Whalswp commited on
Commit
318f1b0
·
verified ·
1 Parent(s): f1a37d5

Add files using upload-large-folder tool

Browse files
Files changed (50) hide show
  1. MGRKD/.launches/mgrkd_da_cls_transformer_bs128_gpu5_20260622_160217/command.txt +1 -0
  2. MGRKD/.launches/mgrkd_da_cls_transformer_bs128_gpu5_20260622_160217/pid +1 -0
  3. MGRKD/.launches/mgrkd_da_cls_transformer_bs128_gpu5_20260622_160217/train.log +0 -0
  4. MGRKD/launch_logs/mgrkd_raw_action_flatten_frozen_mlp_gpu5_bs128_20260622_165623.log +0 -0
  5. MGRKD/logs/mgrkd_da_flatten_action_encoder_gpu4_bs128_20260622_134555.log +0 -0
  6. MGRKD/mgrkd_distance_angle_cls_transformer_action_encoder/default/mgrkd_da_cls_transformer.yaml +67 -0
  7. MGRKD/mgrkd_distance_angle_cls_transformer_action_encoder/default/phase2/experiment_cfg/metadata.json +431 -0
  8. MGRKD/mgrkd_distance_angle_cls_transformer_action_encoder/default/resolved_config.yaml +68 -0
  9. MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/mgrkd_da_flatten_action_encoder.yaml +61 -0
  10. MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase2/config.json +84 -0
  11. MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase2/experiment_cfg/metadata.json +431 -0
  12. MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase2/model.safetensors.index.json +0 -0
  13. MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase2/trainer_state.json +0 -0
  14. MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase3/experiment_cfg/metadata.json +431 -0
  15. MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/resolved_config.yaml +62 -0
  16. MGRKD/mgrkd_distance_angle_flatten_raw_action/default/mgrkd_da_flatten_raw_action.yaml +61 -0
  17. MGRKD/mgrkd_distance_angle_flatten_raw_action/default/phase2/config.json +90 -0
  18. MGRKD/mgrkd_distance_angle_flatten_raw_action/default/phase2/experiment_cfg/metadata.json +431 -0
  19. MGRKD/mgrkd_distance_angle_flatten_raw_action/default/phase2/model.safetensors.index.json +0 -0
  20. MGRKD/mgrkd_distance_angle_flatten_raw_action/default/phase3/experiment_cfg/metadata.json +431 -0
  21. MGRKD/mgrkd_distance_angle_flatten_raw_action/default/resolved_config.yaml +62 -0
  22. rkd_v2_1/rkd_v2_1_distance_angle_full_120k_cosine_decay/default/full_fm_rkd_da_cosine_decay_120k/checkpoint-120000/experiment_cfg/metadata.json +431 -0
  23. rkd_v2_1/rkd_v2_1_distance_angle_full_120k_cosine_decay/default/full_fm_rkd_da_cosine_decay_120k/checkpoint-120000/model.safetensors.index.json +0 -0
  24. rkd_v2_1/rkd_v2_1_distance_angle_full_120k_cosine_decay/default/full_fm_rkd_da_cosine_decay_120k/config.json +77 -0
  25. rkd_v2_1/rkd_v2_1_distance_angle_full_120k_cosine_decay/default/full_fm_rkd_da_cosine_decay_120k/model.safetensors.index.json +0 -0
  26. rkd_v2_2/launch_logs/action_encoder_flatten_gpu4_5_bs32_ngpu2_20260622_124039.log +757 -0
  27. rkd_v2_2/launch_logs/action_encoder_flatten_gpu4_5_bs64_ngpu2.pid +1 -0
  28. rkd_v2_2/launch_logs/action_encoder_flatten_gpu4_5_bs64_ngpu2_20260622_123748.log +865 -0
  29. rkd_v2_2/launch_logs/action_encoder_gpu4.pid +1 -0
  30. rkd_v2_2/launch_logs/action_encoder_gpu4_bs128_20260622_123336.log +351 -0
  31. rkd_v2_2/launch_logs/raw_action_gpu5.pid +1 -0
  32. rkd_v2_2/launch_logs/raw_action_gpu5_bs128_20260622_123336.log +351 -0
  33. rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu45_bs32_20260622_131131.log +764 -0
  34. rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu45_bs32_20260622_131131.log.pid +1 -0
  35. rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu45_bs64_20260622_130844.log +871 -0
  36. rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu45_bs64_20260622_130844.log.pid +1 -0
  37. rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu4_bs128_20260622_130630.log +355 -0
  38. rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu4_bs128_20260622_130630.log.pid +1 -0
  39. rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu45_bs32_20260622_132220.log +763 -0
  40. rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu45_bs32_20260622_132220.log.pid +1 -0
  41. rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu45_bs64_20260622_131850.log +871 -0
  42. rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu45_bs64_20260622_131850.log.pid +1 -0
  43. rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu4_bs128_20260622_131539.log +355 -0
  44. rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu4_bs128_20260622_131539.log.pid +1 -0
  45. rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/experiment_cfg/metadata.json +431 -0
  46. rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/resolved_config.yaml +56 -0
  47. rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/rkd_v2.2_da_flatten_action_encoder.yaml +55 -0
  48. rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2/experiment_cfg/metadata.json +431 -0
  49. rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/resolved_config.yaml +52 -0
  50. rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/rkd_v2.2_da_flatten_raw_action.yaml +51 -0
MGRKD/.launches/mgrkd_da_cls_transformer_bs128_gpu5_20260622_160217/command.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ CUDA_VISIBLE_DEVICES=5 TRANSFORMERS_VIDEO_BACKEND=av /home/ext_minje/miniforge3/envs/robocasa/bin/python my_scripts/gr00t_finetune.py --config /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/MGRKD/mgrkd_da_cls_transformer.yaml --num-gpus 1 --batch-size 128
MGRKD/.launches/mgrkd_da_cls_transformer_bs128_gpu5_20260622_160217/pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2874182
MGRKD/.launches/mgrkd_da_cls_transformer_bs128_gpu5_20260622_160217/train.log ADDED
The diff for this file is too large to render. See raw diff
 
MGRKD/launch_logs/mgrkd_raw_action_flatten_frozen_mlp_gpu5_bs128_20260622_165623.log ADDED
The diff for this file is too large to render. See raw diff
 
MGRKD/logs/mgrkd_da_flatten_action_encoder_gpu4_bs128_20260622_134555.log ADDED
The diff for this file is too large to render. See raw diff
 
MGRKD/mgrkd_distance_angle_cls_transformer_action_encoder/default/mgrkd_da_cls_transformer.yaml ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: mgrkd_distance_angle_cls_transformer_action_encoder_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_MGRKD
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/MGRKD/mgrkd_distance_angle_cls_transformer_action_encoder
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 1
9
+ batch_size: 128
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_mgrkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: action_encoder
29
+ rkd_action_encoder_projector_enabled: true
30
+ rkd_action_encoder_projector_dim: 512
31
+ rkd_action_encoder_projector_pooling: cls_transformer
32
+ rkd_action_encoder_cls_num_heads: 8
33
+ rkd_action_encoder_cls_num_layers: 1
34
+ rkd_action_encoder_cls_ffn_hidden_dim: 2048
35
+ rkd_student_reconstruction_head_enabled: true
36
+ rkd_student_reconstruction_head_dim: 512
37
+ rkd_student_reconstruction_head_hidden_dim: 512
38
+ rkd_distance_loss_weight: 1.0
39
+ rkd_angle_loss_weight: 2.0
40
+ rkd_exclude_diagonal: true
41
+ - name: phase3_fm_mgrkd_da_fixed_0p5
42
+ max_steps: 30000
43
+ save_steps: 0
44
+ trainable:
45
+ tune_llm: false
46
+ tune_visual: false
47
+ tune_projector: true
48
+ tune_diffusion_model: true
49
+ losses:
50
+ rkd_enabled: true
51
+ rkd_fm_loss_weight: 1.0
52
+ rkd_loss_weight: 0.5
53
+ rkd_relation_mode: flatten
54
+ rkd_loss_type: distance_angle
55
+ rkd_teacher_source: action_encoder
56
+ rkd_action_encoder_projector_enabled: true
57
+ rkd_action_encoder_projector_dim: 512
58
+ rkd_action_encoder_projector_pooling: cls_transformer
59
+ rkd_action_encoder_cls_num_heads: 8
60
+ rkd_action_encoder_cls_num_layers: 1
61
+ rkd_action_encoder_cls_ffn_hidden_dim: 2048
62
+ rkd_student_reconstruction_head_enabled: true
63
+ rkd_student_reconstruction_head_dim: 512
64
+ rkd_student_reconstruction_head_hidden_dim: 512
65
+ rkd_distance_loss_weight: 1.0
66
+ rkd_angle_loss_weight: 2.0
67
+ rkd_exclude_diagonal: true
MGRKD/mgrkd_distance_angle_cls_transformer_action_encoder/default/phase2/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,431 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "base_position": {
6
+ "max": [
7
+ 7.3139495849609375,
8
+ 0.4876587688922882,
9
+ 0.7196521759033203
10
+ ],
11
+ "min": [
12
+ -4.94293737411499,
13
+ -6.890198230743408,
14
+ 0.6996827125549316
15
+ ],
16
+ "mean": [
17
+ 2.749288365724937,
18
+ -2.0455216239909983,
19
+ 0.700532551020533
20
+ ],
21
+ "std": [
22
+ 1.6786295897046262,
23
+ 1.312897969434244,
24
+ 0.00133751758906258
25
+ ],
26
+ "q01": [
27
+ -1.4241907881148037,
28
+ -5.1716547935950885,
29
+ 0.6999670898768614
30
+ ],
31
+ "q99": [
32
+ 6.198111545831555,
33
+ -0.5873531334104489,
34
+ 0.7038833014242587
35
+ ]
36
+ },
37
+ "base_rotation": {
38
+ "max": [
39
+ 0.0,
40
+ 0.0,
41
+ 1.0,
42
+ 1.0
43
+ ],
44
+ "min": [
45
+ 0.0,
46
+ 0.0,
47
+ -1.0,
48
+ 0.0
49
+ ],
50
+ "mean": [
51
+ 0.0,
52
+ 0.0,
53
+ 0.23084999587450247,
54
+ 0.6238431088581847
55
+ ],
56
+ "std": [
57
+ 0.0,
58
+ 0.0,
59
+ 0.6556404371644047,
60
+ 0.3572252105732042
61
+ ],
62
+ "q01": [
63
+ 0.0,
64
+ 0.0,
65
+ -0.9999999999999999,
66
+ 1.7482354593273349e-06
67
+ ],
68
+ "q99": [
69
+ 0.0,
70
+ 0.0,
71
+ 0.9999999999999999,
72
+ 0.9999999999999999
73
+ ]
74
+ },
75
+ "end_effector_position_relative": {
76
+ "max": [
77
+ 0.9014528393745422,
78
+ 0.8003877401351929,
79
+ 0.9829942584037781
80
+ ],
81
+ "min": [
82
+ -0.37471598386764526,
83
+ -0.8472502827644348,
84
+ -0.25070279836654663
85
+ ],
86
+ "mean": [
87
+ 0.28667332601863704,
88
+ -0.038315473170876704,
89
+ 0.45974658906777444
90
+ ],
91
+ "std": [
92
+ 0.17067058828305154,
93
+ 0.23569952037784195,
94
+ 0.21950702354181487
95
+ ],
96
+ "q01": [
97
+ 0.015172948963784228,
98
+ -0.44450502063220615,
99
+ 0.23314444103329338
100
+ ],
101
+ "q99": [
102
+ 0.5609069689572412,
103
+ 0.38454179128009547,
104
+ 0.7028102736473852
105
+ ]
106
+ },
107
+ "end_effector_rotation_relative": {
108
+ "max": [
109
+ 0.9999998807907104,
110
+ 0.9984455108642578,
111
+ 0.9434727430343628,
112
+ 0.9062229990959167
113
+ ],
114
+ "min": [
115
+ -0.9999930262565613,
116
+ -0.9988337755203247,
117
+ -0.9618149995803833,
118
+ 2.1872274658107926e-07
119
+ ],
120
+ "mean": [
121
+ -0.25672297401336,
122
+ 0.024425335806687216,
123
+ -0.08740977164483847,
124
+ 0.16608739644018397
125
+ ],
126
+ "std": [
127
+ 0.7932585208392383,
128
+ 0.3102260195580416,
129
+ 0.3772472576315148,
130
+ 0.17451959300158773
131
+ ],
132
+ "q01": [
133
+ -0.99379006987882,
134
+ -0.5478102050038828,
135
+ -0.6101777952971636,
136
+ 0.002274085014134748
137
+ ],
138
+ "q99": [
139
+ 0.8804856038896288,
140
+ 0.5910611092873037,
141
+ 0.5216058042993562,
142
+ 0.5231965974012255
143
+ ]
144
+ },
145
+ "gripper_qpos": {
146
+ "max": [
147
+ 0.055869169533252716,
148
+ 0.010916369967162609
149
+ ],
150
+ "min": [
151
+ -0.011436971835792065,
152
+ -0.05664053186774254
153
+ ],
154
+ "mean": [
155
+ 0.031651551516052555,
156
+ -0.03161481387920421
157
+ ],
158
+ "std": [
159
+ 0.013119769222217526,
160
+ 0.01306078225192402
161
+ ],
162
+ "q01": [
163
+ 0.0060999246446777336,
164
+ -0.04062601653964198
165
+ ],
166
+ "q99": [
167
+ 0.04054891621517283,
168
+ -0.0059163567113456345
169
+ ]
170
+ }
171
+ },
172
+ "action": {
173
+ "base_motion": {
174
+ "max": [
175
+ 1.0,
176
+ 1.0,
177
+ 1.0,
178
+ 0.0
179
+ ],
180
+ "min": [
181
+ -1.0,
182
+ -1.0,
183
+ -1.0,
184
+ 0.0
185
+ ],
186
+ "mean": [
187
+ 0.008144896753718373,
188
+ -0.00018893135146078263,
189
+ -0.0008739062727221845,
190
+ 0.0
191
+ ],
192
+ "std": [
193
+ 0.11030045424364411,
194
+ 0.10148082570313594,
195
+ 0.08975698373467506,
196
+ 0.0
197
+ ],
198
+ "q01": [
199
+ -0.07287373067581518,
200
+ -0.09446994199569948,
201
+ -0.07702818259249572,
202
+ 0.0
203
+ ],
204
+ "q99": [
205
+ 0.04485485414407898,
206
+ 0.09626941605965177,
207
+ 0.07610665906122742,
208
+ 0.0
209
+ ]
210
+ },
211
+ "control_mode": {
212
+ "max": [
213
+ 1.0
214
+ ],
215
+ "min": [
216
+ -1.0
217
+ ],
218
+ "mean": [
219
+ -0.9221974313425939
220
+ ],
221
+ "std": [
222
+ 0.38670194342612674
223
+ ],
224
+ "q01": [
225
+ -1.0
226
+ ],
227
+ "q99": [
228
+ -0.384693946883099
229
+ ]
230
+ },
231
+ "end_effector_position": {
232
+ "max": [
233
+ 1.0,
234
+ 1.0,
235
+ 1.0
236
+ ],
237
+ "min": [
238
+ -1.0,
239
+ -1.0,
240
+ -1.0
241
+ ],
242
+ "mean": [
243
+ 0.01976129923017491,
244
+ -0.020870631344490034,
245
+ -0.05993476517349181
246
+ ],
247
+ "std": [
248
+ 0.4347024990804053,
249
+ 0.4221662976569997,
250
+ 0.3845506843044066
251
+ ],
252
+ "q01": [
253
+ -0.8835792939767807,
254
+ -0.9280919251713778,
255
+ -0.783361071470956
256
+ ],
257
+ "q99": [
258
+ 0.7622084438449307,
259
+ 0.8625125252609077,
260
+ 0.7470974753250706
261
+ ]
262
+ },
263
+ "end_effector_rotation": {
264
+ "max": [
265
+ 1.0,
266
+ 1.0,
267
+ 1.0
268
+ ],
269
+ "min": [
270
+ -1.0,
271
+ -1.0,
272
+ -1.0
273
+ ],
274
+ "mean": [
275
+ 0.008089673068544335,
276
+ -0.026498254627927348,
277
+ 0.0029083551743059005
278
+ ],
279
+ "std": [
280
+ 0.11216734503733773,
281
+ 0.13206002613798629,
282
+ 0.12615870412692864
283
+ ],
284
+ "q01": [
285
+ -0.2814627972846672,
286
+ -0.4342386516115439,
287
+ -0.34489755628340574
288
+ ],
289
+ "q99": [
290
+ 0.31879353266939386,
291
+ 0.28411797683220563,
292
+ 0.3597718671669894
293
+ ]
294
+ },
295
+ "gripper_close": {
296
+ "max": [
297
+ 1.0
298
+ ],
299
+ "min": [
300
+ -1.0
301
+ ],
302
+ "mean": [
303
+ -0.3824228874769045
304
+ ],
305
+ "std": [
306
+ 0.9240015096469786
307
+ ],
308
+ "q01": [
309
+ -1.0
310
+ ],
311
+ "q99": [
312
+ 0.5597272305813592
313
+ ]
314
+ }
315
+ }
316
+ },
317
+ "modalities": {
318
+ "video": {
319
+ "robot0_eye_in_hand": {
320
+ "resolution": [
321
+ 256,
322
+ 256
323
+ ],
324
+ "channels": 3,
325
+ "fps": 20.0
326
+ },
327
+ "robot0_agentview_left": {
328
+ "resolution": [
329
+ 256,
330
+ 256
331
+ ],
332
+ "channels": 3,
333
+ "fps": 20.0
334
+ },
335
+ "robot0_agentview_right": {
336
+ "resolution": [
337
+ 256,
338
+ 256
339
+ ],
340
+ "channels": 3,
341
+ "fps": 20.0
342
+ }
343
+ },
344
+ "state": {
345
+ "base_position": {
346
+ "absolute": true,
347
+ "rotation_type": null,
348
+ "shape": [
349
+ 3
350
+ ],
351
+ "continuous": true
352
+ },
353
+ "base_rotation": {
354
+ "absolute": true,
355
+ "rotation_type": "quaternion",
356
+ "shape": [
357
+ 4
358
+ ],
359
+ "continuous": true
360
+ },
361
+ "end_effector_position_relative": {
362
+ "absolute": true,
363
+ "rotation_type": null,
364
+ "shape": [
365
+ 3
366
+ ],
367
+ "continuous": true
368
+ },
369
+ "end_effector_rotation_relative": {
370
+ "absolute": true,
371
+ "rotation_type": "quaternion",
372
+ "shape": [
373
+ 4
374
+ ],
375
+ "continuous": true
376
+ },
377
+ "gripper_qpos": {
378
+ "absolute": true,
379
+ "rotation_type": null,
380
+ "shape": [
381
+ 2
382
+ ],
383
+ "continuous": true
384
+ }
385
+ },
386
+ "action": {
387
+ "base_motion": {
388
+ "absolute": true,
389
+ "rotation_type": null,
390
+ "shape": [
391
+ 4
392
+ ],
393
+ "continuous": true
394
+ },
395
+ "control_mode": {
396
+ "absolute": true,
397
+ "rotation_type": null,
398
+ "shape": [
399
+ 1
400
+ ],
401
+ "continuous": true
402
+ },
403
+ "end_effector_position": {
404
+ "absolute": true,
405
+ "rotation_type": null,
406
+ "shape": [
407
+ 3
408
+ ],
409
+ "continuous": true
410
+ },
411
+ "end_effector_rotation": {
412
+ "absolute": true,
413
+ "rotation_type": "axis_angle",
414
+ "shape": [
415
+ 3
416
+ ],
417
+ "continuous": true
418
+ },
419
+ "gripper_close": {
420
+ "absolute": true,
421
+ "rotation_type": null,
422
+ "shape": [
423
+ 1
424
+ ],
425
+ "continuous": true
426
+ }
427
+ }
428
+ },
429
+ "embodiment_tag": "new_embodiment"
430
+ }
431
+ }
MGRKD/mgrkd_distance_angle_cls_transformer_action_encoder/default/resolved_config.yaml ADDED
@@ -0,0 +1,68 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: mgrkd_distance_angle_cls_transformer_action_encoder_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_MGRKD
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/MGRKD/mgrkd_distance_angle_cls_transformer_action_encoder
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 1
9
+ batch_size: 128
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_mgrkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: action_encoder
29
+ rkd_action_encoder_projector_enabled: true
30
+ rkd_action_encoder_projector_dim: 512
31
+ rkd_action_encoder_projector_pooling: cls_transformer
32
+ rkd_action_encoder_cls_num_heads: 8
33
+ rkd_action_encoder_cls_num_layers: 1
34
+ rkd_action_encoder_cls_ffn_hidden_dim: 2048
35
+ rkd_student_reconstruction_head_enabled: true
36
+ rkd_student_reconstruction_head_dim: 512
37
+ rkd_student_reconstruction_head_hidden_dim: 512
38
+ rkd_distance_loss_weight: 1.0
39
+ rkd_angle_loss_weight: 2.0
40
+ rkd_exclude_diagonal: true
41
+ - name: phase3_fm_mgrkd_da_fixed_0p5
42
+ max_steps: 30000
43
+ save_steps: 0
44
+ trainable:
45
+ tune_llm: false
46
+ tune_visual: false
47
+ tune_projector: true
48
+ tune_diffusion_model: true
49
+ losses:
50
+ rkd_enabled: true
51
+ rkd_fm_loss_weight: 1.0
52
+ rkd_loss_weight: 0.5
53
+ rkd_relation_mode: flatten
54
+ rkd_loss_type: distance_angle
55
+ rkd_teacher_source: action_encoder
56
+ rkd_action_encoder_projector_enabled: true
57
+ rkd_action_encoder_projector_dim: 512
58
+ rkd_action_encoder_projector_pooling: cls_transformer
59
+ rkd_action_encoder_cls_num_heads: 8
60
+ rkd_action_encoder_cls_num_layers: 1
61
+ rkd_action_encoder_cls_ffn_hidden_dim: 2048
62
+ rkd_student_reconstruction_head_enabled: true
63
+ rkd_student_reconstruction_head_dim: 512
64
+ rkd_student_reconstruction_head_hidden_dim: 512
65
+ rkd_distance_loss_weight: 1.0
66
+ rkd_angle_loss_weight: 2.0
67
+ rkd_exclude_diagonal: true
68
+ resolved_sweep: {}
MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/mgrkd_da_flatten_action_encoder.yaml ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: mgrkd_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_MGRKD
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/MGRKD/mgrkd_distance_angle_flatten_action_encoder
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 1
9
+ batch_size: 128
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_mgrkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: action_encoder
29
+ rkd_action_encoder_projector_enabled: true
30
+ rkd_action_encoder_projector_dim: 512
31
+ rkd_action_encoder_projector_pooling: flatten
32
+ rkd_student_reconstruction_head_enabled: true
33
+ rkd_student_reconstruction_head_dim: 512
34
+ rkd_student_reconstruction_head_hidden_dim: 512
35
+ rkd_distance_loss_weight: 1.0
36
+ rkd_angle_loss_weight: 2.0
37
+ rkd_exclude_diagonal: true
38
+ - name: phase3_fm_mgrkd_da_fixed_0p5
39
+ max_steps: 30000
40
+ save_steps: 0
41
+ trainable:
42
+ tune_llm: false
43
+ tune_visual: false
44
+ tune_projector: true
45
+ tune_diffusion_model: true
46
+ losses:
47
+ rkd_enabled: true
48
+ rkd_fm_loss_weight: 1.0
49
+ rkd_loss_weight: 0.5
50
+ rkd_relation_mode: flatten
51
+ rkd_loss_type: distance_angle
52
+ rkd_teacher_source: action_encoder
53
+ rkd_action_encoder_projector_enabled: true
54
+ rkd_action_encoder_projector_dim: 512
55
+ rkd_action_encoder_projector_pooling: flatten
56
+ rkd_student_reconstruction_head_enabled: true
57
+ rkd_student_reconstruction_head_dim: 512
58
+ rkd_student_reconstruction_head_hidden_dim: 512
59
+ rkd_distance_loss_weight: 1.0
60
+ rkd_angle_loss_weight: 2.0
61
+ rkd_exclude_diagonal: true
MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase2/config.json ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "action_dim": 32,
5
+ "action_horizon": 16,
6
+ "add_pos_embed": true,
7
+ "backbone_embedding_dim": 2048,
8
+ "diffusion_model_cfg": {
9
+ "attention_head_dim": 48,
10
+ "cross_attention_dim": 2048,
11
+ "dropout": 0.2,
12
+ "final_dropout": true,
13
+ "interleave_self_attention": true,
14
+ "norm_type": "ada_norm",
15
+ "num_attention_heads": 32,
16
+ "num_layers": 16,
17
+ "output_dim": 1024,
18
+ "positional_embeddings": null
19
+ },
20
+ "hidden_size": 1024,
21
+ "input_embedding_dim": 1536,
22
+ "max_action_dim": 32,
23
+ "max_state_dim": 64,
24
+ "model_dtype": "float32",
25
+ "noise_beta_alpha": 1.5,
26
+ "noise_beta_beta": 1.0,
27
+ "noise_s": 0.999,
28
+ "num_inference_timesteps": 4,
29
+ "num_target_vision_tokens": 32,
30
+ "num_timestep_buckets": 1000,
31
+ "tune_diffusion_model": true,
32
+ "tune_projector": true,
33
+ "use_vlln": true,
34
+ "vl_self_attention_cfg": {
35
+ "attention_head_dim": 64,
36
+ "dropout": 0.2,
37
+ "final_dropout": true,
38
+ "num_attention_heads": 32,
39
+ "num_layers": 4,
40
+ "positional_embeddings": null
41
+ }
42
+ },
43
+ "action_horizon": 16,
44
+ "architectures": [
45
+ "GR00T_N1_5_RKD"
46
+ ],
47
+ "attn_implementation": null,
48
+ "backbone_cfg": {
49
+ "eagle_path": "NVEagle/eagle_er-qwen3_1_7B-Siglip2_400M_stage1_5_128gpu_er_v7_1mlp_nops",
50
+ "load_bf16": false,
51
+ "project_to_dim": null,
52
+ "reproject_vision": false,
53
+ "select_layer": 12,
54
+ "tune_llm": false,
55
+ "tune_visual": true,
56
+ "use_flash_attention": true
57
+ },
58
+ "compute_dtype": "bfloat16",
59
+ "hidden_size": 2048,
60
+ "model_dtype": "float32",
61
+ "model_type": "gr00t_n1_5",
62
+ "rkd_action_encoder_projector_dim": 512,
63
+ "rkd_action_encoder_projector_enabled": true,
64
+ "rkd_action_encoder_projector_pooling": "flatten",
65
+ "rkd_action_temp": 0.1,
66
+ "rkd_angle_loss_weight": 2.0,
67
+ "rkd_distance_loss_weight": 1.0,
68
+ "rkd_enabled": true,
69
+ "rkd_exclude_diagonal": true,
70
+ "rkd_fm_loss_weight": 0.0,
71
+ "rkd_loss_type": "distance_angle",
72
+ "rkd_loss_weight": 1.0,
73
+ "rkd_loss_weight_end": 0.0,
74
+ "rkd_loss_weight_schedule": null,
75
+ "rkd_loss_weight_start": 0.1,
76
+ "rkd_relation_mode": "flatten",
77
+ "rkd_student_reconstruction_head_dim": 512,
78
+ "rkd_student_reconstruction_head_enabled": true,
79
+ "rkd_student_reconstruction_head_hidden_dim": 512,
80
+ "rkd_teacher_source": "action_encoder",
81
+ "rkd_vlm_temp": 0.07,
82
+ "torch_dtype": "bfloat16",
83
+ "transformers_version": "4.51.3"
84
+ }
MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase2/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,431 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "base_position": {
6
+ "max": [
7
+ 7.3139495849609375,
8
+ 0.4876587688922882,
9
+ 0.7196521759033203
10
+ ],
11
+ "min": [
12
+ -4.94293737411499,
13
+ -6.890198230743408,
14
+ 0.6996827125549316
15
+ ],
16
+ "mean": [
17
+ 2.749288365724937,
18
+ -2.0455216239909983,
19
+ 0.700532551020533
20
+ ],
21
+ "std": [
22
+ 1.6786295897046262,
23
+ 1.312897969434244,
24
+ 0.00133751758906258
25
+ ],
26
+ "q01": [
27
+ -1.4241907881148037,
28
+ -5.1716547935950885,
29
+ 0.6999670898768614
30
+ ],
31
+ "q99": [
32
+ 6.198111545831555,
33
+ -0.5873531334104489,
34
+ 0.7038833014242587
35
+ ]
36
+ },
37
+ "base_rotation": {
38
+ "max": [
39
+ 0.0,
40
+ 0.0,
41
+ 1.0,
42
+ 1.0
43
+ ],
44
+ "min": [
45
+ 0.0,
46
+ 0.0,
47
+ -1.0,
48
+ 0.0
49
+ ],
50
+ "mean": [
51
+ 0.0,
52
+ 0.0,
53
+ 0.23084999587450247,
54
+ 0.6238431088581847
55
+ ],
56
+ "std": [
57
+ 0.0,
58
+ 0.0,
59
+ 0.6556404371644047,
60
+ 0.3572252105732042
61
+ ],
62
+ "q01": [
63
+ 0.0,
64
+ 0.0,
65
+ -0.9999999999999999,
66
+ 1.7482354593273349e-06
67
+ ],
68
+ "q99": [
69
+ 0.0,
70
+ 0.0,
71
+ 0.9999999999999999,
72
+ 0.9999999999999999
73
+ ]
74
+ },
75
+ "end_effector_position_relative": {
76
+ "max": [
77
+ 0.9014528393745422,
78
+ 0.8003877401351929,
79
+ 0.9829942584037781
80
+ ],
81
+ "min": [
82
+ -0.37471598386764526,
83
+ -0.8472502827644348,
84
+ -0.25070279836654663
85
+ ],
86
+ "mean": [
87
+ 0.28667332601863704,
88
+ -0.038315473170876704,
89
+ 0.45974658906777444
90
+ ],
91
+ "std": [
92
+ 0.17067058828305154,
93
+ 0.23569952037784195,
94
+ 0.21950702354181487
95
+ ],
96
+ "q01": [
97
+ 0.015172948963784228,
98
+ -0.44450502063220615,
99
+ 0.23314444103329338
100
+ ],
101
+ "q99": [
102
+ 0.5609069689572412,
103
+ 0.38454179128009547,
104
+ 0.7028102736473852
105
+ ]
106
+ },
107
+ "end_effector_rotation_relative": {
108
+ "max": [
109
+ 0.9999998807907104,
110
+ 0.9984455108642578,
111
+ 0.9434727430343628,
112
+ 0.9062229990959167
113
+ ],
114
+ "min": [
115
+ -0.9999930262565613,
116
+ -0.9988337755203247,
117
+ -0.9618149995803833,
118
+ 2.1872274658107926e-07
119
+ ],
120
+ "mean": [
121
+ -0.25672297401336,
122
+ 0.024425335806687216,
123
+ -0.08740977164483847,
124
+ 0.16608739644018397
125
+ ],
126
+ "std": [
127
+ 0.7932585208392383,
128
+ 0.3102260195580416,
129
+ 0.3772472576315148,
130
+ 0.17451959300158773
131
+ ],
132
+ "q01": [
133
+ -0.99379006987882,
134
+ -0.5478102050038828,
135
+ -0.6101777952971636,
136
+ 0.002274085014134748
137
+ ],
138
+ "q99": [
139
+ 0.8804856038896288,
140
+ 0.5910611092873037,
141
+ 0.5216058042993562,
142
+ 0.5231965974012255
143
+ ]
144
+ },
145
+ "gripper_qpos": {
146
+ "max": [
147
+ 0.055869169533252716,
148
+ 0.010916369967162609
149
+ ],
150
+ "min": [
151
+ -0.011436971835792065,
152
+ -0.05664053186774254
153
+ ],
154
+ "mean": [
155
+ 0.031651551516052555,
156
+ -0.03161481387920421
157
+ ],
158
+ "std": [
159
+ 0.013119769222217526,
160
+ 0.01306078225192402
161
+ ],
162
+ "q01": [
163
+ 0.0060999246446777336,
164
+ -0.04062601653964198
165
+ ],
166
+ "q99": [
167
+ 0.04054891621517283,
168
+ -0.0059163567113456345
169
+ ]
170
+ }
171
+ },
172
+ "action": {
173
+ "base_motion": {
174
+ "max": [
175
+ 1.0,
176
+ 1.0,
177
+ 1.0,
178
+ 0.0
179
+ ],
180
+ "min": [
181
+ -1.0,
182
+ -1.0,
183
+ -1.0,
184
+ 0.0
185
+ ],
186
+ "mean": [
187
+ 0.008144896753718373,
188
+ -0.00018893135146078263,
189
+ -0.0008739062727221845,
190
+ 0.0
191
+ ],
192
+ "std": [
193
+ 0.11030045424364411,
194
+ 0.10148082570313594,
195
+ 0.08975698373467506,
196
+ 0.0
197
+ ],
198
+ "q01": [
199
+ -0.07287373067581518,
200
+ -0.09446994199569948,
201
+ -0.07702818259249572,
202
+ 0.0
203
+ ],
204
+ "q99": [
205
+ 0.04485485414407898,
206
+ 0.09626941605965177,
207
+ 0.07610665906122742,
208
+ 0.0
209
+ ]
210
+ },
211
+ "control_mode": {
212
+ "max": [
213
+ 1.0
214
+ ],
215
+ "min": [
216
+ -1.0
217
+ ],
218
+ "mean": [
219
+ -0.9221974313425939
220
+ ],
221
+ "std": [
222
+ 0.38670194342612674
223
+ ],
224
+ "q01": [
225
+ -1.0
226
+ ],
227
+ "q99": [
228
+ -0.384693946883099
229
+ ]
230
+ },
231
+ "end_effector_position": {
232
+ "max": [
233
+ 1.0,
234
+ 1.0,
235
+ 1.0
236
+ ],
237
+ "min": [
238
+ -1.0,
239
+ -1.0,
240
+ -1.0
241
+ ],
242
+ "mean": [
243
+ 0.01976129923017491,
244
+ -0.020870631344490034,
245
+ -0.05993476517349181
246
+ ],
247
+ "std": [
248
+ 0.4347024990804053,
249
+ 0.4221662976569997,
250
+ 0.3845506843044066
251
+ ],
252
+ "q01": [
253
+ -0.8835792939767807,
254
+ -0.9280919251713778,
255
+ -0.783361071470956
256
+ ],
257
+ "q99": [
258
+ 0.7622084438449307,
259
+ 0.8625125252609077,
260
+ 0.7470974753250706
261
+ ]
262
+ },
263
+ "end_effector_rotation": {
264
+ "max": [
265
+ 1.0,
266
+ 1.0,
267
+ 1.0
268
+ ],
269
+ "min": [
270
+ -1.0,
271
+ -1.0,
272
+ -1.0
273
+ ],
274
+ "mean": [
275
+ 0.008089673068544335,
276
+ -0.026498254627927348,
277
+ 0.0029083551743059005
278
+ ],
279
+ "std": [
280
+ 0.11216734503733773,
281
+ 0.13206002613798629,
282
+ 0.12615870412692864
283
+ ],
284
+ "q01": [
285
+ -0.2814627972846672,
286
+ -0.4342386516115439,
287
+ -0.34489755628340574
288
+ ],
289
+ "q99": [
290
+ 0.31879353266939386,
291
+ 0.28411797683220563,
292
+ 0.3597718671669894
293
+ ]
294
+ },
295
+ "gripper_close": {
296
+ "max": [
297
+ 1.0
298
+ ],
299
+ "min": [
300
+ -1.0
301
+ ],
302
+ "mean": [
303
+ -0.3824228874769045
304
+ ],
305
+ "std": [
306
+ 0.9240015096469786
307
+ ],
308
+ "q01": [
309
+ -1.0
310
+ ],
311
+ "q99": [
312
+ 0.5597272305813592
313
+ ]
314
+ }
315
+ }
316
+ },
317
+ "modalities": {
318
+ "video": {
319
+ "robot0_eye_in_hand": {
320
+ "resolution": [
321
+ 256,
322
+ 256
323
+ ],
324
+ "channels": 3,
325
+ "fps": 20.0
326
+ },
327
+ "robot0_agentview_left": {
328
+ "resolution": [
329
+ 256,
330
+ 256
331
+ ],
332
+ "channels": 3,
333
+ "fps": 20.0
334
+ },
335
+ "robot0_agentview_right": {
336
+ "resolution": [
337
+ 256,
338
+ 256
339
+ ],
340
+ "channels": 3,
341
+ "fps": 20.0
342
+ }
343
+ },
344
+ "state": {
345
+ "base_position": {
346
