Hugging Face's logo

maximellerbach
/
dm05_omx_multicubes_rel

Robotics
LeRobot
Safetensors
dm05
Model card Files
xet
 Community
maximellerbach HF Staff commited on
Commit
5cca39a
·
verified ·
1 Parent(s): 69cf14f

Upload model card and train config

Browse files
Files changed (2) hide show
  1. README.md +167 -0
  2. train_config.json +600 -0
README.md ADDED
@@ -0,0 +1,167 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: Dexmal/DM05
3
+ datasets: maximellerbach/omx_multicubes
4
+ library_name: lerobot
5
+ license: gemma
6
+ model_name: dm05
7
+ pipeline_tag: robotics
8
+ tags:
9
+ - dm05
10
+ - lerobot
11
+ - robotics
12
+ ---
13
+
14
+ # Model Card for dm05
15
+
16
+ <!-- Provide a quick summary of what the model is/does. -->
17
+
18
+
19
+ [DM05](https://huggingface.co/Dexmal/DM05) is Dexmal's Vision-Language-Action model for robot control. It conditions on camera observations, robot state, and language instructions to predict continuous action chunks with a flow-matching action expert.
20
+
21
+
22
+
23
+
24
+ <!-- A short demo is worth more than any description! Record a GIF/video of the policy
25
+ running on your robot, upload it to this repo, and embed it here:
26
+ <p align="center">
27
+ <img src="https://huggingface.co/<hf_user>/<policy_repo_id>/resolve/main/demo.gif" width="60%"/>
28
+ </p>
29
+ -->
30
+
31
+ This policy has been trained and pushed to the Hub using [LeRobot](https://github.com/huggingface/lerobot).
32
+
33
+ Learn how to train and run it in the [LeRobot dm05 guide](https://huggingface.co/docs/lerobot/main/en/dm05), or browse the [full documentation](https://huggingface.co/docs/lerobot/index).
34
+
35
+
36
+ ---
37
+
38
+ ## Model Details
39
+
40
+ - **License:** gemma
41
+ - **Fine-tuned from:** [Dexmal/DM05](https://huggingface.co/Dexmal/DM05)
42
+ - **Robot type:** `omx_follower`
43
+ - **Cameras:** `wrist`, `top`
44
+
45
+
46
+ ## Inputs & Outputs
47
+
48
+ The policy consumes these observation features and produces these action features.
49
+
50
+ **Inputs**
51
+
52
+ | Feature | Type | Shape |
53
+ | --- | --- | --- |
54
+ | `observation.state` | STATE | `(6,)` |
55
+ | `observation.images.wrist` | VISUAL | `(3, 480, 640)` |
56
+ | `observation.images.top` | VISUAL | `(3, 480, 640)` |
57
+
58
+ **Outputs**
59
+
60
+ | Feature | Type | Shape |
61
+ | --- | --- | --- |
62
+ | `action` | ACTION | `(6,)` |
63
+
64
+
65
+ ## Training Dataset
66
+
67
+ - **Repository:** [maximellerbach/omx_multicubes](https://huggingface.co/datasets/maximellerbach/omx_multicubes)
68
+ - **Episodes:** 176
69
+ - **Frames:** 137392
70
+ - **Frame rate:** 30 FPS
71
+ - **Task(s):** "pick all the cubes and place them one by one in the blue square"
72
+
73
+ <a class="flex" href="https://huggingface.co/spaces/lerobot/visualize_dataset?path=maximellerbach/omx_multicubes">
74
+ <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl.svg"/>
75
+ <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl-dark.svg"/>
76
+ </a>
77
+
78
+
79
+ ## Training Configuration
80
+
81
+ | Setting | Value |
82
+ | --- | --- |
83
+ | Training steps | 10000 |
84
+ | Batch size | 4 |
85
+ | Optimizer | adamw |
86
+ | Learning rate | 2.5e-05 |
87
+ | Seed | 1000 |
88
+ | LeRobot version | 0.6.2 |
89
+
90
+ ---
91
+
92
+ ## How to Get Started with the Model
93
+
94
+ New to LeRobot? These guides cover the full workflow:
95
+
96
+ - **[Install LeRobot](https://huggingface.co/docs/lerobot/main/en/installation)** — set up the `lerobot` package.
