Hugging Face's logo

ardalon
/
sequential_t1_task48_fft

Robotics
LeRobot
Safetensors
smolvla
Model card Files
xet
 Community
ardalon commited on
Commit
65424d2
·
verified ·
1 Parent(s): f1d80ed

Upload policy weights, train config and readme

Browse files
Files changed (4) hide show
  1. README.md +63 -0
  2. config.json +111 -0
  3. model.safetensors +3 -0
  4. train_config.json +392 -0
README.md ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: lerobot/smolvla_base
3
+ datasets: daniecraig/libero_90_no_noops_lerobot_v30
4
+ library_name: lerobot
5
+ license: apache-2.0
6
+ model_name: smolvla
7
+ pipeline_tag: robotics
8
+ tags:
9
+ - lerobot
10
+ - smolvla
11
+ - robotics
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 +1,111 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "smolvla",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.images.image": {
6
+ "type": "VISUAL",
7
+ "shape": [
8
+ 256,
9
+ 256,
10
+ 3
11
+ ]
12
+ },
13
+ "observation.images.wrist_image": {
14
+ "type": "VISUAL",
15
+ "shape": [
16
+ 256,
17
+ 256,
18
+ 3
19
+ ]
20
+ },
21
+ "observation.state": {
22
+ "type": "STATE",
23
+ "shape": [
24
+ 8
25
+ ]
26
+ },
27
+ "observation.states.ee_state": {
28
+ "type": "STATE",
29
+ "shape": [
30
+ 6
31
+ ]
32
+ },
33
+ "observation.states.joint_state": {
34
+ "type": "STATE",
35
+ "shape": [
36
+ 7
37
+ ]
38
+ },
39
+ "observation.states.gripper_state": {
40
+ "type": "STATE",
41
+ "shape": [
42
+ 2
43
+ ]
44
+ }
45
+ },
46
+ "output_features": {
47
+ "action": {
48
+ "type": "ACTION",
49
+ "shape": [
50
+ 7
51
+ ]
52
+ }
53
+ },
54
+ "device": "cuda",
55
+ "use_amp": false,
56
+ "use_peft": false,
57
+ "push_to_hub": true,
58
+ "repo_id": "ardalon/sequential_t1_task48_fft",
59
+ "private": null,
60
+ "tags": null,
61
+ "license": null,
62
+ "pretrained_path": "/content/smolvla_base",
63
+ "chunk_size": 50,
64
+ "n_action_steps": 50,
65
+ "normalization_mapping": {
66
+ "VISUAL": "IDENTITY",
67
+ "STATE": "MEAN_STD",
68
+ "ACTION": "MEAN_STD"
69
+ },
70
+ "max_state_dim": 32,
71
+ "max_action_dim": 32,
72
+ "resize_imgs_with_padding": [
73
+ 512,
74
+ 512
75
+ ],
76
+ "empty_cameras": 0,
77
+ "adapt_to_pi_aloha": false,
78
+ "use_delta_joint_actions_aloha": false,
79
+ "tokenizer_max_length": 48,
80
+ "num_steps": 10,
81
+ "use_cache": true,
82
+ "freeze_vision_encoder": true,
83
+ "train_expert_only": true,
84
+ "train_state_proj": true,
85
+ "optimizer_lr": 2e-05,
86
+ "optimizer_betas": [
87
+ 0.9,
88
+ 0.95
89
+ ],
90
+ "optimizer_eps": 1e-08,
91
+ "optimizer_weight_decay": 0.01,
92
+ "optimizer_grad_clip_norm": 10.0,
93
+ "scheduler_warmup_steps": 1000,
94
+ "scheduler_decay_steps": 30000,
95
+ "scheduler_decay_lr": 2.5e-06,
96
+ "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
97
+ "load_vlm_weights": true,
98
+ "add_image_special_tokens": false,
99
