Hugging Face's logo

danielsanjosepro
/
ditmeanflow_stack_cake_v2

Robotics
LeRobot
Safetensors
ditmeanflow
Model card Files
xet
 Community
danielsanjosepro commited on
Commit
669ffea
·
verified ·
1 Parent(s): f2761d8

Upload policy weights, train config and readme

Browse files
Files changed (4) hide show
  1. README.md +62 -0
  2. config.json +130 -0
  3. model.safetensors +3 -0
  4. train_config.json +240 -0
README.md ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ datasets: LSY-lab/stack_cake_v2
3
+ library_name: lerobot
4
+ license: apache-2.0
5
+ model_name: ditmeanflow
6
+ pipeline_tag: robotics
7
+ tags:
8
+ - robotics
9
+ - lerobot
10
+ - ditmeanflow
11
+ ---
12
+
13
+ # Model Card for ditmeanflow
14
+
15
+ <!-- Provide a quick summary of what the model is/does. -->
16
+
17
+
18
+ _Model type not recognized — please update this template._
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
+ python -m lerobot.scripts.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
+ python -m 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 +1,130 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "ditmeanflow",
3
+ "n_obs_steps": 2,
4
+ "normalization_mapping": {
5
+ "VISUAL": "MEAN_STD",
6
+ "STATE": "MIN_MAX",
7
+ "ACTION": "MEAN_STD"
8
+ },
9
+ "input_features": {
10
+ "observation.images.wrist": {
11
+ "type": "VISUAL",
12
+ "shape": [
13
+ 3,
14
+ 256,
15
+ 256
16
+ ]
17
+ },
18
+ "observation.state.cartesian": {
19
+ "type": "STATE",
20
+ "shape": [
21
+ 6
22
+ ]
23
+ },
24
+ "observation.state.gripper": {
25
+ "type": "STATE",
26
+ "shape": [
27
+ 1
28
+ ]
29
+ },
30
+ "observation.state.joints": {
31
+ "type": "STATE",
32
+ "shape": [
33
+ 7
34
+ ]
35
+ },
36
+ "observation.state.sensors_ft_sensor": {
37
+ "type": "STATE",
38
+ "shape": [
39
+ 6
40
+ ]
41
+ },
42
+ "observation.state.sensors_tactile_sensor": {
43
+ "type": "STATE",
44
+ "shape": [
45
+ 15
46
+ ]
47
+ },
48
+ "observation.state.target": {
49
+ "type": "STATE",
50
+ "shape": [
51
+ 6
52
+ ]
53
+ },
54
+ "observation.state": {
55
+ "type": "STATE",
56
+ "shape": [
57
+ 41
58
+ ]
59
+ }
60
+ },
61
+ "output_features": {
62
+ "action": {
63
+ "type": "ACTION",
64
+ "shape": [
65
+ 7
66
+ ]
67
+ }
68
+ },
69
+ "device": "cuda",
70
+ "use_amp": false,
71
+ "push_to_hub": true,
72
+ "repo_id": "danielsanjosepro/ditmeanflow_stack_cake_v2",
73
+ "private": null,
74
+ "tags": null,
75
+ "license": null,
76
+ "horizon": 16,
77
+ "n_action_steps": 8,
78
+ "use_proprioceptive": true,
79
+ "drop_n_last_frames": 7,
80
+ "vision_backbone": "resnet18",
81
+ "crop_shape": [
82
+ 256,
83
+ 256
84
+ ],
85
+ "crop_is_random": true,
86
+ "pretrained_backbone_weights": null,
87
+ "use_group_norm": true,
88
+ "spatial_softmax_num_keypoints": 32,
89
+ "use_separate_rgb_encoder_per_camera": false,
90
+ "frequency_embedding_dim": 256,
91
+ "hidden_dim": 512,
92
+ "num_blocks": 6,
93
+ "num_heads": 16,
94
+ "dropout": 0.1,
95
+ "dim_feedforward": 4096,
96
+ "activation": "gelu",
97
+ "training_noise_sampling": "uniform",
98
+ "clip_sample": true,
99
+ "clip_sample_range": 1.0,
100
+ "num_inference_steps": 100,
101
+ "do_mask_loss_for_padding": false,
102
+ "optimizer_lr": 0.000141,
103
+ "optimizer_betas": [
104
+ 0.95,
105
+ 0.999
106
+ ],
107
+ "optimizer_eps": 1e-08,
108
+ "optimizer_weight_decay": 1e-06,
109
+ "scheduler_name": "cosine",
110
+ "scheduler_warmup_steps": 500,
111
+ "features_to_exclude": [
112
+ "observation.images.primary",
113
+ "observation.images.tactile"
114
+ ],
115
+ "features_to_include": null,
116
+ "use_meanflow": true,
117
+ "do_multi_step_sampling": false,
118
+ "inference_timesteps": 1,
119
+ "time_distribution": "uniform",
120
+ "log_norm_mu": -0.4,
121
+ "log_norm_sigma": 1.0,
122
+ "use_autograd_functional_jvp": false,
123
+ "use_adaptative_loss": true,
124
+ "flow_ratio": 0.5,
125
+ "cfg_prob": 0.1,
126
+ "cfg_omega": 2.0,
127
+ "cfg_w": 2.0,
128
+ "cfg_kappa": 0.0,
129
+ "optimizer_grad_clip_norm": 0.5
130
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9d1d4bc4c6ff39a1150e7d88a5689520721b293882f1dd477c93cfe582e5a351
3
+ size 214435548
train_config.json ADDED
