Hugging Face's logo

lerobot
/
smolvla_robomme

Robotics
LeRobot
Safetensors
smolvla
Model card Files
xet
 Community
pepijn223 HF Staff commited on
Commit
fb11faa
·
verified ·
1 Parent(s): cc31065

Upload policy weights, train config and readme

Browse files
Files changed (4) hide show
  1. README.md +63 -0
  2. config.json +117 -0
  3. model.safetensors +3 -0
  4. train_config.json +304 -0
README.md ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: lerobot/smolvla_base
3
+ datasets: pepijn223/robomme_data_lerobot_video
4
+ library_name: lerobot
5
+ license: apache-2.0
6
+ model_name: smolvla
7
+ pipeline_tag: robotics
8
+ tags:
9
+ - lerobot
10
+ - robotics
11
+ - smolvla
12
+ ---
13
+
14
+ # Model Card for smolvla
15
+
16
+ <!-- Provide a quick summary of what the model is/does. -->
17
+
18
+
19
+ [SmolVLA](https://huggingface.co/papers/2506.01844) is a compact, efficient vision-language-action model that achieves competitive performance at reduced computational costs and can be deployed on consumer-grade hardware.
20
+
21
+
22
+ This policy has been trained and pushed to the Hub using [LeRobot](https://github.com/huggingface/lerobot).
23
+ See the full documentation at [LeRobot Docs](https://huggingface.co/docs/lerobot/index).
24
+
25
+ ---
26
+
27
+ ## How to Get Started with the Model
28
+
29
+ For a complete walkthrough, see the [training guide](https://huggingface.co/docs/lerobot/il_robots#train-a-policy).
30
+ Below is the short version on how to train and run inference/eval:
31
+
32
+ ### Train from scratch
33
+
34
+ ```bash
35
+ lerobot-train \
36
+ --dataset.repo_id=${HF_USER}/<dataset> \
37
+ --policy.type=act \
38
+ --output_dir=outputs/train/<desired_policy_repo_id> \
39
+ --job_name=lerobot_training \
40
+ --policy.device=cuda \
41
+ --policy.repo_id=${HF_USER}/<desired_policy_repo_id>
42
+ --wandb.enable=true
43
+ ```
44
+
45
+ _Writes checkpoints to `outputs/train/<desired_policy_repo_id>/checkpoints/`._
46
+
47
+ ### Evaluate the policy/run inference
48
+
49
+ ```bash
50
+ lerobot-record \
51
+ --robot.type=so100_follower \
52
+ --dataset.repo_id=<hf_user>/eval_<dataset> \
53
+ --policy.path=<hf_user>/<desired_policy_repo_id> \
54
+ --episodes=10
55
+ ```
56
+
57
+ Prefix the dataset repo with **eval\_** and supply `--policy.path` pointing to a local or hub checkpoint.
58
+
59
+ ---
60
+
61
+ ## Model Details
62
+
63
+ - **License:** apache-2.0
config.json ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "smolvla",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 6
9
+ ]
10
+ },
11
+ "observation.images.camera1": {
12
+ "type": "VISUAL",
13
+ "shape": [
14
+ 3,
15
+ 256,
16
+ 256
17
+ ]
18
+ },
19
+ "observation.images.camera2": {
20
+ "type": "VISUAL",
21
+ "shape": [
22
+ 3,
23
+ 256,
24
+ 256
25
+ ]
26
+ },
27
+ "observation.images.camera3": {
28
+ "type": "VISUAL",
29
+ "shape": [
30
+ 3,
31
+ 256,
32
+ 256
33
+ ]
34
+ },
35
+ "observation.images.empty_camera_0": {
36
+ "type": "VISUAL",
37
+ "shape": [
38
+ 3,
39
+ 480,
40
+ 640
41
+ ]
42
+ },
43
+ "observation.images.empty_camera_1": {
44
+ "type": "VISUAL",
45
+ "shape": [
46
+ 3,
47
+ 480,
48
+ 640
49
+ ]
50
+ }
51
+ },
52
+ "output_features": {
53
+ "actions": {
54
+ "type": "ACTION",
55
+ "shape": [
56
+ 8
57
+ ]
58
+ }
59
+ },
60
+ "device": "cuda",
61
+ "use_amp": false,
62
+ "use_peft": false,
63
+ "push_to_hub": true,
64
+ "repo_id": "pepijn223/smolvla_robomme",
65
+ "private": null,
66
+ "tags": null,
67
+ "license": null,
68
+ "pretrained_path": "lerobot/smolvla_base",
