Robotics
LeRobot
Safetensors
diffusion
benmayeux commited on
Commit
342d13e
·
verified ·
1 Parent(s): 600089a

Upload policy weights, train config and readme

Browse files
Files changed (4) hide show
  1. README.md +62 -0
  2. config.json +92 -0
  3. model.safetensors +3 -0
  4. train_config.json +172 -0
README.md ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ datasets: sortingroboticsinc/telti_256_ee_delta
3
+ library_name: lerobot
4
+ license: apache-2.0
5
+ model_name: diffusion
6
+ pipeline_tag: robotics
7
+ tags:
8
+ - lerobot
9
+ - diffusion
10
+ - robotics
11
+ ---
12
+
13
+ # Model Card for diffusion
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+ [Diffusion Policy](https://huggingface.co/papers/2303.04137) treats visuomotor control as a generative diffusion process, producing smooth, multi-step action trajectories that excel at contact-rich manipulation.
19
+
20
+
21
+ This policy has been trained and pushed to the Hub using [LeRobot](https://github.com/huggingface/lerobot).
22
+ See the full documentation at [LeRobot Docs](https://huggingface.co/docs/lerobot/index).
23
+
24
+ ---
25
+
26
+ ## How to Get Started with the Model
27
+
28
+ For a complete walkthrough, see the [training guide](https://huggingface.co/docs/lerobot/il_robots#train-a-policy).
29
+ Below is the short version on how to train and run inference/eval:
30
+
31
+ ### Train from scratch
32
+
33
+ ```bash
34
+ lerobot-train \
35
+ --dataset.repo_id=${HF_USER}/<dataset> \
36
+ --policy.type=act \
37
+ --output_dir=outputs/train/<desired_policy_repo_id> \
38
+ --job_name=lerobot_training \
39
+ --policy.device=cuda \
40
+ --policy.repo_id=${HF_USER}/<desired_policy_repo_id>
41
+ --wandb.enable=true
42
+ ```
43
+
44
+ _Writes checkpoints to `outputs/train/<desired_policy_repo_id>/checkpoints/`._
45
+
46
+ ### Evaluate the policy/run inference
47
+
48
+ ```bash
49
+ lerobot-record \
50
+ --robot.type=so100_follower \
51
+ --dataset.repo_id=<hf_user>/eval_<dataset> \
52
+ --policy.path=<hf_user>/<desired_policy_repo_id> \
53
+ --episodes=10
54
+ ```
55
+
56
+ Prefix the dataset repo with **eval\_** and supply `--policy.path` pointing to a local or hub checkpoint.
57
+
58
+ ---
59
+
60
+ ## Model Details
61
+
62
+ - **License:** apache-2.0
config.json ADDED
@@ -0,0 +1,92 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "diffusion",
3
+ "n_obs_steps": 2,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 38
9
+ ]
10
+ },
11
+ "observation.images.front": {
12
+ "type": "VISUAL",
13
+ "shape": [
14
+ 3,
15
+ 540,
16
+ 960
17
+ ]
18
+ },
19
+ "observation.images.wrist": {
20
+ "type": "VISUAL",
21
+ "shape": [
22
+ 3,
23
+ 480,
24
+ 640
25
+ ]
26
+ }
27
+ },
28
+ "output_features": {
29
+ "action": {
30
+ "type": "ACTION",
31
+ "shape": [
32
+ 7
33
+ ]
34
+ }
35
+ },
36
+ "device": "cuda",
37
+ "use_amp": false,
38
+ "use_peft": false,
39
+ "push_to_hub": true,
40
+ "repo_id": "sortingroboticsinc/telti_diffusion",
41
+ "private": null,
42
+ "tags": null,
43
+ "license": null,
44
+ "pretrained_path": null,
45
+ "horizon": 16,
46
+ "n_action_steps": 8,
47
+ "normalization_mapping": {
48
+ "VISUAL": "MEAN_STD",
49
+ "STATE": "MIN_MAX",
50
+ "ACTION": "MIN_MAX"
51
+ },
52
+ "drop_n_last_frames": 0,
53
+ "vision_backbone": "resnet18",
54
+ "crop_shape": [
55
+ 224,
56
+ 224
57
+ ],
58
+ "resize_shape": null,
59
+ "crop_is_random": true,
60
+ "pretrained_backbone_weights": null,
61
+ "use_group_norm": true,
62
+ "spatial_softmax_num_keypoints": 32,
63
+ "use_separate_rgb_encoder_per_camera": false,
64
+ "down_dims": [
65
+ 512,
66
+ 1024,
67
+ 2048
68
+ ],
69
+ "kernel_size": 5,
70
+ "n_groups": 8,
71
+ "diffusion_step_embed_dim": 128,
72
+ "use_film_scale_modulation": true,
73
+ "noise_scheduler_type": "DDPM",
74
+ "num_train_timesteps": 100,
75
+ "beta_schedule": "squaredcos_cap_v2",
76
+ "beta_start": 0.0001,
77
+ "beta_end": 0.02,
78
+ "prediction_type": "epsilon",
79
+ "clip_sample": true,
80
+ "clip_sample_range": 1.0,
81
+ "num_inference_steps": null,
82
+ "do_mask_loss_for_padding": false,
83
+ "optimizer_lr": 0.0001,
84
+ "optimizer_betas": [
85
+ 0.95,
86