+ "attention_mode": "cross_attn",
100
+ "prefix_length": 0,
101
+ "pad_language_to": "max_length",
102
+ "num_expert_layers": 0,
103
+ "num_vlm_layers": 16,
104
+ "self_attn_every_n_layers": 2,
105
+ "expert_width_multiplier": 0.75,
106
+ "min_period": 0.004,
107
+ "max_period": 4.0,
108
+ "rtc_config": null,
109
+ "compile_model": false,
110
+ "compile_mode": "max-autotune"
111
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3cecc83df49e006d9f235ddf486ff82d3b3bf1372ad91b9ff9c0bd335903e0da
3
+ size 906712520
train_config.json ADDED
@@ -0,0 +1,392 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "daniecraig/libero_90_no_noops_lerobot_v30",
4
+ "root": null,
5
+ "episodes": [
6
+ 3092,
7
+ 3093,
8
+ 3094,
9
+ 3095,
10
+ 3096,
11
+ 3097,
12
+ 3098,
13
+ 3099,
14
+ 3100,
15
+ 3101,
16
+ 3102,
17
+ 3103,
18
+ 3104,
19
+ 3105,
20
+ 3106,
21
+ 3107,
22
+ 3108,
23
+ 3109,
24
+ 3110,
25
+ 3111,
26
+ 3112,
27
+ 3113,
28
+ 3114,
29
+ 3115,
30
+ 3116,
31
+ 3117,
32
+ 3118,
33
+ 3119,
34
+ 3120,
35
+ 3121,
36
+ 3122,
37
+ 3123,
38
+ 3124,
39
+ 3125,
40
+ 3126,
41
+ 3127,
42
+ 3128,
43
+ 3129,
44
+ 3130,
45
+ 3131,
46
+ 3132,
47
+ 3133,
48
+ 3134,
49
+ 3135,
50
+ 3136
51
+ ],
52
+ "image_transforms": {
53
+ "enable": false,
54
+ "max_num_transforms": 3,
55
+ "random_order": false,
56
+ "tfs": {
57
+ "brightness": {
58
+ "weight": 1.0,
59
+ "type": "ColorJitter",
60
+ "kwargs": {
61
+ "brightness": [
62
+ 0.8,
63
+ 1.2
64
+ ]
65
+ }
66
+ },
67
+ "contrast": {
68
+ "weight": 1.0,
69
+ "type": "ColorJitter",
70
+ "kwargs": {
71
+ "contrast": [
72
+ 0.8,
73
+ 1.2
74
+ ]
75
+ }
76
+ },
77
+ "saturation": {
78
+ "weight": 1.0,
79
+ "type": "ColorJitter",
80
+ "kwargs": {
81
+ "saturation": [
82
+ 0.5,
83
+ 1.5
84
+ ]
85
+ }
86
+ },
87
+ "hue": {
88
+ "weight": 1.0,
89
+ "type": "ColorJitter",
90
+ "kwargs": {
91
+ "hue": [
92
+ -0.05,
93
+ 0.05
94
+ ]
95
+ }
96
+ },
97
+ "sharpness": {
98
+ "weight": 1.0,
99
+ "type": "SharpnessJitter",
100
+ "kwargs": {
101
+ "sharpness": [
102
+ 0.5,
103
+ 1.5
104
+ ]
105
+ }
106
+ },
107
+ "affine": {
108
+ "weight": 1.0,
109
+ "type": "RandomAffine",
110
+ "kwargs": {
111
+ "degrees": [
112
+ -5.0,
113
+ 5.0
114
+ ],
115
+ "translate": [
116
+ 0.05,
117
+ 0.05
118
+ ]
119
+ }
120
+ }
121
+ }
122
+ },
123
+ "revision": "main",
124
+ "use_imagenet_stats": true,
125
+ "video_backend": "torchcodec",
126
+ "streaming": false
127
+ },
128
+ "env": {
129
+ "type": "libero",
130
+ "task": "libero_90",
131
+ "fps": 30,
132
+ "features": {
133
+ "action": {
134
+ "type": "ACTION",
135
+ "shape": [
136
+ 7
137
+ ]
138
+ },
139
+ "pixels/agentview_image": {
140
+ "type": "VISUAL",
141
+ "shape": [
142
+ 360,
143
+ 360,
144
+ 3
145
+ ]
146
+ },
147
+ "pixels/robot0_eye_in_hand_image": {
148
+ "type": "VISUAL",
149
+ "shape": [
150
+ 360,
151
+ 360,
152
+ 3
153
+ ]
154
+ },
155