69
+ "chunk_size": 50,
70
+ "n_action_steps": 50,
71
+ "normalization_mapping": {
72
+ "VISUAL": "IDENTITY",
73
+ "STATE": "MEAN_STD",
74
+ "ACTION": "MEAN_STD"
75
+ },
76
+ "max_state_dim": 32,
77
+ "max_action_dim": 32,
78
+ "resize_imgs_with_padding": [
79
+ 512,
80
+ 512
81
+ ],
82
+ "empty_cameras": 2,
83
+ "adapt_to_pi_aloha": false,
84
+ "use_delta_joint_actions_aloha": false,
85
+ "tokenizer_max_length": 48,
86
+ "num_steps": 10,
87
+ "use_cache": true,
88
+ "freeze_vision_encoder": false,
89
+ "train_expert_only": false,
90
+ "train_state_proj": true,
91
+ "optimizer_lr": 0.0001,
92
+ "optimizer_betas": [
93
+ 0.9,
94
+ 0.95
95
+ ],
96
+ "optimizer_eps": 1e-08,
97
+ "optimizer_weight_decay": 1e-10,
98
+ "optimizer_grad_clip_norm": 10.0,
99
+ "scheduler_warmup_steps": 1000,
100
+ "scheduler_decay_steps": 20000,
101
+ "scheduler_decay_lr": 2.5e-06,
102
+ "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
103
+ "load_vlm_weights": true,
104
+ "add_image_special_tokens": false,
105
+ "attention_mode": "cross_attn",
106
+ "prefix_length": 0,
107
+ "pad_language_to": "max_length",
108
+ "num_expert_layers": 0,
109
+ "num_vlm_layers": 16,
110
+ "self_attn_every_n_layers": 2,
111
+ "expert_width_multiplier": 0.75,
112
+ "min_period": 0.004,
113
+ "max_period": 4.0,
114
+ "rtc_config": null,
115
+ "compile_model": false,
116
+ "compile_mode": "max-autotune"
117
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b6fd1652f410bef171b5b5c9145f66a14e11b00cb77473dd1b41f436bdc23bb
3
+ size 906712520
train_config.json ADDED
@@ -0,0 +1,304 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "pepijn223/robomme_data_lerobot_video",
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
+ "affine": {
62
+ "weight": 1.0,
63
+ "type": "RandomAffine",
64
+ "kwargs": {
65
+ "degrees": [
66
+ -5.0,
67
+ 5.0
68
+ ],
69
+ "translate": [
70
+ 0.05,
71
+ 0.05
72
+ ]
73
+ }
74
+ }
75
+ }
76
+ },
77
+ "revision": "main",
78
+ "use_imagenet_stats": false,
79
+ "video_backend": "torchcodec",
80
+ "streaming": false
81
+ },
82
+ "env": {
83
+ "type": "robomme",
84
+ "task": "PickXtimes",
85
+ "fps": 10,
86
+ "features": {
87
+ "action": {
88
+ "type": "ACTION",
89
+ "shape": [
90
+ 8
91
+ ]
92
+ },
93
+ "front_rgb": {
94
+ "type": "VISUAL",
95
+ "shape": [
96
+ 256,
97
+ 256,
98
+ 3
99
+ ]
100
+ },
101
+ "wrist_rgb": {
102
+ "type": "VISUAL",
103
+ "shape": [
104
+ 256,
105
+ 256,
106
+ 3
107
+ ]
108
+ },
109
+ "observation.state": {
110
+ "type": "STATE",
111
+ "shape": [
112
+ 8
113
+ ]
114
+ }
115
+ },
116
+ "features_map": {
117
+ "action": "action",
118
+ "front_rgb": "observation.images.front",
119
+ "wrist_rgb": "observation.images.wrist",
120
+ "observation.state": "observation.state"
121
+ },
122
+ "max_parallel_tasks": 1,
123
+ "disable_env_checker": true,
124
+ "episode_length": 300,
125
+ "action_space": "joint_angle",
126
+ "dataset_split": "test",
127
+ "task_ids": null
128
+ },
129
+ "policy": {
130
+ "type": "smolvla",
131
+ "n_obs_steps": 1,
132
+ "input_features": {
133
+ "observation.state": {
134
+ "type": "STATE",
135
+ "shape": [
136
+ 6
137
+ ]
138
+ },
139
+ "observation.images.camera1": {
140
+ "type": "VISUAL",
141
+ "shape": [
142
+ 3,
143
+ 256,
144
+ 256
145
+ ]
146
+ },
147
+ "observation.images.camera2": {
148
+ "type": "VISUAL",
149
+ "shape": [
150
+ 3,
151
+ 256,
152
+ 256
153
+ ]
154
+ },
155
+ "observation.images.camera3": {
156
+ "type": "VISUAL",
157
+ "shape": [
158
+ 3,
159
+ 256,
160
+ 256
161
+ ]
162
+ },
163
+ "observation.images.empty_camera_0": {