97
+ - **[Hardware setup](https://huggingface.co/docs/lerobot/main/en/hardware_guide)** — assemble, wire, and calibrate your robot and cameras.
98
+ - **[Record data & train a policy](https://huggingface.co/docs/lerobot/en/il_robots)** — the end-to-end imitation-learning walkthrough.
99
+ - **[CLI cheat-sheet](https://huggingface.co/docs/lerobot/main/en/cheat-sheet)** — quick reference for the `lerobot-*` commands.
100
+
101
+ The short version to run and train this policy:
102
+
103
+ ### Run the policy on your robot
104
+
105
+ ```bash
106
+ lerobot-rollout \
107
+ --strategy.type=base \
108
+ --robot.type=omx_follower \
109
+ --robot.port=<your_robot_port> \
110
+ --robot.cameras="{ <camera_1>: {type: opencv, index_or_path: <index_or_path>, width: 640, height: 480, fps: 30}, <camera_2>: {type: opencv, index_or_path: <index_or_path>, width: 640, height: 480, fps: 30}}" \
111
+ --policy.path=maximellerbach/dm05_omx_multicubes_rel \
112
+ --task="pick all the cubes and place them one by one in the blue square" \
113
+ --duration=60
114
+ ```
115
+
116
+ Replace the remaining `<...>` placeholders with your own values: `--robot.port` and the camera names/indices are specific to your machine, and the camera names must match the observation keys this policy was trained on.
117
+
118
+ When `--strategy.type=base` is used the script doesn't record the episodes. Skipping duration will make the policy run indefinitely. For more information look at [rollout documentation](https://huggingface.co/docs/lerobot/main/en/inference).
119
+
120
+ ### Train your own policy
121
+
122
+ This policy type is usually fine-tuned from the pretrained base model [Dexmal/DM05](https://huggingface.co/Dexmal/DM05):
123
+
124
+ ```bash
125
+ lerobot-train \
126
+ --dataset.repo_id=${HF_USER}/<dataset> \
127
+ --policy.path=Dexmal/DM05 \
128
+ --output_dir=outputs/train/<policy_repo_id> \
129
+ --job_name=lerobot_training \
130
+ --policy.device=cuda \
131
+ --policy.repo_id=${HF_USER}/<policy_repo_id> \
132
+ --wandb.enable=true
133
+ ```
134
+
135
+ _Writes checkpoints to `outputs/train/<policy_repo_id>/checkpoints/`._
136
+
137
+ ---
138
+
139
+ ## Evaluation
140
+
141
+ <!-- Report real-robot results here: run the policy several times per task and count the
142
+ successes. Delete the "No evaluation results" line and fill in this table instead:
143
+
144
+ | Task | Trials | Successes | Success rate |
145
+ | ---- | ------ | --------- | ------------ |
146
+ | pick the lego brick | 10 | 8 | 80% |
147
+
148
+ Also worth noting: anything that affects difficulty (new object positions, lighting,
149
+ distractors, a different robot of the same type, ...).
