Hugging Face's logo

Dexmal
/
DM05-Lerobot

Robotics
LeRobot
Safetensors
robot-control
vision-language-action
vla
dm05
dm0.5
opendm
Model card Files
xet
 Community
Dexmal2026 commited on
Commit
cf08a72
·
verified ·
1 Parent(s): 53a6aa4

initial commit

Browse files
Files changed (12) hide show
  1. .gitattributes +1 -0
  2. README.md +100 -0
  3. chat_template.jinja +47 -0
  4. config.json +389 -0
  5. model.safetensors +3 -0
  6. policy_postprocessor.json +38 -0
  7. policy_postprocessor_step_1_unnormalizer_processor.safetensors +3 -0
  8. policy_preprocessor.json +98 -0
  9. policy_preprocessor_step_5_normalizer_processor.safetensors +3 -0
  10. processor_config.json +29 -0
  11. tokenizer.json +3 -0
  12. tokenizer_config.json +28 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,3 +1,103 @@
1
  ---
2
  license: gemma
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: gemma
3
+ library_name: lerobot
4
+ pipeline_tag: robotics
5
+ base_model: Dexmal/DM05
6
+ tags:
7
+ - robotics
8
+ - robot-control
9
+ - vision-language-action
10
+ - vla
11
+ - lerobot
12
+ - dm05
13
+ - dm0.5
14
+ - opendm
15
  ---
16
+
17
+ # DM0.5 for LeRobot
18
+
19
+ ![DM0.5](https://raw.githubusercontent.com/dexmal/opendm/main/docs/image/header.png)
20
+
21
+ `Dexmal/DM05-Lerobot` is the LeRobot-format base checkpoint of
22
+ [DM0.5](https://huggingface.co/Dexmal/DM05), adapted from
23
+ [OpenDM](https://github.com/dexmal/OpenDM). It predicts continuous action chunks from images, robot state,
24
+ and language instructions.
25
+
26
+ This is a base model for supervised fine-tuning, not a LIBERO-, RoboTwin-, or robot-specific checkpoint.
27
+
28
+ ## Fine-tuning
29
+
30
+ ```bash
31
+ lerobot-train \
32
+ --dataset.repo_id=HuggingFaceVLA/libero \
33
+ --policy.path=Dexmal/DM05-Lerobot \
34
+ --policy.add_state=false \
35
+ --policy.chunk_size=10 \
36
+ --policy.n_action_steps=10 \
37
+ --policy.repo_id=your_repo_id \
38
+ --output_dir=outputs/train/dm05-libero \
39
+ --steps=50000 \
40
+ --batch_size=8 \
41
+ --policy.device=cuda
42
+ ```
43
+
44
+ For local-only training, replace `--policy.repo_id=...` with `--policy.push_to_hub=false`.
45
+
46
+ This LIBERO recipe matches OpenDM: it excludes state from the prompt and learns stored actions unchanged. Keep
47
+ `policy.add_state=true` unless the target recipe specifies otherwise. Policy action representation and environment
48
+ control mode are configured independently.
49
+
50
+ DM05 recommends complete state/action statistics in the dataset's `meta/stats.json`. If a non-standard local
51
+ dataset lacks them, generate the file before training:
52
+
53
+ ```bash
54
+ uv run python -m lerobot.policies.dm05.prepare_stats_dm05 \
55
+ --repo-id=org/dataset \
56
+ --root=/path/to/dataset \
57
+ --chunk-size=10 \
58
+ --drop-n-last-frames=1
59
+ ```
60
+
61
+ The command writes `meta/stats.json` in place. Match its chunk, episode selection, and action representation to
62
+ training. For relative targets, add `--use-relative-actions` and use `--force` when replacing absolute-action
63
+ statistics. Without target statistics, checkpoint statistics are reused with a warning and are valid only for the
64
+ same feature contract and distribution.
65
+
66
+ ## Evaluation
67
+
68
+ Evaluate a fine-tuned checkpoint:
69
+
70
+ ```bash
71
+ MUJOCO_GL=egl lerobot-eval \
72
+ --policy.path=/path/to/checkpoint/pretrained_model \
73
+ --env.type=libero \
74
+ --env.task=libero_spatial \
75
+ --env.control_mode=relative \
76
+ --policy.device=cuda
77
+ ```
78
+
79
+ ## Checkpoint contract
80
+
81
+ The base checkpoint uses OpenDM's 14-dimensional state/action schema. Fresh SFT takes its effective feature schema
82
+ and statistics from the target LeRobot dataset.
