Robotics
LeRobot
Safetensors
smolvla
shantanu-tr commited on
Commit
dbd68a3
·
verified ·
1 Parent(s): 5996a0b

Upload policy weights, train config and readme

Browse files
Files changed (4) hide show
  1. README.md +63 -0
  2. config.json +104 -0
  3. model.safetensors +3 -0
  4. train_config.json +216 -0
README.md ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: lerobot/smolvla_base
3
+ datasets: TrossenRoboticsCommunity/bimanual-widowxai-handover-cube_01
4
+ library_name: lerobot
5
+ license: apache-2.0
6
+ model_name: smolvla
7
+ pipeline_tag: robotics
8
+ tags:
9
+ - robotics
10
+ - smolvla
11
+ - lerobot
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
+ python -m lerobot.scripts.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
+ python -m 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 +1,104 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "smolvla",
3
+ "n_obs_steps": 1,
4
+ "normalization_mapping": {
5
+ "VISUAL": "IDENTITY",
6
+ "STATE": "MEAN_STD",
7
+ "ACTION": "MEAN_STD"
8
+ },
9
+ "input_features": {
10
+ "observation.state": {
11
+ "type": "STATE",
12
+ "shape": [
13
+ 14
14
+ ]
15
+ },
16
+ "observation.images.top": {
17
+ "type": "VISUAL",
18
+ "shape": [
19
+ 3,
20
+ 480,
21
+ 640
22
+ ]
23
+ },
24
+ "observation.images.bottom": {
25
+ "type": "VISUAL",
26
+ "shape": [
27
+ 3,
28
+ 480,
29
+ 640
30
+ ]
31
+ },
32
+ "observation.images.right": {
33
+ "type": "VISUAL",
34
+ "shape": [
35
+ 3,
36
+ 480,
37
+ 640
38
+ ]
39
+ },
40
+ "observation.images.left": {
41
+ "type": "VISUAL",
42
+ "shape": [
43
+ 3,
44
+ 480,
45
+ 640
46
+ ]
47
+ }
48
+ },
49
+ "output_features": {
50
+ "action": {
51
+ "type": "ACTION",
52
+ "shape": [
53
+ 14
54
+ ]
55
+ }
56
+ },
57
+ "device": "cuda",
58
+ "use_amp": false,
59
+ "push_to_hub": true,
60
+ "repo_id": "TrossenRoboticsCommunity/bimanual_widowxai_handover_cube_smolvla",
61
+ "private": null,
62
+ "tags": null,
63
+ "license": null,
64
+ "chunk_size": 50,
65
+ "n_action_steps": 50,
66
+ "max_state_dim": 32,
67
+ "max_action_dim": 32,
68
+ "resize_imgs_with_padding": [
69
+ 512,
70
+ 512
71
+ ],
72
+ "empty_cameras": 0,
73
+ "adapt_to_pi_aloha": false,
74
+ "use_delta_joint_actions_aloha": false,
75
+ "tokenizer_max_length": 48,
76
+ "num_steps": 10,
77
+ "use_cache": true,
78
+ "freeze_vision_encoder": true,
79
+ "train_expert_only": true,
80
+ "train_state_proj": true,
81
+ "optimizer_lr": 0.0001,
82
+ "optimizer_betas": [
83
+ 0.9,
84
+ 0.95
85
+ ],
86
+ "optimizer_eps": 1e-08,
87
+ "optimizer_weight_decay": 1e-10,
88
+ "optimizer_grad_clip_norm": 10.0,
89
+ "scheduler_warmup_steps": 1000,
90
+ "scheduler_decay_steps": 30000,
91
+ "scheduler_decay_lr": 2.5e-06,
92
+ "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
93
+ "load_vlm_weights": true,
94
+ "add_image_special_tokens": false,
95
+ "attention_mode": "cross_attn",
96
+ "prefix_length": 0,
97
+ "pad_language_to": "max_length",
98
+ "num_expert_layers": 0,
99
+ "num_vlm_layers": 16,
100
+ "self_attn_every_n_layers": 2,
101
+ "expert_width_multiplier": 0.75,
102
+ "min_period": 0.004,
103
+ "max_period": 4.0
104
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:714c4601e0ee87cc1dd959cb5a443e84cc9e42ef61bf7f1ae5beed88b0b24cbf
3
+ size 906713496
train_config.json ADDED
@@ -0,0 +1,216 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "TrossenRoboticsCommunity/bimanual-widowxai-handover-cube_01",