150
+ -->
151
+
152
+ _No evaluation results have been provided for this policy yet._
153
+
154
+ ---
155
+
156
+ ## Citation
157
+
158
+ If you use this policy, please cite the method linked in the description above, along with LeRobot:
159
+
160
+ ```bibtex
161
+ @misc{cadene2024lerobot,
162
+ author = {Cadene, Remi and Alibert, Simon and Soare, Alexander and Gallouedec, Quentin and Zouitine, Adil and Palma, Steven and Kooijmans, Pepijn and Aractingi, Michel and Shukor, Mustafa and Aubakirova, Dana and Russi, Martino and Capuano, Francesco and Pascal, Caroline and Choghari, Jade and Moss, Jess and Wolf, Thomas},
163
+ title = {LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch},
164
+ howpublished = "\url{https://github.com/huggingface/lerobot}",
165
+ year = {2024}
166
+ }
167
+ ```
train_config.json ADDED
@@ -0,0 +1,600 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset": {
3
+ "repo_id": "maximellerbach/omx_multicubes",
4
+ "repo_type": "dataset",
5
+ "root": null,
6
+ "episodes": null,
7
+ "exclude_episodes": null,
8
+ "image_transforms": {
9
+ "enable": false,
10
+ "max_num_transforms": 3,
11
+ "random_order": false,
12
+ "tfs": {
13
+ "brightness": {
14
+ "weight": 1.0,
15
+ "type": "ColorJitter",
16
+ "kwargs": {
17
+ "brightness": [
18
+ 0.8,
19
+ 1.2
20
+ ]
21
+ }
22
+ },
23
+ "contrast": {
24
+ "weight": 1.0,
25
+ "type": "ColorJitter",
26
+ "kwargs": {
27
+ "contrast": [
28
+ 0.8,
29
+ 1.2
30
+ ]
31
+ }
32
+ },
33
+ "saturation": {
34
+ "weight": 1.0,
35
+ "type": "ColorJitter",
36
+ "kwargs": {
37
+ "saturation": [
38
+ 0.5,
39
+ 1.5
40
+ ]
41
+ }
42
+ },
43
+ "hue": {
44
+ "weight": 1.0,
45
+ "type": "ColorJitter",
46
+ "kwargs": {
47
+ "hue": [
48
+ -0.05,
49
+ 0.05
50
+ ]
51
+ }
52
+ },
53
+ "sharpness": {
54
+ "weight": 1.0,
55
+ "type": "SharpnessJitter",
56
+ "kwargs": {
57
+ "sharpness": [
58
+ 0.5,
59
+ 1.5
60
+ ]
61
+ }
62
+ },
63
+ "affine": {
64
+ "weight": 1.0,
65
+ "type": "RandomAffine",
66
+ "kwargs": {
67
+ "degrees": [
68
+ -5.0,
69
+ 5.0
70
+ ],
71
+ "translate": [
72
+ 0.05,
73
+ 0.05
74
+ ]
75
+ }
76
+ }
77
+ }
78
+ },
79
+ "revision": "dm05-relative-chunk50",
80
+ "use_imagenet_stats": true,
81
+ "video_backend": "pyav",
82
+ "return_uint8": false,
83
+ "depth_output_unit": "mm",
84
+ "streaming": false,
85
+ "eval_split": 0.0
86
+ },
87
+ "env": null,
88
+ "policy": {
89
+ "type": "dm05",
90
+ "n_obs_steps": 1,
91
+ "input_features": {
92
+ "observation.state": {
93
+ "type": "STATE",
94
+ "shape": [
95
+ 6
96
+ ]
97
+ },
98
+ "observation.images.wrist": {
99
+ "type": "VISUAL",
100
+ "shape": [
101
+ 3,
102
+ 480,
103
+ 640
104
+ ]
105
+ },
106
+ "observation.images.top": {
107
+ "type": "VISUAL",
108
+ "shape": [
109
+ 3,
110
+ 480,
111
+ 640
112
+ ]
113
+ }
114
+ },
115
+ "output_features": {
116
+ "action": {
117
+ "type": "ACTION",
118
+ "shape": [
119
+ 6
120
+ ]
121
+ }
122
+ },
123
+ "device": "cuda:0",
124
+ "use_amp": false,
125
+ "use_peft": false,
126
+ "push_to_hub": true,
127
+ "repo_id": "maximellerbach/dm05_omx_multicubes_rel",
128
+ "private": null,
129
+ "tags": null,
130
+ "license": "gemma",
