1e-06,
229
+ "grad_clip_norm": 10.0,
230
+ "betas": [
231
+ 0.95,
232
+ 0.999
233
+ ],
234
+ "eps": 1e-08
235
+ },
236
+ "scheduler": {
237
+ "type": "diffuser",
238
+ "num_warmup_steps": 500,
239
+ "name": "cosine"
240
+ },
241
+ "eval": {
242
+ "n_episodes": 50,
243
+ "batch_size": 50,
244
+ "use_async_envs": false
245
+ },
246
+ "wandb": {
247
+ "enable": true,
248
+ "disable_artifact": false,
249
+ "project": "lerobot",
250
+ "entity": null,
251
+ "notes": null,
252
+ "run_id": null,
253
+ "mode": null
254
+ },
255
+ "peft": null,
256
+ "use_rabc": false,
257
+ "rabc_progress_path": null,
258
+ "rabc_kappa": 0.01,
259
+ "rabc_epsilon": 1e-06,
260
+ "rabc_head_mode": "sparse",
261
+ "rename_map": {},
262
+ "checkpoint_path": "outputs/train/build_block_tower_diffusion_motion_primitives/checkpoints/135000"
263
+ }