+ "absolute": true,
347
+ "rotation_type": null,
348
+ "shape": [
349
+ 3
350
+ ],
351
+ "continuous": true
352
+ },
353
+ "base_rotation": {
354
+ "absolute": true,
355
+ "rotation_type": "quaternion",
356
+ "shape": [
357
+ 4
358
+ ],
359
+ "continuous": true
360
+ },
361
+ "end_effector_position_relative": {
362
+ "absolute": true,
363
+ "rotation_type": null,
364
+ "shape": [
365
+ 3
366
+ ],
367
+ "continuous": true
368
+ },
369
+ "end_effector_rotation_relative": {
370
+ "absolute": true,
371
+ "rotation_type": "quaternion",
372
+ "shape": [
373
+ 4
374
+ ],
375
+ "continuous": true
376
+ },
377
+ "gripper_qpos": {
378
+ "absolute": true,
379
+ "rotation_type": null,
380
+ "shape": [
381
+ 2
382
+ ],
383
+ "continuous": true
384
+ }
385
+ },
386
+ "action": {
387
+ "base_motion": {
388
+ "absolute": true,
389
+ "rotation_type": null,
390
+ "shape": [
391
+ 4
392
+ ],
393
+ "continuous": true
394
+ },
395
+ "control_mode": {
396
+ "absolute": true,
397
+ "rotation_type": null,
398
+ "shape": [
399
+ 1
400
+ ],
401
+ "continuous": true
402
+ },
403
+ "end_effector_position": {
404
+ "absolute": true,
405
+ "rotation_type": null,
406
+ "shape": [
407
+ 3
408
+ ],
409
+ "continuous": true
410
+ },
411
+ "end_effector_rotation": {
412
+ "absolute": true,
413
+ "rotation_type": "axis_angle",
414
+ "shape": [
415
+ 3
416
+ ],
417
+ "continuous": true
418
+ },
419
+ "gripper_close": {
420
+ "absolute": true,
421
+ "rotation_type": null,
422
+ "shape": [
423
+ 1
424
+ ],
425
+ "continuous": true
426
+ }
427
+ }
428
+ },
429
+ "embodiment_tag": "new_embodiment"
430
+ }
431
+ }
MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase2/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase2/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff
 
MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/phase3/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,431 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "base_position": {
6
+ "max": [
7
+ 7.3139495849609375,
8
+ 0.4876587688922882,
9
+ 0.7196521759033203
10
+ ],
11
+ "min": [
12
+ -4.94293737411499,
13
+ -6.890198230743408,
14
+ 0.6996827125549316
15
+ ],
16
+ "mean": [
17
+ 2.749288365724937,
18
+ -2.0455216239909983,
19
+ 0.700532551020533
20
+ ],
21
+ "std": [
22
+ 1.6786295897046262,
23
+ 1.312897969434244,
24
+ 0.00133751758906258
25
+ ],
26
+ "q01": [
27
+ -1.4241907881148037,
28
+ -5.1716547935950885,
29
+ 0.6999670898768614
30
+ ],
31
+ "q99": [
32
+ 6.198111545831555,
33
+ -0.5873531334104489,
34
+ 0.7038833014242587
35
+ ]
36
+ },
37
+ "base_rotation": {
38
+ "max": [
39
+ 0.0,
40
+ 0.0,
41
+ 1.0,
42
+ 1.0
43
+ ],
44
+ "min": [
45
+ 0.0,
46
+ 0.0,
47
+ -1.0,
48
+ 0.0
49
+ ],
50
+ "mean": [
51
+ 0.0,
52
+ 0.0,
53
+ 0.23084999587450247,
54
+ 0.6238431088581847
55
+ ],
56
+ "std": [
57
+ 0.0,
58
+ 0.0,
59
+ 0.6556404371644047,
60
+ 0.3572252105732042
61
+ ],
62
+ "q01": [
63
+ 0.0,
64
+ 0.0,
65
+ -0.9999999999999999,
66
+ 1.7482354593273349e-06
67
+ ],
68
+ "q99": [
69
+ 0.0,
70
+ 0.0,
71
+ 0.9999999999999999,
72
+ 0.9999999999999999
73
+ ]
74
+ },
75
+ "end_effector_position_relative": {
76
+ "max": [
77
+ 0.9014528393745422,
78
+ 0.8003877401351929,
79
+ 0.9829942584037781
80
+ ],
81
+ "min": [
82
+ -0.37471598386764526,
83
+ -0.8472502827644348,
84
+ -0.25070279836654663
85
+ ],
86
+ "mean": [
87
+ 0.28667332601863704,
88
+ -0.038315473170876704,
89
+ 0.45974658906777444
90
+ ],
91
+ "std": [
92
+ 0.17067058828305154,
93
+ 0.23569952037784195,
94
+ 0.21950702354181487
95
+ ],
96
+ "q01": [
97
+ 0.015172948963784228,
98
+ -0.44450502063220615,
99
+ 0.23314444103329338
100
+ ],
101
+ "q99": [
102
+ 0.5609069689572412,
103
+ 0.38454179128009547,
104
+ 0.7028102736473852
105
+ ]
106
+ },
107
+ "end_effector_rotation_relative": {
108
+ "max": [
109
+ 0.9999998807907104,
110
+ 0.9984455108642578,
111
+ 0.9434727430343628,
112
+ 0.9062229990959167
113
+ ],
114
+ "min": [
115
+ -0.9999930262565613,
116
+ -0.9988337755203247,
117
+ -0.9618149995803833,
118
+ 2.1872274658107926e-07
119
+ ],
120
+ "mean": [
121
+ -0.25672297401336,
122
+ 0.024425335806687216,
123
+ -0.08740977164483847,
124
+ 0.16608739644018397
125
+ ],
126
+ "std": [
127
+ 0.7932585208392383,
128
+ 0.3102260195580416,
129
+ 0.3772472576315148,
130
+ 0.17451959300158773
131
+ ],
132
+ "q01": [
133
+ -0.99379006987882,
134
+ -0.5478102050038828,
135
+ -0.6101777952971636,
136
+ 0.002274085014134748
137
+ ],
138
+ "q99": [
139
+ 0.8804856038896288,
140
+ 0.5910611092873037,
141
+ 0.5216058042993562,
142
+ 0.5231965974012255
143
+ ]
144
+ },
145
+ "gripper_qpos": {
146
+ "max": [
147
+ 0.055869169533252716,
148
+ 0.010916369967162609
149
+ ],
150
+ "min": [
151
+ -0.011436971835792065,
152
+ -0.05664053186774254
153
+ ],
154
+ "mean": [
155
+ 0.031651551516052555,
156
+ -0.03161481387920421
157
+ ],
158
+ "std": [
159
+ 0.013119769222217526,
160
+ 0.01306078225192402
161
+ ],
162
+ "q01": [
163
+ 0.0060999246446777336,
164
+ -0.04062601653964198
165
+ ],
166
+ "q99": [
167
+ 0.04054891621517283,
168
+ -0.0059163567113456345
169
+ ]
170
+ }
171
+ },
172
+ "action": {
173
+ "base_motion": {
174
+ "max": [
175
+ 1.0,
176
+ 1.0,
177
+ 1.0,
178
+ 0.0
179
+ ],
180
+ "min": [
181
+ -1.0,
182
+ -1.0,
183
+ -1.0,
184
+ 0.0
185
+ ],
186
+ "mean": [
187
+ 0.008144896753718373,
188
+ -0.00018893135146078263,
189
+ -0.0008739062727221845,
190
+ 0.0
191
+ ],
192
+ "std": [
193
+ 0.11030045424364411,
194
+ 0.10148082570313594,
195
+ 0.08975698373467506,
196
+ 0.0
197
+ ],
198
+ "q01": [
199
+ -0.07287373067581518,
200
+ -0.09446994199569948,
201
+ -0.07702818259249572,
202
+ 0.0
203
+ ],
204
+ "q99": [
205
+ 0.04485485414407898,
206
+ 0.09626941605965177,
207
+ 0.07610665906122742,
208
+ 0.0
209
+ ]
210
+ },
211
+ "control_mode": {
212
+ "max": [
213
+ 1.0
214
+ ],
215
+ "min": [
216
+ -1.0
217
+ ],
218
+ "mean": [
219
+ -0.9221974313425939
220
+ ],
221
+ "std": [
222
+ 0.38670194342612674
223
+ ],
224
+ "q01": [
225
+ -1.0
226
+ ],
227
+ "q99": [
228
+ -0.384693946883099
229
+ ]
230
+ },
231
+ "end_effector_position": {
232
+ "max": [
233
+ 1.0,
234
+ 1.0,
235
+ 1.0
236
+ ],
237
+ "min": [
238
+ -1.0,
239
+ -1.0,
240
+ -1.0
241
+ ],
242
+ "mean": [
243
+ 0.01976129923017491,
244
+ -0.020870631344490034,
245
+ -0.05993476517349181
246
+ ],
247
+ "std": [
248
+ 0.4347024990804053,
249
+ 0.4221662976569997,
250
+ 0.3845506843044066
251
+ ],
252
+ "q01": [
253
+ -0.8835792939767807,
254
+ -0.9280919251713778,
255
+ -0.783361071470956
256
+ ],
257
+ "q99": [
258
+ 0.7622084438449307,
259
+ 0.8625125252609077,
260
+ 0.7470974753250706
261
+ ]
262
+ },
263
+ "end_effector_rotation": {
264
+ "max": [
265
+ 1.0,
266
+ 1.0,
267
+ 1.0
268
+ ],
269
+ "min": [
270
+ -1.0,
271
+ -1.0,
272
+ -1.0
273
+ ],
274
+ "mean": [
275
+ 0.008089673068544335,
276
+ -0.026498254627927348,
277
+ 0.0029083551743059005
278
+ ],
279
+ "std": [
280
+ 0.11216734503733773,
281
+ 0.13206002613798629,
282
+ 0.12615870412692864
283
+ ],
284
+ "q01": [
285
+ -0.2814627972846672,
286
+ -0.4342386516115439,
287
+ -0.34489755628340574
288
+ ],
289
+ "q99": [
290
+ 0.31879353266939386,
291
+ 0.28411797683220563,
292
+ 0.3597718671669894
293
+ ]
294
+ },
295
+ "gripper_close": {
296
+ "max": [
297
+ 1.0
298
+ ],
299
+ "min": [
300
+ -1.0
301
+ ],
302
+ "mean": [
303
+ -0.3824228874769045
304
+ ],
305
+ "std": [
306
+ 0.9240015096469786
307
+ ],
308
+ "q01": [
309
+ -1.0
310
+ ],
311
+ "q99": [
312
+ 0.5597272305813592
313
+ ]
314
+ }
315
+ }
316
+ },
317
+ "modalities": {
318
+ "video": {
319
+ "robot0_eye_in_hand": {
320
+ "resolution": [
321
+ 256,
322
+ 256
323
+ ],
324
+ "channels": 3,
325
+ "fps": 20.0
326
+ },
327
+ "robot0_agentview_left": {
328
+ "resolution": [
329
+ 256,
330
+ 256
331
+ ],
332
+ "channels": 3,
333
+ "fps": 20.0
334
+ },
335
+ "robot0_agentview_right": {
336
+ "resolution": [
337
+ 256,
338
+ 256
339
+ ],
340
+ "channels": 3,
341
+ "fps": 20.0
342
+ }
343
+ },
344
+ "state": {
345
+ "base_position": {
346
+ "absolute": true,
347
+ "rotation_type": null,
348
+ "shape": [
349
+ 3
350
+ ],
351
+ "continuous": true
352
+ },
353
+ "base_rotation": {
354
+ "absolute": true,
355
+ "rotation_type": "quaternion",
356
+ "shape": [
357
+ 4
358
+ ],
359
+ "continuous": true
360
+ },
361
+ "end_effector_position_relative": {
362
+ "absolute": true,
363
+ "rotation_type": null,
364
+ "shape": [
365
+ 3
366
+ ],
367
+ "continuous": true
368
+ },
369
+ "end_effector_rotation_relative": {
370
+ "absolute": true,
371
+ "rotation_type": "quaternion",
372
+ "shape": [
373
+ 4
374
+ ],
375
+ "continuous": true
376
+ },
377
+ "gripper_qpos": {
378
+ "absolute": true,
379
+ "rotation_type": null,
380
+ "shape": [
381
+ 2
382
+ ],
383
+ "continuous": true
384
+ }
385
+ },
386
+ "action": {
387
+ "base_motion": {
388
+ "absolute": true,
389
+ "rotation_type": null,
390
+ "shape": [
391
+ 4
392
+ ],
393
+ "continuous": true
394
+ },
395
+ "control_mode": {
396
+ "absolute": true,
397
+ "rotation_type": null,
398
+ "shape": [
399
+ 1
400
+ ],
401
+ "continuous": true
402
+ },
403
+ "end_effector_position": {
404
+ "absolute": true,
405
+ "rotation_type": null,
406
+ "shape": [
407
+ 3
408
+ ],
409
+ "continuous": true
410
+ },
411
+ "end_effector_rotation": {
412
+ "absolute": true,
413
+ "rotation_type": "axis_angle",
414
+ "shape": [
415
+ 3
416
+ ],
417
+ "continuous": true
418
+ },
419
+ "gripper_close": {
420
+ "absolute": true,
421
+ "rotation_type": null,
422
+ "shape": [
423
+ 1
424
+ ],
425
+ "continuous": true
426
+ }
427
+ }
428
+ },
429
+ "embodiment_tag": "new_embodiment"
430
+ }
431
+ }
MGRKD/mgrkd_distance_angle_flatten_action_encoder/default/resolved_config.yaml ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: mgrkd_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_MGRKD
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/MGRKD/mgrkd_distance_angle_flatten_action_encoder
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 1
9
+ batch_size: 128
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_mgrkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: action_encoder
29
+ rkd_action_encoder_projector_enabled: true
30
+ rkd_action_encoder_projector_dim: 512
31
+ rkd_action_encoder_projector_pooling: flatten
32
+ rkd_student_reconstruction_head_enabled: true
33
+ rkd_student_reconstruction_head_dim: 512
34
+ rkd_student_reconstruction_head_hidden_dim: 512
35
+ rkd_distance_loss_weight: 1.0
36
+ rkd_angle_loss_weight: 2.0
37
+ rkd_exclude_diagonal: true
38
+ - name: phase3_fm_mgrkd_da_fixed_0p5
39
+ max_steps: 30000
40
+ save_steps: 0
41
+ trainable:
42
+ tune_llm: false
43
+ tune_visual: false
44
+ tune_projector: true
45
+ tune_diffusion_model: true
46
+ losses:
47
+ rkd_enabled: true
48
+ rkd_fm_loss_weight: 1.0
49
+ rkd_loss_weight: 0.5
50
+ rkd_relation_mode: flatten
51
+ rkd_loss_type: distance_angle
52
+ rkd_teacher_source: action_encoder
53
+ rkd_action_encoder_projector_enabled: true
54
+ rkd_action_encoder_projector_dim: 512
55
+ rkd_action_encoder_projector_pooling: flatten
56
+ rkd_student_reconstruction_head_enabled: true
57
+ rkd_student_reconstruction_head_dim: 512
58
+ rkd_student_reconstruction_head_hidden_dim: 512
59
+ rkd_distance_loss_weight: 1.0
60
+ rkd_angle_loss_weight: 2.0
61
+ rkd_exclude_diagonal: true
62
+ resolved_sweep: {}
MGRKD/mgrkd_distance_angle_flatten_raw_action/default/mgrkd_da_flatten_raw_action.yaml ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: mgrkd_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_MGRKD
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/MGRKD/mgrkd_distance_angle_flatten_raw_action
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 1
9
+ batch_size: 128
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_mgrkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: raw_action
29
+ rkd_raw_action_projector_enabled: true
30
+ rkd_raw_action_projector_dim: 512
31
+ rkd_raw_action_projector_pooling: flatten
32
+ rkd_student_reconstruction_head_enabled: true
33
+ rkd_student_reconstruction_head_dim: 512
34
+ rkd_student_reconstruction_head_hidden_dim: 512
35
+ rkd_distance_loss_weight: 1.0
36
+ rkd_angle_loss_weight: 2.0
37
+ rkd_exclude_diagonal: true
38
+ - name: phase3_fm_mgrkd_da_fixed_0p5
39
+ max_steps: 30000
40
+ save_steps: 0
41
+ trainable:
42
+ tune_llm: false
43
+ tune_visual: false
44
+ tune_projector: true
45
+ tune_diffusion_model: true
46
+ losses:
47
+ rkd_enabled: true
48
+ rkd_fm_loss_weight: 1.0
49
+ rkd_loss_weight: 0.5
50
+ rkd_relation_mode: flatten
51
+ rkd_loss_type: distance_angle
52
+ rkd_teacher_source: raw_action
53
+ rkd_raw_action_projector_enabled: true
54
+ rkd_raw_action_projector_dim: 512
55
+ rkd_raw_action_projector_pooling: flatten
56
+ rkd_student_reconstruction_head_enabled: true
57
+ rkd_student_reconstruction_head_dim: 512
58
+ rkd_student_reconstruction_head_hidden_dim: 512
59
+ rkd_distance_loss_weight: 1.0
60
+ rkd_angle_loss_weight: 2.0
61
+ rkd_exclude_diagonal: true
MGRKD/mgrkd_distance_angle_flatten_raw_action/default/phase2/config.json ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "action_dim": 32,
5
+ "action_horizon": 16,
6
+ "add_pos_embed": true,
7
+ "backbone_embedding_dim": 2048,
8
+ "diffusion_model_cfg": {
9
+ "attention_head_dim": 48,
10
+ "cross_attention_dim": 2048,
11
+ "dropout": 0.2,
12
+ "final_dropout": true,
13
+ "interleave_self_attention": true,
14
+ "norm_type": "ada_norm",
15
+ "num_attention_heads": 32,
16
+ "num_layers": 16,
17
+ "output_dim": 1024,
18
+ "positional_embeddings": null
19
+ },
20
+ "hidden_size": 1024,
21
+ "input_embedding_dim": 1536,
22
+ "max_action_dim": 32,
23
+ "max_state_dim": 64,
24
+ "model_dtype": "float32",
25
+ "noise_beta_alpha": 1.5,
26
+ "noise_beta_beta": 1.0,
27
+ "noise_s": 0.999,
28
+ "num_inference_timesteps": 4,
29
+ "num_target_vision_tokens": 32,
30
+ "num_timestep_buckets": 1000,
31
+ "tune_diffusion_model": true,
32
+ "tune_projector": true,
33
+ "use_vlln": true,
34
+ "vl_self_attention_cfg": {
35
+ "attention_head_dim": 64,
36
+ "dropout": 0.2,
37
+ "final_dropout": true,
38
+ "num_attention_heads": 32,
39
+ "num_layers": 4,
40
+ "positional_embeddings": null
41
+ }
42
+ },
43
+ "action_horizon": 16,
44
+ "architectures": [
45
+ "GR00T_N1_5_RKD"
46
+ ],
47
+ "attn_implementation": null,
48
+ "backbone_cfg": {
49
+ "eagle_path": "NVEagle/eagle_er-qwen3_1_7B-Siglip2_400M_stage1_5_128gpu_er_v7_1mlp_nops",
50
+ "load_bf16": false,
51
+ "project_to_dim": null,
52
+ "reproject_vision": false,
53
+ "select_layer": 12,
54
+ "tune_llm": false,
55
+ "tune_visual": true,
56
+ "use_flash_attention": true
57
+ },
58
+ "compute_dtype": "bfloat16",
59
+ "hidden_size": 2048,
60
+ "model_dtype": "float32",
61
+ "model_type": "gr00t_n1_5",
62
+ "rkd_action_encoder_cls_ffn_hidden_dim": 2048,
63
+ "rkd_action_encoder_cls_num_heads": 8,
64
+ "rkd_action_encoder_cls_num_layers": 1,
65
+ "rkd_action_encoder_projector_dim": 512,
66
+ "rkd_action_encoder_projector_enabled": false,
67
+ "rkd_action_encoder_projector_pooling": "flatten",
68
+ "rkd_action_temp": 0.1,
69
+ "rkd_angle_loss_weight": 2.0,
70
+ "rkd_distance_loss_weight": 1.0,
71
+ "rkd_enabled": true,
72
+ "rkd_exclude_diagonal": true,
73
+ "rkd_fm_loss_weight": 0.0,
74
+ "rkd_loss_type": "distance_angle",
75
+ "rkd_loss_weight": 1.0,
76
+ "rkd_loss_weight_end": 0.0,
77
+ "rkd_loss_weight_schedule": null,
78
+ "rkd_loss_weight_start": 0.1,
79
+ "rkd_raw_action_projector_dim": 512,
80
+ "rkd_raw_action_projector_enabled": true,
81
+ "rkd_raw_action_projector_pooling": "flatten",
82
+ "rkd_relation_mode": "flatten",
83
+ "rkd_student_reconstruction_head_dim": 512,
84
+ "rkd_student_reconstruction_head_enabled": true,
85
+ "rkd_student_reconstruction_head_hidden_dim": 512,
86
+ "rkd_teacher_source": "raw_action",
87
+ "rkd_vlm_temp": 0.07,
88
+ "torch_dtype": "bfloat16",
89
+ "transformers_version": "4.51.3"
90
+ }
MGRKD/mgrkd_distance_angle_flatten_raw_action/default/phase2/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,431 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "base_position": {
6
+ "max": [
7
+ 7.3139495849609375,
8
+ 0.4876587688922882,
9
+ 0.7196521759033203
10
+ ],
11
+ "min": [
12
+ -4.94293737411499,
13
+ -6.890198230743408,
14
+ 0.6996827125549316
15
+ ],
16
+ "mean": [
17
+ 2.749288365724937,
18
+ -2.0455216239909983,
19
+ 0.700532551020533
20
+ ],
21
+ "std": [
22
+ 1.6786295897046262,
23
+ 1.312897969434244,
24
+ 0.00133751758906258
25
+ ],
26
+ "q01": [
27
+ -1.4241907881148037,
28
+ -5.1716547935950885,
29
+ 0.6999670898768614
30
+ ],
31
+ "q99": [
32
+ 6.198111545831555,
33
+ -0.5873531334104489,
34
+ 0.7038833014242587
35
+ ]
36
+ },
37
+ "base_rotation": {
38
+ "max": [
39
+ 0.0,
40
+ 0.0,
41
+ 1.0,
42
+ 1.0
43
+ ],
44
+ "min": [
45
+ 0.0,
46
+ 0.0,
47
+ -1.0,
48
+ 0.0
49
+ ],
50
+ "mean": [
51
+ 0.0,
52
+ 0.0,
53
+ 0.23084999587450247,
54
+ 0.6238431088581847
55
+ ],
56
+ "std": [
57
+ 0.0,
58
+ 0.0,
59
+ 0.6556404371644047,
60
+ 0.3572252105732042
61
+ ],
62
+ "q01": [
63
+ 0.0,
64
+ 0.0,
65
+ -0.9999999999999999,
66
+ 1.7482354593273349e-06
67
+ ],
68
+ "q99": [
69
+ 0.0,
70
+ 0.0,
71
+ 0.9999999999999999,
72
+ 0.9999999999999999
73
+ ]
74
+ },
75
+ "end_effector_position_relative": {
76
+ "max": [
77
+ 0.9014528393745422,
78
+ 0.8003877401351929,
79
+ 0.9829942584037781
80
+ ],
81
+ "min": [
82
+ -0.37471598386764526,
83
+ -0.8472502827644348,
84
+ -0.25070279836654663
85
+ ],
86
+ "mean": [
87
+ 0.28667332601863704,
88
+ -0.038315473170876704,
89
+ 0.45974658906777444
90
+ ],
91
+ "std": [
92
+ 0.17067058828305154,
93
+ 0.23569952037784195,
94
+ 0.21950702354181487
95
+ ],
96
+ "q01": [
97
+ 0.015172948963784228,
98
+ -0.44450502063220615,
99
+ 0.23314444103329338
100
+ ],
101
+ "q99": [
102
+ 0.5609069689572412,
103
+ 0.38454179128009547,
104
+ 0.7028102736473852
105
+ ]
106
+ },
107
+ "end_effector_rotation_relative": {
108
+ "max": [
109
+ 0.9999998807907104,
110
+ 0.9984455108642578,
111
+ 0.9434727430343628,
112
+ 0.9062229990959167
113
+ ],
114
+ "min": [
115
+ -0.9999930262565613,
116
+ -0.9988337755203247,
117
+ -0.9618149995803833,
118
+ 2.1872274658107926e-07
119
+ ],
120
+ "mean": [
121
+ -0.25672297401336,
122
+ 0.024425335806687216,
123
+ -0.08740977164483847,
124
+ 0.16608739644018397
125
+ ],
126
+ "std": [
127
+ 0.7932585208392383,
128
+ 0.3102260195580416,
129
+ 0.3772472576315148,
130
+ 0.17451959300158773
131
+ ],
132
+ "q01": [
133
+ -0.99379006987882,
134
+ -0.5478102050038828,
135
+ -0.6101777952971636,
136
+ 0.002274085014134748
137
+ ],
138
+ "q99": [
139
+ 0.8804856038896288,
140
+ 0.5910611092873037,
141
+ 0.5216058042993562,
142
+ 0.5231965974012255
143
+ ]
144
+ },
145
+ "gripper_qpos": {
146
+ "max": [
147
+ 0.055869169533252716,
148
+ 0.010916369967162609
149
+ ],
150
+ "min": [
151
+ -0.011436971835792065,
152
+ -0.05664053186774254
153
+ ],
154
+ "mean": [
155
+ 0.031651551516052555,
156
+ -0.03161481387920421
157
+ ],
158
+ "std": [
159
+ 0.013119769222217526,
160
+ 0.01306078225192402
161
+ ],
162
+ "q01": [
163
+ 0.0060999246446777336,
164
+ -0.04062601653964198
165
+ ],
166
+ "q99": [
167
+ 0.04054891621517283,
168
+ -0.0059163567113456345
169
+ ]
170
+ }
171
+ },
172
+ "action": {
173
+ "base_motion": {
174
+ "max": [
175
+ 1.0,
176
+ 1.0,
177
+ 1.0,
178
+ 0.0
179
+ ],
180
+ "min": [
181
+ -1.0,
182
+ -1.0,
183
+ -1.0,
184
+ 0.0
185
+ ],
186
+ "mean": [
187
+ 0.008144896753718373,
188
+ -0.00018893135146078263,
189
+ -0.0008739062727221845,
190
+ 0.0
191
+ ],
192
+ "std": [
193
+ 0.11030045424364411,
194
+ 0.10148082570313594,
195
+ 0.08975698373467506,
196
+ 0.0
197
+ ],
198
+ "q01": [
199
+ -0.07287373067581518,
200
+ -0.09446994199569948,
201
+ -0.07702818259249572,
202
+ 0.0
203
+ ],
204
+ "q99": [
205
+ 0.04485485414407898,
206
+ 0.09626941605965177,
207
+ 0.07610665906122742,
208
+ 0.0
209
+ ]
210
+ },
211
+ "control_mode": {
212
+ "max": [
213
+ 1.0
214
+ ],
215
+ "min": [
216
+ -1.0
217
+ ],
218
+ "mean": [
219
+ -0.9221974313425939
220
+ ],
221
+ "std": [
222
+ 0.38670194342612674
223
+ ],
224
+ "q01": [
225
+ -1.0
226
+ ],
227
+ "q99": [
228
+ -0.384693946883099
229
+ ]
230
+ },
231
+ "end_effector_position": {
232
+ "max": [
233
+ 1.0,
234
+ 1.0,
235
+ 1.0
236
+ ],
237
+ "min": [
238
+ -1.0,
239
+ -1.0,
240
+ -1.0
241
+ ],
242
+ "mean": [
243
+ 0.01976129923017491,
244
+ -0.020870631344490034,
245
+ -0.05993476517349181
246
+ ],
247
+ "std": [
248
+ 0.4347024990804053,
249
+ 0.4221662976569997,
250
+ 0.3845506843044066
251
+ ],
252
+ "q01": [
253
+ -0.8835792939767807,
254
+ -0.9280919251713778,
255
+ -0.783361071470956
256
+ ],
257
+ "q99": [
258
+ 0.7622084438449307,
259
+ 0.8625125252609077,
260
+ 0.7470974753250706
261
+ ]
262
+ },
263
+ "end_effector_rotation": {
264
+ "max": [
265
+ 1.0,
266
+ 1.0,
267
+ 1.0
268
+ ],
269
+ "min": [
270
+ -1.0,
271
+ -1.0,
272
+ -1.0
273
+ ],
274
+ "mean": [
275
+ 0.008089673068544335,
276
+ -0.026498254627927348,
277
+ 0.0029083551743059005
278
+ ],
279
+ "std": [
280
+ 0.11216734503733773,
281
+ 0.13206002613798629,
282
+ 0.12615870412692864
283
+ ],
284
+ "q01": [
285
+ -0.2814627972846672,
286
+ -0.4342386516115439,
287
+ -0.34489755628340574
288
+ ],
289
+ "q99": [
290
+ 0.31879353266939386,
291
+ 0.28411797683220563,
292
+ 0.3597718671669894
293
+ ]
294
+ },
295
+ "gripper_close": {
296
+ "max": [
297
+ 1.0
298
+ ],
299
+ "min": [
300
+ -1.0
301
+ ],
302
+ "mean": [
303
+ -0.3824228874769045
304
+ ],
305
+ "std": [
306
+ 0.9240015096469786
307
+ ],
308
+ "q01": [
309
+ -1.0
310
+ ],
311
+ "q99": [
312
+ 0.5597272305813592
313
+ ]
314
+ }
315
+ }
316
+ },
317
+ "modalities": {
318
+ "video": {
319
+ "robot0_eye_in_hand": {
320
+ "resolution": [
321
+ 256,
322
+ 256
323
+ ],
324
+ "channels": 3,
325
+ "fps": 20.0
326
+ },
327
+ "robot0_agentview_left": {
328
+ "resolution": [
329
+ 256,
330
+ 256
331
+ ],
332
+ "channels": 3,
333
+ "fps": 20.0
334
+ },
335
+ "robot0_agentview_right": {
336
+ "resolution": [
337
+ 256,
338
+ 256
339
+ ],
340
+ "channels": 3,
341
+ "fps": 20.0
342
+ }
343
+ },
344
+ "state": {
345
+ "base_position": {
346
+ "absolute": true,
347
+ "rotation_type": null,
348
+ "shape": [
349
+ 3
350
+ ],
351
+ "continuous": true
352
+ },
353
+ "base_rotation": {
354
+ "absolute": true,
355
+ "rotation_type": "quaternion",
356
+ "shape": [
357
+ 4
358
+ ],
359
+ "continuous": true
360
+ },
361
+ "end_effector_position_relative": {
362
+ "absolute": true,
363
+ "rotation_type": null,
364
+ "shape": [
365
+ 3
366
+ ],
367
+ "continuous": true
368
+ },
369
+ "end_effector_rotation_relative": {
370
+ "absolute": true,
371
+ "rotation_type": "quaternion",
372
+ "shape": [
373
+ 4
374
+ ],
375
+ "continuous": true
376
+ },
377
+ "gripper_qpos": {
378
+ "absolute": true,
379
+ "rotation_type": null,
380
+ "shape": [
381
+ 2
382
+ ],
383
+ "continuous": true
384
+ }
385
+ },
386
+ "action": {
387
+ "base_motion": {
388
+ "absolute": true,
389
+ "rotation_type": null,
390
+ "shape": [
391
+ 4
392
+ ],
393
+ "continuous": true
394
+ },
395
+ "control_mode": {
396
+ "absolute": true,
397
+ "rotation_type": null,
398
+ "shape": [
399
+ 1
400
+ ],
401
+ "continuous": true
402
+ },
403
+ "end_effector_position": {
404
+ "absolute": true,
405
+ "rotation_type": null,
406
+ "shape": [
407
+ 3
408
+ ],
409
+ "continuous": true
410
+ },
411
+ "end_effector_rotation": {
412
+ "absolute": true,
413
+ "rotation_type": "axis_angle",
414
+ "shape": [
415
+ 3
416
+ ],
417
+ "continuous": true
418
+ },
419
+ "gripper_close": {
420
+ "absolute": true,
421
+ "rotation_type": null,
422
+ "shape": [
423
+ 1
424
+ ],
425
+ "continuous": true
426
+ }
427
+ }
428
+ },
429
+ "embodiment_tag": "new_embodiment"
430
+ }
431
+ }
MGRKD/mgrkd_distance_angle_flatten_raw_action/default/phase2/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
MGRKD/mgrkd_distance_angle_flatten_raw_action/default/phase3/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,431 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "base_position": {
6
+ "max": [
7
+ 7.3139495849609375,
8
+ 0.4876587688922882,
9
+ 0.7196521759033203
10
+ ],
11
+ "min": [
12
+ -4.94293737411499,
13
+ -6.890198230743408,
14
+ 0.6996827125549316
15
+ ],
16
+ "mean": [
17
+ 2.749288365724937,
18
+ -2.0455216239909983,
19
+ 0.700532551020533
20
+ ],
21
+ "std": [
22
+ 1.6786295897046262,
23
+ 1.312897969434244,
24
+ 0.00133751758906258
25
+ ],
26
+ "q01": [
27
+ -1.4241907881148037,
28
+ -5.1716547935950885,
29
+ 0.6999670898768614
30
+ ],
31
+ "q99": [
32
+ 6.198111545831555,
33
+ -0.5873531334104489,
34
+ 0.7038833014242587
35
+ ]
36
+ },
37
+ "base_rotation": {
38
+ "max": [
39
+ 0.0,
40
+ 0.0,
41
+ 1.0,
42
+ 1.0
43
+ ],
44
+ "min": [
45
+ 0.0,
46
+ 0.0,
47
+ -1.0,
48
+ 0.0
49
+ ],
50
+ "mean": [
51
+ 0.0,
52
+ 0.0,
53
+ 0.23084999587450247,
54
+ 0.6238431088581847
55
+ ],
56
+ "std": [
57
+ 0.0,
58
+ 0.0,
59
+ 0.6556404371644047,
60
+ 0.3572252105732042
61
+ ],
62
+ "q01": [
63
+ 0.0,
64
+ 0.0,
65
+ -0.9999999999999999,
66
+ 1.7482354593273349e-06
67
+ ],
68
+ "q99": [
69
+ 0.0,
70
+ 0.0,
71
+ 0.9999999999999999,
72
+ 0.9999999999999999
73
+ ]
74
+ },
75
+ "end_effector_position_relative": {
76
+ "max": [
77
+ 0.9014528393745422,
78
+ 0.8003877401351929,
79
+ 0.9829942584037781
80
+ ],
81
+ "min": [
82
+ -0.37471598386764526,
83
+ -0.8472502827644348,
84
+ -0.25070279836654663
85
+ ],
86
+ "mean": [
87
+ 0.28667332601863704,
88
+ -0.038315473170876704,
89
+ 0.45974658906777444
90
+ ],
91
+ "std": [
92
+ 0.17067058828305154,
93
+ 0.23569952037784195,
94
+ 0.21950702354181487
95
+ ],
96
+ "q01": [
97
+ 0.015172948963784228,
98
+ -0.44450502063220615,
99
+ 0.23314444103329338
100
+ ],
101
+ "q99": [
102
+ 0.5609069689572412,
103
+ 0.38454179128009547,
104
+ 0.7028102736473852
105
+ ]
106
+ },
107
+ "end_effector_rotation_relative": {
108
+ "max": [
109
+ 0.9999998807907104,
110
+ 0.9984455108642578,
111
+ 0.9434727430343628,
112
+ 0.9062229990959167
113
+ ],
114
+ "min": [
115
+ -0.9999930262565613,
116
+ -0.9988337755203247,
117
+ -0.9618149995803833,
118
+ 2.1872274658107926e-07
119
+ ],
120
+ "mean": [
121
+ -0.25672297401336,
122
+ 0.024425335806687216,
123
+ -0.08740977164483847,
124
+ 0.16608739644018397
125
+ ],
126
+ "std": [
127
+ 0.7932585208392383,
128
+ 0.3102260195580416,
129
+ 0.3772472576315148,
130
+ 0.17451959300158773
131
+ ],
132
+ "q01": [
133
+ -0.99379006987882,
134
+ -0.5478102050038828,
135
+ -0.6101777952971636,
136
+ 0.002274085014134748
137
+ ],
138
+ "q99": [
139
+ 0.8804856038896288,
140
+ 0.5910611092873037,
141
+ 0.5216058042993562,
142
+ 0.5231965974012255
143
+ ]
144
+ },
145
+ "gripper_qpos": {
146
+ "max": [
147
+ 0.055869169533252716,
148
+ 0.010916369967162609
149
+ ],
150
+ "min": [
151
+ -0.011436971835792065,
152
+ -0.05664053186774254
153
+ ],
154
+ "mean": [
155
+ 0.031651551516052555,
156
+ -0.03161481387920421
157
+ ],
158
+ "std": [
159
+ 0.013119769222217526,
160
+ 0.01306078225192402
161
+ ],
162
+ "q01": [
163
+ 0.0060999246446777336,
164
+ -0.04062601653964198
165
+ ],
166
+ "q99": [
167
+ 0.04054891621517283,
168
+ -0.0059163567113456345
169
+ ]
170
+ }
171
+ },
172
+ "action": {
173
+ "base_motion": {
174
+ "max": [
175
+ 1.0,
176
+ 1.0,
177
+ 1.0,
178
+ 0.0
179
+ ],
180
+ "min": [
181
+ -1.0,
182
+ -1.0,
183
+ -1.0,
184
+ 0.0
185
+ ],
186
+ "mean": [
187
+ 0.008144896753718373,
188
+ -0.00018893135146078263,
189
+ -0.0008739062727221845,
190
+ 0.0
191
+ ],
192
+ "std": [
193
+ 0.11030045424364411,
194
+ 0.10148082570313594,
195
+ 0.08975698373467506,
196
+ 0.0
197
+ ],
198
+ "q01": [
199
+ -0.07287373067581518,
200
+ -0.09446994199569948,
201
+ -0.07702818259249572,
202
+ 0.0
203
+ ],
204
+ "q99": [
205
+ 0.04485485414407898,
206
+ 0.09626941605965177,
207
+ 0.07610665906122742,
208
+ 0.0
209
+ ]
210
+ },
211
+ "control_mode": {
212
+ "max": [
213
+ 1.0
214
+ ],
215
+ "min": [
216
+ -1.0
217
+ ],
218
+ "mean": [
219
+ -0.9221974313425939
220
+ ],
221
+ "std": [
222
+ 0.38670194342612674
223
+ ],
224
+ "q01": [
225
+ -1.0
226
+ ],
227
+ "q99": [
228
+ -0.384693946883099
229
+ ]
230
+ },
231
+ "end_effector_position": {
232
+ "max": [
233
+ 1.0,
234
+ 1.0,
235
+ 1.0
236
+ ],
237
+ "min": [
238
+ -1.0,
239
+ -1.0,
240
+ -1.0
241
+ ],
242
+ "mean": [
243
+ 0.01976129923017491,
244
+ -0.020870631344490034,
245
+ -0.05993476517349181
246
+ ],
247
+ "std": [
248
+ 0.4347024990804053,
249
+ 0.4221662976569997,
250
+ 0.3845506843044066
251
+ ],
252
+ "q01": [
253
+ -0.8835792939767807,
254
+ -0.9280919251713778,
255
+ -0.783361071470956
256
+ ],
257