131
+ "pretrained_path": "Dexmal/DM05-Lerobot",
132
+ "pretrained_revision": null,
133
+ "core_config": {
134
+ "vlm_gradient_checkpointing": true,
135
+ "ae_gradient_checkpointing": true,
136
+ "ae_gradient_checkpointing_layers": 1,
137
+ "freeze_vlm_embedding": false,
138
+ "llm_attn_implementation": "eager",
139
+ "vision_attn_implementation": "sdpa",
140
+ "action_attn_implementation": "sdpa",
141
+ "transformers_version": "5.5.4",
142
+ "architectures": [
143
+ "DM05ForConditionalGeneration"
144
+ ],
145
+ "output_hidden_states": false,
146
+ "return_dict": true,
147
+ "dtype": "bfloat16",
148
+ "chunk_size_feed_forward": 0,
149
+ "is_encoder_decoder": false,
150
+ "id2label": {
151
+ "0": "LABEL_0",
152
+ "1": "LABEL_1"
153
+ },
154
+ "label2id": {
155
+ "LABEL_0": 0,
156
+ "LABEL_1": 1
157
+ },
158
+ "problem_type": null,
159
+ "vlm_config": {
160
+ "architectures": [
161
+ "Gemma3ForConditionalGeneration"
162
+ ],
163
+ "output_hidden_states": false,
164
+ "return_dict": true,
165
+ "dtype": "bfloat16",
166
+ "chunk_size_feed_forward": 0,
167
+ "is_encoder_decoder": false,
168
+ "id2label": {
169
+ "0": "LABEL_0",
170
+ "1": "LABEL_1"
171
+ },
172
+ "label2id": {
173
+ "LABEL_0": 0,
174
+ "LABEL_1": 1
175
+ },
176
+ "problem_type": null,
177
+ "text_config": {
178
+ "architectures": null,
179
+ "output_hidden_states": false,
180
+ "return_dict": true,
181
+ "dtype": "bfloat16",
182
+ "chunk_size_feed_forward": 0,
183
+ "is_encoder_decoder": false,
184
+ "id2label": {
185
+ "0": "LABEL_0",
186
+ "1": "LABEL_1"
187
+ },
188
+ "label2id": {
189
+ "LABEL_0": 0,
190
+ "LABEL_1": 1
191
+ },
192
+ "problem_type": null,
193
+ "vocab_size": 262208,
194
+ "hidden_size": 2560,
195
+ "intermediate_size": 10240,
196
+ "num_hidden_layers": 34,
197
+ "num_attention_heads": 8,
198
+ "num_key_value_heads": 4,
199
+ "head_dim": 256,
200
+ "hidden_activation": "gelu_pytorch_tanh",
201
+ "max_position_embeddings": 131072,
202
+ "initializer_range": 0.02,
203
+ "rms_norm_eps": 1e-06,
204
+ "use_cache": true,
205
+ "pad_token_id": 0,
206
+ "eos_token_id": 1,
207
+ "bos_token_id": 2,
208
+ "tie_word_embeddings": true,
209
+ "rope_parameters": {
210
+ "full_attention": {
211
+ "factor": 8.0,
212
+ "rope_theta": 1000000.0,
213
+ "rope_type": "linear"
214
+ },
215
+ "sliding_attention": {
216
+ "rope_theta": 10000.0,
217
+ "rope_type": "default"
218
+ }
219
+ },
220
+ "attention_bias": false,
221
+ "attention_dropout": 0.0,
222
+ "query_pre_attn_scalar": 256,
223
+ "sliding_window": 4096,
224
+ "layer_types": [
225
+ "sliding_attention",
226
+ "sliding_attention",
227
+ "sliding_attention",
228
+ "sliding_attention",
229
+ "sliding_attention",
230
+ "full_attention",
231
+ "sliding_attention",
232
+ "sliding_attention",
233
+ "sliding_attention",
234
+ "sliding_attention",
235
+ "sliding_attention",
236
+ "full_attention",
237
+ "sliding_attention",
238
+ "sliding_attention",
239
+ "sliding_attention",
240
+ "sliding_attention",
241
+ "sliding_attention",
242
+ "full_attention",
243
+ "sliding_attention",
244
+ "sliding_attention",
245
+ "sliding_attention",
246