83
+
84
+ Load the complete checkpoint directory with `DM05Policy.from_pretrained()` or `--policy.path`; policy config,
85
+ tokenizer, preprocessing state, and weights are all required.
86
+
87
+ ## Resources
88
+
89
+ - [DM0.5 technical blog](https://www.dexmal.com/blog/dm0.5/index_en.html)
90
+ - [OpenDM repository](https://github.com/dexmal/OpenDM)
91
+ - [LeRobot repository](https://github.com/huggingface/lerobot)
92
+
93
+ ## Citation
94
+
95
+ ```bibtex
96
+ @misc{dm05,
97
+ title = {{DM0.5}: An Open-World Foundation Model for General-Purpose Embodied Intelligence},
98
+ author = {{Dexmal Team}},
99
+ month = {July},
100
+ year = {2026},
101
+ url = {https://www.dexmal.com/blog/dm0.5/index_en.html}
102
+ }
103
+ ```
chat_template.jinja ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {{ bos_token }}
2
+ {%- if messages[0]['role'] == 'system' -%}
3
+ {%- if messages[0]['content'] is string -%}
4
+ {%- set first_user_prefix = messages[0]['content'] + '
5
+
6
+ ' -%}
7
+ {%- else -%}
8
+ {%- set first_user_prefix = messages[0]['content'][0]['text'] + '
9
+
10
+ ' -%}
11
+ {%- endif -%}
12
+ {%- set loop_messages = messages[1:] -%}
13
+ {%- else -%}
14
+ {%- set first_user_prefix = "" -%}
15
+ {%- set loop_messages = messages -%}
16
+ {%- endif -%}
17
+ {%- for message in loop_messages -%}
18
+ {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}
19
+ {{ raise_exception("Conversation roles must alternate user/assistant/user/assistant/...") }}
20
+ {%- endif -%}
21
+ {%- if (message['role'] == 'assistant') -%}
22
+ {%- set role = "model" -%}
23
+ {%- else -%}
24
+ {%- set role = message['role'] -%}
25
+ {%- endif -%}
26
+ {{ '<start_of_turn>' + role + '
27
+ ' + (first_user_prefix if loop.first else "") }}
28
+ {%- if message['content'] is string -%}
29
+ {{ message['content'] | trim }}
30
+ {%- elif message['content'] is iterable -%}
31
+ {%- for item in message['content'] -%}
32
+ {%- if item['type'] == 'image' -%}
33
+ {{ '<start_of_image>' }}
34
+ {%- elif item['type'] == 'text' -%}
35
+ {{ item['text'] | trim }}
36
+ {%- endif -%}
37
+ {%- endfor -%}
38
+ {%- else -%}
39
+ {{ raise_exception("Invalid content type") }}
40
+ {%- endif -%}
41
+ {{ '<end_of_turn>
42
+ ' }}
43
+ {%- endfor -%}
44
+ {%- if add_generation_prompt -%}
45
+ {{'<start_of_turn>model
46
+ '}}
47
+ {%- endif -%}
config.json ADDED
@@ -0,0 +1,389 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "type": "dm05",
3
+ "n_obs_steps": 1,
4
+ "input_features": {
5
+ "observation.state": {
6
+ "type": "STATE",
7
+ "shape": [
8
+ 14
9
+ ]
10
+ }
11
+ },
12
+ "output_features": {
13
+ "action": {
14
+ "type": "ACTION",
15
+ "shape": [
16
+ 14
17
+ ]
18
+ }
19
+ },
20
+ "device": "cpu",
21
+ "use_amp": false,
22
+ "use_peft": false,
23
+ "push_to_hub": true,
24
+ "repo_id": null,
25
+ "private": null,
26
+ "tags": null,
27
+ "license": "gemma",
28
+ "pretrained_path": null,
29
+ "pretrained_revision": null,
30
+ "core_config": {
31
+ "vlm_gradient_checkpointing": true,
32
+ "ae_gradient_checkpointing": true,
33
+ "ae_gradient_checkpointing_layers": 1,
34
+ "freeze_vlm_embedding": false,
35