+ "q99": [
258
+ 0.7622084438449307,
259
+ 0.8625125252609077,
260
+ 0.7470974753250706
261
+ ]
262
+ },
263
+ "end_effector_rotation": {
264
+ "max": [
265
+ 1.0,
266
+ 1.0,
267
+ 1.0
268
+ ],
269
+ "min": [
270
+ -1.0,
271
+ -1.0,
272
+ -1.0
273
+ ],
274
+ "mean": [
275
+ 0.008089673068544335,
276
+ -0.026498254627927348,
277
+ 0.0029083551743059005
278
+ ],
279
+ "std": [
280
+ 0.11216734503733773,
281
+ 0.13206002613798629,
282
+ 0.12615870412692864
283
+ ],
284
+ "q01": [
285
+ -0.2814627972846672,
286
+ -0.4342386516115439,
287
+ -0.34489755628340574
288
+ ],
289
+ "q99": [
290
+ 0.31879353266939386,
291
+ 0.28411797683220563,
292
+ 0.3597718671669894
293
+ ]
294
+ },
295
+ "gripper_close": {
296
+ "max": [
297
+ 1.0
298
+ ],
299
+ "min": [
300
+ -1.0
301
+ ],
302
+ "mean": [
303
+ -0.3824228874769045
304
+ ],
305
+ "std": [
306
+ 0.9240015096469786
307
+ ],
308
+ "q01": [
309
+ -1.0
310
+ ],
311
+ "q99": [
312
+ 0.5597272305813592
313
+ ]
314
+ }
315
+ }
316
+ },
317
+ "modalities": {
318
+ "video": {
319
+ "robot0_eye_in_hand": {
320
+ "resolution": [
321
+ 256,
322
+ 256
323
+ ],
324
+ "channels": 3,
325
+ "fps": 20.0
326
+ },
327
+ "robot0_agentview_left": {
328
+ "resolution": [
329
+ 256,
330
+ 256
331
+ ],
332
+ "channels": 3,
333
+ "fps": 20.0
334
+ },
335
+ "robot0_agentview_right": {
336
+ "resolution": [
337
+ 256,
338
+ 256
339
+ ],
340
+ "channels": 3,
341
+ "fps": 20.0
342
+ }
343
+ },
344
+ "state": {
345
+ "base_position": {
346
+ "absolute": true,
347
+ "rotation_type": null,
348
+ "shape": [
349
+ 3
350
+ ],
351
+ "continuous": true
352
+ },
353
+ "base_rotation": {
354
+ "absolute": true,
355
+ "rotation_type": "quaternion",
356
+ "shape": [
357
+ 4
358
+ ],
359
+ "continuous": true
360
+ },
361
+ "end_effector_position_relative": {
362
+ "absolute": true,
363
+ "rotation_type": null,
364
+ "shape": [
365
+ 3
366
+ ],
367
+ "continuous": true
368
+ },
369
+ "end_effector_rotation_relative": {
370
+ "absolute": true,
371
+ "rotation_type": "quaternion",
372
+ "shape": [
373
+ 4
374
+ ],
375
+ "continuous": true
376
+ },
377
+ "gripper_qpos": {
378
+ "absolute": true,
379
+ "rotation_type": null,
380
+ "shape": [
381
+ 2
382
+ ],
383
+ "continuous": true
384
+ }
385
+ },
386
+ "action": {
387
+ "base_motion": {
388
+ "absolute": true,
389
+ "rotation_type": null,
390
+ "shape": [
391
+ 4
392
+ ],
393
+ "continuous": true
394
+ },
395
+ "control_mode": {
396
+ "absolute": true,
397
+ "rotation_type": null,
398
+ "shape": [
399
+ 1
400
+ ],
401
+ "continuous": true
402
+ },
403
+ "end_effector_position": {
404
+ "absolute": true,
405
+ "rotation_type": null,
406
+ "shape": [
407
+ 3
408
+ ],
409
+ "continuous": true
410
+ },
411
+ "end_effector_rotation": {
412
+ "absolute": true,
413
+ "rotation_type": "axis_angle",
414
+ "shape": [
415
+ 3
416
+ ],
417
+ "continuous": true
418
+ },
419
+ "gripper_close": {
420
+ "absolute": true,
421
+ "rotation_type": null,
422
+ "shape": [
423
+ 1
424
+ ],
425
+ "continuous": true
426
+ }
427
+ }
428
+ },
429
+ "embodiment_tag": "new_embodiment"
430
+ }
431
+ }
MGRKD/mgrkd_distance_angle_flatten_raw_action/default/resolved_config.yaml ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: mgrkd_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_MGRKD
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/MGRKD/mgrkd_distance_angle_flatten_raw_action
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 1
9
+ batch_size: 128
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_mgrkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: raw_action
29
+ rkd_raw_action_projector_enabled: true
30
+ rkd_raw_action_projector_dim: 512
31
+ rkd_raw_action_projector_pooling: flatten
32
+ rkd_student_reconstruction_head_enabled: true
33
+ rkd_student_reconstruction_head_dim: 512
34
+ rkd_student_reconstruction_head_hidden_dim: 512
35
+ rkd_distance_loss_weight: 1.0
36
+ rkd_angle_loss_weight: 2.0
37
+ rkd_exclude_diagonal: true
38
+ - name: phase3_fm_mgrkd_da_fixed_0p5
39
+ max_steps: 30000
40
+ save_steps: 0
41
+ trainable:
42
+ tune_llm: false
43
+ tune_visual: false
44
+ tune_projector: true
45
+ tune_diffusion_model: true
46
+ losses:
47
+ rkd_enabled: true
48
+ rkd_fm_loss_weight: 1.0
49
+ rkd_loss_weight: 0.5
50
+ rkd_relation_mode: flatten
51
+ rkd_loss_type: distance_angle
52
+ rkd_teacher_source: raw_action
53
+ rkd_raw_action_projector_enabled: true
54
+ rkd_raw_action_projector_dim: 512
55
+ rkd_raw_action_projector_pooling: flatten
56
+ rkd_student_reconstruction_head_enabled: true
57
+ rkd_student_reconstruction_head_dim: 512
58
+ rkd_student_reconstruction_head_hidden_dim: 512
59
+ rkd_distance_loss_weight: 1.0
60
+ rkd_angle_loss_weight: 2.0
61
+ rkd_exclude_diagonal: true
62
+ resolved_sweep: {}
rkd_v2_1/rkd_v2_1_distance_angle_full_120k_cosine_decay/default/full_fm_rkd_da_cosine_decay_120k/checkpoint-120000/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,431 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "base_position": {
6
+ "max": [
7
+ 7.3139495849609375,
8
+ 0.4876587688922882,
9
+ 0.7196521759033203
10
+ ],
11
+ "min": [
12
+ -4.94293737411499,
13
+ -6.890198230743408,
14
+ 0.6996827125549316
15
+ ],
16
+ "mean": [
17
+ 2.749288365724937,
18
+ -2.0455216239909983,
19
+ 0.700532551020533
20
+ ],
21
+ "std": [
22
+ 1.6786295897046262,
23
+ 1.312897969434244,
24
+ 0.00133751758906258
25
+ ],
26
+ "q01": [
27
+ -1.4241907881148037,
28
+ -5.1716547935950885,
29
+ 0.6999670898768614
30
+ ],
31
+ "q99": [
32
+ 6.198111545831555,
33
+ -0.5873531334104489,
34
+ 0.7038833014242587
35
+ ]
36
+ },
37
+ "base_rotation": {
38
+ "max": [
39
+ 0.0,
40
+ 0.0,
41
+ 1.0,
42
+ 1.0
43
+ ],
44
+ "min": [
45
+ 0.0,
46
+ 0.0,
47
+ -1.0,
48
+ 0.0
49
+ ],
50
+ "mean": [
51
+ 0.0,
52
+ 0.0,
53
+ 0.23084999587450247,
54
+ 0.6238431088581847
55
+ ],
56
+ "std": [
57
+ 0.0,
58
+ 0.0,
59
+ 0.6556404371644047,
60
+ 0.3572252105732042
61
+ ],
62
+ "q01": [
63
+ 0.0,
64
+ 0.0,
65
+ -0.9999999999999999,
66
+ 1.7482354593273349e-06
67
+ ],
68
+ "q99": [
69
+ 0.0,
70
+ 0.0,
71
+ 0.9999999999999999,
72
+ 0.9999999999999999
73
+ ]
74
+ },
75
+ "end_effector_position_relative": {
76
+ "max": [
77
+ 0.9014528393745422,
78
+ 0.8003877401351929,
79
+ 0.9829942584037781
80
+ ],
81
+ "min": [
82
+ -0.37471598386764526,
83
+ -0.8472502827644348,
84
+ -0.25070279836654663
85
+ ],
86
+ "mean": [
87
+ 0.28667332601863704,
88
+ -0.038315473170876704,
89
+ 0.45974658906777444
90
+ ],
91
+ "std": [
92
+ 0.17067058828305154,
93
+ 0.23569952037784195,
94
+ 0.21950702354181487
95
+ ],
96
+ "q01": [
97
+ 0.015172948963784228,
98
+ -0.44450502063220615,
99
+ 0.23314444103329338
100
+ ],
101
+ "q99": [
102
+ 0.5609069689572412,
103
+ 0.38454179128009547,
104
+ 0.7028102736473852
105
+ ]
106
+ },
107
+ "end_effector_rotation_relative": {
108
+ "max": [
109
+ 0.9999998807907104,
110
+ 0.9984455108642578,
111
+ 0.9434727430343628,
112
+ 0.9062229990959167
113
+ ],
114
+ "min": [
115
+ -0.9999930262565613,
116
+ -0.9988337755203247,
117
+ -0.9618149995803833,
118
+ 2.1872274658107926e-07
119
+ ],
120
+ "mean": [
121
+ -0.25672297401336,
122
+ 0.024425335806687216,
123
+ -0.08740977164483847,
124
+ 0.16608739644018397
125
+ ],
126
+ "std": [
127
+ 0.7932585208392383,
128
+ 0.3102260195580416,
129
+ 0.3772472576315148,
130
+ 0.17451959300158773
131
+ ],
132
+ "q01": [
133
+ -0.99379006987882,
134
+ -0.5478102050038828,
135
+ -0.6101777952971636,
136
+ 0.002274085014134748
137
+ ],
138
+ "q99": [
139
+ 0.8804856038896288,
140
+ 0.5910611092873037,
141
+ 0.5216058042993562,
142
+ 0.5231965974012255
143
+ ]
144
+ },
145
+ "gripper_qpos": {
146
+ "max": [
147
+ 0.055869169533252716,
148
+ 0.010916369967162609
149
+ ],
150
+ "min": [
151
+ -0.011436971835792065,
152
+ -0.05664053186774254
153
+ ],
154
+ "mean": [
155
+ 0.031651551516052555,
156
+ -0.03161481387920421
157
+ ],
158
+ "std": [
159
+ 0.013119769222217526,
160
+ 0.01306078225192402
161
+ ],
162
+ "q01": [
163
+ 0.0060999246446777336,
164
+ -0.04062601653964198
165
+ ],
166
+ "q99": [
167
+ 0.04054891621517283,
168
+ -0.0059163567113456345
169
+ ]
170
+ }
171
+ },
172
+ "action": {
173
+ "base_motion": {
174
+ "max": [
175
+ 1.0,
176
+ 1.0,
177
+ 1.0,
178
+ 0.0
179
+ ],
180
+ "min": [
181
+ -1.0,
182
+ -1.0,
183
+ -1.0,
184
+ 0.0
185
+ ],
186
+ "mean": [
187
+ 0.008144896753718373,
188
+ -0.00018893135146078263,
189
+ -0.0008739062727221845,
190
+ 0.0
191
+ ],
192
+ "std": [
193
+ 0.11030045424364411,
194
+ 0.10148082570313594,
195
+ 0.08975698373467506,
196
+ 0.0
197
+ ],
198
+ "q01": [
199
+ -0.07287373067581518,
200
+ -0.09446994199569948,
201
+ -0.07702818259249572,
202
+ 0.0
203
+ ],
204
+ "q99": [
205
+ 0.04485485414407898,
206
+ 0.09626941605965177,
207
+ 0.07610665906122742,
208
+ 0.0
209
+ ]
210
+ },
211
+ "control_mode": {
212
+ "max": [
213
+ 1.0
214
+ ],
215
+ "min": [
216
+ -1.0
217
+ ],
218
+ "mean": [
219
+ -0.9221974313425939
220
+ ],
221
+ "std": [
222
+ 0.38670194342612674
223
+ ],
224
+ "q01": [
225
+ -1.0
226
+ ],
227
+ "q99": [
228
+ -0.384693946883099
229
+ ]
230
+ },
231
+ "end_effector_position": {
232
+ "max": [
233
+ 1.0,
234
+ 1.0,
235
+ 1.0
236
+ ],
237
+ "min": [
238
+ -1.0,
239
+ -1.0,
240
+ -1.0
241
+ ],
242
+ "mean": [
243
+ 0.01976129923017491,
244
+ -0.020870631344490034,
245
+ -0.05993476517349181
246
+ ],
247
+ "std": [
248
+ 0.4347024990804053,
249
+ 0.4221662976569997,
250
+ 0.3845506843044066
251
+ ],
252
+ "q01": [
253
+ -0.8835792939767807,
254
+ -0.9280919251713778,
255
+ -0.783361071470956
256
+ ],
257
+ "q99": [
258
+ 0.7622084438449307,
259
+ 0.8625125252609077,
260
+ 0.7470974753250706
261
+ ]
262
+ },
263
+ "end_effector_rotation": {
264
+ "max": [
265
+ 1.0,
266
+ 1.0,
267
+ 1.0
268
+ ],
269
+ "min": [
270
+ -1.0,
271
+ -1.0,
272
+ -1.0
273
+ ],
274
+ "mean": [
275
+ 0.008089673068544335,
276
+ -0.026498254627927348,
277
+ 0.0029083551743059005
278
+ ],
279
+ "std": [
280
+ 0.11216734503733773,
281
+ 0.13206002613798629,
282
+ 0.12615870412692864
283
+ ],
284
+ "q01": [
285
+ -0.2814627972846672,
286
+ -0.4342386516115439,
287
+ -0.34489755628340574
288
+ ],
289
+ "q99": [
290
+ 0.31879353266939386,
291
+ 0.28411797683220563,
292
+ 0.3597718671669894
293
+ ]
294
+ },
295
+ "gripper_close": {
296
+ "max": [
297
+ 1.0
298
+ ],
299
+ "min": [
300
+ -1.0
301
+ ],
302
+ "mean": [
303
+ -0.3824228874769045
304
+ ],
305
+ "std": [
306
+ 0.9240015096469786
307
+ ],
308
+ "q01": [
309
+ -1.0
310
+ ],
311
+ "q99": [
312
+ 0.5597272305813592
313
+ ]
314
+ }
315
+ }
316
+ },
317
+ "modalities": {
318
+ "video": {
319
+ "robot0_eye_in_hand": {
320
+ "resolution": [
321
+ 256,
322
+ 256
323
+ ],
324
+ "channels": 3,
325
+ "fps": 20.0
326
+ },
327
+ "robot0_agentview_left": {
328
+ "resolution": [
329
+ 256,
330
+ 256
331
+ ],
332
+ "channels": 3,
333
+ "fps": 20.0
334
+ },
335
+ "robot0_agentview_right": {
336
+ "resolution": [
337
+ 256,
338
+ 256
339
+ ],
340
+ "channels": 3,
341
+ "fps": 20.0
342
+ }
343
+ },
344
+ "state": {
345
+ "base_position": {
346
+ "absolute": true,
347
+ "rotation_type": null,
348
+ "shape": [
349
+ 3
350
+ ],
351
+ "continuous": true
352
+ },
353
+ "base_rotation": {
354
+ "absolute": true,
355
+ "rotation_type": "quaternion",
356
+ "shape": [
357
+ 4
358
+ ],
359
+ "continuous": true
360
+ },
361
+ "end_effector_position_relative": {
362
+ "absolute": true,
363
+ "rotation_type": null,
364
+ "shape": [
365
+ 3
366
+ ],
367
+ "continuous": true
368
+ },
369
+ "end_effector_rotation_relative": {
370
+ "absolute": true,
371
+ "rotation_type": "quaternion",
372
+ "shape": [
373
+ 4
374
+ ],
375
+ "continuous": true
376
+ },
377
+ "gripper_qpos": {
378
+ "absolute": true,
379
+ "rotation_type": null,
380
+ "shape": [
381
+ 2
382
+ ],
383
+ "continuous": true
384
+ }
385
+ },
386
+ "action": {
387
+ "base_motion": {
388
+ "absolute": true,
389
+ "rotation_type": null,
390
+ "shape": [
391
+ 4
392
+ ],
393
+ "continuous": true
394
+ },
395
+ "control_mode": {
396
+ "absolute": true,
397
+ "rotation_type": null,
398
+ "shape": [
399
+ 1
400
+ ],
401
+ "continuous": true
402
+ },
403
+ "end_effector_position": {
404
+ "absolute": true,
405
+ "rotation_type": null,
406
+ "shape": [
407
+ 3
408
+ ],
409
+ "continuous": true
410
+ },
411
+ "end_effector_rotation": {
412
+ "absolute": true,
413
+ "rotation_type": "axis_angle",
414
+ "shape": [
415
+ 3
416
+ ],
417
+ "continuous": true
418
+ },
419
+ "gripper_close": {
420
+ "absolute": true,
421
+ "rotation_type": null,
422
+ "shape": [
423
+ 1
424
+ ],
425
+ "continuous": true
426
+ }
427
+ }
428
+ },
429
+ "embodiment_tag": "new_embodiment"
430
+ }
431
+ }
rkd_v2_1/rkd_v2_1_distance_angle_full_120k_cosine_decay/default/full_fm_rkd_da_cosine_decay_120k/checkpoint-120000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
rkd_v2_1/rkd_v2_1_distance_angle_full_120k_cosine_decay/default/full_fm_rkd_da_cosine_decay_120k/config.json ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_dim": 32,
3
+ "action_head_cfg": {
4
+ "action_dim": 32,
5
+ "action_horizon": 16,
6
+ "add_pos_embed": true,
7
+ "backbone_embedding_dim": 2048,
8
+ "diffusion_model_cfg": {
9
+ "attention_head_dim": 48,
10
+ "cross_attention_dim": 2048,
11
+ "dropout": 0.2,
12
+ "final_dropout": true,
13
+ "interleave_self_attention": true,
14
+ "norm_type": "ada_norm",
15
+ "num_attention_heads": 32,
16
+ "num_layers": 16,
17
+ "output_dim": 1024,
18
+ "positional_embeddings": null
19
+ },
20
+ "hidden_size": 1024,
21
+ "input_embedding_dim": 1536,
22
+ "max_action_dim": 32,
23
+ "max_state_dim": 64,
24
+ "model_dtype": "float32",
25
+ "noise_beta_alpha": 1.5,
26
+ "noise_beta_beta": 1.0,
27
+ "noise_s": 0.999,
28
+ "num_inference_timesteps": 4,
29
+ "num_target_vision_tokens": 32,
30
+ "num_timestep_buckets": 1000,
31
+ "tune_diffusion_model": true,
32
+ "tune_projector": true,
33
+ "use_vlln": true,
34
+ "vl_self_attention_cfg": {
35
+ "attention_head_dim": 64,
36
+ "dropout": 0.2,
37
+ "final_dropout": true,
38
+ "num_attention_heads": 32,
39
+ "num_layers": 4,
40
+ "positional_embeddings": null
41
+ }
42
+ },
43
+ "action_horizon": 16,
44
+ "architectures": [
45
+ "GR00T_N1_5_RKD"
46
+ ],
47
+ "attn_implementation": null,
48
+ "backbone_cfg": {
49
+ "eagle_path": "NVEagle/eagle_er-qwen3_1_7B-Siglip2_400M_stage1_5_128gpu_er_v7_1mlp_nops",
50
+ "load_bf16": false,
51
+ "project_to_dim": null,
52
+ "reproject_vision": false,
53
+ "select_layer": 12,
54
+ "tune_llm": false,
55
+ "tune_visual": true,
56
+ "use_flash_attention": true
57
+ },
58
+ "compute_dtype": "bfloat16",
59
+ "hidden_size": 2048,
60
+ "model_dtype": "float32",
61
+ "model_type": "gr00t_n1_5",
62
+ "rkd_action_temp": 0.1,
63
+ "rkd_angle_loss_weight": 2.0,
64
+ "rkd_distance_loss_weight": 1.0,
65
+ "rkd_enabled": true,
66
+ "rkd_exclude_diagonal": true,
67
+ "rkd_fm_loss_weight": 1.0,
68
+ "rkd_loss_type": "distance_angle",
69
+ "rkd_loss_weight": 0.0,
70
+ "rkd_loss_weight_end": 0.0,
71
+ "rkd_loss_weight_schedule": "cosine_decay",
72
+ "rkd_loss_weight_start": 1.0,
73
+ "rkd_relation_mode": "token_pair_mean",
74
+ "rkd_vlm_temp": 0.07,
75
+ "torch_dtype": "bfloat16",
76
+ "transformers_version": "4.51.3"
77
+ }
rkd_v2_1/rkd_v2_1_distance_angle_full_120k_cosine_decay/default/full_fm_rkd_da_cosine_decay_120k/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
rkd_v2_2/launch_logs/action_encoder_flatten_gpu4_5_bs32_ngpu2_20260622_124039.log ADDED
@@ -0,0 +1,757 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s][rank1]: Traceback (most recent call last):
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [robosuite WARNING] No private macro file found! (macros.py:57)
2
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
3
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
4
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
5
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
6
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
7
+ check_for_updates()
8
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
9
+
10
+ *****************************************
11
+ Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ *****************************************
13
+ [robosuite WARNING] No private macro file found! (macros.py:57)
14
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
15
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
16
+ [robosuite WARNING] No private macro file found! (macros.py:57)
17
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
18
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
19
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
20
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
21
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
22
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
23
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
24
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
25
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
26
+ check_for_updates()
27
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
28
+ check_for_updates()
29
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
30
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
31
+
32
+ ==================================================
33
+ GR00T FINE-TUNING CONFIGURATION:
34
+ ==================================================
35
+ config: experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
36
+ dataset_soup: None
37
+ output_dir: /tmp/gr00t
38
+ output_root: None
39
+ data_config: panda_omron
40
+ batch_size: 32
41
+ max_steps: 300000
42
+ num_gpus: 2
43
+ save_steps: 20000
44
+ run_name: None
45
+ save_total_limit: 100
46
+ seed: 42
47
+ base_model_path: nvidia/GR00T-N1.5-3B
48
+ tune_llm: False
49
+ tune_visual: False
50
+ tune_projector: True
51
+ tune_diffusion_model: True
52
+ resume: False
53
+ learning_rate: 3e-05
54
+ weight_decay: 1e-05
55
+ warmup_ratio: 0.05
56
+ lora_rank: 0
57
+ lora_alpha: 16
58
+ lora_dropout: 0.1
59
+ lora_full_model: False
60
+ dataloader_num_workers: 8
61
+ report_to: wandb
62
+ embodiment_tag: new_embodiment
63
+ video_backend: opencv
64
+ balance_dataset_weights: True
65
+ balance_trajectory_weights: True
66
+ ds_weights_alpha: 0.4
67
+ ==================================================
68
+
69
+ Using 2 GPUs
70
+
71
+ ================================================================================
72
+ Starting sweep branch: default
73
+ Sweep vars: {}
74
+ ================================================================================
75
+
76
+ --------------------------------------------------------------------------------
77
+ Running phase 1: phase2_rkd_da_only
78
+ Policy type: groot_rkd_v2
79
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
80
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
81
+ Trainable preset: processing_line_only
82
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
83
+ --------------------------------------------------------------------------------
84
+
85
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
86
+ Using 100 subset demos for filter_key: 100_demos
87
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
88
+ self.statistics[key] = torch.tensor(value)
89
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
90
+ Using 100 subset demos for filter_key: 100_demos
91
+
92
+ ==================================================
93
+ GR00T FINE-TUNING CONFIGURATION:
94
+ ==================================================
95
+ config: experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
96
+ dataset_soup: None
97
+ output_dir: /tmp/gr00t
98
+ output_root: None
99
+ data_config: panda_omron
100
+ batch_size: 32
101
+ max_steps: 300000
102
+ num_gpus: 2
103
+ save_steps: 20000
104
+ run_name: None
105
+ save_total_limit: 100
106
+ seed: 42
107
+ base_model_path: nvidia/GR00T-N1.5-3B
108
+ tune_llm: False
109
+ tune_visual: False
110
+ tune_projector: True
111
+ tune_diffusion_model: True
112
+ resume: False
113
+ learning_rate: 3e-05
114
+ weight_decay: 1e-05
115
+ warmup_ratio: 0.05
116
+ lora_rank: 0
117
+ lora_alpha: 16
118
+ lora_dropout: 0.1
119
+ lora_full_model: False
120
+ dataloader_num_workers: 8
121
+ report_to: wandb
122
+ embodiment_tag: new_embodiment
123
+ video_backend: opencv
124
+ balance_dataset_weights: True
125
+ balance_trajectory_weights: True
126
+ ds_weights_alpha: 0.4
127
+ ==================================================
128
+
129
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
130
+ Using 100 subset demos for filter_key: 100_demos
131
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
132
+ Using 100 subset demos for filter_key: 100_demos
133
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
134
+ Using 100 subset demos for filter_key: 100_demos
135
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
136
+ Using 100 subset demos for filter_key: 100_demos
137
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
138
+ Using 100 subset demos for filter_key: 100_demos
139
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
140
+ Using 100 subset demos for filter_key: 100_demos
141
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
142
+ Using 100 subset demos for filter_key: 100_demos
143
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
144
+ Using 100 subset demos for filter_key: 100_demos
145
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
146
+ Using 100 subset demos for filter_key: 100_demos
147
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
148
+ Using 100 subset demos for filter_key: 100_demos
149
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
150
+ Using 100 subset demos for filter_key: 100_demos
151
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
152
+ Using 100 subset demos for filter_key: 100_demos
153
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
154
+ Using 100 subset demos for filter_key: 100_demos
155
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
156
+ Using 100 subset demos for filter_key: 100_demos
157
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
158
+ Using 100 subset demos for filter_key: 100_demos
159
+ Using 2 GPUs
160
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
161
+ Using 100 subset demos for filter_key: 100_demos
162
+
163
+ ================================================================================
164
+ Starting sweep branch: default
165
+ Sweep vars: {}
166
+ ================================================================================
167
+
168
+ --------------------------------------------------------------------------------
169
+ Running phase 1: phase2_rkd_da_only
170
+ Policy type: groot_rkd_v2
171
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
172
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
173
+ Trainable preset: processing_line_only
174
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
175
+ --------------------------------------------------------------------------------
176
+
177
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
178
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
179
+ Using 100 subset demos for filter_key: 100_demos
180
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
181
+ Using 100 subset demos for filter_key: 100_demos
182
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
183
+ self.statistics[key] = torch.tensor(value)
184
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
185
+ Using 100 subset demos for filter_key: 100_demos
186
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
187
+ Using 100 subset demos for filter_key: 100_demos
188
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
189
+ Using 100 subset demos for filter_key: 100_demos
190
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
191
+ Using 100 subset demos for filter_key: 100_demos
192
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
193
+ Using 100 subset demos for filter_key: 100_demos
194
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
195
+ Using 100 subset demos for filter_key: 100_demos
196
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
197
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
198
+ Using 100 subset demos for filter_key: 100_demos
199
+ Using 100 subset demos for filter_key: 100_demos
200
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
201
+ Using 100 subset demos for filter_key: 100_demos
202
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
203
+ Using 100 subset demos for filter_key: 100_demos
204
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
205
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
206
+ Using 100 subset demos for filter_key: 100_demos
207
+ Using 100 subset demos for filter_key: 100_demos
208
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENTInitialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
209
+
210
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
211
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
212
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
213
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
214
+ 0.75517122 0.7973985 ]
215
+ Using 100 subset demos for filter_key: 100_demos
216
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
217
+ Using 100 subset demos for filter_key: 100_demos
218
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
219
+ Using 100 subset demos for filter_key: 100_demos
220
+ Loaded 26 datasets
221
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
222
+ Using 100 subset demos for filter_key: 100_demos
223
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
224
+ Using 100 subset demos for filter_key: 100_demos
225
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
226
+ Using 100 subset demos for filter_key: 100_demos
227
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
228
+ Tune backbone vision tower: False
229
+ Tune backbone LLM: False
230
+ Tune action head projector: False
231
+ Tune action head DiT: False
232
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
233
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
234
+ Using 100 subset demos for filter_key: 100_demos
235
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
236
+ Using 100 subset demos for filter_key: 100_demos
237
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
238
+ Using 100 subset demos for filter_key: 100_demos
239
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
240
+ Using 100 subset demos for filter_key: 100_demos
241
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
242
+ Using 100 subset demos for filter_key: 100_demos
243
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
244
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
245
+ Using 100 subset demos for filter_key: 100_demos
246
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
247
+ Using 100 subset demos for filter_key: 100_demos
248
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
249
+ Using 100 subset demos for filter_key: 100_demos
250
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
251
+ Using 100 subset demos for filter_key: 100_demos
252
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
253
+ Using 100 subset demos for filter_key: 100_demos
254
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
255
+ Using 100 subset demos for filter_key: 100_demos
256
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
257
+ Using 100 subset demos for filter_key: 100_demos
258
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
259
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
260
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
261
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
262
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
263
+ 0.75517122 0.7973985 ]
264
+ Loaded 26 datasets
265
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
266
+ Tune backbone vision tower: False
267
+ Tune backbone LLM: False
268
+ Tune action head projector: False
269
+ Tune action head DiT: False
270
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
271
+ Tune backbone llm: False
272
+ Tune backbone visual: True
273
+ Total number of DiT parameters: 550386688
274
+ Tune backbone llm: False
275
+ Tune backbone visual: True
276
+ Total number of DiT parameters: 550386688
277
+ Total number of SelfAttentionTransformer parameters: 201433088
278
+ Tune action head projector: True
279
+ Tune action head diffusion model: True
280
+
281
+ Tune backbone llm: False
282
+ Tune backbone visual: False
283
+ Warning: No backbone trainable parameters found.