+ "sliding_attention",
247
+ "sliding_attention",
248
+ "full_attention",
249
+ "sliding_attention",
250
+ "sliding_attention",
251
+ "sliding_attention",
252
+ "sliding_attention",
253
+ "sliding_attention",
254
+ "full_attention",
255
+ "sliding_attention",
256
+ "sliding_attention",
257
+ "sliding_attention",
258
+ "sliding_attention"
259
+ ],
260
+ "final_logit_softcapping": null,
261
+ "attn_logit_softcapping": null,
262
+ "use_bidirectional_attention": false,
263
+ "_sliding_window_pattern": 6,
264
+ "_name_or_path": "",
265
+ "model_type": "gemma3_text",
266
+ "attn_implementation": "eager",
267
+ "output_attentions": false
268
+ },
269
+ "vision_config": {
270
+ "architectures": null,
271
+ "output_hidden_states": false,
272
+ "return_dict": true,
273
+ "dtype": "bfloat16",
274
+ "chunk_size_feed_forward": 0,
275
+ "is_encoder_decoder": false,
276
+ "id2label": {
277
+ "0": "LABEL_0",
278
+ "1": "LABEL_1"
279
+ },
280
+ "label2id": {
281
+ "LABEL_0": 0,
282
+ "LABEL_1": 1
283
+ },
284
+ "problem_type": null,
285
+ "hidden_size": 1152,
286
+ "intermediate_size": 4304,
287
+ "num_hidden_layers": 27,
288
+ "num_attention_heads": 16,
289
+ "num_channels": 3,
290
+ "image_size": 448,
291
+ "patch_size": 14,
292
+ "hidden_act": "gelu_pytorch_tanh",
293
+ "layer_norm_eps": 1e-06,
294
+ "attention_dropout": 0.0,
295
+ "_name_or_path": "",
296
+ "model_type": "siglip_vision_model",
297
+ "vision_use_head": false,
298
+ "attn_implementation": "sdpa",
299
+ "output_attentions": false
300
+ },
301
+ "mm_tokens_per_image": 256,
302
+ "boi_token_index": 255999,
303
+ "eoi_token_index": 256000,
304
+ "image_token_index": 262144,
305
+ "initializer_range": 0.02,
306
+ "tie_word_embeddings": true,
307
+ "_name_or_path": "",
308
+ "eos_token_id": [
309
+ 1,
310
+ 106
311
+ ],
312
+ "model_type": "gemma3",
313
+ "attn_implementation": {
314
+ "": "eager",
315
+ "text_config": "eager",
316
+ "vision_config": "sdpa"
317
+ },
318
+ "output_attentions": false
319
+ },
320
+ "action_config": {
321
+ "architectures": null,
322
+ "output_hidden_states": false,
323
+ "return_dict": true,
324
+ "dtype": null,
325
+ "chunk_size_feed_forward": 0,
326
+ "is_encoder_decoder": false,
327
+ "id2label": {
328
+ "0": "LABEL_0",
329
+ "1": "LABEL_1"
330
+ },
331
+ "label2id": {
332
+ "LABEL_0": 0,
333
+ "LABEL_1": 1
334
+ },
335
+ "problem_type": null,
336
+ "vocab_size": 262208,
337
+ "hidden_size": 1024,
338
+ "intermediate_size": 4096,
339
+ "num_hidden_layers": 34,
340
+ "num_attention_heads": 8,
341
+ "num_key_value_heads": 4,
342
+ "head_dim": 256,
343
+ "hidden_activation": "gelu_pytorch_tanh",
344
+ "max_position_embeddings": 131072,
345
+ "initializer_range": 0.02,
346
+ "rms_norm_eps": 1e-06,
347
+ "use_cache": true,
348
+ "pad_token_id": 0,
349
+ "eos_token_id": 1,
350
+ "bos_token_id": 2,
351
+ "tie_word_embeddings": true,
352
+ "rope_parameters": {
353
+ "full_attention": {
354
+ "factor": 8.0,
355
+ "rope_theta": 1000000.0,
356
+ "rope_type": "linear"
357
+ },
358
+ "sliding_attention": {
359
+ "rope_theta": 10000.0,
360
+ "rope_type": "default"
361
+ }
362
+ },
363