+ "llm_attn_implementation": "eager",
36
+ "vision_attn_implementation": "sdpa",
37
+ "action_attn_implementation": "sdpa",
38
+ "transformers_version": "5.5.4",
39
+ "architectures": [
40
+ "DM05ForConditionalGeneration"
41
+ ],
42
+ "output_hidden_states": false,
43
+ "return_dict": true,
44
+ "dtype": "bfloat16",
45
+ "chunk_size_feed_forward": 0,
46
+ "is_encoder_decoder": false,
47
+ "id2label": {
48
+ "0": "LABEL_0",
49
+ "1": "LABEL_1"
50
+ },
51
+ "label2id": {
52
+ "LABEL_0": 0,
53
+ "LABEL_1": 1
54
+ },
55
+ "problem_type": null,
56
+ "vlm_config": {
57
+ "architectures": [
58
+ "Gemma3ForConditionalGeneration"
59
+ ],
60
+ "output_hidden_states": false,
61
+ "return_dict": true,
62
+ "dtype": "bfloat16",
63
+ "chunk_size_feed_forward": 0,
64
+ "is_encoder_decoder": false,
65
+ "id2label": {
66
+ "0": "LABEL_0",
67
+ "1": "LABEL_1"
68
+ },
69
+ "label2id": {
70
+ "LABEL_0": 0,
71
+ "LABEL_1": 1
72
+ },
73
+ "problem_type": null,
74
+ "text_config": {
75
+ "architectures": null,
76
+ "output_hidden_states": false,
77
+ "return_dict": true,
78
+ "dtype": "bfloat16",
79
+ "chunk_size_feed_forward": 0,
80
+ "is_encoder_decoder": false,
81
+ "id2label": {
82
+ "0": "LABEL_0",
83
+ "1": "LABEL_1"
84
+ },
85
+ "label2id": {
86
+ "LABEL_0": 0,
87
+ "LABEL_1": 1
88
+ },
89
+ "problem_type": null,
90
+ "vocab_size": 262208,
91
+ "hidden_size": 2560,
92
+ "intermediate_size": 10240,
93
+ "num_hidden_layers": 34,
94
+ "num_attention_heads": 8,
95
+ "num_key_value_heads": 4,
96
+ "head_dim": 256,
97
+ "hidden_activation": "gelu_pytorch_tanh",
98
+ "max_position_embeddings": 131072,
99
+ "initializer_range": 0.02,
100
+ "rms_norm_eps": 1e-06,
101
+ "use_cache": true,
102
+ "pad_token_id": 0,
103
+ "eos_token_id": 1,
104
+ "bos_token_id": 2,
105
+ "tie_word_embeddings": true,
106
+ "rope_parameters": {
107
+ "full_attention": {
108
+ "factor": 8.0,
109
+ "rope_theta": 1000000.0,
110
+ "rope_type": "linear"
111
+ },
112
+ "sliding_attention": {
113
+ "rope_theta": 10000.0,
114
+ "rope_type": "default"
115
+ }
116
+ },
117
+ "attention_bias": false,
118
+ "attention_dropout": 0.0,
119
+ "query_pre_attn_scalar": 256,
120
+ "sliding_window": 4096,
121
+ "layer_types": [
122
+ "sliding_attention",
123
+ "sliding_attention",
124
+ "sliding_attention",
125
+ "sliding_attention",
126
+ "sliding_attention",
127
+ "full_attention",
128
+ "sliding_attention",
129
+ "sliding_attention",
130
+ "sliding_attention",
131
+ "sliding_attention",
132
+ "sliding_attention",
133
+ "full_attention",
134
+ "sliding_attention",
135
+ "sliding_attention",
136
+ "sliding_attention",
137
+ "sliding_attention",
138
+ "sliding_attention",
139
+ "full_attention",
140
+ "sliding_attention",
141
+ "sliding_attention",
142
+ "sliding_attention",
143
+ "sliding_attention",
144
+ "sliding_attention",
145
+ "full_attention",
146
+ "sliding_attention",
147
+ "sliding_attention",
148
+ "sliding_attention",
149
+ "sliding_attention",
150
+ "sliding_attention",
151
+ "full_attention",
152
+ "sliding_attention",
153
+ "sliding_attention",
154
+ "sliding_attention",
155
+ "sliding_attention"
156
+ ],
157
+ "final_logit_softcapping": null,