284
+ Tune action head projector: False
285
+ Tune action head diffusion model: False
286
+ Action head trainable parameter: future_tokens.weight
287
+ Action head trainable parameter: vlln.weight
288
+ Action head trainable parameter: vlln.bias
289
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
290
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
291
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
292
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
293
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
294
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
295
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
296
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
297
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
298
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
299
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
300
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
301
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
302
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
303
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
304
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
305
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
306
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
307
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
308
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
309
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
310
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
311
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
312
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
313
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
314
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
315
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
316
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
317
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
318
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
319
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
320
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
321
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
322
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
323
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
324
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
325
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
326
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
327
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
328
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
329
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
330
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
331
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
332
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
333
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
334
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
335
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
336
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
337
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
338
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
339
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
340
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
341
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
342
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
343
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
344
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
345
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
346
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
347
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
348
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
349
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
350
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
351
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
352
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
353
+ Applied trainable preset: processing_line_only
354
+ Trainable parameter tensors after preset: 66
355
+ trainable: action_head.vlln.weight
356
+ trainable: action_head.vlln.bias
357
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
358
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
359
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
360
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
361
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
362
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
363
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
364
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
365
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
366
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
367
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
368
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
369
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
370
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
371
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
372
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
373
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
374
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
375
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
376
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
377
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
378
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
379
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
380
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
381
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
382
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
383
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
384
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
385
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
386
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
387
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
388
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
389
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
390
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
391
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
392
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
393
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
394
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
395
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
396
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
397
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
398
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
399
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
400
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
401
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
402
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
403
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
404
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
405
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
406
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
407
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
408
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
409
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
410
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
411
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
412
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
413
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
414
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
415
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
416
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
417
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
418
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
419
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
420
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
421
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2724163520, 'trainable_param_ratio': 0.0739446007998815}
422
+ Total number of SelfAttentionTransformer parameters: 201433088
423
+ Tune action head projector: True
424
+ Tune action head diffusion model: True
425
+
426
+ Tune backbone llm: False
427
+ Tune backbone visual: False
428
+ Warning: No backbone trainable parameters found.
429
+ Tune action head projector: False
430
+ Tune action head diffusion model: False
431
+ Action head trainable parameter: future_tokens.weight
432
+ Action head trainable parameter: vlln.weight
433
+ Action head trainable parameter: vlln.bias
434
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
435
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
436
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
437
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
438
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
439
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
440
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
441
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
442
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
443
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
444
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
445
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
446
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
447
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
448
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
449
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
450
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
451
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
452
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
453
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
454
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
455
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
456
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
457
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
458
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
459
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
460
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
461
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
462
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
463
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
464
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
465
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
466
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
467
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
468
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
469
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
470
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
471
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
472
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
473
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
474
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
475
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
476
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
477
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
478
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
479
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
480
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
481
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
482
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
483
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
484
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
485
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
486
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
487
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
488
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
489
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
490
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
491
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
492
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
493
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
494
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
495
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
496
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
497
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
498
+ Applied trainable preset: processing_line_only
499
+ Trainable parameter tensors after preset: 66
500
+ trainable: action_head.vlln.weight
501
+ trainable: action_head.vlln.bias
502
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
503
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
504
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
505
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
506
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
507
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
508
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
509
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
510
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
511
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
512
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
513
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
514
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
515
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
516
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
517
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
518
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
519
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
520
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
521
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
522
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
523
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
524
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
525
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
526
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
527
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
528
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
529
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
530
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
531
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
532
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
533
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
534
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
535
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
536
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
537
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
538
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
539
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
540
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
541
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
542
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
543
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
544
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
545
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
546
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
547
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
548
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
549
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
550
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
551
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
552
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
553
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
554
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
555
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
556
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
557
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
558
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
559
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
560
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
561
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
562
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
563
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
564
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
565
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
566
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2724163520, 'trainable_param_ratio': 0.0739446007998815}
567
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
568
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
569
+ train dataloader length: 6873
570
+ train dataset length: 439854
571
+ GPU memory before training: 7.076685905456543 GB
572
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
573
+ train dataloader length: 6873
574
+ train dataset length: 439854
575
+ GPU memory before training: 7.076685905456543 GB
576
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
577
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
578
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
579
+ wandb: setting up run 8e8ggss5
580
+ wandb: Tracking run with wandb version 0.25.0
581
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_124109-8e8ggss5
582
+ wandb: Run `wandb offline` to turn off syncing.
583
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
584
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2
585
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2/runs/8e8ggss5
586
+
587
  0%| | 0/30000 [00:00<?, ?it/s][rank1]: Traceback (most recent call last):
588
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1039, in <module>
589
+ [rank1]: run_yaml_experiment(
590
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 988, in run_yaml_experiment
591
+ [rank1]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
592
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 718, in main
593
+ [rank1]: experiment.train()
594
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
595
+ [rank1]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
596
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
597
+ [rank1]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
598
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
599
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
600
+ [rank1]: return inner_training_loop(
601
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^
602
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
603
+ [rank1]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
604
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
605
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3782, in training_step
606
+ [rank1]: self.accelerator.backward(loss, **kwargs)
607
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/accelerator.py", line 2838, in backward
608
+ [rank1]: loss.backward(**kwargs)
609
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/_tensor.py", line 648, in backward
610
+ [rank1]: torch.autograd.backward(
611
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/__init__.py", line 353, in backward
612
+ [rank1]: _engine_run_backward(
613
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
614
+ [rank1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
615
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
616
+ [rank1]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.09 GiB. GPU 1 has a total capacity of 139.80 GiB of which 12.59 GiB is free. Process 1961377 has 11.53 GiB memory in use. Including non-PyTorch memory, this process has 102.42 GiB memory in use. Of the allocated memory 83.76 GiB is allocated by PyTorch, and 17.12 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
617
+ wandb: updating run metadata
618
+ wandb: uploading config.yaml
619
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2/runs/8e8ggss5
620
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2
621
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
622
+ wandb: Find logs at: ./wandb/run-20260622_124109-8e8ggss5/logs
623
+ Traceback (most recent call last):
624
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1039, in <module>
625
+ run_yaml_experiment(
626
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 988, in run_yaml_experiment
627
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
628
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 718, in main
629
+ experiment.train()
630
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
631
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
632
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
633
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
634
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
635
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
636
+ return inner_training_loop(
637
+ ^^^^^^^^^^^^^^^^^^^^
638
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
639
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
640
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
641
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3782, in training_step
642
+ self.accelerator.backward(loss, **kwargs)
643
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/accelerator.py", line 2838, in backward
644
+ loss.backward(**kwargs)
645
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/_tensor.py", line 648, in backward
646
+ torch.autograd.backward(
647
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/__init__.py", line 353, in backward
648
+ _engine_run_backward(
649
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
650
+ return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
651
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
652
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.09 GiB. GPU 0 has a total capacity of 139.80 GiB of which 12.49 GiB is free. Including non-PyTorch memory, this process has 127.29 GiB memory in use. Of the allocated memory 122.90 GiB is allocated by PyTorch, and 2.85 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
653
+ [rank0]: Traceback (most recent call last):
654
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1039, in <module>
655
+ [rank0]: run_yaml_experiment(
656
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 988, in run_yaml_experiment
657
+ [rank0]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
658
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 718, in main
659
+ [rank0]: experiment.train()
660
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
661
+ [rank0]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
662
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
663
+ [rank0]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
664
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
665
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
666
+ [rank0]: return inner_training_loop(
667
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^
668
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
669
+ [rank0]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
670
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
671
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3782, in training_step
672
+ [rank0]: self.accelerator.backward(loss, **kwargs)
673
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/accelerator.py", line 2838, in backward
674
+ [rank0]: loss.backward(**kwargs)
675
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/_tensor.py", line 648, in backward
676
+ [rank0]: torch.autograd.backward(
677
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/__init__.py", line 353, in backward
678
+ [rank0]: _engine_run_backward(
679
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
680
+ [rank0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
681
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
682
+ [rank0]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.09 GiB. GPU 0 has a total capacity of 139.80 GiB of which 12.49 GiB is free. Including non-PyTorch memory, this process has 127.29 GiB memory in use. Of the allocated memory 122.90 GiB is allocated by PyTorch, and 2.85 GiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
683
+ W0622 12:41:20.375000 2402310 site-packages/torch/distributed/elastic/multiprocessing/api.py:900] Sending process 2402406 closing signal SIGTERM
684
+ E0622 12:41:20.892000 2402310 site-packages/torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1) local_rank: 1 (pid: 2402408) of binary: /home/ext_minje/miniforge3/envs/robocasa/bin/python
685
+ Traceback (most recent call last):
686
+ File "<frozen runpy>", line 198, in _run_module_as_main
687
+ File "<frozen runpy>", line 88, in _run_code
688
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 896, in <module>
689
+ main()
690
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
691
+ return f(*args, **kwargs)
692
+ ^^^^^^^^^^^^^^^^^^
693
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 892, in main
694
+ run(args)
695
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 883, in run
696
+ elastic_launch(
697
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 139, in __call__
698
+ return launch_agent(self._config, self._entrypoint, list(args))
699
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
700
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
701
+ raise ChildFailedError(
702
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
703
+ ============================================================
704
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py FAILED
705
+ ------------------------------------------------------------
706
+ Failures:
707
+ <NO_OTHER_FAILURES>
708
+ ------------------------------------------------------------
709
+ Root Cause (first observed failure):
710
+ [0]:
711
+ time : 2026-06-22_12:41:20
712
+ host : DGX-H200-01
713
+ rank : 1 (local_rank: 1)
714
+ exitcode : 1 (pid: 2402408)
715
+ error_file: <N/A>
716
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
717
+ ============================================================
718
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
719
+
720
+ ==================================================
721
+ GR00T FINE-TUNING CONFIGURATION:
722
+ ==================================================
723
+ config: experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
724
+ dataset_soup: None
725
+ output_dir: /tmp/gr00t
726
+ output_root: None
727
+ data_config: panda_omron
728
+ batch_size: 32
729
+ max_steps: 300000
730
+ num_gpus: 2
731
+ save_steps: 20000
732
+ run_name: None
733
+ save_total_limit: 100
734
+ seed: 42
735
+ base_model_path: nvidia/GR00T-N1.5-3B
736
+ tune_llm: False
737
+ tune_visual: False
738
+ tune_projector: True
739
+ tune_diffusion_model: True
740
+ resume: False
741
+ learning_rate: 3e-05
742
+ weight_decay: 1e-05
743
+ warmup_ratio: 0.05
744
+ lora_rank: 0
745
+ lora_alpha: 16
746
+ lora_dropout: 0.1
747
+ lora_full_model: False
748
+ dataloader_num_workers: 8
749
+ report_to: wandb
750
+ embodiment_tag: new_embodiment
751
+ video_backend: opencv
752
+ balance_dataset_weights: True
753
+ balance_trajectory_weights: True
754
+ ds_weights_alpha: 0.4
755
+ ==================================================
756
+
757
+ Using 2 GPUs
758
+ Running torchrun command: ['/home/ext_minje/miniforge3/envs/robocasa/bin/python', '-m', 'torch.distributed.run', '--standalone', '--nproc_per_node=2', '--nnodes=1', '/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py', '--config', 'experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml', '--batch-size', '32', '--num-gpus', '2']
rkd_v2_2/launch_logs/action_encoder_flatten_gpu4_5_bs64_ngpu2.pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2324826
rkd_v2_2/launch_logs/action_encoder_flatten_gpu4_5_bs64_ngpu2_20260622_123748.log ADDED
@@ -0,0 +1,865 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s][rank1]: Traceback (most recent call last):
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [robosuite WARNING] No private macro file found! (macros.py:57)
2
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
3
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
4
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
5
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
6
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
7
+ check_for_updates()
8
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
9
+
10
+ *****************************************
11
+ Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
12
+ *****************************************
13
+ [robosuite WARNING] No private macro file found! (macros.py:57)
14
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
15
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
16
+ [robosuite WARNING] No private macro file found! (macros.py:57)
17
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
18
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
19
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
20
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
21
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
22
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
23
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
24
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
25
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
26
+ check_for_updates()
27
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
28
+ check_for_updates()
29
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
30
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
31
+
32
+ ==================================================
33
+ GR00T FINE-TUNING CONFIGURATION:
34
+ ==================================================
35
+ config: experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
36
+ dataset_soup: None
37
+ output_dir: /tmp/gr00t
38
+ output_root: None
39
+ data_config: panda_omron
40
+ batch_size: 64
41
+ max_steps: 300000
42
+ num_gpus: 2
43
+ save_steps: 20000
44
+ run_name: None
45
+ save_total_limit: 100
46
+ seed: 42
47
+ base_model_path: nvidia/GR00T-N1.5-3B
48
+ tune_llm: False
49
+ tune_visual: False
50
+ tune_projector: True
51
+ tune_diffusion_model: True
52
+ resume: False
53
+ learning_rate: 3e-05
54
+ weight_decay: 1e-05
55
+ warmup_ratio: 0.05
56
+ lora_rank: 0
57
+ lora_alpha: 16
58
+ lora_dropout: 0.1
59
+ lora_full_model: False
60
+ dataloader_num_workers: 8
61
+ report_to: wandb
62
+ embodiment_tag: new_embodiment
63
+ video_backend: opencv
64
+ balance_dataset_weights: True
65
+ balance_trajectory_weights: True
66
+ ds_weights_alpha: 0.4
67
+ ==================================================
68
+
69
+ Using 2 GPUs
70
+
71
+ ================================================================================
72
+ Starting sweep branch: default
73
+ Sweep vars: {}
74
+ ================================================================================
75
+
76
+ --------------------------------------------------------------------------------
77
+ Running phase 1: phase2_rkd_da_only
78
+ Policy type: groot_rkd_v2
79
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
80
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
81
+ Trainable preset: processing_line_only
82
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
83
+ --------------------------------------------------------------------------------
84
+
85
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
86
+ Using 100 subset demos for filter_key: 100_demos
87
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
88
+ self.statistics[key] = torch.tensor(value)
89
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
90
+ Using 100 subset demos for filter_key: 100_demos
91
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
92
+ Using 100 subset demos for filter_key: 100_demos
93
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
94
+ Using 100 subset demos for filter_key: 100_demos
95
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
96
+ Using 100 subset demos for filter_key: 100_demos
97
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
98
+ Using 100 subset demos for filter_key: 100_demos
99
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
100
+ Using 100 subset demos for filter_key: 100_demos
101
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
102
+ Using 100 subset demos for filter_key: 100_demos
103
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
104
+ Using 100 subset demos for filter_key: 100_demos
105
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
106
+ Using 100 subset demos for filter_key: 100_demos
107
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
108
+ Using 100 subset demos for filter_key: 100_demos
109
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
110
+ Using 100 subset demos for filter_key: 100_demos
111
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
112
+ Using 100 subset demos for filter_key: 100_demos
113
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
114
+ Using 100 subset demos for filter_key: 100_demos
115
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
116
+ Using 100 subset demos for filter_key: 100_demos
117
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
118
+ Using 100 subset demos for filter_key: 100_demos
119
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
120
+ Using 100 subset demos for filter_key: 100_demos
121
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
122
+ Using 100 subset demos for filter_key: 100_demos
123
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
124
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
125
+ Using 100 subset demos for filter_key: 100_demos
126
+
127
+ ==================================================
128
+ GR00T FINE-TUNING CONFIGURATION:
129
+ ==================================================
130
+ config: experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
131
+ dataset_soup: None
132
+ output_dir: /tmp/gr00t
133
+ output_root: None
134
+ data_config: panda_omron
135
+ batch_size: 64
136
+ max_steps: 300000
137
+ num_gpus: 2
138
+ save_steps: 20000
139
+ run_name: None
140
+ save_total_limit: 100
141
+ seed: 42
142
+ base_model_path: nvidia/GR00T-N1.5-3B
143
+ tune_llm: False
144
+ tune_visual: False
145
+ tune_projector: True
146
+ tune_diffusion_model: True
147
+ resume: False
148
+ learning_rate: 3e-05
149
+ weight_decay: 1e-05
150
+ warmup_ratio: 0.05
151
+ lora_rank: 0
152
+ lora_alpha: 16
153
+ lora_dropout: 0.1
154
+ lora_full_model: False
155
+ dataloader_num_workers: 8
156
+ report_to: wandb
157
+ embodiment_tag: new_embodiment
158
+ video_backend: opencv
159
+ balance_dataset_weights: True
160
+ balance_trajectory_weights: True
161
+ ds_weights_alpha: 0.4
162
+ ==================================================
163
+
164
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
165
+ Using 100 subset demos for filter_key: 100_demos
166
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
167
+ Using 100 subset demos for filter_key: 100_demos
168
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
169
+ Using 100 subset demos for filter_key: 100_demos
170
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
171
+ Using 100 subset demos for filter_key: 100_demos
172
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
173
+ Using 100 subset demos for filter_key: 100_demos
174
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
175
+ Using 100 subset demos for filter_key: 100_demos
176
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
177
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
178
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
179
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
180
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
181
+ 0.75517122 0.7973985 ]
182
+ Loaded 26 datasets
183
+ Using 2 GPUs
184
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
185
+ Tune backbone vision tower: False
186
+ Tune backbone LLM: False
187
+ Tune action head projector: False
188
+ Tune action head DiT: False
189
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
190
+
191
+ ================================================================================
192
+ Starting sweep branch: default
193
+ Sweep vars: {}
194
+ ================================================================================
195
+
196
+ --------------------------------------------------------------------------------
197
+ Running phase 1: phase2_rkd_da_only
198
+ Policy type: groot_rkd_v2
199
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
200
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
201
+ Trainable preset: processing_line_only
202
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
203
+ --------------------------------------------------------------------------------
204
+
205
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
206
+ Using 100 subset demos for filter_key: 100_demos
207
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
208
+ self.statistics[key] = torch.tensor(value)
209
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
210
+ Using 100 subset demos for filter_key: 100_demos
211
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
212
+ Using 100 subset demos for filter_key: 100_demos
213
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
214
+ Using 100 subset demos for filter_key: 100_demos
215
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
216
+ Using 100 subset demos for filter_key: 100_demos
217
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
218
+ Using 100 subset demos for filter_key: 100_demos
219
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
220
+ Using 100 subset demos for filter_key: 100_demos
221
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
222
+ Using 100 subset demos for filter_key: 100_demos
223
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
224
+ Using 100 subset demos for filter_key: 100_demos
225
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
226
+ Using 100 subset demos for filter_key: 100_demos
227
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
228
+ Using 100 subset demos for filter_key: 100_demos
229
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
230
+ Using 100 subset demos for filter_key: 100_demos
231
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
232
+ Using 100 subset demos for filter_key: 100_demos
233
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
234
+ Using 100 subset demos for filter_key: 100_demos
235
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
236
+ Using 100 subset demos for filter_key: 100_demos
237
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
238
+ Using 100 subset demos for filter_key: 100_demos
239
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
240
+ Using 100 subset demos for filter_key: 100_demos
241
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
242
+ Using 100 subset demos for filter_key: 100_demos
243
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
244
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
245
+ Using 100 subset demos for filter_key: 100_demos
246
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
247
+ Using 100 subset demos for filter_key: 100_demos
248
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
249
+ Using 100 subset demos for filter_key: 100_demos
250
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
251
+ Using 100 subset demos for filter_key: 100_demos
252
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
253
+ Using 100 subset demos for filter_key: 100_demos
254
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
255
+ Using 100 subset demos for filter_key: 100_demos
256
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
257
+ Using 100 subset demos for filter_key: 100_demos
258
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
259
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
260
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
261
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
262
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
263
+ 0.75517122 0.7973985 ]
264
+ Loaded 26 datasets
265
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
266
+ Tune backbone vision tower: False
267
+ Tune backbone LLM: False
268
+ Tune action head projector: False
269
+ Tune action head DiT: False
270
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
271
+ Tune backbone llm: False
272
+ Tune backbone visual: True
273
+ Total number of DiT parameters: 550386688
274
+ Tune backbone llm: False
275
+ Tune backbone visual: True
276
+ Total number of DiT parameters: 550386688
277
+ Total number of SelfAttentionTransformer parameters: 201433088
278
+ Tune action head projector: True
279
+ Tune action head diffusion model: True
280
+
281
+ Tune backbone llm: False
282
+ Tune backbone visual: False
283
+ Warning: No backbone trainable parameters found.
284
+ Tune action head projector: False
285
+ Tune action head diffusion model: False
286
+ Action head trainable parameter: future_tokens.weight
287
+ Action head trainable parameter: vlln.weight
288
+ Action head trainable parameter: vlln.bias
289
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
290
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
291
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
292
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
293
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
294
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
295
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
296
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
297
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
298
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
299
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
300
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
301
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
302
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
303
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
304
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
305
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
306
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
307
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
308
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
309
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
310
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
311
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
312
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
313
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
314
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
315
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
316
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
317
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
318
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
319
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
320
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
321
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
322
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
323
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
324
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
325
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
326
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
327
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
328
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
329
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
330
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
331
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
332
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
333
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
334
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
335
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
336
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
337
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
338
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
339
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
340
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
341
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
342
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
343
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
344
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
345
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
346
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
347
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
348
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
349
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
350
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
351
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
352
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
353
+ Applied trainable preset: processing_line_only
354
+ Trainable parameter tensors after preset: 66
355
+ trainable: action_head.vlln.weight
356
+ trainable: action_head.vlln.bias
357
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
358
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
359
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
360
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
361
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
362
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
363
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
364
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
365
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
366
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
367
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
368
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
369
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
370
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
371
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
372
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
373
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
374
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
375
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
376
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
377
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
378
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
379
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
380
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
381
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
382
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
383
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
384
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
385
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
386
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
387
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
388
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
389
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
390
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
391
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
392
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
393
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
394
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
395
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
396
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
397
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
398
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
399
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
400
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
401
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
402
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
403
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
404
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
405
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
406
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
407
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
408
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
409
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
410
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
411
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
412
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
413
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
414
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
415
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
416
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
417
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
418
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
419
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
420
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
421
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2724163520, 'trainable_param_ratio': 0.0739446007998815}
422
+ Total number of SelfAttentionTransformer parameters: 201433088
423
+ Tune action head projector: True
424
+ Tune action head diffusion model: True
425
+
426
+ Tune backbone llm: False
427
+ Tune backbone visual: False
428
+ Warning: No backbone trainable parameters found.
429
+ Tune action head projector: False
430
+ Tune action head diffusion model: False
431
+ Action head trainable parameter: future_tokens.weight
432
+ Action head trainable parameter: vlln.weight
433
+ Action head trainable parameter: vlln.bias
434
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
435
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
436
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
437
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
438
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
439
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
440
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
441
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
442
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
443
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
444
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
445
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
446
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
447
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
448
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
449
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
450
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
451
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
452
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
453
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
454
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
455
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
456
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
457
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
458
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
459
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
460
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
461
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
462
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
463
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
464
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
465
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
466
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
467
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
468
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
469
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
470
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
471
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
472
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
473
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
474
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
475
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
476
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
477
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
478
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
479
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
480
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
481
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
482
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
483
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
484
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
485
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
486
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
487
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
488
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
489
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
490
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
491
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
492
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
493
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
494
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
495
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
496
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
497
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
498
+ Applied trainable preset: processing_line_only
499
+ Trainable parameter tensors after preset: 66
500
+ trainable: action_head.vlln.weight
501
+ trainable: action_head.vlln.bias
502
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
503
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
504
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
505
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
506
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
507
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
508
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
509
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
510
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
511
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
512
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
513
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
514
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
515
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
516
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
517
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
518
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
519
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
520
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
521
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
522
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
523
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
524
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
525
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
526
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
527
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
528
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
529
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
530
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
531
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
532
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
533
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
534
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
535
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
536
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
537
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
538
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
539
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
540
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
541
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
542
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
543
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
544
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
545
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
546
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
547
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
548
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
549
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
550
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
551
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
552
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
553
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
554
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
555
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
556
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
557
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
558
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
559
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
560
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
561
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
562
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
563
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
564
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
565
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
566
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2724163520, 'trainable_param_ratio': 0.0739446007998815}
567
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
568
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
569
+ train dataloader length: 3437
570
+ train dataset length: 439854
571
+ GPU memory before training: 7.076685905456543 GB
572
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
573
+ train dataloader length: 3437
574
+ train dataset length: 439854
575
+ GPU memory before training: 7.076685905456543 GB
576
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
577
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
578
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
579
+ wandb: setting up run voonzmye
580
+ wandb: Tracking run with wandb version 0.25.0
581
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_123820-voonzmye
582
+ wandb: Run `wandb offline` to turn off syncing.
583
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
584
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2
585
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2/runs/voonzmye
586
+
587
  0%| | 0/30000 [00:00<?, ?it/s][rank1]: Traceback (most recent call last):
588
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1039, in <module>
589
+ [rank1]: run_yaml_experiment(
590
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 988, in run_yaml_experiment
591
+ [rank1]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
592
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 718, in main
593
+ [rank1]: experiment.train()
594
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
595
+ [rank1]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
596
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
597
+ [rank1]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
598
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
599
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
600
+ [rank1]: return inner_training_loop(
601
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^
602
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
603
+ [rank1]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
604
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
605
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
606
+ [rank1]: loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
607
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
608
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
609
+ [rank1]: outputs = model(inputs)
610
+ [rank1]: ^^^^^^^^^^^^^
611
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
612
+ [rank1]: return self._call_impl(*args, **kwargs)
613
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
614
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
615
+ [rank1]: return forward_call(*args, **kwargs)
616
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
617
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1637, in forward
618
+ [rank1]: else self._run_ddp_forward(*inputs, **kwargs)
619
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
620
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1464, in _run_ddp_forward
621
+ [rank1]: return self.module(*inputs, **kwargs) # type: ignore[index]
622
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
623
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
624
+ [rank1]: return self._call_impl(*args, **kwargs)
625
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
626
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
627
+ [rank1]: return forward_call(*args, **kwargs)
628
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
629
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
630
+ [rank1]: return model_forward(*args, **kwargs)
631
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
632
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
633
+ [rank1]: return convert_to_fp32(self.model_forward(*args, **kwargs))
634
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
635
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
636
+ [rank1]: return func(*args, **kwargs)
637
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^
638
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 509, in forward
639
+ [rank1]: self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
640
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 450, in _add_rkd_loss
641
+ [rank1]: rkd_loss, rkd_metrics = self._compute_rkd_loss(
642
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^
643
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 361, in _compute_rkd_loss
644
+ [rank1]: angle_loss = rkd_angle_loss(student_vector, teacher_vector)
645
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
646
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
647
+ [rank1]: student_angle = _angle_relation(student, eps=eps)
648
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
649
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
650
+ [rank1]: diff = x[:, None, :] - x[None, :, :]
651
+ [rank1]: ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
652
+ [rank1]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 1 has a total capacity of 139.80 GiB of which 77.70 GiB is free. Process 1961377 has 11.53 GiB memory in use. Including non-PyTorch memory, this process has 35.94 GiB memory in use. Of the allocated memory 34.24 GiB is allocated by PyTorch, and 225.09 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
653
+ wandb: updating run metadata
654
+ wandb: uploading wandb-summary.json; uploading config.yaml
655
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2/runs/voonzmye
656
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2
657
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
658
+ wandb: Find logs at: ./wandb/run-20260622_123820-voonzmye/logs
659
+ Traceback (most recent call last):
660
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1039, in <module>
661
+ run_yaml_experiment(
662
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 988, in run_yaml_experiment
663
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
664
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 718, in main
665
+ experiment.train()
666
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
667
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
668
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
669
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
670
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
671
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
672
+ return inner_training_loop(
673
+ ^^^^^^^^^^^^^^^^^^^^
674
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
675
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
676
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
677
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
678
+ loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
679
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
680
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
681
+ outputs = model(inputs)
682
+ ^^^^^^^^^^^^^
683
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
684
+ return self._call_impl(*args, **kwargs)
685
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
686
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
687
+ return forward_call(*args, **kwargs)
688
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
689
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1637, in forward
690
+ else self._run_ddp_forward(*inputs, **kwargs)
691
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
692
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1464, in _run_ddp_forward
693
+ return self.module(*inputs, **kwargs) # type: ignore[index]
694
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
695
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
696
+ return self._call_impl(*args, **kwargs)
697
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
698
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
699
+ return forward_call(*args, **kwargs)
700
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
701
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
702
+ return model_forward(*args, **kwargs)
703
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
704
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
705
+ return convert_to_fp32(self.model_forward(*args, **kwargs))
706
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
707
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
708
+ return func(*args, **kwargs)
709
+ ^^^^^^^^^^^^^^^^^^^^^
710
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 509, in forward
711
+ self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
712
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 450, in _add_rkd_loss
713
+ rkd_loss, rkd_metrics = self._compute_rkd_loss(
714
+ ^^^^^^^^^^^^^^^^^^^^^^^
715
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 361, in _compute_rkd_loss
716
+ angle_loss = rkd_angle_loss(student_vector, teacher_vector)
717
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
718
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
719
+ student_angle = _angle_relation(student, eps=eps)
720
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
721
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
722
+ diff = x[:, None, :] - x[None, :, :]
723
+ ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
724
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 103.87 GiB is free. Including non-PyTorch memory, this process has 35.91 GiB memory in use. Of the allocated memory 34.24 GiB is allocated by PyTorch, and 196.48 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
725
+ [rank0]: Traceback (most recent call last):
726
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1039, in <module>
727
+ [rank0]: run_yaml_experiment(
728
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 988, in run_yaml_experiment
729
+ [rank0]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
730
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 718, in main
731
+ [rank0]: experiment.train()
732
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
733
+ [rank0]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
734
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
735
+ [rank0]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
736
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
737
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
738
+ [rank0]: return inner_training_loop(
739
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^
740
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
741
+ [rank0]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
742
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
743
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
744
+ [rank0]: loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
745
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
746
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
747
+ [rank0]: outputs = model(inputs)
748
+ [rank0]: ^^^^^^^^^^^^^
749
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
750
+ [rank0]: return self._call_impl(*args, **kwargs)
751
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
752
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
753
+ [rank0]: return forward_call(*args, **kwargs)
754
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
755
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1637, in forward
756
+ [rank0]: else self._run_ddp_forward(*inputs, **kwargs)
757
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
758
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1464, in _run_ddp_forward
759
+ [rank0]: return self.module(*inputs, **kwargs) # type: ignore[index]
760
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
761
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
762
+ [rank0]: return self._call_impl(*args, **kwargs)
763
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
764
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
765
+ [rank0]: return forward_call(*args, **kwargs)
766
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
767
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
768
+ [rank0]: return model_forward(*args, **kwargs)
769
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
770
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
771
+ [rank0]: return convert_to_fp32(self.model_forward(*args, **kwargs))
772
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
773
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
774
+ [rank0]: return func(*args, **kwargs)
775
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^
776
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 509, in forward
777
+ [rank0]: self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
778
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 450, in _add_rkd_loss
779
+ [rank0]: rkd_loss, rkd_metrics = self._compute_rkd_loss(
780
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^
781
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 361, in _compute_rkd_loss
782
+ [rank0]: angle_loss = rkd_angle_loss(student_vector, teacher_vector)
783
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
784
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
785
+ [rank0]: student_angle = _angle_relation(student, eps=eps)
786
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
787
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
788
+ [rank0]: diff = x[:, None, :] - x[None, :, :]
789
+ [rank0]: ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
790
+ [rank0]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 103.87 GiB is free. Including non-PyTorch memory, this process has 35.91 GiB memory in use. Of the allocated memory 34.24 GiB is allocated by PyTorch, and 196.48 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
791
+ W0622 12:38:35.764000 2325119 site-packages/torch/distributed/elastic/multiprocessing/api.py:900] Sending process 2325210 closing signal SIGTERM
792
+ E0622 12:38:36.279000 2325119 site-packages/torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1) local_rank: 1 (pid: 2325211) of binary: /home/ext_minje/miniforge3/envs/robocasa/bin/python
793
+ Traceback (most recent call last):
794
+ File "<frozen runpy>", line 198, in _run_module_as_main
795
+ File "<frozen runpy>", line 88, in _run_code
796
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 896, in <module>
797
+ main()
798
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
799
+ return f(*args, **kwargs)
800
+ ^^^^^^^^^^^^^^^^^^
801
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 892, in main
802
+ run(args)
803
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 883, in run
804
+ elastic_launch(
805
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 139, in __call__
806
+ return launch_agent(self._config, self._entrypoint, list(args))
807
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
808
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
809
+ raise ChildFailedError(
810
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
811
+ ============================================================
812
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py FAILED
813
+ ------------------------------------------------------------
814
+ Failures:
815
+ <NO_OTHER_FAILURES>
816
+ ------------------------------------------------------------
817
+ Root Cause (first observed failure):
818
+ [0]:
819
+ time : 2026-06-22_12:38:35
820
+ host : DGX-H200-01
821
+ rank : 1 (local_rank: 1)
822
+ exitcode : 1 (pid: 2325211)
823
+ error_file: <N/A>
824
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
825
+ ============================================================
826
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
827
+
828
+ ==================================================
829
+ GR00T FINE-TUNING CONFIGURATION:
830
+ ==================================================
831
+ config: experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
832
+ dataset_soup: None
833
+ output_dir: /tmp/gr00t
834
+ output_root: None
835
+ data_config: panda_omron
836
+ batch_size: 64
837
+ max_steps: 300000
838
+ num_gpus: 2
839
+ save_steps: 20000
840
+ run_name: None
841
+ save_total_limit: 100
842
+ seed: 42
843
+ base_model_path: nvidia/GR00T-N1.5-3B
844
+ tune_llm: False
845
+ tune_visual: False
846
+ tune_projector: True
847
+ tune_diffusion_model: True
848
+ resume: False
849
+ learning_rate: 3e-05
850
+ weight_decay: 1e-05
851
+ warmup_ratio: 0.05
852
+ lora_rank: 0
853
+ lora_alpha: 16
854
+ lora_dropout: 0.1
855
+ lora_full_model: False
856
+ dataloader_num_workers: 8
857
+ report_to: wandb
858
+ embodiment_tag: new_embodiment
859
+ video_backend: opencv
860
+ balance_dataset_weights: True
861
+ balance_trajectory_weights: True
862
+ ds_weights_alpha: 0.4
863
+ ==================================================
864
+
865
+ Using 2 GPUs
866
+ Running torchrun command: ['/home/ext_minje/miniforge3/envs/robocasa/bin/python', '-m', 'torch.distributed.run', '--standalone', '--nproc_per_node=2', '--nnodes=1', '/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py', '--config', 'experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml', '--batch-size', '64', '--num-gpus', '2']
rkd_v2_2/launch_logs/action_encoder_gpu4.pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2196189
rkd_v2_2/launch_logs/action_encoder_gpu4_bs128_20260622_123336.log ADDED
@@ -0,0 +1,351 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [robosuite WARNING] No private macro file found! (macros.py:57)
2
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
3
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
4
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
5
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
6
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
7
+ check_for_updates()
8
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
9
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
10
+
11
+ ==================================================
12
+ GR00T FINE-TUNING CONFIGURATION:
13
+ ==================================================
14
+ config: experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
15
+ dataset_soup: None
16
+ output_dir: /tmp/gr00t
17
+ output_root: None
18
+ data_config: panda_omron
19
+ batch_size: 128
20
+ max_steps: 300000
21
+ num_gpus: 1
22
+ save_steps: 20000
23
+ run_name: None
24
+ save_total_limit: 100
25
+ seed: 42
26
+ base_model_path: nvidia/GR00T-N1.5-3B
27
+ tune_llm: False
28
+ tune_visual: False
29
+ tune_projector: True
30
+ tune_diffusion_model: True
31
+ resume: False
32
+ learning_rate: 3e-05
33
+ weight_decay: 1e-05
34
+ warmup_ratio: 0.05
35
+ lora_rank: 0
36
+ lora_alpha: 16
37
+ lora_dropout: 0.1
38
+ lora_full_model: False
39
+ dataloader_num_workers: 8
40
+ report_to: wandb
41
+ embodiment_tag: new_embodiment
42
+ video_backend: opencv
43
+ balance_dataset_weights: True
44
+ balance_trajectory_weights: True
45
+ ds_weights_alpha: 0.4
46
+ ==================================================
47
+
48
+ Using 1 GPUs
49
+
50
+ ================================================================================
51
+ Starting sweep branch: default
52
+ Sweep vars: {}
53
+ ================================================================================
54
+
55
+ --------------------------------------------------------------------------------
56
+ Running phase 1: phase2_rkd_da_only
57
+ Policy type: groot_rkd_v2
58
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
59
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
60
+ Trainable preset: processing_line_only
61
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
62
+ --------------------------------------------------------------------------------
63
+
64
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
65
+ self.statistics[key] = torch.tensor(value)
66
+
67
+ Using 100 subset demos for filter_key: 100_demos
68
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
69
+ Using 100 subset demos for filter_key: 100_demos
70
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
71
+ Using 100 subset demos for filter_key: 100_demos
72
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
73
+ Using 100 subset demos for filter_key: 100_demos
74
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
75
+ Using 100 subset demos for filter_key: 100_demos
76
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
77
+ Using 100 subset demos for filter_key: 100_demos
78
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
79
+ Using 100 subset demos for filter_key: 100_demos
80
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
81
+ Using 100 subset demos for filter_key: 100_demos
82
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
83
+ Using 100 subset demos for filter_key: 100_demos
84
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
85
+ Using 100 subset demos for filter_key: 100_demos
86
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
87
+ Using 100 subset demos for filter_key: 100_demos
88
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
89
+ Using 100 subset demos for filter_key: 100_demos
90
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
91
+ Using 100 subset demos for filter_key: 100_demos
92
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
93
+ Using 100 subset demos for filter_key: 100_demos
94
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
95
+ Using 100 subset demos for filter_key: 100_demos
96
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
97
+ Using 100 subset demos for filter_key: 100_demos
98
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
99
+ Using 100 subset demos for filter_key: 100_demos
100
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
101
+ Using 100 subset demos for filter_key: 100_demos
102
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
103
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
104
+ Using 100 subset demos for filter_key: 100_demos
105
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
106
+ Using 100 subset demos for filter_key: 100_demos
107
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
108
+ Using 100 subset demos for filter_key: 100_demos
109
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
110
+ Using 100 subset demos for filter_key: 100_demos
111
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
112
+ Using 100 subset demos for filter_key: 100_demos
113
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
114
+ Using 100 subset demos for filter_key: 100_demos
115
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
116
+ Using 100 subset demos for filter_key: 100_demos
117
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
118
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
119
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
120
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
121
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
122
+ 0.75517122 0.7973985 ]
123
+ Loaded 26 datasets
124
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
125
+ Tune backbone vision tower: False
126
+ Tune backbone LLM: False
127
+ Tune action head projector: False
128
+ Tune action head DiT: False
129
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
130
+ Tune backbone llm: False
131
+ Tune backbone visual: True
132
+ Total number of DiT parameters: 550386688
133
+ Total number of SelfAttentionTransformer parameters: 201433088
134
+ Tune action head projector: True
135
+ Tune action head diffusion model: True
136
+
137
+ Tune backbone llm: False
138
+ Tune backbone visual: False
139
+ Warning: No backbone trainable parameters found.