+ 0.999
87
+ ],
88
+ "optimizer_eps": 1e-08,
89
+ "optimizer_weight_decay": 1e-06,
90
+ "scheduler_name": "cosine",
91
+ "scheduler_warmup_steps": 500
92
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3ccdc31af316480953582654e2ef409b922fc030aa6320dabb2d2d0acd4ff166
3
+ size 1073871076
train_config.json ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "sortingroboticsinc/telti_256_ee_delta",
4
+ "root": null,
5
+ "episodes": null,
6
+ "image_transforms": {
7
+ "enable": true,
8
+ "max_num_transforms": 3,
9
+ "random_order": false,
10
+ "tfs": {
11
+ "resize": {
12
+ "weight": 1.0,
13
+ "type": "Resize",
14
+ "kwargs": {
15
+ "size": [
16
+ 256,
17
+ 256
18
+ ]
19
+ }
20
+ }
21
+ }
22
+ },
23
+ "revision": null,
24
+ "use_imagenet_stats": true,
25
+ "video_backend": "torchcodec",
26
+ "streaming": false
27
+ },
28
+ "env": null,
29
+ "policy": {
30
+ "type": "diffusion",
31
+ "n_obs_steps": 2,
32
+ "input_features": {
33
+ "observation.state": {
34
+ "type": "STATE",
35
+ "shape": [
36
+ 38
37
+ ]
38
+ },
39
+ "observation.images.front": {
40
+ "type": "VISUAL",
41
+ "shape": [
42
+ 3,
43
+ 540,
44
+ 960
45
+ ]
46
+ },
47
+ "observation.images.wrist": {
48
+ "type": "VISUAL",
49
+ "shape": [
50
+ 3,
51
+ 480,
52
+ 640
53
+ ]
54
+ }
55
+ },
56
+ "output_features": {
57
+ "action": {
58
+ "type": "ACTION",
59
+ "shape": [
60
+ 7
61
+ ]
62
+ }
63
+ },
64
+ "device": "cuda",
65
+ "use_amp": false,
66
+ "use_peft": false,
67
+ "push_to_hub": true,
68
+ "repo_id": "sortingroboticsinc/telti_diffusion",
69
+ "private": null,
70
+ "tags": null,
71
+ "license": null,
72
+ "pretrained_path": null,
73
+ "horizon": 16,
74
+ "n_action_steps": 8,
75
+ "normalization_mapping": {
76
+ "VISUAL": "MEAN_STD",
77
+ "STATE": "MIN_MAX",
78
+ "ACTION": "MIN_MAX"
79
+ },
80
+ "drop_n_last_frames": 0,
81
+ "vision_backbone": "resnet18",
82
+ "crop_shape": [
83
+ 224,
84
+ 224
85
+ ],
86
+ "resize_shape": null,
87
+ "crop_is_random": true,
88
+ "pretrained_backbone_weights": null,
89
+ "use_group_norm": true,
90
+ "spatial_softmax_num_keypoints": 32,
91
+ "use_separate_rgb_encoder_per_camera": false,
92
+ "down_dims": [
93
+ 512,
94
+ 1024,
95
+ 2048
96
+ ],
97
+ "kernel_size": 5,
98
+ "n_groups": 8,
99
+ "diffusion_step_embed_dim": 128,
100
+ "use_film_scale_modulation": true,
101
+ "noise_scheduler_type": "DDPM",
102
+ "num_train_timesteps": 100,
103
+ "beta_schedule": "squaredcos_cap_v2",
104
+ "beta_start": 0.0001,
105
+ "beta_end": 0.02,
106
+ "prediction_type": "epsilon",
107
+ "clip_sample": true,
108
+ "clip_sample_range": 1.0,
109
+ "num_inference_steps": null,
110
+ "do_mask_loss_for_padding": false,
111
+ "optimizer_lr": 0.0001,
112
+ "optimizer_betas": [
113
+ 0.95,
114
+ 0.999
115
+ ],
116
+ "optimizer_eps": 1e-08,
117
+ "optimizer_weight_decay": 1e-06,
118
+ "scheduler_name": "cosine",
119
+ "scheduler_warmup_steps": 500
120
+ },
121
+ "output_dir": "outputs/train/2026-02-06/16-53-24_telti_diffusion",
122
+ "job_name": "telti_diffusion",
123
+ "resume": false,
124
+ "seed": 1000,
125
+ "num_workers": 4,
126
+ "batch_size": 64,
127
+ "steps": 100000,
128
+ "eval_freq": 20000,
129
+ "log_freq": 200,
130
+ "tolerance_s": 0.0001,
131
+ "save_checkpoint": true,
132
+ "save_freq": 20000,
133
+ "use_policy_training_preset": true,
134
+ "optimizer": {
135
+ "type": "adam",
136
+ "lr": 0.0001,
137
+ "weight_decay": 1e-06,
138
+ "grad_clip_norm": 10.0,
139
+ "betas": [
140
+ 0.95,
141
+ 0.999
142
+ ],
143
+ "eps": 1e-08
144
+ },
145
+ "scheduler": {
146
+ "type": "diffuser",
147
+ "num_warmup_steps": 500,
148
+ "name": "cosine"
149
+ },
150
+ "eval": {
151
+ "n_episodes": 50,
152
+ "batch_size": 50,
153
+ "use_async_envs": false
154
+ },
155
+ "wandb": {
156
+ "enable": true,
157
+ "disable_artifact": false,
158
+ "project": "lerobot",
159
+ "entity": null,
160
+ "notes": null,
161
+ "run_id": "pk3j8f5p",
162
+ "mode": null
163
+ },
164
+ "peft": null,
165
+ "use_rabc": false,
166
+ "rabc_progress_path": null,
167
+ "rabc_kappa": 0.01,
168
+ "rabc_epsilon": 1e-06,
169
+ "rabc_head_mode": "sparse",
170
+ "rename_map": {},
171
+ "checkpoint_path": null
172
+ }