+ "robot_state/eef/pos": {
156
+ "type": "STATE",
157
+ "shape": [
158
+ 3
159
+ ]
160
+ },
161
+ "robot_state/eef/quat": {
162
+ "type": "STATE",
163
+ "shape": [
164
+ 4
165
+ ]
166
+ },
167
+ "robot_state/eef/mat": {
168
+ "type": "STATE",
169
+ "shape": [
170
+ 3,
171
+ 3
172
+ ]
173
+ },
174
+ "robot_state/gripper/qpos": {
175
+ "type": "STATE",
176
+ "shape": [
177
+ 2
178
+ ]
179
+ },
180
+ "robot_state/gripper/qvel": {
181
+ "type": "STATE",
182
+ "shape": [
183
+ 2
184
+ ]
185
+ },
186
+ "robot_state/joints/pos": {
187
+ "type": "STATE",
188
+ "shape": [
189
+ 7
190
+ ]
191
+ },
192
+ "robot_state/joints/vel": {
193
+ "type": "STATE",
194
+ "shape": [
195
+ 7
196
+ ]
197
+ }
198
+ },
199
+ "features_map": {
200
+ "action": "action",
201
+ "robot_state/eef/pos": "observation.state.eef_pos",
202
+ "robot_state/eef/quat": "observation.state.eef_quat",
203
+ "robot_state/eef/mat": "observation.state.eef_mat",
204
+ "robot_state/gripper/qpos": "observation.state.gripper_qpos",
205
+ "robot_state/gripper/qvel": "observation.state.gripper_qvel",
206
+ "robot_state/joints/pos": "observation.state.joint_pos",
207
+ "robot_state/joints/vel": "observation.state.joint_vel",
208
+ "pixels/agentview_image": "observation.images.image",
209
+ "pixels/robot0_eye_in_hand_image": "observation.images.image2"
210
+ },
211
+ "max_parallel_tasks": 1,
212
+ "disable_env_checker": true,
213
+ "task_ids": [
214
+ 48
215
+ ],
216
+ "episode_length": null,
217
+ "obs_type": "pixels_agent_pos",
218
+ "render_mode": "rgb_array",
219
+ "camera_name": "agentview_image,robot0_eye_in_hand_image",
220
+ "init_states": true,
221
+ "camera_name_mapping": null,
222
+ "observation_height": 360,
223
+ "observation_width": 360,
224
+ "control_mode": "relative"
225
+ },
226
+ "policy": {
227
+ "type": "smolvla",
228
+ "n_obs_steps": 1,
229
+ "input_features": {
230
+ "observation.images.image": {
231
+ "type": "VISUAL",
232
+ "shape": [
233
+ 256,
234
+ 256,
235
+ 3
236
+ ]
237
+ },
238
+ "observation.images.wrist_image": {
239
+ "type": "VISUAL",
240
+ "shape": [
241
+ 256,
242
+ 256,
243
+ 3
244
+ ]
245
+ },
246
+ "observation.state": {
247
+ "type": "STATE",
248
+ "shape": [
249
+ 8
250
+ ]
251
+ },
252
+ "observation.states.ee_state": {
253
+ "type": "STATE",
254
+ "shape": [
255
+ 6
256
+ ]
257
+ },
258
+ "observation.states.joint_state": {
259
+ "type": "STATE",
260
+ "shape": [
261
+ 7
262
+ ]
263
+ },
264
+ "observation.states.gripper_state": {
265
+ "type": "STATE",
266
+ "shape": [
267
+ 2
268
+ ]
269
+ }
270
+ },
271
+ "output_features": {
272
+ "action": {
273
+ "type": "ACTION",
274
+ "shape": [
275
+ 7
276
+ ]
277
+ }
278
+ },
279
+ "device": "cuda",
280
+ "use_amp": false,
281
+ "use_peft": false,
282
+ "push_to_hub": true,
283
+ "repo_id": "ardalon/sequential_t1_task48_fft",
284
+ "private": null,
285
+ "tags": null,
286