140
+ Tune action head projector: False
141
+ Tune action head diffusion model: False
142
+ Action head trainable parameter: future_tokens.weight
143
+ Action head trainable parameter: vlln.weight
144
+ Action head trainable parameter: vlln.bias
145
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
146
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
147
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
148
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
149
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
150
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
151
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
152
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
153
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
154
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
155
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
156
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
157
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
158
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
159
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
160
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
161
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
162
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
163
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
164
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
165
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
166
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
167
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
168
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
169
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
170
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
171
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
172
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
173
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
174
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
175
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
176
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
177
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
178
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
179
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
180
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
181
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
182
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
183
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
184
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
185
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
186
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
187
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
188
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
189
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
190
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
191
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
192
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
193
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
194
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
195
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
196
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
197
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
198
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
199
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
200
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
201
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
202
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
203
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
204
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
205
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
206
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
207
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
208
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
209
+ Applied trainable preset: processing_line_only
210
+ Trainable parameter tensors after preset: 66
211
+ trainable: action_head.vlln.weight
212
+ trainable: action_head.vlln.bias
213
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
214
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
215
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
216
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
217
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
218
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
219
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
220
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
221
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
222
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
223
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
224
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
225
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
226
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
227
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
228
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
229
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
230
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
231
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
232
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
233
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
234
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
235
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
236
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
237
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
238
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
239
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
240
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
241
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
242
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
243
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
244
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
245
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
246
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
247
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
248
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
249
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
250
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
251
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
252
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
253
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
254
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
255
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
256
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
257
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
258
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
259
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
260
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
261
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
262
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
263
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
264
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
265
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
266
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
267
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
268
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
269
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
270
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
271
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
272
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
273
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
274
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
275
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
276
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
277
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2724163520, 'trainable_param_ratio': 0.0739446007998815}
278
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
279
+ train dataloader length: 3437
280
+ train dataset length: 439854
281
+ GPU memory before training: 7.076685905456543 GB
282
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
283
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
284
+ wandb: setting up run q8utcfl2
285
+ wandb: Tracking run with wandb version 0.25.0
286
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_123354-q8utcfl2
287
+ wandb: Run `wandb offline` to turn off syncing.
288
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
289
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2
290
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2/runs/q8utcfl2
291
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
292
+
293
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
294
+ wandb: uploading config.yaml
295
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2/runs/q8utcfl2
296
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2
297
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
298
+ wandb: Find logs at: ./wandb/run-20260622_123354-q8utcfl2/logs
299
+ Traceback (most recent call last):
300
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1029, in <module>
301
+ run_yaml_experiment(
302
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 988, in run_yaml_experiment
303
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
304
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 718, in main
305
+ experiment.train()
306
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
307
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
308
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
309
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
310
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
311
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
312
+ return inner_training_loop(
313
+ ^^^^^^^^^^^^^^^^^^^^
314
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
315
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
316
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
317
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
318
+ loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
319
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
320
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
321
+ outputs = model(inputs)
322
+ ^^^^^^^^^^^^^
323
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
324
+ return self._call_impl(*args, **kwargs)
325
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
326
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
327
+ return forward_call(*args, **kwargs)
328
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
329
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
330
+ return model_forward(*args, **kwargs)
331
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
332
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
333
+ return convert_to_fp32(self.model_forward(*args, **kwargs))
334
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
335
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
336
+ return func(*args, **kwargs)
337
+ ^^^^^^^^^^^^^^^^^^^^^
338
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 509, in forward
339
+ self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
340
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 450, in _add_rkd_loss
341
+ rkd_loss, rkd_metrics = self._compute_rkd_loss(
342
+ ^^^^^^^^^^^^^^^^^^^^^^^
343
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 361, in _compute_rkd_loss
344
+ angle_loss = rkd_angle_loss(student_vector, teacher_vector)
345
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
346
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
347
+ student_angle = _angle_relation(student, eps=eps)
348
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
349
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
350
+ diff = x[:, None, :] - x[None, :, :]
351
+ ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
352
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 80.75 GiB is free. Including non-PyTorch memory, this process has 59.03 GiB memory in use. Of the allocated memory 57.80 GiB is allocated by PyTorch, and 573.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
rkd_v2_2/launch_logs/raw_action_gpu5.pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2196191
rkd_v2_2/launch_logs/raw_action_gpu5_bs128_20260622_123336.log ADDED
@@ -0,0 +1,351 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [robosuite WARNING] No private macro file found! (macros.py:57)
2
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
3
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
4
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
5
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
6
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
7
+ check_for_updates()
8
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
9
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
10
+
11
+ ==================================================
12
+ GR00T FINE-TUNING CONFIGURATION:
13
+ ==================================================
14
+ config: experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml
15
+ dataset_soup: None
16
+ output_dir: /tmp/gr00t
17
+ output_root: None
18
+ data_config: panda_omron
19
+ batch_size: 128
20
+ max_steps: 300000
21
+ num_gpus: 1
22
+ save_steps: 20000
23
+ run_name: None
24
+ save_total_limit: 100
25
+ seed: 42
26
+ base_model_path: nvidia/GR00T-N1.5-3B
27
+ tune_llm: False
28
+ tune_visual: False
29
+ tune_projector: True
30
+ tune_diffusion_model: True
31
+ resume: False
32
+ learning_rate: 3e-05
33
+ weight_decay: 1e-05
34
+ warmup_ratio: 0.05
35
+ lora_rank: 0
36
+ lora_alpha: 16
37
+ lora_dropout: 0.1
38
+ lora_full_model: False
39
+ dataloader_num_workers: 8
40
+ report_to: wandb
41
+ embodiment_tag: new_embodiment
42
+ video_backend: opencv
43
+ balance_dataset_weights: True
44
+ balance_trajectory_weights: True
45
+ ds_weights_alpha: 0.4
46
+ ==================================================
47
+
48
+ Using 1 GPUs
49
+
50
+ ================================================================================
51
+ Starting sweep branch: default
52
+ Sweep vars: {}
53
+ ================================================================================
54
+
55
+ --------------------------------------------------------------------------------
56
+ Running phase 1: phase2_rkd_da_only
57
+ Policy type: groot_rkd_v2
58
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
59
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2
60
+ Trainable preset: processing_line_only
61
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'raw_action', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
62
+ --------------------------------------------------------------------------------
63
+
64
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
65
+ self.statistics[key] = torch.tensor(value)
66
+
67
+ Using 100 subset demos for filter_key: 100_demos
68
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
69
+ Using 100 subset demos for filter_key: 100_demos
70
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
71
+ Using 100 subset demos for filter_key: 100_demos
72
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
73
+ Using 100 subset demos for filter_key: 100_demos
74
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
75
+ Using 100 subset demos for filter_key: 100_demos
76
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
77
+ Using 100 subset demos for filter_key: 100_demos
78
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
79
+ Using 100 subset demos for filter_key: 100_demos
80
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
81
+ Using 100 subset demos for filter_key: 100_demos
82
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
83
+ Using 100 subset demos for filter_key: 100_demos
84
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
85
+ Using 100 subset demos for filter_key: 100_demos
86
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
87
+ Using 100 subset demos for filter_key: 100_demos
88
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
89
+ Using 100 subset demos for filter_key: 100_demos
90
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
91
+ Using 100 subset demos for filter_key: 100_demos
92
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
93
+ Using 100 subset demos for filter_key: 100_demos
94
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
95
+ Using 100 subset demos for filter_key: 100_demos
96
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
97
+ Using 100 subset demos for filter_key: 100_demos
98
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
99
+ Using 100 subset demos for filter_key: 100_demos
100
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
101
+ Using 100 subset demos for filter_key: 100_demos
102
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
103
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
104
+ Using 100 subset demos for filter_key: 100_demos
105
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
106
+ Using 100 subset demos for filter_key: 100_demos
107
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
108
+ Using 100 subset demos for filter_key: 100_demos
109
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
110
+ Using 100 subset demos for filter_key: 100_demos
111
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
112
+ Using 100 subset demos for filter_key: 100_demos
113
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
114
+ Using 100 subset demos for filter_key: 100_demos
115
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
116
+ Using 100 subset demos for filter_key: 100_demos
117
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
118
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
119
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
120
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
121
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
122
+ 0.75517122 0.7973985 ]
123
+ Loaded 26 datasets
124
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
125
+ Tune backbone vision tower: False
126
+ Tune backbone LLM: False
127
+ Tune action head projector: False
128
+ Tune action head DiT: False
129
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
130
+ Tune backbone llm: False
131
+ Tune backbone visual: True
132
+ Total number of DiT parameters: 550386688
133
+ Total number of SelfAttentionTransformer parameters: 201433088
134
+ Tune action head projector: True
135
+ Tune action head diffusion model: True
136
+
137
+ Tune backbone llm: False
138
+ Tune backbone visual: False
139
+ Warning: No backbone trainable parameters found.
140
+ Tune action head projector: False
141
+ Tune action head diffusion model: False
142
+ Action head trainable parameter: future_tokens.weight
143
+ Action head trainable parameter: vlln.weight
144
+ Action head trainable parameter: vlln.bias
145
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
146
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
147
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
148
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
149
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
150
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
151
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
152
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
153
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
154
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
155
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
156
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
157
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
158
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
159
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
160
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
161
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
162
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
163
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
164
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
165
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
166
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
167
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
168
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
169
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
170
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
171
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
172
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
173
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
174
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
175
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
176
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
177
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
178
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
179
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
180
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
181
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
182
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
183
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
184
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
185
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
186
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
187
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
188
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
189
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
190
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
191
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
192
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
193
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
194
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
195
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
196
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
197
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
198
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
199
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
200
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
201
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
202
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
203
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
204
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
205
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
206
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
207
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
208
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
209
+ Applied trainable preset: processing_line_only
210
+ Trainable parameter tensors after preset: 66
211
+ trainable: action_head.vlln.weight
212
+ trainable: action_head.vlln.bias
213
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
214
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
215
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
216
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
217
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
218
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
219
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
220
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
221
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
222
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
223
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
224
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
225
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
226
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
227
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
228
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
229
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
230
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
231
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
232
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
233
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
234
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
235
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
236
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
237
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
238
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
239
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
240
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
241
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
242
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
243
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
244
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
245
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
246
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
247
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
248
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
249
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
250
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
251
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
252
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
253
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
254
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
255
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
256
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
257
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
258
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
259
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
260
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
261
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
262
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
263
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
264
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
265
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
266
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
267
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
268
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
269
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
270
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
271
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
272
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
273
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
274
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
275
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
276
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
277
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2724163520, 'trainable_param_ratio': 0.0739446007998815}
278
+ Run name: rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
279
+ train dataloader length: 3437
280
+ train dataset length: 439854
281
+ GPU memory before training: 7.076685905456543 GB
282
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
283
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
284
+ wandb: setting up run 27d83vgq
285
+ wandb: Tracking run with wandb version 0.25.0
286
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_123354-27d83vgq
287
+ wandb: Run `wandb offline` to turn off syncing.
288
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
289
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2
290
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2/runs/27d83vgq
291
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2/runs
292
+
293
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
294
+ wandb: uploading config.yaml
295
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2/runs/27d83vgq
296
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/robocasa_rkd_v2_2
297
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
298
+ wandb: Find logs at: ./wandb/run-20260622_123354-27d83vgq/logs
299
+ Traceback (most recent call last):
300
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1029, in <module>
301
+ run_yaml_experiment(
302
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 988, in run_yaml_experiment
303
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
304
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 718, in main
305
+ experiment.train()
306
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
307
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
308
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
309
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
310
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
311
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
312
+ return inner_training_loop(
313
+ ^^^^^^^^^^^^^^^^^^^^
314
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
315
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
316
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
317
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
318
+ loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
319
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
320
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
321
+ outputs = model(inputs)
322
+ ^^^^^^^^^^^^^
323
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
324
+ return self._call_impl(*args, **kwargs)
325
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
326
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
327
+ return forward_call(*args, **kwargs)
328
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
329
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
330
+ return model_forward(*args, **kwargs)
331
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
332
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
333
+ return convert_to_fp32(self.model_forward(*args, **kwargs))
334
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
335
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
336
+ return func(*args, **kwargs)
337
+ ^^^^^^^^^^^^^^^^^^^^^
338
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 509, in forward
339
+ self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
340
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 450, in _add_rkd_loss
341
+ rkd_loss, rkd_metrics = self._compute_rkd_loss(
342
+ ^^^^^^^^^^^^^^^^^^^^^^^
343
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 361, in _compute_rkd_loss
344
+ angle_loss = rkd_angle_loss(student_vector, teacher_vector)
345
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
346
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
347
+ student_angle = _angle_relation(student, eps=eps)
348
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
349
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
350
+ diff = x[:, None, :] - x[None, :, :]
351
+ ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
352
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 57.77 GiB is free. Process 1961377 has 11.53 GiB memory in use. Including non-PyTorch memory, this process has 58.72 GiB memory in use. Of the allocated memory 57.77 GiB is allocated by PyTorch, and 280.87 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu45_bs32_20260622_131131.log ADDED
@@ -0,0 +1,764 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CMD: CUDA_VISIBLE_DEVICES=4,5 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True /home/ext_minje/miniforge3/envs/robocasa/bin/python /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py --config /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml --num-gpus 2 --batch-size 32
2
+ [robosuite WARNING] No private macro file found! (macros.py:57)
3
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
4
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
5
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
6
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
7
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
8
+ check_for_updates()
9
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
10
+
11
+ *****************************************
12
+ Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
13
+ *****************************************
14
+ [robosuite WARNING] No private macro file found! (macros.py:57)
15
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
16
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
17
+ [robosuite WARNING] No private macro file found! (macros.py:57)
18
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
19
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
20
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
21
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
22
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
23
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
24
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
25
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
26
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
27
+ check_for_updates()
28
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
29
+ check_for_updates()
30
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
31
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
32
+
33
+ ==================================================
34
+ GR00T FINE-TUNING CONFIGURATION:
35
+ ==================================================
36
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
37
+ dataset_soup: None
38
+ output_dir: /tmp/gr00t
39
+ output_root: None
40
+ data_config: panda_omron
41
+ batch_size: 32
42
+ max_steps: 300000
43
+ num_gpus: 2
44
+ save_steps: 20000
45
+ run_name: None
46
+ save_total_limit: 100
47
+ seed: 42
48
+ base_model_path: nvidia/GR00T-N1.5-3B
49
+ tune_llm: False
50
+ tune_visual: False
51
+ tune_projector: True
52
+ tune_diffusion_model: True
53
+ resume: False
54
+ learning_rate: 3e-05
55
+ weight_decay: 1e-05
56
+ warmup_ratio: 0.05
57
+ lora_rank: 0
58
+ lora_alpha: 16
59
+ lora_dropout: 0.1
60
+ lora_full_model: False
61
+ dataloader_num_workers: 8
62
+ report_to: wandb
63
+ embodiment_tag: new_embodiment
64
+ video_backend: opencv
65
+ balance_dataset_weights: True
66
+ balance_trajectory_weights: True
67
+ ds_weights_alpha: 0.4
68
+ ==================================================
69
+
70
+ Using 2 GPUs
71
+
72
+ ================================================================================
73
+ Starting sweep branch: default
74
+ Sweep vars: {}
75
+ ================================================================================
76
+
77
+ --------------------------------------------------------------------------------
78
+ Running phase 1: phase2_rkd_da_only
79
+ Policy type: groot_rkd_v2
80
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
81
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
82
+ Trainable preset: processing_line_only
83
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_action_encoder_projector_enabled': True, 'rkd_action_encoder_projector_dim': 512, 'rkd_action_encoder_projector_pooling': 'flatten', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
84
+ --------------------------------------------------------------------------------
85
+
86
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
87
+ Using 100 subset demos for filter_key: 100_demos
88
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
89
+ self.statistics[key] = torch.tensor(value)
90
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
91
+ Using 100 subset demos for filter_key: 100_demos
92
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
93
+ Using 100 subset demos for filter_key: 100_demos
94
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
95
+ Using 100 subset demos for filter_key: 100_demos
96
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
97
+ Using 100 subset demos for filter_key: 100_demos
98
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
99
+ Using 100 subset demos for filter_key: 100_demos
100
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
101
+ Using 100 subset demos for filter_key: 100_demos
102
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
103
+ Using 100 subset demos for filter_key: 100_demos
104
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
105
+ Using 100 subset demos for filter_key: 100_demos
106
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
107
+ Using 100 subset demos for filter_key: 100_demos
108
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
109
+ Using 100 subset demos for filter_key: 100_demos
110
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
111
+ Using 100 subset demos for filter_key: 100_demos
112
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
113
+ Using 100 subset demos for filter_key: 100_demos
114
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
115
+ Using 100 subset demos for filter_key: 100_demos
116
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
117
+ Using 100 subset demos for filter_key: 100_demos
118
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
119
+ Using 100 subset demos for filter_key: 100_demos
120
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
121
+ Using 100 subset demos for filter_key: 100_demos
122
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
123
+ Using 100 subset demos for filter_key: 100_demos
124
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
125
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
126
+ Using 100 subset demos for filter_key: 100_demos
127
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
128
+ Using 100 subset demos for filter_key: 100_demos
129
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
130
+ Using 100 subset demos for filter_key: 100_demos
131
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
132
+ Using 100 subset demos for filter_key: 100_demos
133
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
134
+ Using 100 subset demos for filter_key: 100_demos
135
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
136
+ Using 100 subset demos for filter_key: 100_demos
137
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
138
+ Using 100 subset demos for filter_key: 100_demos
139
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
140
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
141
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
142
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
143
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
144
+ 0.75517122 0.7973985 ]
145
+ Loaded 26 datasets
146
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
147
+ Tune backbone vision tower: False
148
+ Tune backbone LLM: False
149
+ Tune action head projector: False
150
+ Tune action head DiT: False
151
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
152
+
153
+ ==================================================
154
+ GR00T FINE-TUNING CONFIGURATION:
155
+ ==================================================
156
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
157
+ dataset_soup: None
158
+ output_dir: /tmp/gr00t
159
+ output_root: None
160
+ data_config: panda_omron
161
+ batch_size: 32
162
+ max_steps: 300000
163
+ num_gpus: 2
164
+ save_steps: 20000
165
+ run_name: None
166
+ save_total_limit: 100
167
+ seed: 42
168
+ base_model_path: nvidia/GR00T-N1.5-3B
169
+ tune_llm: False
170
+ tune_visual: False
171
+ tune_projector: True
172
+ tune_diffusion_model: True
173
+ resume: False
174
+ learning_rate: 3e-05
175
+ weight_decay: 1e-05
176
+ warmup_ratio: 0.05
177
+ lora_rank: 0
178
+ lora_alpha: 16
179
+ lora_dropout: 0.1
180
+ lora_full_model: False
181
+ dataloader_num_workers: 8
182
+ report_to: wandb
183
+ embodiment_tag: new_embodiment
184
+ video_backend: opencv
185
+ balance_dataset_weights: True
186
+ balance_trajectory_weights: True
187
+ ds_weights_alpha: 0.4
188
+ ==================================================
189
+
190
+ Using 2 GPUs
191
+
192
+ ================================================================================
193
+ Starting sweep branch: default
194
+ Sweep vars: {}
195
+ ================================================================================
196
+
197
+ --------------------------------------------------------------------------------
198
+ Running phase 1: phase2_rkd_da_only
199
+ Policy type: groot_rkd_v2
200
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
201
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
202
+ Trainable preset: processing_line_only
203
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_action_encoder_projector_enabled': True, 'rkd_action_encoder_projector_dim': 512, 'rkd_action_encoder_projector_pooling': 'flatten', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
204
+ --------------------------------------------------------------------------------
205
+
206
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
207
+ Using 100 subset demos for filter_key: 100_demos
208
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
209
+ self.statistics[key] = torch.tensor(value)
210
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
211
+ Using 100 subset demos for filter_key: 100_demos
212
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
213
+ Using 100 subset demos for filter_key: 100_demos
214
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
215
+ Using 100 subset demos for filter_key: 100_demos
216
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
217
+ Using 100 subset demos for filter_key: 100_demos
218
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
219
+ Using 100 subset demos for filter_key: 100_demos
220
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
221
+ Using 100 subset demos for filter_key: 100_demos
222
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
223
+ Using 100 subset demos for filter_key: 100_demos
224
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
225
+ Using 100 subset demos for filter_key: 100_demos
226
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
227
+ Using 100 subset demos for filter_key: 100_demos
228
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
229
+ Using 100 subset demos for filter_key: 100_demos
230
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
231
+ Using 100 subset demos for filter_key: 100_demos
232
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
233
+ Using 100 subset demos for filter_key: 100_demos
234
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
235
+ Using 100 subset demos for filter_key: 100_demos
236
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
237
+ Using 100 subset demos for filter_key: 100_demos
238
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
239
+ Using 100 subset demos for filter_key: 100_demos
240
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
241
+ Using 100 subset demos for filter_key: 100_demos
242
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
243
+ Using 100 subset demos for filter_key: 100_demos
244
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
245
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
246
+ Using 100 subset demos for filter_key: 100_demos
247
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
248
+ Using 100 subset demos for filter_key: 100_demos
249
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
250
+ Using 100 subset demos for filter_key: 100_demos
251
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
252
+ Using 100 subset demos for filter_key: 100_demos
253
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
254
+ Using 100 subset demos for filter_key: 100_demos
255
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
256
+ Using 100 subset demos for filter_key: 100_demos
257
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
258
+ Using 100 subset demos for filter_key: 100_demos
259
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
260
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
261
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
262
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
263
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
264
+ 0.75517122 0.7973985 ]
265
+ Loaded 26 datasets
266
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
267
+ Tune backbone vision tower: False
268
+ Tune backbone LLM: False
269
+ Tune action head projector: False
270
+ Tune action head DiT: False
271
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
272
+ Tune backbone llm: False
273
+ Tune backbone visual: True
274
+ Total number of DiT parameters: 550386688
275
+ Tune backbone llm: False
276
+ Tune backbone visual: True
277
+ Total number of DiT parameters: 550386688
278
+ Total number of SelfAttentionTransformer parameters: 201433088
279
+ Tune action head projector: True
280
+ Tune action head diffusion model: True
281
+
282
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
283
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
284
+ Tune backbone llm: False
285
+ Tune backbone visual: False
286
+ Warning: No backbone trainable parameters found.
287
+ Tune action head projector: False
288
+ Tune action head diffusion model: False
289
+ Action head trainable parameter: future_tokens.weight
290
+ Action head trainable parameter: vlln.weight
291
+ Action head trainable parameter: vlln.bias
292
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
293
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
294
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
295
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
296
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
297
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
298
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
299
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
300
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
301
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
302
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
303
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
304
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
305
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
306
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
307
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
308
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
309
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
310
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
311
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
312
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
313
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
314
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
315
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
316
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
317
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
318
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
319
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
320
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
321
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
322
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
323
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
324
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
325
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
326
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
327
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
328
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
329
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
330
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
331
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
332
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
333
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
334
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
335
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
336
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
337
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
338
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
339
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
340
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
341
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
342
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
343
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
344
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
345
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
346
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
347
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
348
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
349
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
350
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
351
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
352
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
353
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
354
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
355
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
356
+ Applied trainable preset: processing_line_only
357
+ Trainable parameter tensors after preset: 66
358
+ trainable: action_head.vlln.weight
359
+ trainable: action_head.vlln.bias
360
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
361
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
362
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
363
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
364
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
365
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
366
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
367
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
368
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
369
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
370
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
371
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
372
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
373
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
374
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
375
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
376
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
377
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
378
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
379
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
380
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
381
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
382
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
383
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
384
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
385
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
386
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
387
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
388
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
389
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
390
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
391
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
392
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
393
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
394
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
395
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
396
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
397
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
398
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
399
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
400
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
401
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
402
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
403
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
404
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
405
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
406
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
407
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
408
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
409
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
410
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
411
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
412
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
413
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
414
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
415
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
416
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
417
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
418
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
419
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
420
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
421
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
422
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
423
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
424
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
425
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
426
+ Total number of SelfAttentionTransformer parameters: 201433088
427
+ Tune action head projector: True
428
+ Tune action head diffusion model: True
429
+
430
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
431
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
432
+ Tune backbone llm: False
433
+ Tune backbone visual: False
434
+ Warning: No backbone trainable parameters found.
435
+ Tune action head projector: False
436
+ Tune action head diffusion model: False
437
+ Action head trainable parameter: future_tokens.weight
438
+ Action head trainable parameter: vlln.weight
439
+ Action head trainable parameter: vlln.bias
440
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
441
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
442
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
443
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
444
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
445
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
446
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
447
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
448
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
449
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
450
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
451
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
452
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
453
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
454
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
455
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
456
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
457
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
458
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
459
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
460
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
461
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
462
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
463
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
464
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
465
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
466
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
467
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
468
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
469
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
470
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
471
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
472
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
473
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
474
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
475
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
476
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
477
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
478
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
479
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
480
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
481
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
482
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
483
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
484
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
485
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
486
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
487
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
488
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
489
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
490
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
491
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
492
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
493
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
494
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
495
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
496
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
497
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
498
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
499
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
500
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
501
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
502
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
503
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
504
+ Applied trainable preset: processing_line_only
505
+ Trainable parameter tensors after preset: 66
506
+ trainable: action_head.vlln.weight
507
+ trainable: action_head.vlln.bias
508
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
509
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
510
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
511
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
512
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
513
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
514
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
515
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
516
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
517
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
518
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
519
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
520
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
521
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
522
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
523
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
524
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
525
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
526
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
527
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
528
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
529
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
530
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
531
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
532
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
533
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
534
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
535
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
536
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
537
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
538
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
539
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
540
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
541
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
542
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
543
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
544
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
545
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
546
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
547
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
548
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
549
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
550
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
551
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
552
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
553
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
554
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
555
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
556
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
557
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
558
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
559
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
560
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
561
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
562
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
563
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
564
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
565
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
566
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
567
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
568
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
569
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
570
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
571
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
572
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
573
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
574
+ train dataloader length: 6873
575
+ train dataset length: 439854
576
+ GPU memory before training: 7.111904144287109 GB
577
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
578
+ train dataloader length: 6873
579
+ train dataset length: 439854
580
+ GPU memory before training: 7.111904144287109 GB
581
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
582
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
583
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
584
+ wandb: setting up run mtiqrjf1
585
+ wandb: Tracking run with wandb version 0.25.0
586
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_131153-mtiqrjf1
587
+ wandb: Run `wandb offline` to turn off syncing.