+ "attention_bias": false,
364
+ "attention_dropout": 0.0,
365
+ "query_pre_attn_scalar": 256,
366
+ "sliding_window": 4096,
367
+ "layer_types": [
368
+ "sliding_attention",
369
+ "sliding_attention",
370
+ "sliding_attention",
371
+ "sliding_attention",
372
+ "sliding_attention",
373
+ "full_attention",
374
+ "sliding_attention",
375
+ "sliding_attention",
376
+ "sliding_attention",
377
+ "sliding_attention",
378
+ "sliding_attention",
379
+ "full_attention",
380
+ "sliding_attention",
381
+ "sliding_attention",
382
+ "sliding_attention",
383
+ "sliding_attention",
384
+ "sliding_attention",
385
+ "full_attention",
386
+ "sliding_attention",
387
+ "sliding_attention",
388
+ "sliding_attention",
389
+ "sliding_attention",
390
+ "sliding_attention",
391
+ "full_attention",
392
+ "sliding_attention",
393
+ "sliding_attention",
394
+ "sliding_attention",
395
+ "sliding_attention",
396
+ "sliding_attention",
397
+ "full_attention",
398
+ "sliding_attention",
399
+ "sliding_attention",
400
+ "sliding_attention",
401
+ "sliding_attention"
402
+ ],
403
+ "final_logit_softcapping": null,
404
+ "attn_logit_softcapping": null,
405
+ "use_bidirectional_attention": false,
406
+ "_sliding_window_pattern": 6,
407
+ "_name_or_path": "",
408
+ "model_type": "gemma3_text",
409
+ "attn_implementation": "sdpa",
410
+ "action_attn_implementation": "sdpa",
411
+ "ae_gradient_checkpointing_layers": 1,
412
+ "output_attentions": false
413
+ },
414
+ "processor_config": null,
415
+ "action_dim": 32,
416
+ "chunk_size": 50,
417
+ "bf16": true,
418
+ "_name_or_path": ".",
419
+ "bos_token_id": 2,
420
+ "eos_token_id": 1,
421
+ "model_type": "dexbotic_dm05",
422
+ "pad_token_id": 0,
423
+ "tie_word_embeddings": true,
424
+ "gradient_checkpointing": true,
425
+ "output_attentions": false
426
+ },
427
+ "chunk_size": 50,
428
+ "n_action_steps": 50,
429
+ "drop_n_last_frames": 1,
430
+ "max_state_dim": 32,
431
+ "max_action_dim": 32,
432
+ "image_resolution": [
433
+ 448,
434
+ 448
435
+ ],
436
+ "empty_cameras": 0,
437
+ "add_state": true,
438
+ "pretrained_name_or_path": ".",
439
+ "processor_name_or_path": null,
440
+ "image_keys": null,
441
+ "n_bins": 256,
442
+ "tokenizer_max_length": 1024,
443
+ "diffusion_steps": 10,
444
+ "compile_model": false,
445
+ "compile_suffix_pad_length": 1024,
446
+ "llm_attn_implementation": "eager",
447
+ "vision_attn_implementation": "sdpa",
448
+ "action_attn_implementation": "sdpa",
449
+ "use_liger_kernel": false,
450
+ "dtype": "bfloat16",
451
+ "vlm_gradient_checkpointing": true,
452
+ "ae_gradient_checkpointing": true,
453
+ "ae_gradient_checkpointing_layers": 1,
454
+ "freeze_vlm_embedding": true,
455
+ "use_relative_actions": true,
456
+ "relative_exclude_joints": [
457
+ "gripper"
458
+ ],
459
+ "action_feature_names": [
460
+ "shoulder_pan.pos",
461
+ "shoulder_lift.pos",
462
+ "elbow_flex.pos",
463
+ "wrist_flex.pos",
464
+ "wrist_roll.pos",
465
+ "gripper.pos"
466
+ ],
467
+ "norm_clip": true,
468
+ "normalization_mapping": {
469
+ "VISUAL": "IDENTITY",
470
+ "STATE": "QUANTILES",
471
+ "ACTION": "QUANTILES"
472
+ },
473