158
+ "attn_logit_softcapping": null,
159
+ "use_bidirectional_attention": false,
160
+ "_sliding_window_pattern": 6,
161
+ "_name_or_path": "",
162
+ "model_type": "gemma3_text",
163
+ "attn_implementation": "eager",
164
+ "output_attentions": false
165
+ },
166
+ "vision_config": {
167
+ "architectures": null,
168
+ "output_hidden_states": false,
169
+ "return_dict": true,
170
+ "dtype": "bfloat16",
171
+ "chunk_size_feed_forward": 0,
172
+ "is_encoder_decoder": false,
173
+ "id2label": {
174
+ "0": "LABEL_0",
175
+ "1": "LABEL_1"
176
+ },
177
+ "label2id": {
178
+ "LABEL_0": 0,
179
+ "LABEL_1": 1
180
+ },
181
+ "problem_type": null,
182
+ "hidden_size": 1152,
183
+ "intermediate_size": 4304,
184
+ "num_hidden_layers": 27,
185
+ "num_attention_heads": 16,
186
+ "num_channels": 3,
187
+ "image_size": 448,
188
+ "patch_size": 14,
189
+ "hidden_act": "gelu_pytorch_tanh",
190
+ "layer_norm_eps": 1e-06,
191
+ "attention_dropout": 0.0,
192
+ "_name_or_path": "",
193
+ "model_type": "siglip_vision_model",
194
+ "vision_use_head": false,
195
+ "attn_implementation": "sdpa",
196
+ "output_attentions": false
197
+ },
198
+ "mm_tokens_per_image": 256,
199
+ "boi_token_index": 255999,
200
+ "eoi_token_index": 256000,
201
+ "image_token_index": 262144,
202
+ "initializer_range": 0.02,
203
+ "tie_word_embeddings": true,
204
+ "_name_or_path": "",
205
+ "eos_token_id": [
206
+ 1,
207
+ 106
208
+ ],
209
+ "model_type": "gemma3",
210
+ "attn_implementation": {
211
+ "": "eager",
212
+ "text_config": "eager",
213
+ "vision_config": "sdpa"
214
+ },
215
+ "output_attentions": false
216
+ },
217
+ "action_config": {
218
+ "architectures": null,
219
+ "output_hidden_states": false,
220
+ "return_dict": true,
221
+ "dtype": null,
222
+ "chunk_size_feed_forward": 0,
223
+ "is_encoder_decoder": false,
224
+ "id2label": {
225
+ "0": "LABEL_0",
226
+ "1": "LABEL_1"
227
+ },
228
+ "label2id": {
229
+ "LABEL_0": 0,
230
+ "LABEL_1": 1
231
+ },
232
+ "problem_type": null,
233
+ "vocab_size": 262208,
234
+ "hidden_size": 1024,
235
+ "intermediate_size": 4096,
236
+ "num_hidden_layers": 34,
237
+ "num_attention_heads": 8,
238
+ "num_key_value_heads": 4,
239
+ "head_dim": 256,
240
+ "hidden_activation": "gelu_pytorch_tanh",
241
+ "max_position_embeddings": 131072,
242
+ "initializer_range": 0.02,
243
+ "rms_norm_eps": 1e-06,
244
+ "use_cache": true,
245
+ "pad_token_id": 0,
246
+ "eos_token_id": 1,
247
+ "bos_token_id": 2,
248
+ "tie_word_embeddings": true,
249
+ "rope_parameters": {
250
+ "full_attention": {
251
+ "factor": 8.0,
252
+ "rope_theta": 1000000.0,
253
+ "rope_type": "linear"
254
+ },
255
+ "sliding_attention": {
256
+ "rope_theta": 10000.0,
257
+ "rope_type": "default"
258
+ }
259
+ },
260
+ "attention_bias": false,
261
+ "attention_dropout": 0.0,
262
+ "query_pre_attn_scalar": 256,
263
+ "sliding_window": 4096,
264
+ "layer_types": [
265
+ "sliding_attention",
266
+ "sliding_attention",
267
+ "sliding_attention",
268
+ "sliding_attention",
269
+ "sliding_attention",
270
+ "full_attention",
271
+ "sliding_attention",
272
+ "sliding_attention",
273
+ "sliding_attention",
274