588
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
589
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/huggingface
590
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/mtiqrjf1
591
+
592
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
593
+ [rank1]: Traceback (most recent call last):
594
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
595
+ [rank1]: run_yaml_experiment(
596
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
597
+ [rank1]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
598
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
599
+ [rank1]: experiment.train()
600
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
601
+ [rank1]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
602
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
603
+ [rank1]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
604
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
605
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
606
+ [rank1]: return inner_training_loop(
607
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^
608
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
609
+ [rank1]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
610
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
611
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3782, in training_step
612
+ [rank1]: self.accelerator.backward(loss, **kwargs)
613
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/accelerator.py", line 2838, in backward
614
+ [rank1]: loss.backward(**kwargs)
615
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/_tensor.py", line 648, in backward
616
+ [rank1]: torch.autograd.backward(
617
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/__init__.py", line 353, in backward
618
+ [rank1]: _engine_run_backward(
619
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
620
+ [rank1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
621
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
622
+ [rank1]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.09 GiB. GPU 1 has a total capacity of 139.80 GiB of which 17.41 GiB is free. Process 1961377 has 11.53 GiB memory in use. Including non-PyTorch memory, this process has 98.58 GiB memory in use. Of the allocated memory 96.84 GiB is allocated by PyTorch, and 202.72 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
623
+ wandb: uploading output.log; uploading wandb-summary.json; uploading config.yaml
624
+ wandb: uploading summary
625
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/mtiqrjf1
626
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface
627
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
628
+ wandb: Find logs at: ./wandb/run-20260622_131153-mtiqrjf1/logs
629
+ Traceback (most recent call last):
630
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
631
+ run_yaml_experiment(
632
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
633
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
634
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
635
+ experiment.train()
636
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
637
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
638
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
639
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
640
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
641
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
642
+ return inner_training_loop(
643
+ ^^^^^^^^^^^^^^^^^^^^
644
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
645
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
646
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
647
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3782, in training_step
648
+ self.accelerator.backward(loss, **kwargs)
649
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/accelerator.py", line 2838, in backward
650
+ loss.backward(**kwargs)
651
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/_tensor.py", line 648, in backward
652
+ torch.autograd.backward(
653
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/__init__.py", line 353, in backward
654
+ _engine_run_backward(
655
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
656
+ return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
657
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
658
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.09 GiB. GPU 0 has a total capacity of 139.80 GiB of which 15.12 GiB is free. Including non-PyTorch memory, this process has 124.66 GiB memory in use. Of the allocated memory 122.94 GiB is allocated by PyTorch, and 182.72 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
659
+ [rank0]: Traceback (most recent call last):
660
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
661
+ [rank0]: run_yaml_experiment(
662
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
663
+ [rank0]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
664
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
665
+ [rank0]: experiment.train()
666
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
667
+ [rank0]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
668
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
669
+ [rank0]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
670
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
671
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
672
+ [rank0]: return inner_training_loop(
673
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^
674
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
675
+ [rank0]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
676
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
677
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3782, in training_step
678
+ [rank0]: self.accelerator.backward(loss, **kwargs)
679
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/accelerator.py", line 2838, in backward
680
+ [rank0]: loss.backward(**kwargs)
681
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/_tensor.py", line 648, in backward
682
+ [rank0]: torch.autograd.backward(
683
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/__init__.py", line 353, in backward
684
+ [rank0]: _engine_run_backward(
685
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
686
+ [rank0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
687
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
688
+ [rank0]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.09 GiB. GPU 0 has a total capacity of 139.80 GiB of which 15.12 GiB is free. Including non-PyTorch memory, this process has 124.66 GiB memory in use. Of the allocated memory 122.94 GiB is allocated by PyTorch, and 182.72 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
689
+ [rank0]:[W622 13:12:05.494464865 ProcessGroupNCCL.cpp:1479] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
690
+ W0622 13:12:05.945000 2646358 site-packages/torch/distributed/elastic/multiprocessing/api.py:900] Sending process 2646464 closing signal SIGTERM
691
+ E0622 13:12:07.415000 2646358 site-packages/torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1) local_rank: 1 (pid: 2646465) of binary: /home/ext_minje/miniforge3/envs/robocasa/bin/python
692
+ Traceback (most recent call last):
693
+ File "<frozen runpy>", line 198, in _run_module_as_main
694
+ File "<frozen runpy>", line 88, in _run_code
695
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 896, in <module>
696
+ main()
697
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
698
+ return f(*args, **kwargs)
699
+ ^^^^^^^^^^^^^^^^^^
700
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 892, in main
701
+ run(args)
702
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 883, in run
703
+ elastic_launch(
704
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 139, in __call__
705
+ return launch_agent(self._config, self._entrypoint, list(args))
706
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
707
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
708
+ raise ChildFailedError(
709
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
710
+ ============================================================
711
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py FAILED
712
+ ------------------------------------------------------------
713
+ Failures:
714
+ <NO_OTHER_FAILURES>
715
+ ------------------------------------------------------------
716
+ Root Cause (first observed failure):
717
+ [0]:
718
+ time : 2026-06-22_13:12:05
719
+ host : DGX-H200-01
720
+ rank : 1 (local_rank: 1)
721
+ exitcode : 1 (pid: 2646465)
722
+ error_file: <N/A>
723
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
724
+ ============================================================
725
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
726
+
727
+ ==================================================
728
+ GR00T FINE-TUNING CONFIGURATION:
729
+ ==================================================
730
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
731
+ dataset_soup: None
732
+ output_dir: /tmp/gr00t
733
+ output_root: None
734
+ data_config: panda_omron
735
+ batch_size: 32
736
+ max_steps: 300000
737
+ num_gpus: 2
738
+ save_steps: 20000
739
+ run_name: None
740
+ save_total_limit: 100
741
+ seed: 42
742
+ base_model_path: nvidia/GR00T-N1.5-3B
743
+ tune_llm: False
744
+ tune_visual: False
745
+ tune_projector: True
746
+ tune_diffusion_model: True
747
+ resume: False
748
+ learning_rate: 3e-05
749
+ weight_decay: 1e-05
750
+ warmup_ratio: 0.05
751
+ lora_rank: 0
752
+ lora_alpha: 16
753
+ lora_dropout: 0.1
754
+ lora_full_model: False
755
+ dataloader_num_workers: 8
756
+ report_to: wandb
757
+ embodiment_tag: new_embodiment
758
+ video_backend: opencv
759
+ balance_dataset_weights: True
760
+ balance_trajectory_weights: True
761
+ ds_weights_alpha: 0.4
762
+ ==================================================
763
+
764
+ Using 2 GPUs
765
+ Running torchrun command: ['/home/ext_minje/miniforge3/envs/robocasa/bin/python', '-m', 'torch.distributed.run', '--standalone', '--nproc_per_node=2', '--nnodes=1', '/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py', '--config', '/home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml', '--batch-size', '32', '--num-gpus', '2']
rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu45_bs32_20260622_131131.log.pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2646051
rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu45_bs64_20260622_130844.log ADDED
@@ -0,0 +1,871 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s][rank1]: Traceback (most recent call last):
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CMD: CUDA_VISIBLE_DEVICES=4,5 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True /home/ext_minje/miniforge3/envs/robocasa/bin/python /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py --config /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml --num-gpus 2 --batch-size 64
2
+ [robosuite WARNING] No private macro file found! (macros.py:57)
3
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
4
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
5
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
6
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
7
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
8
+ check_for_updates()
9
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
10
+
11
+ *****************************************
12
+ Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
13
+ *****************************************
14
+ [robosuite WARNING] No private macro file found! (macros.py:57)
15
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
16
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
17
+ [robosuite WARNING] No private macro file found! (macros.py:57)
18
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
19
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
20
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
21
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
22
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
23
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
24
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
25
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
26
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
27
+ check_for_updates()
28
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
29
+ check_for_updates()
30
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
31
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
32
+
33
+ ==================================================
34
+ GR00T FINE-TUNING CONFIGURATION:
35
+ ==================================================
36
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
37
+ dataset_soup: None
38
+ output_dir: /tmp/gr00t
39
+ output_root: None
40
+ data_config: panda_omron
41
+ batch_size: 64
42
+ max_steps: 300000
43
+ num_gpus: 2
44
+ save_steps: 20000
45
+ run_name: None
46
+ save_total_limit: 100
47
+ seed: 42
48
+ base_model_path: nvidia/GR00T-N1.5-3B
49
+ tune_llm: False
50
+ tune_visual: False
51
+ tune_projector: True
52
+ tune_diffusion_model: True
53
+ resume: False
54
+ learning_rate: 3e-05
55
+ weight_decay: 1e-05
56
+ warmup_ratio: 0.05
57
+ lora_rank: 0
58
+ lora_alpha: 16
59
+ lora_dropout: 0.1
60
+ lora_full_model: False
61
+ dataloader_num_workers: 8
62
+ report_to: wandb
63
+ embodiment_tag: new_embodiment
64
+ video_backend: opencv
65
+ balance_dataset_weights: True
66
+ balance_trajectory_weights: True
67
+ ds_weights_alpha: 0.4
68
+ ==================================================
69
+
70
+
71
+ ==================================================
72
+ GR00T FINE-TUNING CONFIGURATION:
73
+ ==================================================
74
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
75
+ dataset_soup: None
76
+ output_dir: /tmp/gr00t
77
+ output_root: None
78
+ data_config: panda_omron
79
+ batch_size: 64
80
+ max_steps: 300000
81
+ num_gpus: 2
82
+ save_steps: 20000
83
+ run_name: None
84
+ save_total_limit: 100
85
+ seed: 42
86
+ base_model_path: nvidia/GR00T-N1.5-3B
87
+ tune_llm: False
88
+ tune_visual: False
89
+ tune_projector: True
90
+ tune_diffusion_model: True
91
+ resume: False
92
+ learning_rate: 3e-05
93
+ weight_decay: 1e-05
94
+ warmup_ratio: 0.05
95
+ lora_rank: 0
96
+ lora_alpha: 16
97
+ lora_dropout: 0.1
98
+ lora_full_model: False
99
+ dataloader_num_workers: 8
100
+ report_to: wandb
101
+ embodiment_tag: new_embodiment
102
+ video_backend: opencv
103
+ balance_dataset_weights: True
104
+ balance_trajectory_weights: True
105
+ ds_weights_alpha: 0.4
106
+ ==================================================
107
+
108
+ Using 2 GPUs
109
+
110
+ ================================================================================
111
+ Starting sweep branch: default
112
+ Sweep vars: {}
113
+ ================================================================================
114
+
115
+ --------------------------------------------------------------------------------
116
+ Running phase 1: phase2_rkd_da_only
117
+ Policy type: groot_rkd_v2
118
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
119
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
120
+ Trainable preset: processing_line_only
121
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_action_encoder_projector_enabled': True, 'rkd_action_encoder_projector_dim': 512, 'rkd_action_encoder_projector_pooling': 'flatten', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
122
+ --------------------------------------------------------------------------------
123
+
124
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
125
+ Using 100 subset demos for filter_key: 100_demos
126
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
127
+ self.statistics[key] = torch.tensor(value)
128
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
129
+ Using 2 GPUs
130
+ Using 100 subset demos for filter_key: 100_demos
131
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
132
+ Using 100 subset demos for filter_key: 100_demos
133
+
134
+ ================================================================================
135
+ Starting sweep branch: default
136
+ Sweep vars: {}
137
+ ================================================================================
138
+
139
+ --------------------------------------------------------------------------------
140
+ Running phase 1: phase2_rkd_da_only
141
+ Policy type: groot_rkd_v2
142
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
143
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
144
+ Trainable preset: processing_line_only
145
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_action_encoder_projector_enabled': True, 'rkd_action_encoder_projector_dim': 512, 'rkd_action_encoder_projector_pooling': 'flatten', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
146
+ --------------------------------------------------------------------------------
147
+
148
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
149
+ Using 100 subset demos for filter_key: 100_demos
150
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
151
+ Using 100 subset demos for filter_key: 100_demos
152
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
153
+ self.statistics[key] = torch.tensor(value)
154
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
155
+ Using 100 subset demos for filter_key: 100_demos
156
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
157
+ Using 100 subset demos for filter_key: 100_demos
158
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
159
+ Using 100 subset demos for filter_key: 100_demos
160
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
161
+ Using 100 subset demos for filter_key: 100_demos
162
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
163
+ Using 100 subset demos for filter_key: 100_demos
164
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
165
+ Using 100 subset demos for filter_key: 100_demos
166
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
167
+ Using 100 subset demos for filter_key: 100_demos
168
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
169
+ Using 100 subset demos for filter_key: 100_demos
170
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
171
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
172
+ Using 100 subset demos for filter_key: 100_demos
173
+ Using 100 subset demos for filter_key: 100_demos
174
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
175
+ Using 100 subset demos for filter_key: 100_demos
176
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
177
+ Using 100 subset demos for filter_key: 100_demos
178
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
179
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
180
+ Using 100 subset demos for filter_key: 100_demos
181
+ Using 100 subset demos for filter_key: 100_demos
182
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
183
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
184
+ Using 100 subset demos for filter_key: 100_demos
185
+ Using 100 subset demos for filter_key: 100_demos
186
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
187
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
188
+ Using 100 subset demos for filter_key: 100_demos
189
+ Using 100 subset demos for filter_key: 100_demos
190
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
191
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
192
+ Using 100 subset demos for filter_key: 100_demos
193
+ Using 100 subset demos for filter_key: 100_demos
194
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
195
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
196
+ Using 100 subset demos for filter_key: 100_demos
197
+ Using 100 subset demos for filter_key: 100_demos
198
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
199
+ Using 100 subset demos for filter_key: 100_demos
200
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
201
+ Using 100 subset demos for filter_key: 100_demos
202
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
203
+ Using 100 subset demos for filter_key: 100_demos
204
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
205
+ Using 100 subset demos for filter_key: 100_demos
206
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
207
+ Using 100 subset demos for filter_key: 100_demos
208
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
209
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
210
+ Using 100 subset demos for filter_key: 100_demos
211
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
212
+ Using 100 subset demos for filter_key: 100_demos
213
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
214
+ Using 100 subset demos for filter_key: 100_demos
215
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
216
+ Using 100 subset demos for filter_key: 100_demos
217
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
218
+ Using 100 subset demos for filter_key: 100_demos
219
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
220
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
221
+ Using 100 subset demos for filter_key: 100_demos
222
+ Using 100 subset demos for filter_key: 100_demos
223
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
224
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
225
+ Using 100 subset demos for filter_key: 100_demos
226
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
227
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
228
+ Using 100 subset demos for filter_key: 100_demos
229
+ Using 100 subset demos for filter_key: 100_demos
230
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
231
+ Using 100 subset demos for filter_key: 100_demos
232
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
233
+ Using 100 subset demos for filter_key: 100_demos
234
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
235
+ Using 100 subset demos for filter_key: 100_demos
236
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
237
+ Using 100 subset demos for filter_key: 100_demos
238
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
239
+ Using 100 subset demos for filter_key: 100_demos
240
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
241
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
242
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
243
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
244
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
245
+ 0.75517122 0.7973985 ]
246
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
247
+ Using 100 subset demos for filter_key: 100_demos
248
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
249
+ Using 100 subset demos for filter_key: 100_demos
250
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
251
+ Using 100 subset demos for filter_key: 100_demos
252
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
253
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
254
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
255
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
256
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
257
+ 0.75517122 0.7973985 ]
258
+ Loaded 26 datasets
259
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
260
+ Tune backbone vision tower: False
261
+ Tune backbone LLM: False
262
+ Tune action head projector: False
263
+ Tune action head DiT: False
264
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
265
+ Loaded 26 datasets
266
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
267
+ Tune backbone vision tower: False
268
+ Tune backbone LLM: False
269
+ Tune action head projector: False
270
+ Tune action head DiT: False
271
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
272
+ Tune backbone llm: False
273
+ Tune backbone visual: True
274
+ Total number of DiT parameters: 550386688
275
+ Tune backbone llm: False
276
+ Tune backbone visual: True
277
+ Total number of DiT parameters: 550386688
278
+ Total number of SelfAttentionTransformer parameters: 201433088
279
+ Tune action head projector: True
280
+ Tune action head diffusion model: True
281
+ Total number of SelfAttentionTransformer parameters: 201433088
282
+ Tune action head projector: True
283
+ Tune action head diffusion model: True
284
+
285
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
286
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
287
+ Tune backbone llm: False
288
+ Tune backbone visual: False
289
+ Warning: No backbone trainable parameters found.
290
+ Tune action head projector: False
291
+ Tune action head diffusion model: False
292
+ Action head trainable parameter: future_tokens.weight
293
+ Action head trainable parameter: vlln.weight
294
+ Action head trainable parameter: vlln.bias
295
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
296
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
297
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
298
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
299
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
300
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
301
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
302
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
303
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
304
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
305
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
306
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
307
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
308
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
309
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
310
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
311
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
312
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
313
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
314
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
315
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
316
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
317
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
318
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
319
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
320
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
321
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
322
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
323
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
324
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
325
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
326
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
327
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
328
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
329
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
330
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
331
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
332
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
333
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
334
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
335
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
336
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
337
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
338
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
339
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
340
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
341
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
342
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
343
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
344
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
345
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
346
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
347
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
348
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
349
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
350
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
351
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
352
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
353
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
354
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
355
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
356
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
357
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
358
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
359
+ Applied trainable preset: processing_line_only
360
+ Trainable parameter tensors after preset: 66
361
+ trainable: action_head.vlln.weight
362
+ trainable: action_head.vlln.bias
363
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
364
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
365
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
366
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
367
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
368
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
369
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
370
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
371
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
372
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
373
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
374
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
375
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
376
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
377
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
378
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
379
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
380
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
381
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
382
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
383
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
384
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
385
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
386
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
387
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
388
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
389
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
390
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
391
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
392
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
393
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
394
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
395
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
396
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
397
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
398
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
399
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
400
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
401
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
402
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
403
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
404
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
405
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
406
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
407
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
408
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
409
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
410
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
411
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
412
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
413
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
414
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
415
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
416
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
417
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
418
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
419
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
420
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
421
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
422
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
423
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
424
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
425
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
426
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
427
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
428
+
429
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
430
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
431
+ Tune backbone llm: False
432
+ Tune backbone visual: False
433
+ Warning: No backbone trainable parameters found.
434
+ Tune action head projector: False
435
+ Tune action head diffusion model: False
436
+ Action head trainable parameter: future_tokens.weight
437
+ Action head trainable parameter: vlln.weight
438
+ Action head trainable parameter: vlln.bias
439
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
440
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
441
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
442
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
443
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
444
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
445
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
446
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
447
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
448
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
449
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
450
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
451
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
452
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
453
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
454
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
455
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
456
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
457
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
458
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
459
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
460
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
461
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
462
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
463
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
464
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
465
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
466
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
467
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
468
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
469
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
470
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
471
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
472
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
473
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
474
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
475
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
476
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
477
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
478
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
479
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
480
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
481
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
482
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
483
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
484
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
485
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
486
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
487
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
488
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
489
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
490
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
491
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
492
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
493
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
494
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
495
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
496
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
497
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
498
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
499
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
500
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
501
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
502
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
503
+ Applied trainable preset: processing_line_only
504
+ Trainable parameter tensors after preset: 66
505
+ trainable: action_head.vlln.weight
506
+ trainable: action_head.vlln.bias
507
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
508
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
509
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
510
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
511
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
512
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
513
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
514
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
515
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
516
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
517
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
518
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
519
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
520
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
521
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
522
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
523
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
524
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
525
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
526
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
527
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
528
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
529
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
530
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
531
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
532
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
533
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
534
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
535
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
536
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
537
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
538
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
539
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
540
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
541
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
542
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
543
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
544
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
545
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
546
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
547
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
548
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
549
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
550
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
551
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
552
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
553
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
554
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
555
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
556
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
557
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
558
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
559
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
560
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
561
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
562
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
563
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
564
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
565
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
566
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
567
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
568
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
569
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
570
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
571
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
572
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
573
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
574
+ train dataloader length: 3437
575
+ train dataset length: 439854
576
+ GPU memory before training: 7.111904144287109 GB
577
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
578
+ train dataloader length: 3437
579
+ train dataset length: 439854
580
+ GPU memory before training: 7.111904144287109 GB
581
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
582
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
583
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
584
+ wandb: setting up run iru7hsxj
585
+ wandb: Tracking run with wandb version 0.25.0
586
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_130913-iru7hsxj
587
+ wandb: Run `wandb offline` to turn off syncing.
588
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
589
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/huggingface
590
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/iru7hsxj
591
+
592
  0%| | 0/30000 [00:00<?, ?it/s][rank1]: Traceback (most recent call last):
593
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
594
+ [rank1]: run_yaml_experiment(
595
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
596
+ [rank1]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
597
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
598
+ [rank1]: experiment.train()
599
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
600
+ [rank1]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
601
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
602
+ [rank1]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
603
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
604
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
605
+ [rank1]: return inner_training_loop(
606
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^
607
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
608
+ [rank1]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
609
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
610
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
611
+ [rank1]: loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
612
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
613
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
614
+ [rank1]: outputs = model(inputs)
615
+ [rank1]: ^^^^^^^^^^^^^
616
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
617
+ [rank1]: return self._call_impl(*args, **kwargs)
618
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
619
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
620
+ [rank1]: return forward_call(*args, **kwargs)
621
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
622
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1637, in forward
623
+ [rank1]: else self._run_ddp_forward(*inputs, **kwargs)
624
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
625
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1464, in _run_ddp_forward
626
+ [rank1]: return self.module(*inputs, **kwargs) # type: ignore[index]
627
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
628
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
629
+ [rank1]: return self._call_impl(*args, **kwargs)
630
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
631
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
632
+ [rank1]: return forward_call(*args, **kwargs)
633
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
634
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
635
+ [rank1]: return model_forward(*args, **kwargs)
636
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
637
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
638
+ [rank1]: return convert_to_fp32(self.model_forward(*args, **kwargs))
639
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
640
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
641
+ [rank1]: return func(*args, **kwargs)
642
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^
643
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 597, in forward
644
+ [rank1]: self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
645
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 529, in _add_rkd_loss
646
+ [rank1]: rkd_loss, rkd_metrics = self._compute_rkd_loss(
647
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^
648
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 436, in _compute_rkd_loss
649
+ [rank1]: angle_loss = rkd_angle_loss(student_vector, teacher_vector)
650
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
651
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
652
+ [rank1]: student_angle = _angle_relation(student, eps=eps)
653
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
654
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
655
+ [rank1]: diff = x[:, None, :] - x[None, :, :]
656
+ [rank1]: ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
657
+ [rank1]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 1 has a total capacity of 139.80 GiB of which 80.45 GiB is free. Process 1961377 has 11.53 GiB memory in use. Including non-PyTorch memory, this process has 35.91 GiB memory in use. Of the allocated memory 34.24 GiB is allocated by PyTorch, and 196.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
658
+ wandb: updating run metadata
659
+ wandb: uploading output.log; uploading wandb-summary.json; uploading config.yaml
660
+ wandb: uploading output.log; uploading config.yaml
661
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/iru7hsxj
662
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface
663
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
664
+ wandb: Find logs at: ./wandb/run-20260622_130913-iru7hsxj/logs
665
+ Traceback (most recent call last):
666
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
667
+ run_yaml_experiment(
668
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
669
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
670
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
671
+ experiment.train()
672
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
673
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
674
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
675
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
676
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
677
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
678
+ return inner_training_loop(
679
+ ^^^^^^^^^^^^^^^^^^^^
680
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
681
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
682
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
683
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
684
+ loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
685
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
686
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
687
+ outputs = model(inputs)
688
+ ^^^^^^^^^^^^^
689
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
690
+ return self._call_impl(*args, **kwargs)
691
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
692
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
693
+ return forward_call(*args, **kwargs)
694
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
695
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1637, in forward
696
+ else self._run_ddp_forward(*inputs, **kwargs)
697
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
698
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1464, in _run_ddp_forward
699
+ return self.module(*inputs, **kwargs) # type: ignore[index]
700
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
701
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
702
+ return self._call_impl(*args, **kwargs)
703
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
704
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
705
+ return forward_call(*args, **kwargs)
706
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
707
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
708
+ return model_forward(*args, **kwargs)
709
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
710
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
711
+ return convert_to_fp32(self.model_forward(*args, **kwargs))
712
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
713
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
714
+ return func(*args, **kwargs)
715
+ ^^^^^^^^^^^^^^^^^^^^^
716
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 597, in forward
717
+ self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
718
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 529, in _add_rkd_loss
719
+ rkd_loss, rkd_metrics = self._compute_rkd_loss(
720
+ ^^^^^^^^^^^^^^^^^^^^^^^
721
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 436, in _compute_rkd_loss
722
+ angle_loss = rkd_angle_loss(student_vector, teacher_vector)
723
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
724
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
725
+ student_angle = _angle_relation(student, eps=eps)
726
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
727
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
728
+ diff = x[:, None, :] - x[None, :, :]
729
+ ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
730
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 103.84 GiB is free. Including non-PyTorch memory, this process has 35.93 GiB memory in use. Of the allocated memory 34.24 GiB is allocated by PyTorch, and 216.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
731
+ [rank0]: Traceback (most recent call last):
732
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
733
+ [rank0]: run_yaml_experiment(
734
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
735
+ [rank0]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
736
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
737
+ [rank0]: experiment.train()
738
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
739
+ [rank0]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
740
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
741
+ [rank0]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
742
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
743
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
744
+ [rank0]: return inner_training_loop(
745
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^
746
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
747
+ [rank0]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
748
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
749
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
750
+ [rank0]: loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
751
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
752
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
753
+ [rank0]: outputs = model(inputs)
754
+ [rank0]: ^^^^^^^^^^^^^
755
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
756
+ [rank0]: return self._call_impl(*args, **kwargs)
757
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
758
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
759
+ [rank0]: return forward_call(*args, **kwargs)
760
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
761
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1637, in forward
762
+ [rank0]: else self._run_ddp_forward(*inputs, **kwargs)
763
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
764
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1464, in _run_ddp_forward
765
+ [rank0]: return self.module(*inputs, **kwargs) # type: ignore[index]
766
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
767
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
768
+ [rank0]: return self._call_impl(*args, **kwargs)
769
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
770
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
771
+ [rank0]: return forward_call(*args, **kwargs)
772
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
773
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
774
+ [rank0]: return model_forward(*args, **kwargs)
775
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
776
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
777
+ [rank0]: return convert_to_fp32(self.model_forward(*args, **kwargs))
778
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
779
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
780
+ [rank0]: return func(*args, **kwargs)
781
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^
782
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 597, in forward
783
+ [rank0]: self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
784
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 529, in _add_rkd_loss
785
+ [rank0]: rkd_loss, rkd_metrics = self._compute_rkd_loss(
786
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^
787
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 436, in _compute_rkd_loss
788
+ [rank0]: angle_loss = rkd_angle_loss(student_vector, teacher_vector)
789
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
790
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
791
+ [rank0]: student_angle = _angle_relation(student, eps=eps)
792
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
793
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
794
+ [rank0]: diff = x[:, None, :] - x[None, :, :]
795
+ [rank0]: ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
796
+ [rank0]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 103.84 GiB is free. Including non-PyTorch memory, this process has 35.93 GiB memory in use. Of the allocated memory 34.24 GiB is allocated by PyTorch, and 216.89 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
797
+ W0622 13:09:28.695000 2547300 site-packages/torch/distributed/elastic/multiprocessing/api.py:900] Sending process 2547426 closing signal SIGTERM
798
+ E0622 13:09:29.312000 2547300 site-packages/torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1) local_rank: 1 (pid: 2547427) of binary: /home/ext_minje/miniforge3/envs/robocasa/bin/python
799
+ Traceback (most recent call last):
800
+ File "<frozen runpy>", line 198, in _run_module_as_main
801
+ File "<frozen runpy>", line 88, in _run_code
802
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 896, in <module>
803
+ main()
804
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
805
+ return f(*args, **kwargs)
806
+ ^^^^^^^^^^^^^^^^^^
807
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 892, in main
808
+ run(args)
809
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 883, in run
810
+ elastic_launch(
811
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 139, in __call__
812
+ return launch_agent(self._config, self._entrypoint, list(args))
813
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
814
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
815
+ raise ChildFailedError(
816
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
817
+ ============================================================
818
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py FAILED
819
+ ------------------------------------------------------------
820
+ Failures:
821
+ <NO_OTHER_FAILURES>
822
+ ------------------------------------------------------------
823
+ Root Cause (first observed failure):
824
+ [0]:
825
+ time : 2026-06-22_13:09:28
826
+ host : DGX-H200-01
827
+ rank : 1 (local_rank: 1)
828
+ exitcode : 1 (pid: 2547427)
829
+ error_file: <N/A>
830
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
831
+ ============================================================
832
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
833
+
834
+ ==================================================
835
+ GR00T FINE-TUNING CONFIGURATION:
836
+ ==================================================
837
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
838
+ dataset_soup: None
839
+ output_dir: /tmp/gr00t
840
+ output_root: None
841
+ data_config: panda_omron
842
+ batch_size: 64
843
+ max_steps: 300000
844
+ num_gpus: 2
845
+ save_steps: 20000
846
+ run_name: None
847
+ save_total_limit: 100
848
+ seed: 42
849
+ base_model_path: nvidia/GR00T-N1.5-3B
850
+ tune_llm: False
851
+ tune_visual: False
852
+ tune_projector: True
853
+ tune_diffusion_model: True
854
+ resume: False
855
+ learning_rate: 3e-05
856
+ weight_decay: 1e-05
857
+ warmup_ratio: 0.05
858
+ lora_rank: 0
859
+ lora_alpha: 16
860
+ lora_dropout: 0.1
861
+ lora_full_model: False
862
+ dataloader_num_workers: 8
863
+ report_to: wandb
864
+ embodiment_tag: new_embodiment
865
+ video_backend: opencv
866
+ balance_dataset_weights: True
867
+ balance_trajectory_weights: True
868
+ ds_weights_alpha: 0.4
869
+ ==================================================
870
+
871
+ Using 2 GPUs
872
+ Running torchrun command: ['/home/ext_minje/miniforge3/envs/robocasa/bin/python', '-m', 'torch.distributed.run', '--standalone', '--nproc_per_node=2', '--nnodes=1', '/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py', '--config', '/home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml', '--batch-size', '64', '--num-gpus', '2']
rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu45_bs64_20260622_130844.log.pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2545729
rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu4_bs128_20260622_130630.log ADDED
@@ -0,0 +1,355 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CMD: CUDA_VISIBLE_DEVICES=4 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True /home/ext_minje/miniforge3/envs/robocasa/bin/python /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py --config /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml --num-gpus 1 --batch-size 128
2
+ [robosuite WARNING] No private macro file found! (macros.py:57)
3
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
4
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
5
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
6
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
7
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
8
+ check_for_updates()
9
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
10
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
11
+
12
+ ==================================================
13
+ GR00T FINE-TUNING CONFIGURATION:
14
+ ==================================================
15
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_action_encoder.yaml
16
+ dataset_soup: None
17
+ output_dir: /tmp/gr00t
18
+ output_root: None
19
+ data_config: panda_omron
20
+ batch_size: 128
21
+ max_steps: 300000
22
+ num_gpus: 1
23
+ save_steps: 20000
24
+ run_name: None
25
+ save_total_limit: 100
26
+ seed: 42
27
+ base_model_path: nvidia/GR00T-N1.5-3B
28
+ tune_llm: False
29
+ tune_visual: False
30
+ tune_projector: True
31
+ tune_diffusion_model: True
32
+ resume: False
33
+ learning_rate: 3e-05
34
+ weight_decay: 1e-05
35
+ warmup_ratio: 0.05
36
+ lora_rank: 0
37
+ lora_alpha: 16
38
+ lora_dropout: 0.1
39
+ lora_full_model: False
40
+ dataloader_num_workers: 8
41
+ report_to: wandb
42
+ embodiment_tag: new_embodiment
43
+ video_backend: opencv
44
+ balance_dataset_weights: True
45
+ balance_trajectory_weights: True
46
+ ds_weights_alpha: 0.4
47
+ ==================================================
48
+
49
+ Using 1 GPUs
50
+
51
+ ================================================================================
52
+ Starting sweep branch: default
53
+ Sweep vars: {}
54
+ ================================================================================
55
+
56
+ --------------------------------------------------------------------------------
57
+ Running phase 1: phase2_rkd_da_only
58
+ Policy type: groot_rkd_v2
59
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
60
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2
61
+ Trainable preset: processing_line_only
62
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'action_encoder', 'rkd_action_encoder_projector_enabled': True, 'rkd_action_encoder_projector_dim': 512, 'rkd_action_encoder_projector_pooling': 'flatten', 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
63
+ --------------------------------------------------------------------------------
64
+
65
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
66
+ self.statistics[key] = torch.tensor(value)
67
+
68
+ Using 100 subset demos for filter_key: 100_demos
69
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
70
+ Using 100 subset demos for filter_key: 100_demos
71
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
72
+ Using 100 subset demos for filter_key: 100_demos
73
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
74
+ Using 100 subset demos for filter_key: 100_demos
75
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
76
+ Using 100 subset demos for filter_key: 100_demos
77
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
78
+ Using 100 subset demos for filter_key: 100_demos
79
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
80
+ Using 100 subset demos for filter_key: 100_demos
81
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
82
+ Using 100 subset demos for filter_key: 100_demos
83
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
84
+ Using 100 subset demos for filter_key: 100_demos
85
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
86
+ Using 100 subset demos for filter_key: 100_demos
87
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
88
+ Using 100 subset demos for filter_key: 100_demos
89
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
90
+ Using 100 subset demos for filter_key: 100_demos
91
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
92
+ Using 100 subset demos for filter_key: 100_demos
93
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
94
+ Using 100 subset demos for filter_key: 100_demos
95
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
96
+ Using 100 subset demos for filter_key: 100_demos
97
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
98
+ Using 100 subset demos for filter_key: 100_demos
99
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
100
+ Using 100 subset demos for filter_key: 100_demos
101
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
102
+ Using 100 subset demos for filter_key: 100_demos
103
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
104
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
105
+ Using 100 subset demos for filter_key: 100_demos
106
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
107
+ Using 100 subset demos for filter_key: 100_demos
108
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
109
+ Using 100 subset demos for filter_key: 100_demos
110
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
111
+ Using 100 subset demos for filter_key: 100_demos
112
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
113
+ Using 100 subset demos for filter_key: 100_demos
114
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
115
+ Using 100 subset demos for filter_key: 100_demos
116
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
117
+ Using 100 subset demos for filter_key: 100_demos
118
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
119
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
120
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
121
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
122
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
123
+ 0.75517122 0.7973985 ]
124
+ Loaded 26 datasets
125
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
126
+ Tune backbone vision tower: False
127
+ Tune backbone LLM: False
128
+ Tune action head projector: False
129
+ Tune action head DiT: False
130
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
131
+ Tune backbone llm: False
132
+ Tune backbone visual: True
133
+ Total number of DiT parameters: 550386688
134
+ Total number of SelfAttentionTransformer parameters: 201433088
135
+ Tune action head projector: True
136
+ Tune action head diffusion model: True
137
+
138
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
139
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
140
+ Tune backbone llm: False
141
+ Tune backbone visual: False
142
+ Warning: No backbone trainable parameters found.
143
+ Tune action head projector: False
144
+ Tune action head diffusion model: False
145
+ Action head trainable parameter: future_tokens.weight
146
+ Action head trainable parameter: vlln.weight
147
+ Action head trainable parameter: vlln.bias
148
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
149
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
150
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
151
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
152
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
153
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
154
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
155
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
156
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
157
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
158
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
159
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
160
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
161
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
162
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
163
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
164
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
165
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
166
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
167
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
168
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
169
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
170
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
171
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
172
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
173
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
174
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
175
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
176
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
177
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
178
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
179
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
180
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
181
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
182
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
183
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
184
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
185
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
186
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
187
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
188
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
189
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
190
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
191
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
192
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
193
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
194
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
195
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
196
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
197
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
198
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
199
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
200
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
201
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
202
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
203
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
204
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
205
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
206
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
207
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
208
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
209
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
210
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
211
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
212
+ Applied trainable preset: processing_line_only
213
+ Trainable parameter tensors after preset: 66
214
+ trainable: action_head.vlln.weight
215
+ trainable: action_head.vlln.bias
216
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
217
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
218
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
219
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
220
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
221
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
222
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
223
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
224
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
225
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
226
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
227
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
228
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
229
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
230
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
231
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
232
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
233
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
234
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
235
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
236
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
237
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
238
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
239
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
240
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
241
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
242
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
243
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
244
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
245
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
246
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
247
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
248
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
249
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
250
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
251
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
252
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
253
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
254
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
255
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
256
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
257
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
258
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
259
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
260
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
261
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
262
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
263
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
264
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
265
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
266
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
267
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
268
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
269
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
270
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
271
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
272
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
273
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
274
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
275
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
276
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
277
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
278
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
279
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
280
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
281
+ Run name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
282
+ train dataloader length: 3437
283
+ train dataset length: 439854
284
+ GPU memory before training: 7.111904144287109 GB
285
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
286
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
287
+ wandb: setting up run 27n8xn3s
288
+ wandb: Tracking run with wandb version 0.25.0
289
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_130650-27n8xn3s
290
+ wandb: Run `wandb offline` to turn off syncing.