+ "license": null,
287
+ "pretrained_path": "/content/smolvla_base",
288
+ "chunk_size": 50,
289
+ "n_action_steps": 50,
290
+ "normalization_mapping": {
291
+ "VISUAL": "IDENTITY",
292
+ "STATE": "MEAN_STD",
293
+ "ACTION": "MEAN_STD"
294
+ },
295
+ "max_state_dim": 32,
296
+ "max_action_dim": 32,
297
+ "resize_imgs_with_padding": [
298
+ 512,
299
+ 512
300
+ ],
301
+ "empty_cameras": 0,
302
+ "adapt_to_pi_aloha": false,
303
+ "use_delta_joint_actions_aloha": false,
304
+ "tokenizer_max_length": 48,
305
+ "num_steps": 10,
306
+ "use_cache": true,
307
+ "freeze_vision_encoder": true,
308
+ "train_expert_only": true,
309
+ "train_state_proj": true,
310
+ "optimizer_lr": 2e-05,
311
+ "optimizer_betas": [
312
+ 0.9,
313
+ 0.95
314
+ ],
315
+ "optimizer_eps": 1e-08,
316
+ "optimizer_weight_decay": 0.01,
317
+ "optimizer_grad_clip_norm": 10.0,
318
+ "scheduler_warmup_steps": 1000,
319
+ "scheduler_decay_steps": 30000,
320
+ "scheduler_decay_lr": 2.5e-06,
321
+ "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
322
+ "load_vlm_weights": true,
323
+ "add_image_special_tokens": false,
324
+ "attention_mode": "cross_attn",
325
+ "prefix_length": 0,
326
+ "pad_language_to": "max_length",
327
+ "num_expert_layers": 0,
328
+ "num_vlm_layers": 16,
329
+ "self_attn_every_n_layers": 2,
330
+ "expert_width_multiplier": 0.75,
331
+ "min_period": 0.004,
332
+ "max_period": 4.0,
333
+ "rtc_config": null,
334
+ "compile_model": false,
335
+ "compile_mode": "max-autotune"
336
+ },
337
+ "output_dir": "outputs/train/sequential_t1_task48_fft",
338
+ "job_name": "sequential_t1_task48_fft",
339
+ "resume": false,
340
+ "seed": 42,
341
+ "cudnn_deterministic": false,
342
+ "num_workers": 4,
343
+ "batch_size": 16,
344
+ "steps": 3000,
345
+ "eval_freq": 0,
346
+ "log_freq": 1,
347
+ "tolerance_s": 0.0001,
348
+ "save_checkpoint": true,
349
+ "save_freq": 500,
350
+ "use_policy_training_preset": true,
351
+ "optimizer": {
352
+ "type": "adamw",
353
+ "lr": 2e-05,
354
+ "weight_decay": 0.01,
355
+ "grad_clip_norm": 10.0,
356
+ "betas": [
357
+ 0.9,
358
+ 0.95
359
+ ],
360
+ "eps": 1e-08
361
+ },
362
+ "scheduler": {
363
+ "type": "cosine_decay_with_warmup",
364
+ "num_warmup_steps": 1000,
365
+ "num_decay_steps": 30000,
366
+ "peak_lr": 2e-05,
367
+ "decay_lr": 2.5e-06
368
+ },
369
+ "eval": {
370
+ "n_episodes": 50,
371
+ "batch_size": 50,
372
+ "use_async_envs": false
373
+ },
374
+ "wandb": {
375
+ "enable": true,
376
+ "disable_artifact": true,
377
+ "project": "lerobot-smolvla",
378
+ "entity": null,
379
+ "notes": null,
380
+ "run_id": "ooipb6on",
381
+ "mode": null,
382
+ "add_tags": true
383
+ },
384
+ "peft": null,
385
+ "use_rabc": false,
386
+ "rabc_progress_path": null,
387
+ "rabc_kappa": 0.01,
388
+ "rabc_epsilon": 1e-06,
389
+ "rabc_head_mode": "sparse",
390
+ "rename_map": {},
391
+ "checkpoint_path": null
392
+ }