@@ -0,0 +1,240 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "LSY-lab/stack_cake_v2",
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": "torchcodec"
66
+ },
67
+ "env": null,
68
+ "policy": {
69
+ "type": "ditmeanflow",
70
+ "n_obs_steps": 2,
71
+ "normalization_mapping": {
72
+ "VISUAL": "MEAN_STD",
73
+ "STATE": "MIN_MAX",
74
+ "ACTION": "MEAN_STD"
75
+ },
76
+ "input_features": {
77
+ "observation.images.wrist": {
78
+ "type": "VISUAL",
79
+ "shape": [
80
+ 3,
81
+ 256,
82
+ 256
83
+ ]
84
+ },
85
+ "observation.state.cartesian": {
86
+ "type": "STATE",
87
+ "shape": [
88
+ 6
89
+ ]
90
+ },
91
+ "observation.state.gripper": {
92
+ "type": "STATE",
93
+ "shape": [
94
+ 1
95
+ ]
96
+ },
97
+ "observation.state.joints": {
98
+ "type": "STATE",
99
+ "shape": [
100
+ 7
101
+ ]
102
+ },
103
+ "observation.state.sensors_ft_sensor": {
104
+ "type": "STATE",
105
+ "shape": [
106
+ 6
107
+ ]
108
+ },
109
+ "observation.state.sensors_tactile_sensor": {
110
+ "type": "STATE",
111
+ "shape": [
112
+ 15
113
+ ]
114
+ },
115
+ "observation.state.target": {
116
+ "type": "STATE",
117
+ "shape": [
118
+ 6
119
+ ]
120
+ },
121
+ "observation.state": {
122
+ "type": "STATE",
123
+ "shape": [
124
+ 41
125
+ ]
126
+ }
127
+ },
128
+ "output_features": {
129
+ "action": {
130
+ "type": "ACTION",
131
+ "shape": [
132
+ 7
133
+ ]
134
+ }
135
+ },
136
+ "device": "cuda",
137
+ "use_amp": false,
138
+ "push_to_hub": true,
139
+ "repo_id": "danielsanjosepro/ditmeanflow_stack_cake_v2",
140
+ "private": null,
141
+ "tags": null,
142
+ "license": null,
143
+ "horizon": 16,
144
+ "n_action_steps": 8,
145
+ "use_proprioceptive": true,
146
+ "drop_n_last_frames": 7,
147
+ "vision_backbone": "resnet18",
148
+ "crop_shape": [
149
+ 256,
150
+ 256
151
+ ],
152
+ "crop_is_random": true,
153
+ "pretrained_backbone_weights": null,
154
+ "use_group_norm": true,
155
+ "spatial_softmax_num_keypoints": 32,
156
+ "use_separate_rgb_encoder_per_camera": false,
157
+ "frequency_embedding_dim": 256,
158
+ "hidden_dim": 512,
159
+ "num_blocks": 6,
160
+ "num_heads": 16,
161
+ "dropout": 0.1,
162
+ "dim_feedforward": 4096,
163
+ "activation": "gelu",
164
+ "training_noise_sampling": "uniform",
165
+ "clip_sample": true,
166
+ "clip_sample_range": 1.0,
167
+ "num_inference_steps": 100,
168
+ "do_mask_loss_for_padding": false,
169
+ "optimizer_lr": 0.000141,
170
+ "optimizer_betas": [
171
+ 0.95,
172
+ 0.999
173
+ ],
174
+ "optimizer_eps": 1e-08,
175
+ "optimizer_weight_decay": 1e-06,
176
+ "scheduler_name": "cosine",
177
+ "scheduler_warmup_steps": 500,
178
+ "features_to_exclude": [
179
+ "observation.images.primary",
180
+ "observation.images.tactile"
181
+ ],
182
+ "features_to_include": null,
183
+ "use_meanflow": true,
184
+ "do_multi_step_sampling": false,
185
+ "inference_timesteps": 1,
186
+ "time_distribution": "uniform",
187
+ "log_norm_mu": -0.4,
188
+ "log_norm_sigma": 1.0,
189
+ "use_autograd_functional_jvp": false,
190
+ "use_adaptative_loss": true,
191
+ "flow_ratio": 0.5,
192
+ "cfg_prob": 0.1,
193
+ "cfg_omega": 2.0,
194
+ "cfg_w": 2.0,
195
+ "cfg_kappa": 0.0,
196
+ "optimizer_grad_clip_norm": 0.5
197
+ },
198
+ "output_dir": "outputs/train/2025-11-24/18-17-10_config/train_meanditflow_stack_cake_v2.json",
199
+ "job_name": "config/train_meanditflow_stack_cake_v2.json",
200
+ "resume": false,
201
+ "seed": 1000,
202
+ "num_workers": 4,
203
+ "batch_size": 128,
204
+ "steps": 20000,
205
+ "eval_freq": 1000,
206
+ "log_freq": 100,
207
+ "save_checkpoint": true,
208
+ "save_freq": 5000,
209
+ "use_policy_training_preset": true,
210
+ "optimizer": {
211
+ "type": "adam",
212
+ "lr": 0.000141,
213
+ "weight_decay": 1e-06,
214
+ "grad_clip_norm": 10.0,
215
+ "betas": [
216
+ 0.95,
217
+ 0.999
218
+ ],
219
+ "eps": 1e-08
220
+ },
221
+ "scheduler": {
222
+ "type": "diffuser",
223
+ "num_warmup_steps": 500,
224
+ "name": "cosine"
225
+ },
226
+ "eval": {
227
+ "n_episodes": 50,
228
+ "batch_size": 50,
229
+ "use_async_envs": false
230
+ },
231
+ "wandb": {
232
+ "enable": true,
233
+ "disable_artifact": false,
234
+ "project": "policies",
235
+ "entity": null,
236
+ "notes": null,
237
+ "run_id": "fqc0qhnp",
238
+ "mode": null
239
+ }
240
+ }