+ "sliding_attention",
275
+ "sliding_attention",
276
+ "full_attention",
277
+ "sliding_attention",
278
+ "sliding_attention",
279
+ "sliding_attention",
280
+ "sliding_attention",
281
+ "sliding_attention",
282
+ "full_attention",
283
+ "sliding_attention",
284
+ "sliding_attention",
285
+ "sliding_attention",
286
+ "sliding_attention",
287
+ "sliding_attention",
288
+ "full_attention",
289
+ "sliding_attention",
290
+ "sliding_attention",
291
+ "sliding_attention",
292
+ "sliding_attention",
293
+ "sliding_attention",
294
+ "full_attention",
295
+ "sliding_attention",
296
+ "sliding_attention",
297
+ "sliding_attention",
298
+ "sliding_attention"
299
+ ],
300
+ "final_logit_softcapping": null,
301
+ "attn_logit_softcapping": null,
302
+ "use_bidirectional_attention": false,
303
+ "_sliding_window_pattern": 6,
304
+ "_name_or_path": "",
305
+ "model_type": "gemma3_text",
306
+ "attn_implementation": "sdpa",
307
+ "action_attn_implementation": "sdpa",
308
+ "ae_gradient_checkpointing_layers": 1,
309
+ "output_attentions": false
310
+ },
311
+ "processor_config": null,
312
+ "action_dim": 32,
313
+ "chunk_size": 50,
314
+ "bf16": true,
315
+ "_name_or_path": ".",
316
+ "bos_token_id": 2,
317
+ "eos_token_id": 1,
318
+ "model_type": "dexbotic_dm05",
319
+ "pad_token_id": 0,
320
+ "tie_word_embeddings": true,
321
+ "gradient_checkpointing": true,
322
+ "output_attentions": false
323
+ },
324
+ "chunk_size": 50,
325
+ "n_action_steps": 50,
326
+ "drop_n_last_frames": 1,
327
+ "max_state_dim": 32,
328
+ "max_action_dim": 32,
329
+ "image_resolution": [
330
+ 448,
331
+ 448
332
+ ],
333
+ "empty_cameras": 0,
334
+ "add_state": true,
335
+ "pretrained_name_or_path": ".",
336
+ "processor_name_or_path": ".",
337
+ "image_keys": null,
338
+ "n_bins": 256,
339
+ "tokenizer_max_length": 1024,
340
+ "diffusion_steps": 10,
341
+ "compile_model": false,
342
+ "compile_suffix_pad_length": 1024,
343
+ "llm_attn_implementation": "eager",
344
+ "vision_attn_implementation": "sdpa",
345
+ "action_attn_implementation": "sdpa",
346
+ "use_liger_kernel": false,
347
+ "dtype": "bfloat16",
348
+ "vlm_gradient_checkpointing": true,
349
+ "ae_gradient_checkpointing": true,
350
+ "ae_gradient_checkpointing_layers": 1,
351
+ "freeze_vlm_embedding": true,
352
+ "use_relative_actions": false,
353
+ "relative_exclude_joints": [
354
+ "gripper"
355
+ ],
356
+ "action_feature_names": [
357
+ "joint_0",
358
+ "joint_1",
359
+ "joint_2",
360
+ "joint_3",
361
+ "joint_4",
362
+ "joint_5",
363
+ "gripper_6",
364
+ "joint_7",
365
+ "joint_8",
366
+ "joint_9",
367
+ "joint_10",
368
+ "joint_11",
369
+ "joint_12",
370
+ "gripper_13"
371
+ ],
372
+ "norm_clip": true,
373
+ "normalization_mapping": {
374
+ "VISUAL": "IDENTITY",
375
+ "STATE": "QUANTILES",
376
+ "ACTION": "QUANTILES"
377
+ },
378
+ "optimizer_lr": 2.5e-05,
379
+ "optimizer_betas": [
380
+ 0.9,
381
+ 0.95
382
+ ],
383
+ "optimizer_eps": 1e-08,
384
+ "optimizer_weight_decay": 1e-10,
385
+ "optimizer_grad_clip_norm": 1.0,
386
+ "scheduler_warmup_steps": 1000,
387
+ "scheduler_decay_steps": 50000,
388
+ "scheduler_decay_lr": 2.5e-06
389