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": null,
68
+ "policy": {
69
+ "type": "smolvla",
70
+ "n_obs_steps": 1,
71
+ "normalization_mapping": {
72
+ "VISUAL": "IDENTITY",
73
+ "STATE": "MEAN_STD",
74
+ "ACTION": "MEAN_STD"
75
+ },
76
+ "input_features": {
77
+ "observation.state": {
78
+ "type": "STATE",
79
+ "shape": [
80
+ 14
81
+ ]
82
+ },
83
+ "observation.images.top": {
84
+ "type": "VISUAL",
85
+ "shape": [
86
+ 3,
87
+ 480,
88
+ 640
89
+ ]
90
+ },
91
+ "observation.images.bottom": {
92
+ "type": "VISUAL",
93
+ "shape": [
94
+ 3,
95
+ 480,
96
+ 640
97
+ ]
98
+ },
99
+ "observation.images.right": {
100
+ "type": "VISUAL",
101
+ "shape": [
102
+ 3,
103
+ 480,
104
+ 640
105
+ ]
106
+ },
107
+ "observation.images.left": {
108
+ "type": "VISUAL",
109
+ "shape": [
110
+ 3,
111
+ 480,
112
+ 640
113
+ ]
114
+ }
115
+ },
116
+ "output_features": {
117
+ "action": {
118
+ "type": "ACTION",
119
+ "shape": [
120
+ 14
121
+ ]
122
+ }
123
+ },
124
+ "device": "cuda",
125
+ "use_amp": false,
126
+ "push_to_hub": true,
127
+ "repo_id": "TrossenRoboticsCommunity/bimanual_widowxai_handover_cube_smolvla",
128
+ "private": null,
129
+ "tags": null,
130
+ "license": null,
131
+ "chunk_size": 50,
132
+ "n_action_steps": 50,
133
+ "max_state_dim": 32,
134
+ "max_action_dim": 32,
135
+ "resize_imgs_with_padding": [
136
+ 512,
137
+ 512
138
+ ],
139
+ "empty_cameras": 0,
140
+ "adapt_to_pi_aloha": false,
141
+ "use_delta_joint_actions_aloha": false,
142
+ "tokenizer_max_length": 48,
143
+ "num_steps": 10,
144
+ "use_cache": true,
145
+ "freeze_vision_encoder": true,
146
+ "train_expert_only": true,
147
+ "train_state_proj": true,
148
+ "optimizer_lr": 0.0001,
149
+ "optimizer_betas": [
150
+ 0.9,
151
+ 0.95
152
+ ],
153
+ "optimizer_eps": 1e-08,
154
+ "optimizer_weight_decay": 1e-10,
155
+ "optimizer_grad_clip_norm": 10.0,
156
+ "scheduler_warmup_steps": 1000,
157
+ "scheduler_decay_steps": 30000,
158
+ "scheduler_decay_lr": 2.5e-06,
159
+ "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
160
+ "load_vlm_weights": true,
161
+ "add_image_special_tokens": false,
162
+ "attention_mode": "cross_attn",
163
+ "prefix_length": 0,
164
+ "pad_language_to": "max_length",
165
+ "num_expert_layers": 0,
166
+ "num_vlm_layers": 16,
167
+ "self_attn_every_n_layers": 2,
168
+ "expert_width_multiplier": 0.75,
169
+ "min_period": 0.004,
170
+ "max_period": 4.0
171
+ },
172
+ "output_dir": "outputs/train/bimanual_widowxai__smolvla",
173
+ "job_name": "bimanual_smolvla_training",
174
+ "resume": false,
175
+ "seed": 1000,
176
+ "num_workers": 4,
177
+ "batch_size": 4,
178
+ "steps": 20000,
179
+ "eval_freq": 20000,
180
+ "log_freq": 200,
181
+ "save_checkpoint": true,
182
+ "save_freq": 20000,
183
+ "use_policy_training_preset": true,
184
+ "optimizer": {
185
+ "type": "adamw",
186
+ "lr": 0.0001,
187
+ "weight_decay": 1e-10,
188
+ "grad_clip_norm": 10.0,
189
+ "betas": [
190
+ 0.9,
191
+ 0.95
192
+ ],
193
+ "eps": 1e-08
194
+ },
195
+ "scheduler": {
196
+ "type": "cosine_decay_with_warmup",
197
+ "num_warmup_steps": 1000,
198
+ "num_decay_steps": 30000,
199
+ "peak_lr": 0.0001,
200
+ "decay_lr": 2.5e-06
201
+ },
202
+ "eval": {
203
+ "n_episodes": 50,
204
+ "batch_size": 50,
205
+ "use_async_envs": false
206
+ },
207
+ "wandb": {
208
+ "enable": false,
209
+ "disable_artifact": false,
210
+ "project": "lerobot",
211
+ "entity": null,
212
+ "notes": null,
213
+ "run_id": null,
214
+ "mode": null
215
+ }
216
+ }