+ "optimizer_lr": 2.5e-05,
474
+ "optimizer_betas": [
475
+ 0.9,
476
+ 0.95
477
+ ],
478
+ "optimizer_eps": 1e-08,
479
+ "optimizer_weight_decay": 1e-10,
480
+ "optimizer_grad_clip_norm": 1.0,
481
+ "scheduler_warmup_steps": 500,
482
+ "scheduler_decay_steps": 10000,
483
+ "scheduler_decay_lr": 2.5e-06
484
+ },
485
+ "reward_model": null,
486
+ "output_dir": "/tmp/train/dm05_omx_multicubes_rel",
487
+ "job_name": "dm05_omx_multicubes_rel",
488
+ "resume": false,
489
+ "seed": 1000,
490
+ "cudnn_deterministic": false,
491
+ "num_workers": 8,
492
+ "batch_size": 4,
493
+ "prefetch_factor": 4,
494
+ "persistent_workers": true,
495
+ "dataloader_multiprocessing_context": "spawn",
496
+ "steps": 10000,
497
+ "env_eval_freq": 20000,
498
+ "log_freq": 25,
499
+ "eval_steps": 0,
500
+ "max_eval_samples": 0,
501
+ "tolerance_s": 0.0001,
502
+ "save_checkpoint": true,
503
+ "save_freq": 5000,
504
+ "checkpoint_format": "safetensors",
505
+ "use_policy_training_preset": true,
506
+ "optimizer": {
507
+ "type": "adamw",
508
+ "lr": 2.5e-05,
509
+ "weight_decay": 1e-10,
510
+ "grad_clip_norm": 1.0,
511
+ "betas": [
512
+ 0.9,
513
+ 0.95
514
+ ],
515
+ "eps": 1e-08
516
+ },
517
+ "scheduler": {
518
+ "type": "cosine_decay_with_warmup",
519
+ "num_warmup_steps": 500,
520
+ "num_decay_steps": 10000,
521
+ "peak_lr": 2.5e-05,
522
+ "decay_lr": 2.5e-06
523
+ },
524
+ "parallelism": {
525
+ "dp_replicate": 1,
526
+ "dp_shard": 1,
527
+ "context_parallel": {
528
+ "ring_degree": 1,
529
+ "ulysses_degree": 1
530
+ },
531
+ "cfg_parallel": 1
532
+ },
533
+ "accelerator": {
534
+ "mixed_precision": "no",
535
+ "gradient_accumulation": {
536
+ "steps": 1
537
+ },
538
+ "fsdp": {
539
+ "reshard_after_forward": true,
540
+ "wrap_modules": null,
541
+ "min_num_params": null,
542
+ "cpu_offload": false,
543
+ "ignored_modules": null
544
+ },
545
+ "ddp": {
546
+ "find_unused_parameters": true,
547
+ "gradient_as_bucket_view": false,
548
+ "static_graph": false
549
+ },
550
+ "compile": {
551
+ "enabled": false,
552
+ "backend": "inductor",
553
+ "mode": null,
554
+ "regional": true
555
+ },
556
+ "activation_checkpointing": {
557
+ "mode": "none"
558
+ }
559
+ },
560
+ "eval": {
561
+ "n_episodes": 50,
562
+ "batch_size": 50,
563
+ "use_async_envs": true,
564
+ "recording": false,
565
+ "recording_repo_id": null,
566
+ "recording_private": false
567
+ },
568
+ "ema": {
569
+ "enable": false,
570
+ "decay": null,
571
+ "update_after_step": 0,
572
+ "inv_gamma": 1.0,
573
+ "power": 0.75,
574
+ "min_decay": 0.0,
575
+ "max_decay": 0.9999,
576
+ "use_for_eval": true
577
+ },
578
+ "wandb": {
579
+ "enable": true,
580
+ "disable_artifact": true,
581
+ "project": "lerobot",
582
+ "entity": null,
583
+ "notes": null,
584
+ "run_id": "qcwvqsuq",
585
+ "mode": null,
586
+ "add_tags": true
587
+ },
588
+ "peft": null,
589
+ "job": {
590
+ "target": null,
591
+ "image": "huggingface/lerobot-gpu:latest",
592
+ "timeout": "2d",
593
+ "detach": false,
594
+ "tags": []
595
+ },
596
+ "save_checkpoint_to_hub": true,
597
+ "sample_weighting": null,
598
+ "rename_map": {},
599
+ "checkpoint_path": null
600
+ }