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a5f81e751ecb52c018f634ffd3c6a61a420f70f1faa09eadecfb3bb5b1c8226
3
+ size 10315934288
policy_postprocessor.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_postprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "device_processor",
6
+ "config": {
7
+ "device": "cpu",
8
+ "float_dtype": "float32"
9
+ }
10
+ },
11
+ {
12
+ "registry_name": "unnormalizer_processor",
13
+ "config": {
14
+ "eps": 1e-06,
15
+ "features": {
16
+ "action": {
17
+ "type": "ACTION",
18
+ "shape": [
19
+ 14
20
+ ]
21
+ }
22
+ },
23
+ "norm_map": {
24
+ "VISUAL": "IDENTITY",
25
+ "STATE": "QUANTILES",
26
+ "ACTION": "QUANTILES"
27
+ }
28
+ },
29
+ "state_file": "policy_postprocessor_step_1_unnormalizer_processor.safetensors"
30
+ },
31
+ {
32
+ "registry_name": "absolute_actions_processor",
33
+ "config": {
34
+ "enabled": false
35
+ }
36
+ }
37
+ ]
38
+ }
policy_postprocessor_step_1_unnormalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:553b9d4453e67e79632ec99a65940bbe2f23d267783726b82e0b34ff841c72f9
3
+ size 1040
policy_preprocessor.json ADDED
@@ -0,0 +1,98 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "name": "policy_preprocessor",
3
+ "steps": [
4
+ {
5
+ "registry_name": "rename_observations_processor",
6
+ "config": {
7
+ "rename_map": {}
8
+ }
9
+ },
10
+ {
11
+ "registry_name": "to_batch_processor",
12
+ "config": {}
13
+ },
14
+ {
15
+ "registry_name": "dm05_task_processor",
16
+ "config": {
17
+ "default_task": "Execute the robot action."
18
+ }
19
+ },
20
+ {
21
+ "registry_name": "dm05_action_reference_probe_processor",
22
+ "config": {
23
+ "action_dim": 14
24
+ }
25
+ },
26
+ {
27
+ "registry_name": "relative_actions_processor",
28
+ "config": {
29
+ "enabled": false,
30
+ "exclude_joints": [
31
+ "gripper"
32
+ ],
33
+ "action_names": [
34
+ "joint_0",
35
+ "joint_1",
36
+ "joint_2",
37
+ "joint_3",
38
+ "joint_4",
39
+ "joint_5",
40
+ "gripper_6",
41
+ "joint_7",
42
+ "joint_8",
43
+ "joint_9",
44
+ "joint_10",
45
+ "joint_11",
46
+ "joint_12",
47
+ "gripper_13"
48
+ ]
49
+ }
50
+ },
51
+ {
52
+ "registry_name": "normalizer_processor",
53
+ "config": {
54
+ "eps": 1e-06,
55
+ "features": {
56
+ "observation.state": {
57
+ "type": "STATE",
58
+ "shape": [
59
+ 14
60
+ ]
61
+ },
62
+ "action": {
63
+ "type": "ACTION",
64
+ "shape": [
65
+ 14
66
+ ]
67
+ }
68
+ },
69
+ "norm_map": {
70
+ "VISUAL": "IDENTITY",
71
+ "STATE": "QUANTILES",
72
+ "ACTION": "QUANTILES"
73
+ },
74
+ "normalize_observation_keys": [
75
+ "observation.state"
76
+ ]
77
+ },
78
+ "state_file": "policy_preprocessor_step_5_normalizer_processor.safetensors"
79
+ },
80
+ {
81
+ "registry_name": "dm05_action_reference_extract_processor",
82
+ "config": {}
83
+ },
84
+ {
85
+ "registry_name": "dm05_images_to_pil_processor",
86
+ "config": {
87
+ "image_keys": null
88
+ }
89
+ },
90
+ {
91
+ "registry_name": "device_processor",
92
+ "config": {
93
+ "device": "cpu",
94
+ "float_dtype": null
95
+ }
96
+ }
97
+ ]
98
+ }
policy_preprocessor_step_5_normalizer_processor.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:553b9d4453e67e79632ec99a65940bbe2f23d267783726b82e0b34ff841c72f9