291
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2
292
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/huggingface
293
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/27n8xn3s
294
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/runs
295
+
296
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
297
+ wandb: uploading summary
298
+ wandb: uploading config.yaml
299
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/27n8xn3s
300
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface
301
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
302
+ wandb: Find logs at: ./wandb/run-20260622_130650-27n8xn3s/logs
303
+ Traceback (most recent call last):
304
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1032, in <module>
305
+ run_yaml_experiment(
306
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
307
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
308
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
309
+ experiment.train()
310
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
311
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
312
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
313
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
314
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
315
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
316
+ return inner_training_loop(
317
+ ^^^^^^^^^^^^^^^^^^^^
318
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
319
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
320
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
321
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
322
+ loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
323
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
324
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
325
+ outputs = model(inputs)
326
+ ^^^^^^^^^^^^^
327
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
328
+ return self._call_impl(*args, **kwargs)
329
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
330
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
331
+ return forward_call(*args, **kwargs)
332
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
333
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
334
+ return model_forward(*args, **kwargs)
335
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
336
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
337
+ return convert_to_fp32(self.model_forward(*args, **kwargs))
338
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
339
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
340
+ return func(*args, **kwargs)
341
+ ^^^^^^^^^^^^^^^^^^^^^
342
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 597, in forward
343
+ self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
344
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 529, in _add_rkd_loss
345
+ rkd_loss, rkd_metrics = self._compute_rkd_loss(
346
+ ^^^^^^^^^^^^^^^^^^^^^^^
347
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 436, in _compute_rkd_loss
348
+ angle_loss = rkd_angle_loss(student_vector, teacher_vector)
349
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
350
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
351
+ student_angle = _angle_relation(student, eps=eps)
352
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
353
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
354
+ diff = x[:, None, :] - x[None, :, :]
355
+ ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
356
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 81.16 GiB is free. Including non-PyTorch memory, this process has 58.62 GiB memory in use. Of the allocated memory 57.79 GiB is allocated by PyTorch, and 165.44 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
rkd_v2_2/logs/rkd_v2_2_flatten_action_encoder_gpu4_bs128_20260622_130630.log.pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2478154
rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu45_bs32_20260622_132220.log ADDED
@@ -0,0 +1,763 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CMD: CUDA_VISIBLE_DEVICES=4,5 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True /home/ext_minje/miniforge3/envs/robocasa/bin/python /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py --config /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml --num-gpus 2 --batch-size 32
2
+ [robosuite WARNING] No private macro file found! (macros.py:57)
3
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
4
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
5
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
6
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
7
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
8
+ check_for_updates()
9
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
10
+
11
+ *****************************************
12
+ Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
13
+ *****************************************
14
+ [robosuite WARNING] No private macro file found! (macros.py:57)
15
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
16
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
17
+ [robosuite WARNING] No private macro file found! (macros.py:57)
18
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
19
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
20
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
21
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
22
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
23
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
24
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
25
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
26
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
27
+ check_for_updates()
28
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
29
+ check_for_updates()
30
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
31
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
32
+
33
+ ==================================================
34
+ GR00T FINE-TUNING CONFIGURATION:
35
+ ==================================================
36
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml
37
+ dataset_soup: None
38
+ output_dir: /tmp/gr00t
39
+ output_root: None
40
+ data_config: panda_omron
41
+ batch_size: 32
42
+ max_steps: 300000
43
+ num_gpus: 2
44
+ save_steps: 20000
45
+ run_name: None
46
+ save_total_limit: 100
47
+ seed: 42
48
+ base_model_path: nvidia/GR00T-N1.5-3B
49
+ tune_llm: False
50
+ tune_visual: False
51
+ tune_projector: True
52
+ tune_diffusion_model: True
53
+ resume: False
54
+ learning_rate: 3e-05
55
+ weight_decay: 1e-05
56
+ warmup_ratio: 0.05
57
+ lora_rank: 0
58
+ lora_alpha: 16
59
+ lora_dropout: 0.1
60
+ lora_full_model: False
61
+ dataloader_num_workers: 8
62
+ report_to: wandb
63
+ embodiment_tag: new_embodiment
64
+ video_backend: opencv
65
+ balance_dataset_weights: True
66
+ balance_trajectory_weights: True
67
+ ds_weights_alpha: 0.4
68
+ ==================================================
69
+
70
+ Using 2 GPUs
71
+
72
+ ================================================================================
73
+ Starting sweep branch: default
74
+ Sweep vars: {}
75
+ ================================================================================
76
+
77
+ --------------------------------------------------------------------------------
78
+ Running phase 1: phase2_rkd_da_only
79
+ Policy type: groot_rkd_v2
80
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
81
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2
82
+ Trainable preset: processing_line_only
83
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'raw_action', 'rkd_action_encoder_projector_enabled': False, 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
84
+ --------------------------------------------------------------------------------
85
+
86
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
87
+ Using 100 subset demos for filter_key: 100_demos
88
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
89
+ self.statistics[key] = torch.tensor(value)
90
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
91
+ Using 100 subset demos for filter_key: 100_demos
92
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
93
+ Using 100 subset demos for filter_key: 100_demos
94
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
95
+ Using 100 subset demos for filter_key: 100_demos
96
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
97
+ Using 100 subset demos for filter_key: 100_demos
98
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
99
+ Using 100 subset demos for filter_key: 100_demos
100
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
101
+ Using 100 subset demos for filter_key: 100_demos
102
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
103
+ Using 100 subset demos for filter_key: 100_demos
104
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
105
+ Using 100 subset demos for filter_key: 100_demos
106
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
107
+ Using 100 subset demos for filter_key: 100_demos
108
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
109
+ Using 100 subset demos for filter_key: 100_demos
110
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
111
+ Using 100 subset demos for filter_key: 100_demos
112
+
113
+ ==================================================
114
+ GR00T FINE-TUNING CONFIGURATION:
115
+ ==================================================
116
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml
117
+ dataset_soup: None
118
+ output_dir: /tmp/gr00t
119
+ output_root: None
120
+ data_config: panda_omron
121
+ batch_size: 32
122
+ max_steps: 300000
123
+ num_gpus: 2
124
+ save_steps: 20000
125
+ run_name: None
126
+ save_total_limit: 100
127
+ seed: 42
128
+ base_model_path: nvidia/GR00T-N1.5-3B
129
+ tune_llm: False
130
+ tune_visual: False
131
+ tune_projector: True
132
+ tune_diffusion_model: True
133
+ resume: False
134
+ learning_rate: 3e-05
135
+ weight_decay: 1e-05
136
+ warmup_ratio: 0.05
137
+ lora_rank: 0
138
+ lora_alpha: 16
139
+ lora_dropout: 0.1
140
+ lora_full_model: False
141
+ dataloader_num_workers: 8
142
+ report_to: wandb
143
+ embodiment_tag: new_embodiment
144
+ video_backend: opencv
145
+ balance_dataset_weights: True
146
+ balance_trajectory_weights: True
147
+ ds_weights_alpha: 0.4
148
+ ==================================================
149
+
150
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
151
+ Using 100 subset demos for filter_key: 100_demos
152
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
153
+ Using 100 subset demos for filter_key: 100_demos
154
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
155
+ Using 100 subset demos for filter_key: 100_demos
156
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
157
+ Using 100 subset demos for filter_key: 100_demos
158
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
159
+ Using 100 subset demos for filter_key: 100_demos
160
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
161
+ Using 100 subset demos for filter_key: 100_demos
162
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
163
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
164
+ Using 100 subset demos for filter_key: 100_demos
165
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
166
+ Using 100 subset demos for filter_key: 100_demos
167
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
168
+ Using 100 subset demos for filter_key: 100_demos
169
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
170
+ Using 100 subset demos for filter_key: 100_demos
171
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
172
+ Using 100 subset demos for filter_key: 100_demos
173
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
174
+ Using 100 subset demos for filter_key: 100_demos
175
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
176
+ Using 100 subset demos for filter_key: 100_demos
177
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
178
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
179
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
180
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
181
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
182
+ 0.75517122 0.7973985 ]
183
+ Loaded 26 datasets
184
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
185
+ Tune backbone vision tower: False
186
+ Tune backbone LLM: False
187
+ Tune action head projector: False
188
+ Tune action head DiT: False
189
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
190
+ Using 2 GPUs
191
+
192
+ ================================================================================
193
+ Starting sweep branch: default
194
+ Sweep vars: {}
195
+ ================================================================================
196
+
197
+ --------------------------------------------------------------------------------
198
+ Running phase 1: phase2_rkd_da_only
199
+ Policy type: groot_rkd_v2
200
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
201
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2
202
+ Trainable preset: processing_line_only
203
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'raw_action', 'rkd_action_encoder_projector_enabled': False, 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
204
+ --------------------------------------------------------------------------------
205
+
206
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
207
+ Using 100 subset demos for filter_key: 100_demos
208
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
209
+ self.statistics[key] = torch.tensor(value)
210
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
211
+ Using 100 subset demos for filter_key: 100_demos
212
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
213
+ Using 100 subset demos for filter_key: 100_demos
214
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
215
+ Using 100 subset demos for filter_key: 100_demos
216
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
217
+ Using 100 subset demos for filter_key: 100_demos
218
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
219
+ Using 100 subset demos for filter_key: 100_demos
220
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
221
+ Using 100 subset demos for filter_key: 100_demos
222
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
223
+ Using 100 subset demos for filter_key: 100_demos
224
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
225
+ Using 100 subset demos for filter_key: 100_demos
226
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
227
+ Using 100 subset demos for filter_key: 100_demos
228
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
229
+ Using 100 subset demos for filter_key: 100_demos
230
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
231
+ Using 100 subset demos for filter_key: 100_demos
232
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
233
+ Using 100 subset demos for filter_key: 100_demos
234
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
235
+ Using 100 subset demos for filter_key: 100_demos
236
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
237
+ Using 100 subset demos for filter_key: 100_demos
238
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
239
+ Using 100 subset demos for filter_key: 100_demos
240
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
241
+ Using 100 subset demos for filter_key: 100_demos
242
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
243
+ Using 100 subset demos for filter_key: 100_demos
244
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
245
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
246
+ Using 100 subset demos for filter_key: 100_demos
247
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
248
+ Using 100 subset demos for filter_key: 100_demos
249
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
250
+ Using 100 subset demos for filter_key: 100_demos
251
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
252
+ Using 100 subset demos for filter_key: 100_demos
253
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
254
+ Using 100 subset demos for filter_key: 100_demos
255
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
256
+ Using 100 subset demos for filter_key: 100_demos
257
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
258
+ Using 100 subset demos for filter_key: 100_demos
259
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
260
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
261
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
262
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
263
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
264
+ 0.75517122 0.7973985 ]
265
+ Loaded 26 datasets
266
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
267
+ Tune backbone vision tower: False
268
+ Tune backbone LLM: False
269
+ Tune action head projector: False
270
+ Tune action head DiT: False
271
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
272
+ Tune backbone llm: False
273
+ Tune backbone visual: True
274
+ Total number of DiT parameters: 550386688
275
+ Tune backbone llm: False
276
+ Tune backbone visual: True
277
+ Total number of DiT parameters: 550386688
278
+ Total number of SelfAttentionTransformer parameters: 201433088
279
+ Tune action head projector: True
280
+ Tune action head diffusion model: True
281
+
282
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
283
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
284
+ Tune backbone llm: False
285
+ Tune backbone visual: False
286
+ Warning: No backbone trainable parameters found.
287
+ Tune action head projector: False
288
+ Tune action head diffusion model: False
289
+ Action head trainable parameter: future_tokens.weight
290
+ Action head trainable parameter: vlln.weight
291
+ Action head trainable parameter: vlln.bias
292
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
293
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
294
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
295
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
296
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
297
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
298
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
299
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
300
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
301
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
302
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
303
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
304
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
305
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
306
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
307
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
308
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
309
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
310
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
311
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
312
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
313
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
314
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
315
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
316
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
317
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
318
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
319
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
320
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
321
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
322
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
323
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
324
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
325
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
326
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
327
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
328
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
329
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
330
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
331
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
332
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
333
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
334
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
335
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
336
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
337
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
338
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
339
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
340
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
341
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
342
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
343
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
344
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
345
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
346
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
347
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
348
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
349
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
350
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
351
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
352
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
353
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
354
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
355
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
356
+ Applied trainable preset: processing_line_only
357
+ Trainable parameter tensors after preset: 66
358
+ trainable: action_head.vlln.weight
359
+ trainable: action_head.vlln.bias
360
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
361
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
362
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
363
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
364
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
365
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
366
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
367
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
368
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
369
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
370
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
371
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
372
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
373
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
374
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
375
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
376
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
377
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
378
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
379
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
380
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
381
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
382
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
383
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
384
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
385
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
386
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
387
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
388
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
389
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
390
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
391
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
392
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
393
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
394
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
395
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
396
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
397
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
398
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
399
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
400
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
401
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
402
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
403
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
404
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
405
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
406
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
407
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
408
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
409
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
410
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
411
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
412
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
413
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
414
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
415
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
416
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
417
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
418
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
419
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
420
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
421
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
422
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
423
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
424
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
425
+ Total number of SelfAttentionTransformer parameters: 201433088
426
+ Tune action head projector: True
427
+ Tune action head diffusion model: True
428
+
429
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
430
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
431
+ Tune backbone llm: False
432
+ Tune backbone visual: False
433
+ Warning: No backbone trainable parameters found.
434
+ Tune action head projector: False
435
+ Tune action head diffusion model: False
436
+ Action head trainable parameter: future_tokens.weight
437
+ Action head trainable parameter: vlln.weight
438
+ Action head trainable parameter: vlln.bias
439
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
440
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
441
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
442
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
443
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
444
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
445
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
446
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
447
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
448
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
449
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
450
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
451
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
452
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
453
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
454
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
455
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
456
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
457
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
458
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
459
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
460
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
461
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
462
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
463
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
464
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
465
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
466
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
467
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
468
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
469
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
470
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
471
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
472
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
473
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
474
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
475
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
476
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
477
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
478
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
479
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
480
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
481
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
482
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
483
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
484
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
485
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
486
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
487
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
488
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
489
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
490
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
491
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
492
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
493
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
494
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
495
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
496
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
497
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
498
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
499
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
500
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
501
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
502
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
503
+ Applied trainable preset: processing_line_only
504
+ Trainable parameter tensors after preset: 66
505
+ trainable: action_head.vlln.weight
506
+ trainable: action_head.vlln.bias
507
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
508
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
509
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
510
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
511
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
512
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
513
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
514
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
515
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
516
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
517
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
518
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
519
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
520
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
521
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
522
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
523
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
524
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
525
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
526
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
527
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
528
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
529
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
530
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
531
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
532
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
533
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
534
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
535
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
536
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
537
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
538
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
539
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
540
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
541
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
542
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
543
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
544
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
545
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
546
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
547
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
548
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
549
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
550
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
551
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
552
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
553
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
554
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
555
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
556
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
557
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
558
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
559
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
560
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
561
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
562
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
563
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
564
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
565
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
566
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
567
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
568
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
569
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
570
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
571
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
572
+ Run name: rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
573
+ Run name: rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
574
+ train dataloader length: 6873
575
+ train dataset length: 439854
576
+ GPU memory before training: 7.111904144287109 GB
577
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2/runs
578
+ train dataloader length: 6873
579
+ train dataset length: 439854
580
+ GPU memory before training: 7.111904144287109 GB
581
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2/runs
582
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
583
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
584
+ wandb: setting up run r9a8rv7x
585
+ wandb: Tracking run with wandb version 0.25.0
586
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_132246-r9a8rv7x
587
+ wandb: Run `wandb offline` to turn off syncing.
588
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
589
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/huggingface
590
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/r9a8rv7x
591
+
592
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
593
+ [rank1]: Traceback (most recent call last):
594
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
595
+ [rank1]: run_yaml_experiment(
596
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
597
+ [rank1]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
598
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
599
+ [rank1]: experiment.train()
600
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
601
+ [rank1]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
602
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
603
+ [rank1]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
604
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
605
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
606
+ [rank1]: return inner_training_loop(
607
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^
608
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
609
+ [rank1]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
610
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
611
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3782, in training_step
612
+ [rank1]: self.accelerator.backward(loss, **kwargs)
613
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/accelerator.py", line 2838, in backward
614
+ [rank1]: loss.backward(**kwargs)
615
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/_tensor.py", line 648, in backward
616
+ [rank1]: torch.autograd.backward(
617
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/__init__.py", line 353, in backward
618
+ [rank1]: _engine_run_backward(
619
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
620
+ [rank1]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
621
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
622
+ [rank1]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.09 GiB. GPU 1 has a total capacity of 139.80 GiB of which 18.08 GiB is free. Process 1961377 has 11.53 GiB memory in use. Including non-PyTorch memory, this process has 98.56 GiB memory in use. Of the allocated memory 96.84 GiB is allocated by PyTorch, and 184.72 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
623
+ wandb: uploading config.yaml
624
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/r9a8rv7x
625
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface
626
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
627
+ wandb: Find logs at: ./wandb/run-20260622_132246-r9a8rv7x/logs
628
+ Traceback (most recent call last):
629
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
630
+ run_yaml_experiment(
631
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
632
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
633
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
634
+ experiment.train()
635
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
636
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
637
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
638
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
639
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
640
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
641
+ return inner_training_loop(
642
+ ^^^^^^^^^^^^^^^^^^^^
643
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
644
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
645
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
646
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3782, in training_step
647
+ self.accelerator.backward(loss, **kwargs)
648
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/accelerator.py", line 2838, in backward
649
+ loss.backward(**kwargs)
650
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/_tensor.py", line 648, in backward
651
+ torch.autograd.backward(
652
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/__init__.py", line 353, in backward
653
+ _engine_run_backward(
654
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
655
+ return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
656
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
657
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.09 GiB. GPU 0 has a total capacity of 139.80 GiB of which 15.18 GiB is free. Including non-PyTorch memory, this process has 124.60 GiB memory in use. Of the allocated memory 122.94 GiB is allocated by PyTorch, and 124.72 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
658
+ [rank0]: Traceback (most recent call last):
659
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
660
+ [rank0]: run_yaml_experiment(
661
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
662
+ [rank0]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
663
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
664
+ [rank0]: experiment.train()
665
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
666
+ [rank0]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
667
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
668
+ [rank0]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
669
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
670
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
671
+ [rank0]: return inner_training_loop(
672
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^
673
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
674
+ [rank0]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
675
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
676
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3782, in training_step
677
+ [rank0]: self.accelerator.backward(loss, **kwargs)
678
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/accelerator.py", line 2838, in backward
679
+ [rank0]: loss.backward(**kwargs)
680
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/_tensor.py", line 648, in backward
681
+ [rank0]: torch.autograd.backward(
682
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/__init__.py", line 353, in backward
683
+ [rank0]: _engine_run_backward(
684
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/autograd/graph.py", line 824, in _engine_run_backward
685
+ [rank0]: return Variable._execution_engine.run_backward( # Calls into the C++ engine to run the backward pass
686
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
687
+ [rank0]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.09 GiB. GPU 0 has a total capacity of 139.80 GiB of which 15.18 GiB is free. Including non-PyTorch memory, this process has 124.60 GiB memory in use. Of the allocated memory 122.94 GiB is allocated by PyTorch, and 124.72 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
688
+ [rank0]:[W622 13:22:59.831661402 ProcessGroupNCCL.cpp:1479] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
689
+ W0622 13:23:01.163000 2934444 site-packages/torch/distributed/elastic/multiprocessing/api.py:900] Sending process 2934554 closing signal SIGTERM
690
+ E0622 13:23:02.681000 2934444 site-packages/torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1) local_rank: 1 (pid: 2934555) of binary: /home/ext_minje/miniforge3/envs/robocasa/bin/python
691
+ Traceback (most recent call last):
692
+ File "<frozen runpy>", line 198, in _run_module_as_main
693
+ File "<frozen runpy>", line 88, in _run_code
694
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 896, in <module>
695
+ main()
696
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
697
+ return f(*args, **kwargs)
698
+ ^^^^^^^^^^^^^^^^^^
699
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 892, in main
700
+ run(args)
701
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 883, in run
702
+ elastic_launch(
703
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 139, in __call__
704
+ return launch_agent(self._config, self._entrypoint, list(args))
705
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
706
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
707
+ raise ChildFailedError(
708
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
709
+ ============================================================
710
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py FAILED
711
+ ------------------------------------------------------------
712
+ Failures:
713
+ <NO_OTHER_FAILURES>
714
+ ------------------------------------------------------------
715
+ Root Cause (first observed failure):
716
+ [0]:
717
+ time : 2026-06-22_13:23:01
718
+ host : DGX-H200-01
719
+ rank : 1 (local_rank: 1)
720
+ exitcode : 1 (pid: 2934555)
721
+ error_file: <N/A>
722
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
723
+ ============================================================
724
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
725
+
726
+ ==================================================
727
+ GR00T FINE-TUNING CONFIGURATION:
728
+ ==================================================
729
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml
730
+ dataset_soup: None
731
+ output_dir: /tmp/gr00t
732
+ output_root: None
733
+ data_config: panda_omron
734
+ batch_size: 32
735
+ max_steps: 300000
736
+ num_gpus: 2
737
+ save_steps: 20000
738
+ run_name: None
739
+ save_total_limit: 100
740
+ seed: 42
741
+ base_model_path: nvidia/GR00T-N1.5-3B
742
+ tune_llm: False
743
+ tune_visual: False
744
+ tune_projector: True
745
+ tune_diffusion_model: True
746
+ resume: False
747
+ learning_rate: 3e-05
748
+ weight_decay: 1e-05
749
+ warmup_ratio: 0.05
750
+ lora_rank: 0
751
+ lora_alpha: 16
752
+ lora_dropout: 0.1
753
+ lora_full_model: False
754
+ dataloader_num_workers: 8
755
+ report_to: wandb
756
+ embodiment_tag: new_embodiment
757
+ video_backend: opencv
758
+ balance_dataset_weights: True
759
+ balance_trajectory_weights: True
760
+ ds_weights_alpha: 0.4
761
+ ==================================================
762
+
763
+ Using 2 GPUs
764
+ Running torchrun command: ['/home/ext_minje/miniforge3/envs/robocasa/bin/python', '-m', 'torch.distributed.run', '--standalone', '--nproc_per_node=2', '--nnodes=1', '/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py', '--config', '/home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml', '--batch-size', '32', '--num-gpus', '2']
rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu45_bs32_20260622_132220.log.pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2934174
rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu45_bs64_20260622_131850.log ADDED
@@ -0,0 +1,871 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s][rank1]: Traceback (most recent call last):
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CMD: CUDA_VISIBLE_DEVICES=4,5 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True /home/ext_minje/miniforge3/envs/robocasa/bin/python /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py --config /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml --num-gpus 2 --batch-size 64
2
+ [robosuite WARNING] No private macro file found! (macros.py:57)
3
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
4
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
5
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
6
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
7
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
8
+ check_for_updates()
9
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
10
+
11
+ *****************************************
12
+ Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
13
+ *****************************************
14
+ [robosuite WARNING] No private macro file found! (macros.py:57)
15
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
16
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
17
+ [robosuite WARNING] No private macro file found! (macros.py:57)
18
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
19
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
20
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
21
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
22
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
23
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
24
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
25
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
26
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
27
+ check_for_updates()
28
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
29
+ check_for_updates()
30
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
31
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
32
+
33
+ ==================================================
34
+ GR00T FINE-TUNING CONFIGURATION:
35
+ ==================================================
36
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml
37
+ dataset_soup: None
38
+ output_dir: /tmp/gr00t
39
+ output_root: None
40
+ data_config: panda_omron
41
+ batch_size: 64
42
+ max_steps: 300000
43
+ num_gpus: 2
44
+ save_steps: 20000
45
+ run_name: None
46
+ save_total_limit: 100
47
+ seed: 42
48
+ base_model_path: nvidia/GR00T-N1.5-3B
49
+ tune_llm: False
50
+ tune_visual: False
51
+ tune_projector: True
52
+ tune_diffusion_model: True
53
+ resume: False
54
+ learning_rate: 3e-05
55
+ weight_decay: 1e-05
56
+ warmup_ratio: 0.05
57
+ lora_rank: 0
58
+ lora_alpha: 16
59
+ lora_dropout: 0.1
60
+ lora_full_model: False
61
+ dataloader_num_workers: 8
62
+ report_to: wandb
63
+ embodiment_tag: new_embodiment
64
+ video_backend: opencv
65
+ balance_dataset_weights: True
66
+ balance_trajectory_weights: True
67
+ ds_weights_alpha: 0.4
68
+ ==================================================
69
+
70
+ Using 2 GPUs
71
+
72
+ ================================================================================
73
+ Starting sweep branch: default
74
+ Sweep vars: {}
75
+ ================================================================================
76
+
77
+ --------------------------------------------------------------------------------
78
+ Running phase 1: phase2_rkd_da_only
79
+ Policy type: groot_rkd_v2
80
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
81
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2
82
+ Trainable preset: processing_line_only
83
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'raw_action', 'rkd_action_encoder_projector_enabled': False, 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
84
+ --------------------------------------------------------------------------------
85
+
86
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
87
+ Using 100 subset demos for filter_key: 100_demos
88
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
89
+ self.statistics[key] = torch.tensor(value)
90
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
91
+ Using 100 subset demos for filter_key: 100_demos
92
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
93
+ Using 100 subset demos for filter_key: 100_demos
94
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
95
+ Using 100 subset demos for filter_key: 100_demos
96
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
97
+ Using 100 subset demos for filter_key: 100_demos
98
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
99
+ Using 100 subset demos for filter_key: 100_demos
100
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
101
+ Using 100 subset demos for filter_key: 100_demos
102
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
103
+ Using 100 subset demos for filter_key: 100_demos
104
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
105
+ Using 100 subset demos for filter_key: 100_demos
106
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
107
+ Using 100 subset demos for filter_key: 100_demos
108
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
109
+ Using 100 subset demos for filter_key: 100_demos
110
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
111
+ Using 100 subset demos for filter_key: 100_demos
112
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
113
+ Using 100 subset demos for filter_key: 100_demos
114
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
115
+ Using 100 subset demos for filter_key: 100_demos
116
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
117
+ Using 100 subset demos for filter_key: 100_demos
118
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
119
+ Using 100 subset demos for filter_key: 100_demos
120
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
121
+ Using 100 subset demos for filter_key: 100_demos
122
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
123
+ Using 100 subset demos for filter_key: 100_demos
124
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
125
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
126
+ Using 100 subset demos for filter_key: 100_demos
127
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
128
+ Using 100 subset demos for filter_key: 100_demos
129
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
130
+ Using 100 subset demos for filter_key: 100_demos
131
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
132
+ Using 100 subset demos for filter_key: 100_demos
133
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
134
+ Using 100 subset demos for filter_key: 100_demos
135
+
136
+ ==================================================
137
+ GR00T FINE-TUNING CONFIGURATION:
138
+ ==================================================
139
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml
140
+ dataset_soup: None
141
+ output_dir: /tmp/gr00t
142
+ output_root: None
143
+ data_config: panda_omron
144
+ batch_size: 64
145
+ max_steps: 300000
146
+ num_gpus: 2
147
+ save_steps: 20000
148
+ run_name: None
149
+ save_total_limit: 100
150
+ seed: 42
151
+ base_model_path: nvidia/GR00T-N1.5-3B
152
+ tune_llm: False
153
+ tune_visual: False
154
+ tune_projector: True
155
+ tune_diffusion_model: True
156
+ resume: False
157
+ learning_rate: 3e-05
158
+ weight_decay: 1e-05
159
+ warmup_ratio: 0.05
160
+ lora_rank: 0
161
+ lora_alpha: 16
162
+ lora_dropout: 0.1
163
+ lora_full_model: False
164
+ dataloader_num_workers: 8
165
+ report_to: wandb
166
+ embodiment_tag: new_embodiment
167
+ video_backend: opencv
168
+ balance_dataset_weights: True
169
+ balance_trajectory_weights: True
170
+ ds_weights_alpha: 0.4
171
+ ==================================================
172
+
173
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
174
+ Using 100 subset demos for filter_key: 100_demos
175
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
176
+ Using 100 subset demos for filter_key: 100_demos
177
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
178
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
179
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
180
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
181
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
182
+ 0.75517122 0.7973985 ]
183
+ Loaded 26 datasets
184
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
185
+ Tune backbone vision tower: False
186
+ Tune backbone LLM: False
187
+ Tune action head projector: False
188
+ Tune action head DiT: False
189
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
190
+ Using 2 GPUs
191
+
192
+ ================================================================================
193
+ Starting sweep branch: default
194
+ Sweep vars: {}
195
+ ================================================================================
196
+
197
+ --------------------------------------------------------------------------------
198
+ Running phase 1: phase2_rkd_da_only
199
+ Policy type: groot_rkd_v2
200
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
201
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2
202
+ Trainable preset: processing_line_only
203
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'raw_action', 'rkd_action_encoder_projector_enabled': False, 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
204
+ --------------------------------------------------------------------------------
205
+
206
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]
207
+ Using 100 subset demos for filter_key: 100_demos
208
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
209
+ self.statistics[key] = torch.tensor(value)
210
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
211
+ Using 100 subset demos for filter_key: 100_demos
212
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
213
+ Using 100 subset demos for filter_key: 100_demos
214
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
215
+ Using 100 subset demos for filter_key: 100_demos
216
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
217
+ Using 100 subset demos for filter_key: 100_demos
218
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
219
+ Using 100 subset demos for filter_key: 100_demos
220
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
221
+ Using 100 subset demos for filter_key: 100_demos
222
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
223
+ Using 100 subset demos for filter_key: 100_demos
224
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
225
+ Using 100 subset demos for filter_key: 100_demos
226
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
227
+ Using 100 subset demos for filter_key: 100_demos
228
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
229
+ Using 100 subset demos for filter_key: 100_demos
230
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
231
+ Using 100 subset demos for filter_key: 100_demos
232
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
233
+ Using 100 subset demos for filter_key: 100_demos
234
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
235
+ Using 100 subset demos for filter_key: 100_demos
236
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
237
+ Using 100 subset demos for filter_key: 100_demos
238
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
239
+ Using 100 subset demos for filter_key: 100_demos
240
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
241
+ Using 100 subset demos for filter_key: 100_demos
242
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
243
+ Using 100 subset demos for filter_key: 100_demos
244
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
245
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
246
+ Using 100 subset demos for filter_key: 100_demos
247
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
248
+ Using 100 subset demos for filter_key: 100_demos
249
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
250
+ Using 100 subset demos for filter_key: 100_demos
251
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
252
+ Using 100 subset demos for filter_key: 100_demos
253
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
254
+ Using 100 subset demos for filter_key: 100_demos
255
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
256
+ Using 100 subset demos for filter_key: 100_demos
257
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
258
+ Using 100 subset demos for filter_key: 100_demos
259
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
260
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
261
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
262
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
263
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
264
+ 0.75517122 0.7973985 ]
265
+ Loaded 26 datasets
266
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
267
+ Tune backbone vision tower: False
268
+ Tune backbone LLM: False
269
+ Tune action head projector: False
270
+ Tune action head DiT: False
271
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
272
+ Tune backbone llm: False
273
+ Tune backbone visual: True
274
+ Total number of DiT parameters: 550386688
275
+ Tune backbone llm: False
276
+ Tune backbone visual: True
277
+ Total number of DiT parameters: 550386688
278
+ Total number of SelfAttentionTransformer parameters: 201433088
279
+ Tune action head projector: True
280
+ Tune action head diffusion model: True
281
+
282
+ Tune action head projector: True
283
+ Tune action head diffusion model: True
284
+
285
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
286
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
287
+ Tune backbone llm: False
288
+ Tune backbone visual: False
289
+ Warning: No backbone trainable parameters found.
290
+ Tune action head projector: False
291
+ Tune action head diffusion model: False
292
+ Action head trainable parameter: future_tokens.weight
293
+ Action head trainable parameter: vlln.weight
294
+ Action head trainable parameter: vlln.bias
295
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
296
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
297
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
298
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
299
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
300
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
301
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
302
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
303
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
304
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
305
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
306
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
307
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
308
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
309
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
310
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
311
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
312
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
313
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
314
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
315
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
316
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
317
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
318
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
319
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
320
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
321
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
322
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
323
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
324
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
325
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
326
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
327
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
328
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
329
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
330
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
331
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
332
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
333
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
334
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
335
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
336
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
337
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
338
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
339
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
340
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
341
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
342
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
343
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
344
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
345
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
346
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
347
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
348
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
349
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
350
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
351
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
352
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
353
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
354
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
355
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
356
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
357
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
358
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
359
+ Applied trainable preset: processing_line_only
360
+ Trainable parameter tensors after preset: 66
361
+ trainable: action_head.vlln.weight
362
+ trainable: action_head.vlln.bias
363
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
364
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
365
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
366
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
367
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
368
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
369
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
370
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
371
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
372
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
373
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
374
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
375
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
376
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
377
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
378
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
379
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
380
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
381
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
382
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
383
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
384
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
385
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
386
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
387
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
388
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
389
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
390
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
391
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
392
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
393
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
394
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
395
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
396
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
397
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
398
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
399
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
400
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
401
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
402
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
403
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
404
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
405
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
406
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
407
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
408
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
409
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
410
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
411
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
412
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
413
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
414
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
415
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
416
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
417
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
418
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
419
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
420
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
421
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
422
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
423
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
424
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
425
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
426
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
427
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
428
+
429
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
430
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
431
+ Tune backbone llm: False
432
+ Tune backbone visual: False
433
+ Warning: No backbone trainable parameters found.
434
+ Tune action head projector: False
435
+ Tune action head diffusion model: False
436
+ Action head trainable parameter: future_tokens.weight
437
+ Action head trainable parameter: vlln.weight
438
+ Action head trainable parameter: vlln.bias
439
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
440
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
441
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
442
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
443
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
444
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
445
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
446
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
447
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
448
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
449
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
450
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
451
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
452
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
453
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
454
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
455
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
456
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
457
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
458
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
459
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
460
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
461
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
462
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
463
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
464
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
465
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
466
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
467
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
468
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
469
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
470
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
471
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
472
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
473
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
474
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
475
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
476
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
477
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
478
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
479
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
480
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
481
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
482
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
483
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
484
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
485
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
486
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
487
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
488
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
489
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
490
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
491
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
492
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
493
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
494
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
495
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
496
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
497
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
498
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
499
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
500
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
501
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
502
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
503
+ Applied trainable preset: processing_line_only
504
+ Trainable parameter tensors after preset: 66
505
+ trainable: action_head.vlln.weight
506
+ trainable: action_head.vlln.bias
507
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
508
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
509
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
510
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
511
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
512
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
513
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
514
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
515
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
516
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
517
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
518
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
519
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
520
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
521
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
522
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
523
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
524
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
525
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
526
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
527
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
528
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
529
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
530
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
531
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
532
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
533
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
534
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
535
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
536
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
537
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
538
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
539
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
540
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
541
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
542
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
543
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
544
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
545
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
546
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
547
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
548
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
549
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
550
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
551
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
552
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
553
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
554
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
555
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
556
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
557
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
558
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
559
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
560
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
561
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
562
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
563
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
564
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
565
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
566
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
567
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
568
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
569
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
570
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
571
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
572
+ Run name: rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
573
+ Run name: rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
574
+ train dataloader length: 3437
575
+ train dataset length: 439854
576
+ GPU memory before training: 7.111904144287109 GB
577
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2/runs
578
+ train dataloader length: 3437
579
+ train dataset length: 439854
580
+ GPU memory before training: 7.111904144287109 GB
581
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2/runs
582
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
583
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
584
+ wandb: setting up run qsfx6i9d
585
+ wandb: Tracking run with wandb version 0.25.0
586
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_131918-qsfx6i9d
587
+ wandb: Run `wandb offline` to turn off syncing.