164
+ "type": "VISUAL",
165
+ "shape": [
166
+ 3,
167
+ 480,
168
+ 640
169
+ ]
170
+ },
171
+ "observation.images.empty_camera_1": {
172
+ "type": "VISUAL",
173
+ "shape": [
174
+ 3,
175
+ 480,
176
+ 640
177
+ ]
178
+ }
179
+ },
180
+ "output_features": {
181
+ "actions": {
182
+ "type": "ACTION",
183
+ "shape": [
184
+ 8
185
+ ]
186
+ }
187
+ },
188
+ "device": "cuda",
189
+ "use_amp": false,
190
+ "use_peft": false,
191
+ "push_to_hub": true,
192
+ "repo_id": "pepijn223/smolvla_robomme",
193
+ "private": null,
194
+ "tags": null,
195
+ "license": null,
196
+ "pretrained_path": "lerobot/smolvla_base",
197
+ "chunk_size": 50,
198
+ "n_action_steps": 50,
199
+ "normalization_mapping": {
200
+ "VISUAL": "IDENTITY",
201
+ "STATE": "MEAN_STD",
202
+ "ACTION": "MEAN_STD"
203
+ },
204
+ "max_state_dim": 32,
205
+ "max_action_dim": 32,
206
+ "resize_imgs_with_padding": [
207
+ 512,
208
+ 512
209
+ ],
210
+ "empty_cameras": 2,
211
+ "adapt_to_pi_aloha": false,
212
+ "use_delta_joint_actions_aloha": false,
213
+ "tokenizer_max_length": 48,
214
+ "num_steps": 10,
215
+ "use_cache": true,
216
+ "freeze_vision_encoder": false,
217
+ "train_expert_only": false,
218
+ "train_state_proj": true,
219
+ "optimizer_lr": 0.0001,
220
+ "optimizer_betas": [
221
+ 0.9,
222
+ 0.95
223
+ ],
224
+ "optimizer_eps": 1e-08,
225
+ "optimizer_weight_decay": 1e-10,
226
+ "optimizer_grad_clip_norm": 10.0,
227
+ "scheduler_warmup_steps": 1000,
228
+ "scheduler_decay_steps": 20000,
229
+ "scheduler_decay_lr": 2.5e-06,
230
+ "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
231
+ "load_vlm_weights": true,
232
+ "add_image_special_tokens": false,
233
+ "attention_mode": "cross_attn",
234
+ "prefix_length": 0,
235
+ "pad_language_to": "max_length",
236
+ "num_expert_layers": 0,
237
+ "num_vlm_layers": 16,
238
+ "self_attn_every_n_layers": 2,
239
+ "expert_width_multiplier": 0.75,
240
+ "min_period": 0.004,
241
+ "max_period": 4.0,
242
+ "rtc_config": null,
243
+ "compile_model": false,
244
+ "compile_mode": "max-autotune"
245
+ },
246
+ "output_dir": "outputs/train/smolvla_robomme",
247
+ "job_name": "robomme_smolvla",
248
+ "resume": false,
249
+ "seed": 1000,
250
+ "cudnn_deterministic": false,
251
+ "num_workers": 4,
252
+ "batch_size": 32,
253
+ "steps": 20000,
254
+ "eval_freq": 0,
255
+ "log_freq": 200,
256
+ "tolerance_s": 0.0001,
257
+ "save_checkpoint": true,
258
+ "save_freq": 20000,
259
+ "use_policy_training_preset": true,
260
+ "optimizer": {
261
+ "type": "adamw",
262
+ "lr": 0.0001,
263
+ "weight_decay": 1e-10,
264
+ "grad_clip_norm": 10.0,
265
+ "betas": [
266
+ 0.9,
267
+ 0.95
268
+ ],
269
+ "eps": 1e-08
270
+ },
271
+ "scheduler": {
272
+ "type": "cosine_decay_with_warmup",
273
+ "num_warmup_steps": 1000,
274
+ "num_decay_steps": 20000,
275
+ "peak_lr": 0.0001,
276
+ "decay_lr": 2.5e-06
277
+ },
278
+ "eval": {
279
+ "n_episodes": 50,
280
+ "batch_size": 50,
281
+ "use_async_envs": true
282
+ },
283
+ "wandb": {
284
+ "enable": false,
285
+ "disable_artifact": false,
286
+ "project": "lerobot",
287
+ "entity": null,
288
+ "notes": null,
289
+ "run_id": null,
290
+ "mode": null,
291
+ "add_tags": true
292
+ },
293
+ "peft": null,
294
+ "use_rabc": false,
295
+ "rabc_progress_path": null,
296
+ "rabc_kappa": 0.01,
297
+ "rabc_epsilon": 1e-06,
298
+ "rabc_head_mode": "sparse",
299
+ "rename_map": {
300
+ "observation.images.image": "observation.images.camera1",
301
+ "observation.images.wrist_image": "observation.images.camera2"
302
+ },
303
+ "checkpoint_path": null
304
+ }