3
+ size 1040
processor_config.json ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_processor": {
3
+ "data_format": "channels_first",
4
+ "do_convert_rgb": null,
5
+ "do_normalize": true,
6
+ "do_rescale": true,
7
+ "do_resize": true,
8
+ "image_mean": [
9
+ 0.5,
10
+ 0.5,
11
+ 0.5
12
+ ],
13
+ "image_processor_type": "Gemma3ImageProcessor",
14
+ "image_seq_length": 256,
15
+ "image_std": [
16
+ 0.5,
17
+ 0.5,
18
+ 0.5
19
+ ],
20
+ "resample": 2,
21
+ "rescale_factor": 0.00392156862745098,
22
+ "size": {
23
+ "height": 448,
24
+ "width": 448
25
+ }
26
+ },
27
+ "image_seq_length": 256,
28
+ "processor_class": "Gemma3Processor"
29
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:daab2354f8a74e70d70b4d1f804939b68a8c9624dd06cb7858e52dd8970e9726
3
+ size 33384567
tokenizer_config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "boi_token": "<start_of_image>",
4
+ "boi_token_id": 255999,
5
+ "bos_token": "<bos>",
6
+ "chat_template": "{{ bos_token }}\n{%- if messages[0]['role'] == 'system' -%}\n {%- if messages[0]['content'] is string -%}\n {%- set first_user_prefix = messages[0]['content'] + '\n\n' -%}\n {%- else -%}\n {%- set first_user_prefix = messages[0]['content'][0]['text'] + '\n\n' -%}\n {%- endif -%}\n {%- set loop_messages = messages[1:] -%}\n{%- else -%}\n {%- set first_user_prefix = \"\" -%}\n {%- set loop_messages = messages -%}\n{%- endif -%}\n{%- for message in loop_messages -%}\n {%- if (message['role'] == 'user') != (loop.index0 % 2 == 0) -%}\n {{ raise_exception(\"Conversation roles must alternate user/assistant/user/assistant/...\") }}\n {%- endif -%}\n {%- if (message['role'] == 'assistant') -%}\n {%- set role = \"model\" -%}\n {%- else -%}\n {%- set role = message['role'] -%}\n {%- endif -%}\n {{ '<start_of_turn>' + role + '\n' + (first_user_prefix if loop.first else \"\") }}\n {%- if message['content'] is string -%}\n {{ message['content'] | trim }}\n {%- elif message['content'] is iterable -%}\n {%- for item in message['content'] -%}\n {%- if item['type'] == 'image' -%}\n {{ '<start_of_image>' }}\n {%- elif item['type'] == 'text' -%}\n {{ item['text'] | trim }}\n {%- endif -%}\n {%- endfor -%}\n {%- else -%}\n {{ raise_exception(\"Invalid content type\") }}\n {%- endif -%}\n {{ '<end_of_turn>\n' }}\n{%- endfor -%}\n{%- if add_generation_prompt -%}\n {{'<start_of_turn>model\n'}}\n{%- endif -%}\n",
7
+ "clean_up_tokenization_spaces": false,
8
+ "eoi_token": "<end_of_image>",
9
+ "eoi_token_id": 256000,
10
+ "eos_token": "<eos>",
11
+ "image_token": "<image_soft_token>",
12
+ "image_token_id": 262144,
13
+ "is_local": true,
14
+ "mask_token": "<mask>",
15
+ "model_max_length": 1000000000000000019884624838656,
16
+ "model_specific_special_tokens": {
17
+ "boi_token": "<start_of_image>",
18
+ "eoi_token": "<end_of_image>",
19
+ "image_token": "<image_soft_token>"
20
+ },
21
+ "pad_token": "<pad>",
22
+ "processor_class": "Gemma3Processor",
23
+ "sp_model_kwargs": null,
24
+ "spaces_between_special_tokens": false,
25
+ "tokenizer_class": "GemmaTokenizer",
26
+ "unk_token": "<unk>",
27
+ "use_default_system_prompt": false
28
+ }