588
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
589
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/huggingface
590
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/qsfx6i9d
591
+
592
  0%| | 0/30000 [00:00<?, ?it/s][rank1]: Traceback (most recent call last):
593
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
594
+ [rank1]: run_yaml_experiment(
595
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
596
+ [rank1]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
597
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
598
+ [rank1]: experiment.train()
599
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
600
+ [rank1]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
601
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
602
+ [rank1]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
603
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
604
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
605
+ [rank1]: return inner_training_loop(
606
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^
607
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
608
+ [rank1]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
609
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
610
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
611
+ [rank1]: loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
612
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
613
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
614
+ [rank1]: outputs = model(inputs)
615
+ [rank1]: ^^^^^^^^^^^^^
616
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
617
+ [rank1]: return self._call_impl(*args, **kwargs)
618
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
619
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
620
+ [rank1]: return forward_call(*args, **kwargs)
621
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
622
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1637, in forward
623
+ [rank1]: else self._run_ddp_forward(*inputs, **kwargs)
624
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
625
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1464, in _run_ddp_forward
626
+ [rank1]: return self.module(*inputs, **kwargs) # type: ignore[index]
627
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
628
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
629
+ [rank1]: return self._call_impl(*args, **kwargs)
630
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
631
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
632
+ [rank1]: return forward_call(*args, **kwargs)
633
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
634
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
635
+ [rank1]: return model_forward(*args, **kwargs)
636
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
637
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
638
+ [rank1]: return convert_to_fp32(self.model_forward(*args, **kwargs))
639
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
640
+ [rank1]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
641
+ [rank1]: return func(*args, **kwargs)
642
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^
643
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 597, in forward
644
+ [rank1]: self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
645
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 529, in _add_rkd_loss
646
+ [rank1]: rkd_loss, rkd_metrics = self._compute_rkd_loss(
647
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^
648
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 436, in _compute_rkd_loss
649
+ [rank1]: angle_loss = rkd_angle_loss(student_vector, teacher_vector)
650
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
651
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
652
+ [rank1]: student_angle = _angle_relation(student, eps=eps)
653
+ [rank1]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
654
+ [rank1]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
655
+ [rank1]: diff = x[:, None, :] - x[None, :, :]
656
+ [rank1]: ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
657
+ [rank1]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 1 has a total capacity of 139.80 GiB of which 77.26 GiB is free. Process 1961377 has 11.53 GiB memory in use. Including non-PyTorch memory, this process has 35.91 GiB memory in use. Of the allocated memory 34.24 GiB is allocated by PyTorch, and 196.95 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
658
+ wandb: updating run metadata
659
+ wandb: uploading config.yaml
660
+ wandb: uploading data
661
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/qsfx6i9d
662
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface
663
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
664
+ wandb: Find logs at: ./wandb/run-20260622_131918-qsfx6i9d/logs
665
+ Traceback (most recent call last):
666
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
667
+ run_yaml_experiment(
668
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
669
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
670
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
671
+ experiment.train()
672
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
673
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
674
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
675
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
676
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
677
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
678
+ return inner_training_loop(
679
+ ^^^^^^^^^^^^^^^^^^^^
680
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
681
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
682
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
683
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
684
+ loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
685
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
686
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
687
+ outputs = model(inputs)
688
+ ^^^^^^^^^^^^^
689
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
690
+ return self._call_impl(*args, **kwargs)
691
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
692
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
693
+ return forward_call(*args, **kwargs)
694
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
695
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1637, in forward
696
+ else self._run_ddp_forward(*inputs, **kwargs)
697
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
698
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1464, in _run_ddp_forward
699
+ return self.module(*inputs, **kwargs) # type: ignore[index]
700
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
701
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
702
+ return self._call_impl(*args, **kwargs)
703
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
704
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
705
+ return forward_call(*args, **kwargs)
706
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
707
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
708
+ return model_forward(*args, **kwargs)
709
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
710
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
711
+ return convert_to_fp32(self.model_forward(*args, **kwargs))
712
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
713
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
714
+ return func(*args, **kwargs)
715
+ ^^^^^^^^^^^^^^^^^^^^^
716
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 597, in forward
717
+ self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
718
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 529, in _add_rkd_loss
719
+ rkd_loss, rkd_metrics = self._compute_rkd_loss(
720
+ ^^^^^^^^^^^^^^^^^^^^^^^
721
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 436, in _compute_rkd_loss
722
+ angle_loss = rkd_angle_loss(student_vector, teacher_vector)
723
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
724
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
725
+ student_angle = _angle_relation(student, eps=eps)
726
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
727
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
728
+ diff = x[:, None, :] - x[None, :, :]
729
+ ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
730
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 103.88 GiB is free. Including non-PyTorch memory, this process has 35.89 GiB memory in use. Of the allocated memory 34.24 GiB is allocated by PyTorch, and 176.95 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
731
+ [rank0]: Traceback (most recent call last):
732
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1042, in <module>
733
+ [rank0]: run_yaml_experiment(
734
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
735
+ [rank0]: main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
736
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
737
+ [rank0]: experiment.train()
738
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
739
+ [rank0]: self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
740
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
741
+ [rank0]: return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
742
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
743
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
744
+ [rank0]: return inner_training_loop(
745
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^
746
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
747
+ [rank0]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
748
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
749
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
750
+ [rank0]: loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
751
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
752
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
753
+ [rank0]: outputs = model(inputs)
754
+ [rank0]: ^^^^^^^^^^^^^
755
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
756
+ [rank0]: return self._call_impl(*args, **kwargs)
757
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
758
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
759
+ [rank0]: return forward_call(*args, **kwargs)
760
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
761
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1637, in forward
762
+ [rank0]: else self._run_ddp_forward(*inputs, **kwargs)
763
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
764
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/parallel/distributed.py", line 1464, in _run_ddp_forward
765
+ [rank0]: return self.module(*inputs, **kwargs) # type: ignore[index]
766
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
767
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
768
+ [rank0]: return self._call_impl(*args, **kwargs)
769
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
770
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
771
+ [rank0]: return forward_call(*args, **kwargs)
772
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
773
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
774
+ [rank0]: return model_forward(*args, **kwargs)
775
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
776
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
777
+ [rank0]: return convert_to_fp32(self.model_forward(*args, **kwargs))
778
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
779
+ [rank0]: File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
780
+ [rank0]: return func(*args, **kwargs)
781
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^
782
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 597, in forward
783
+ [rank0]: self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
784
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 529, in _add_rkd_loss
785
+ [rank0]: rkd_loss, rkd_metrics = self._compute_rkd_loss(
786
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^
787
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 436, in _compute_rkd_loss
788
+ [rank0]: angle_loss = rkd_angle_loss(student_vector, teacher_vector)
789
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
790
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
791
+ [rank0]: student_angle = _angle_relation(student, eps=eps)
792
+ [rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
793
+ [rank0]: File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
794
+ [rank0]: diff = x[:, None, :] - x[None, :, :]
795
+ [rank0]: ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
796
+ [rank0]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 103.88 GiB is free. Including non-PyTorch memory, this process has 35.89 GiB memory in use. Of the allocated memory 34.24 GiB is allocated by PyTorch, and 176.95 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
797
+ W0622 13:19:36.923000 2826467 site-packages/torch/distributed/elastic/multiprocessing/api.py:900] Sending process 2826545 closing signal SIGTERM
798
+ E0622 13:19:37.488000 2826467 site-packages/torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1) local_rank: 1 (pid: 2826546) of binary: /home/ext_minje/miniforge3/envs/robocasa/bin/python
799
+ Traceback (most recent call last):
800
+ File "<frozen runpy>", line 198, in _run_module_as_main
801
+ File "<frozen runpy>", line 88, in _run_code
802
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 896, in <module>
803
+ main()
804
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
805
+ return f(*args, **kwargs)
806
+ ^^^^^^^^^^^^^^^^^^
807
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 892, in main
808
+ run(args)
809
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/run.py", line 883, in run
810
+ elastic_launch(
811
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 139, in __call__
812
+ return launch_agent(self._config, self._entrypoint, list(args))
813
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
814
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
815
+ raise ChildFailedError(
816
+ torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
817
+ ============================================================
818
+ /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py FAILED
819
+ ------------------------------------------------------------
820
+ Failures:
821
+ <NO_OTHER_FAILURES>
822
+ ------------------------------------------------------------
823
+ Root Cause (first observed failure):
824
+ [0]:
825
+ time : 2026-06-22_13:19:36
826
+ host : DGX-H200-01
827
+ rank : 1 (local_rank: 1)
828
+ exitcode : 1 (pid: 2826546)
829
+ error_file: <N/A>
830
+ traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
831
+ ============================================================
832
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
833
+
834
+ ==================================================
835
+ GR00T FINE-TUNING CONFIGURATION:
836
+ ==================================================
837
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml
838
+ dataset_soup: None
839
+ output_dir: /tmp/gr00t
840
+ output_root: None
841
+ data_config: panda_omron
842
+ batch_size: 64
843
+ max_steps: 300000
844
+ num_gpus: 2
845
+ save_steps: 20000
846
+ run_name: None
847
+ save_total_limit: 100
848
+ seed: 42
849
+ base_model_path: nvidia/GR00T-N1.5-3B
850
+ tune_llm: False
851
+ tune_visual: False
852
+ tune_projector: True
853
+ tune_diffusion_model: True
854
+ resume: False
855
+ learning_rate: 3e-05
856
+ weight_decay: 1e-05
857
+ warmup_ratio: 0.05
858
+ lora_rank: 0
859
+ lora_alpha: 16
860
+ lora_dropout: 0.1
861
+ lora_full_model: False
862
+ dataloader_num_workers: 8
863
+ report_to: wandb
864
+ embodiment_tag: new_embodiment
865
+ video_backend: opencv
866
+ balance_dataset_weights: True
867
+ balance_trajectory_weights: True
868
+ ds_weights_alpha: 0.4
869
+ ==================================================
870
+
871
+ Using 2 GPUs
872
+ Running torchrun command: ['/home/ext_minje/miniforge3/envs/robocasa/bin/python', '-m', 'torch.distributed.run', '--standalone', '--nproc_per_node=2', '--nnodes=1', '/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py', '--config', '/home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml', '--batch-size', '64', '--num-gpus', '2']
rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu45_bs64_20260622_131850.log.pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2826133
rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu4_bs128_20260622_131539.log ADDED
@@ -0,0 +1,355 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ CMD: CUDA_VISIBLE_DEVICES=4 PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True /home/ext_minje/miniforge3/envs/robocasa/bin/python /home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py --config /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml --num-gpus 1 --batch-size 128
2
+ [robosuite WARNING] No private macro file found! (macros.py:57)
3
+ [robosuite WARNING] It is recommended to use a private macro file (macros.py:58)
4
+ [robosuite WARNING] To setup, run: python /home/ext_minje/clvla/benchmarks/robocasa_v2/robosuite/robosuite/scripts/setup_macros.py (macros.py:59)
5
+ [robosuite WARNING] Could not import robosuite_models. Some robots may not be available. If you want to use these robots, please install robosuite_models from source (https://github.com/ARISE-Initiative/robosuite_models) or through pip install. (__init__.py:30)
6
+ [robosuite WARNING] Could not load the mink-based whole-body IK. Make sure you install related import properly (e.g. pip install mink==0.0.5), otherwise you will not be able to use the default IK controller setting for GR1 robot. (__init__.py:40)
7
+ /home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
8
+ check_for_updates()
9
+ `use_fast` is set to `True` but the image processor class does not have a fast version. Falling back to the slow version.
10
+ WARNING: mimicgen environments not imported since mimicgen is not installed!
11
+
12
+ ==================================================
13
+ GR00T FINE-TUNING CONFIGURATION:
14
+ ==================================================
15
+ config: /home/ext_minje/clvla/benchmarks/robocasa_v2/experiment_cfg/processing_line_only_v2/RKD_v2.2/rkd_v2.2_da_flatten_raw_action.yaml
16
+ dataset_soup: None
17
+ output_dir: /tmp/gr00t
18
+ output_root: None
19
+ data_config: panda_omron
20
+ batch_size: 128
21
+ max_steps: 300000
22
+ num_gpus: 1
23
+ save_steps: 20000
24
+ run_name: None
25
+ save_total_limit: 100
26
+ seed: 42
27
+ base_model_path: nvidia/GR00T-N1.5-3B
28
+ tune_llm: False
29
+ tune_visual: False
30
+ tune_projector: True
31
+ tune_diffusion_model: True
32
+ resume: False
33
+ learning_rate: 3e-05
34
+ weight_decay: 1e-05
35
+ warmup_ratio: 0.05
36
+ lora_rank: 0
37
+ lora_alpha: 16
38
+ lora_dropout: 0.1
39
+ lora_full_model: False
40
+ dataloader_num_workers: 8
41
+ report_to: wandb
42
+ embodiment_tag: new_embodiment
43
+ video_backend: opencv
44
+ balance_dataset_weights: True
45
+ balance_trajectory_weights: True
46
+ ds_weights_alpha: 0.4
47
+ ==================================================
48
+
49
+ Using 1 GPUs
50
+
51
+ ================================================================================
52
+ Starting sweep branch: default
53
+ Sweep vars: {}
54
+ ================================================================================
55
+
56
+ --------------------------------------------------------------------------------
57
+ Running phase 1: phase2_rkd_da_only
58
+ Policy type: groot_rkd_v2
59
+ Base model path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
60
+ Output dir: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2
61
+ Trainable preset: processing_line_only
62
+ Policy overrides: {'rkd_enabled': True, 'rkd_fm_loss_weight': 0.0, 'rkd_loss_weight': 1.0, 'rkd_relation_mode': 'flatten', 'rkd_loss_type': 'distance_angle', 'rkd_teacher_source': 'raw_action', 'rkd_action_encoder_projector_enabled': False, 'rkd_distance_loss_weight': 1.0, 'rkd_angle_loss_weight': 2.0, 'rkd_exclude_diagonal': True}
63
+ --------------------------------------------------------------------------------
64
+
65
+ [{'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseBlenderLid/20250822/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseBlenderLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseFridge/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'CloseFridge', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CloseToasterOvenDoor/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'CloseToasterOvenDoor', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/CoffeeSetupMug/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'CoffeeSetupMug', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/NavigateKitchen/20250821/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'NavigateKitchen', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenCabinet/20250819/lerobot', 'horizon': 1050, 'filter_key': '100_demos', 'task': 'OpenCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenDrawer/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'OpenDrawer', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenStandMixerHead/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'OpenStandMixerHead', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToCabinet/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToCabinet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCounterToStove/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceCounterToStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceDrawerToCounter/20250820/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'PickPlaceDrawerToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceSinkToCounter/20250819/lerobot', 'horizon': 900, 'filter_key': '100_demos', 'task': 'PickPlaceSinkToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceToasterToCounter/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'PickPlaceToasterToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/PickPlaceCabinetToCounter/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'PickPlaceCabinetToCounter', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnElectricKettle/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnElectricKettle', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnMicrowave/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffMicrowave/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffMicrowave', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/OpenElectricKettleLid/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'OpenElectricKettleLid', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/SlideDishwasherRack/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'SlideDishwasherRack', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnSinkFaucet/20250819/lerobot', 'horizon': 600, 'filter_key': '100_demos', 'task': 'TurnOnSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnSinkSpout/20250820/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnSinkSpout', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffSinkFaucet/20250819/lerobot', 'horizon': 300, 'filter_key': '100_demos', 'task': 'TurnOffSinkFaucet', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustWaterTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustWaterTemperature', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOffStove/20250819/lerobot', 'horizon': 750, 'filter_key': '100_demos', 'task': 'TurnOffStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/TurnOnStove/20250819/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'TurnOnStove', 'split': 'pretrain', 'source': 'human'}, {'path': '/home/ext_minje/clvla/benchmarks/robocasa_v2/robocasa/datasets/v1.0/pretrain/atomic/AdjustToasterOvenTemperature/20250820/lerobot', 'horizon': 450, 'filter_key': '100_demos', 'task': 'AdjustToasterOvenTemperature', 'split': 'pretrain', 'source': 'human'}]/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/data/transform/state_action.py:257: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.detach().clone() or sourceTensor.detach().clone().requires_grad_(True), rather than torch.tensor(sourceTensor).
66
+ self.statistics[key] = torch.tensor(value)
67
+
68
+ Using 100 subset demos for filter_key: 100_demos
69
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
70
+ Using 100 subset demos for filter_key: 100_demos
71
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
72
+ Using 100 subset demos for filter_key: 100_demos
73
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
74
+ Using 100 subset demos for filter_key: 100_demos
75
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
76
+ Using 100 subset demos for filter_key: 100_demos
77
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
78
+ Using 100 subset demos for filter_key: 100_demos
79
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
80
+ Using 100 subset demos for filter_key: 100_demos
81
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
82
+ Using 100 subset demos for filter_key: 100_demos
83
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
84
+ Using 100 subset demos for filter_key: 100_demos
85
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
86
+ Using 100 subset demos for filter_key: 100_demos
87
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
88
+ Using 100 subset demos for filter_key: 100_demos
89
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
90
+ Using 100 subset demos for filter_key: 100_demos
91
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
92
+ Using 100 subset demos for filter_key: 100_demos
93
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
94
+ Using 100 subset demos for filter_key: 100_demos
95
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
96
+ Using 100 subset demos for filter_key: 100_demos
97
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
98
+ Using 100 subset demos for filter_key: 100_demos
99
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
100
+ Using 100 subset demos for filter_key: 100_demos
101
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
102
+ Using 100 subset demos for filter_key: 100_demos
103
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
104
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
105
+ Using 100 subset demos for filter_key: 100_demos
106
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
107
+ Using 100 subset demos for filter_key: 100_demos
108
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
109
+ Using 100 subset demos for filter_key: 100_demos
110
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
111
+ Using 100 subset demos for filter_key: 100_demos
112
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
113
+ Using 100 subset demos for filter_key: 100_demos
114
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
115
+ Using 100 subset demos for filter_key: 100_demos
116
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
117
+ Using 100 subset demos for filter_key: 100_demos
118
+ Initialized dataset lerobot with EmbodimentTag.NEW_EMBODIMENT
119
+ dataset weights: [1. 0.88494669 0.76443286 0.83888607 0.73785464 1.00501644
120
+ 0.80029956 0.66065525 0.83967491 0.83515002 0.95047759 0.86791692
121
+ 0.88329477 0.78510653 0.64381843 0.67404766 0.69697218 0.60443684
122
+ 0.78448103 0.83447486 0.61146215 0.64384058 0.79545025 0.93845856
123
+ 0.75517122 0.7973985 ]
124
+ Loaded 26 datasets
125
+ Loading pretrained dual brain from /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
126
+ Tune backbone vision tower: False
127
+ Tune backbone LLM: False
128
+ Tune action head projector: False
129
+ Tune action head DiT: False
130
+ Model not found or avail in the huggingface hub. Loading from local path: /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000
131
+ Tune backbone llm: False
132
+ Tune backbone visual: True
133
+ Total number of DiT parameters: 550386688
134
+ Total number of SelfAttentionTransformer parameters: 201433088
135
+ Tune action head projector: True
136
+ Tune action head diffusion model: True
137
+
138
+ Some weights of GR00T_N1_5_RKD were not initialized from the model checkpoint at /home/ext_minje/groot_robocasa/Atomic26_baseline/checkpoint-60000 and are newly initialized: ['rkd_action_encoder_projector.bias', 'rkd_action_encoder_projector.weight']
139
+ You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
140
+ Tune backbone llm: False
141
+ Tune backbone visual: False
142
+ Warning: No backbone trainable parameters found.
143
+ Tune action head projector: False
144
+ Tune action head diffusion model: False
145
+ Action head trainable parameter: future_tokens.weight
146
+ Action head trainable parameter: vlln.weight
147
+ Action head trainable parameter: vlln.bias
148
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.weight
149
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm1.bias
150
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.weight
151
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_q.bias
152
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.weight
153
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_k.bias
154
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.weight
155
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_v.bias
156
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
157
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
158
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.weight
159
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.norm3.bias
160
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
161
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
162
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.weight
163
+ Action head trainable parameter: vl_self_attention.transformer_blocks.0.ff.net.2.bias
164
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.weight
165
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm1.bias
166
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.weight
167
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_q.bias
168
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.weight
169
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_k.bias
170
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.weight
171
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_v.bias
172
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
173
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
174
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.weight
175
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.norm3.bias
176
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
177
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
178
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.weight
179
+ Action head trainable parameter: vl_self_attention.transformer_blocks.1.ff.net.2.bias
180
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.weight
181
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm1.bias
182
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.weight
183
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_q.bias
184
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.weight
185
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_k.bias
186
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.weight
187
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_v.bias
188
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
189
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
190
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.weight
191
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.norm3.bias
192
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
193
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
194
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.weight
195
+ Action head trainable parameter: vl_self_attention.transformer_blocks.2.ff.net.2.bias
196
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.weight
197
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm1.bias
198
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.weight
199
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_q.bias
200
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.weight
201
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_k.bias
202
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.weight
203
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_v.bias
204
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
205
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
206
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.weight
207
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.norm3.bias
208
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
209
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
210
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.weight
211
+ Action head trainable parameter: vl_self_attention.transformer_blocks.3.ff.net.2.bias
212
+ Applied trainable preset: processing_line_only
213
+ Trainable parameter tensors after preset: 66
214
+ trainable: action_head.vlln.weight
215
+ trainable: action_head.vlln.bias
216
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.weight
217
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm1.bias
218
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.weight
219
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_q.bias
220
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.weight
221
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_k.bias
222
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.weight
223
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_v.bias
224
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.weight
225
+ trainable: action_head.vl_self_attention.transformer_blocks.0.attn1.to_out.0.bias
226
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.weight
227
+ trainable: action_head.vl_self_attention.transformer_blocks.0.norm3.bias
228
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.weight
229
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.0.proj.bias
230
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.weight
231
+ trainable: action_head.vl_self_attention.transformer_blocks.0.ff.net.2.bias
232
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.weight
233
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm1.bias
234
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.weight
235
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_q.bias
236
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.weight
237
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_k.bias
238
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.weight
239
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_v.bias
240
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.weight
241
+ trainable: action_head.vl_self_attention.transformer_blocks.1.attn1.to_out.0.bias
242
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.weight
243
+ trainable: action_head.vl_self_attention.transformer_blocks.1.norm3.bias
244
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.weight
245
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.0.proj.bias
246
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.weight
247
+ trainable: action_head.vl_self_attention.transformer_blocks.1.ff.net.2.bias
248
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.weight
249
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm1.bias
250
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.weight
251
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_q.bias
252
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.weight
253
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_k.bias
254
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.weight
255
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_v.bias
256
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.weight
257
+ trainable: action_head.vl_self_attention.transformer_blocks.2.attn1.to_out.0.bias
258
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.weight
259
+ trainable: action_head.vl_self_attention.transformer_blocks.2.norm3.bias
260
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.weight
261
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.0.proj.bias
262
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.weight
263
+ trainable: action_head.vl_self_attention.transformer_blocks.2.ff.net.2.bias
264
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.weight
265
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm1.bias
266
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.weight
267
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_q.bias
268
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.weight
269
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_k.bias
270
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.weight
271
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_v.bias
272
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.weight
273
+ trainable: action_head.vl_self_attention.transformer_blocks.3.attn1.to_out.0.bias
274
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.weight
275
+ trainable: action_head.vl_self_attention.transformer_blocks.3.norm3.bias
276
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.weight
277
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.0.proj.bias
278
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.weight
279
+ trainable: action_head.vl_self_attention.transformer_blocks.3.ff.net.2.bias
280
+ Trainable summary: {'trainable_preset': 'processing_line_only', 'trainable_modules': 'backbone.eagle_linear(0), action_head.vlln(4,096), action_head.vl_self_attention(201,433,088)', 'trainable_module_names': 'backbone.eagle_linear, action_head.vlln, action_head.vl_self_attention', 'trainable_module_param_counts': {'backbone.eagle_linear': 0, 'action_head.vlln': 4096, 'action_head.vl_self_attention': 201433088}, 'trainable_param_count': 201437184, 'total_param_count': 2736746944, 'trainable_param_ratio': 0.07360460726616601}
281
+ Run name: rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
282
+ train dataloader length: 3437
283
+ train dataset length: 439854
284
+ GPU memory before training: 7.111904144287109 GB
285
+ wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from /home/ext_minje/.netrc.
286
+ wandb: Currently logged in as: minje227_hyu (minje227_hyu-hanyang-university) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
287
+ wandb: setting up run jhlxs19d
288
+ wandb: Tracking run with wandb version 0.25.0
289
+ wandb: Run data is saved locally in /home/ext_minje/clvla/benchmarks/robocasa_v2/wandb/run-20260622_131555-jhlxs19d
290
+ wandb: Run `wandb offline` to turn off syncing.
291
+ wandb: Syncing run rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2
292
+ wandb: ⭐️ View project at https://wandb.ai/minje227_hyu-hanyang-university/huggingface
293
+ wandb: 🚀 View run at https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/jhlxs19d
294
+ TensorBoard logs will be saved to: /home/ext_minje/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2/runs
295
+
296
  0%| | 0/30000 [00:00<?, ?it/s]wandb: updating run metadata
297
+ wandb: uploading wandb-summary.json; uploading config.yaml; uploading output.log
298
+ wandb: uploading summary
299
+ wandb: 🚀 View run rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5_phase2 at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface/runs/jhlxs19d
300
+ wandb: ⭐️ View project at: https://wandb.ai/minje227_hyu-hanyang-university/huggingface
301
+ wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
302
+ wandb: Find logs at: ./wandb/run-20260622_131555-jhlxs19d/logs
303
+ Traceback (most recent call last):
304
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 1032, in <module>
305
+ run_yaml_experiment(
306
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 991, in run_yaml_experiment
307
+ main(args, policy_type=policy_type, policy_overrides=policy_overrides, trainable_preset=preset)
308
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_scripts/gr00t_finetune.py", line 721, in main
309
+ experiment.train()
310
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/runner.py", line 172, in train
311
+ self.trainer.train(resume_from_checkpoint=self.resume_from_checkpoint)
312
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 176, in train
313
+ return super().train(resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)
314
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
315
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2245, in train
316
+ return inner_training_loop(
317
+ ^^^^^^^^^^^^^^^^^^^^
318
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 2560, in _inner_training_loop
319
+ tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
320
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
321
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/transformers/trainer.py", line 3736, in training_step
322
+ loss = self.compute_loss(model, inputs, num_items_in_batch=num_items_in_batch)
323
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
324
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/Isaac-GR00T/gr00t/experiment/trainer.py", line 77, in compute_loss
325
+ outputs = model(inputs)
326
+ ^^^^^^^^^^^^^
327
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1751, in _wrapped_call_impl
328
+ return self._call_impl(*args, **kwargs)
329
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
330
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/nn/modules/module.py", line 1762, in _call_impl
331
+ return forward_call(*args, **kwargs)
332
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
333
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 823, in forward
334
+ return model_forward(*args, **kwargs)
335
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
336
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/accelerate/utils/operations.py", line 811, in __call__
337
+ return convert_to_fp32(self.model_forward(*args, **kwargs))
338
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
339
+ File "/home/ext_minje/miniforge3/envs/robocasa/lib/python3.11/site-packages/torch/amp/autocast_mode.py", line 44, in decorate_autocast
340
+ return func(*args, **kwargs)
341
+ ^^^^^^^^^^^^^^^^^^^^^
342
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 597, in forward
343
+ self._add_rkd_loss(action_head_outputs, backbone_outputs, action_inputs)
344
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 529, in _add_rkd_loss
345
+ rkd_loss, rkd_metrics = self._compute_rkd_loss(
346
+ ^^^^^^^^^^^^^^^^^^^^^^^
347
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/gr00t_n1.py", line 436, in _compute_rkd_loss
348
+ angle_loss = rkd_angle_loss(student_vector, teacher_vector)
349
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
350
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 167, in rkd_angle_loss
351
+ student_angle = _angle_relation(student, eps=eps)
352
+ ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
353
+ File "/home/ext_minje/clvla/benchmarks/robocasa_v2/my_policies/groot_rkd_v2/rkd_utils.py", line 184, in _angle_relation
354
+ diff = x[:, None, :] - x[None, :, :]
355
+ ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~
356
+ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 104.38 GiB. GPU 0 has a total capacity of 139.80 GiB of which 81.16 GiB is free. Including non-PyTorch memory, this process has 58.62 GiB memory in use. Of the allocated memory 57.79 GiB is allocated by PyTorch, and 165.56 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)
rkd_v2_2/logs/rkd_v2_2_flatten_raw_action_gpu4_bs128_20260622_131539.log.pid ADDED
@@ -0,0 +1 @@
 
 
1
+ 2713001
rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/phase2/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,431 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "base_position": {
6
+ "max": [
7
+ 7.3139495849609375,
8
+ 0.4876587688922882,
9
+ 0.7196521759033203
10
+ ],
11
+ "min": [
12
+ -4.94293737411499,
13
+ -6.890198230743408,
14
+ 0.6996827125549316
15
+ ],
16
+ "mean": [
17
+ 2.749288365724937,
18
+ -2.0455216239909983,
19
+ 0.700532551020533
20
+ ],
21
+ "std": [
22
+ 1.6786295897046262,
23
+ 1.312897969434244,
24
+ 0.00133751758906258
25
+ ],
26
+ "q01": [
27
+ -1.4241907881148037,
28
+ -5.1716547935950885,
29
+ 0.6999670898768614
30
+ ],
31
+ "q99": [
32
+ 6.198111545831555,
33
+ -0.5873531334104489,
34
+ 0.7038833014242587
35
+ ]
36
+ },
37
+ "base_rotation": {
38
+ "max": [
39
+ 0.0,
40
+ 0.0,
41
+ 1.0,
42
+ 1.0
43
+ ],
44
+ "min": [
45
+ 0.0,
46
+ 0.0,
47
+ -1.0,
48
+ 0.0
49
+ ],
50
+ "mean": [
51
+ 0.0,
52
+ 0.0,
53
+ 0.23084999587450247,
54
+ 0.6238431088581847
55
+ ],
56
+ "std": [
57
+ 0.0,
58
+ 0.0,
59
+ 0.6556404371644047,
60
+ 0.3572252105732042
61
+ ],
62
+ "q01": [
63
+ 0.0,
64
+ 0.0,
65
+ -0.9999999999999999,
66
+ 1.7482354593273349e-06
67
+ ],
68
+ "q99": [
69
+ 0.0,
70
+ 0.0,
71
+ 0.9999999999999999,
72
+ 0.9999999999999999
73
+ ]
74
+ },
75
+ "end_effector_position_relative": {
76
+ "max": [
77
+ 0.9014528393745422,
78
+ 0.8003877401351929,
79
+ 0.9829942584037781
80
+ ],
81
+ "min": [
82
+ -0.37471598386764526,
83
+ -0.8472502827644348,
84
+ -0.25070279836654663
85
+ ],
86
+ "mean": [
87
+ 0.28667332601863704,
88
+ -0.038315473170876704,
89
+ 0.45974658906777444
90
+ ],
91
+ "std": [
92
+ 0.17067058828305154,
93
+ 0.23569952037784195,
94
+ 0.21950702354181487
95
+ ],
96
+ "q01": [
97
+ 0.015172948963784228,
98
+ -0.44450502063220615,
99
+ 0.23314444103329338
100
+ ],
101
+ "q99": [
102
+ 0.5609069689572412,
103
+ 0.38454179128009547,
104
+ 0.7028102736473852
105
+ ]
106
+ },
107
+ "end_effector_rotation_relative": {
108
+ "max": [
109
+ 0.9999998807907104,
110
+ 0.9984455108642578,
111
+ 0.9434727430343628,
112
+ 0.9062229990959167
113
+ ],
114
+ "min": [
115
+ -0.9999930262565613,
116
+ -0.9988337755203247,
117
+ -0.9618149995803833,
118
+ 2.1872274658107926e-07
119
+ ],
120
+ "mean": [
121
+ -0.25672297401336,
122
+ 0.024425335806687216,
123
+ -0.08740977164483847,
124
+ 0.16608739644018397
125
+ ],
126
+ "std": [
127
+ 0.7932585208392383,
128
+ 0.3102260195580416,
129
+ 0.3772472576315148,
130
+ 0.17451959300158773
131
+ ],
132
+ "q01": [
133
+ -0.99379006987882,
134
+ -0.5478102050038828,
135
+ -0.6101777952971636,
136
+ 0.002274085014134748
137
+ ],
138
+ "q99": [
139
+ 0.8804856038896288,
140
+ 0.5910611092873037,
141
+ 0.5216058042993562,
142
+ 0.5231965974012255
143
+ ]
144
+ },
145
+ "gripper_qpos": {
146
+ "max": [
147
+ 0.055869169533252716,
148
+ 0.010916369967162609
149
+ ],
150
+ "min": [
151
+ -0.011436971835792065,
152
+ -0.05664053186774254
153
+ ],
154
+ "mean": [
155
+ 0.031651551516052555,
156
+ -0.03161481387920421
157
+ ],
158
+ "std": [
159
+ 0.013119769222217526,
160
+ 0.01306078225192402
161
+ ],
162
+ "q01": [
163
+ 0.0060999246446777336,
164
+ -0.04062601653964198
165
+ ],
166
+ "q99": [
167
+ 0.04054891621517283,
168
+ -0.0059163567113456345
169
+ ]
170
+ }
171
+ },
172
+ "action": {
173
+ "base_motion": {
174
+ "max": [
175
+ 1.0,
176
+ 1.0,
177
+ 1.0,
178
+ 0.0
179
+ ],
180
+ "min": [
181
+ -1.0,
182
+ -1.0,
183
+ -1.0,
184
+ 0.0
185
+ ],
186
+ "mean": [
187
+ 0.008144896753718373,
188
+ -0.00018893135146078263,
189
+ -0.0008739062727221845,
190
+ 0.0
191
+ ],
192
+ "std": [
193
+ 0.11030045424364411,
194
+ 0.10148082570313594,
195
+ 0.08975698373467506,
196
+ 0.0
197
+ ],
198
+ "q01": [
199
+ -0.07287373067581518,
200
+ -0.09446994199569948,
201
+ -0.07702818259249572,
202
+ 0.0
203
+ ],
204
+ "q99": [
205
+ 0.04485485414407898,
206
+ 0.09626941605965177,
207
+ 0.07610665906122742,
208
+ 0.0
209
+ ]
210
+ },
211
+ "control_mode": {
212
+ "max": [
213
+ 1.0
214
+ ],
215
+ "min": [
216
+ -1.0
217
+ ],
218
+ "mean": [
219
+ -0.9221974313425939
220
+ ],
221
+ "std": [
222
+ 0.38670194342612674
223
+ ],
224
+ "q01": [
225
+ -1.0
226
+ ],
227
+ "q99": [
228
+ -0.384693946883099
229
+ ]
230
+ },
231
+ "end_effector_position": {
232
+ "max": [
233
+ 1.0,
234
+ 1.0,
235
+ 1.0
236
+ ],
237
+ "min": [
238
+ -1.0,
239
+ -1.0,
240
+ -1.0
241
+ ],
242
+ "mean": [
243
+ 0.01976129923017491,
244
+ -0.020870631344490034,
245
+ -0.05993476517349181
246
+ ],
247
+ "std": [
248
+ 0.4347024990804053,
249
+ 0.4221662976569997,
250
+ 0.3845506843044066
251
+ ],
252
+ "q01": [
253
+ -0.8835792939767807,
254
+ -0.9280919251713778,
255
+ -0.783361071470956
256
+ ],
257
+ "q99": [
258
+ 0.7622084438449307,
259
+ 0.8625125252609077,
260
+ 0.7470974753250706
261
+ ]
262
+ },
263
+ "end_effector_rotation": {
264
+ "max": [
265
+ 1.0,
266
+ 1.0,
267
+ 1.0
268
+ ],
269
+ "min": [
270
+ -1.0,
271
+ -1.0,
272
+ -1.0
273
+ ],
274
+ "mean": [
275
+ 0.008089673068544335,
276
+ -0.026498254627927348,
277
+ 0.0029083551743059005
278
+ ],
279
+ "std": [
280
+ 0.11216734503733773,
281
+ 0.13206002613798629,
282
+ 0.12615870412692864
283
+ ],
284
+ "q01": [
285
+ -0.2814627972846672,
286
+ -0.4342386516115439,
287
+ -0.34489755628340574
288
+ ],
289
+ "q99": [
290
+ 0.31879353266939386,
291
+ 0.28411797683220563,
292
+ 0.3597718671669894
293
+ ]
294
+ },
295
+ "gripper_close": {
296
+ "max": [
297
+ 1.0
298
+ ],
299
+ "min": [
300
+ -1.0
301
+ ],
302
+ "mean": [
303
+ -0.3824228874769045
304
+ ],
305
+ "std": [
306
+ 0.9240015096469786
307
+ ],
308
+ "q01": [
309
+ -1.0
310
+ ],
311
+ "q99": [
312
+ 0.5597272305813592
313
+ ]
314
+ }
315
+ }
316
+ },
317
+ "modalities": {
318
+ "video": {
319
+ "robot0_eye_in_hand": {
320
+ "resolution": [
321
+ 256,
322
+ 256
323
+ ],
324
+ "channels": 3,
325
+ "fps": 20.0
326
+ },
327
+ "robot0_agentview_left": {
328
+ "resolution": [
329
+ 256,
330
+ 256
331
+ ],
332
+ "channels": 3,
333
+ "fps": 20.0
334
+ },
335
+ "robot0_agentview_right": {
336
+ "resolution": [
337
+ 256,
338
+ 256
339
+ ],
340
+ "channels": 3,
341
+ "fps": 20.0
342
+ }
343
+ },
344
+ "state": {
345
+ "base_position": {
346
+ "absolute": true,
347
+ "rotation_type": null,
348
+ "shape": [
349
+ 3
350
+ ],
351
+ "continuous": true
352
+ },
353
+ "base_rotation": {
354
+ "absolute": true,
355
+ "rotation_type": "quaternion",
356
+ "shape": [
357
+ 4
358
+ ],
359
+ "continuous": true
360
+ },
361
+ "end_effector_position_relative": {
362
+ "absolute": true,
363
+ "rotation_type": null,
364
+ "shape": [
365
+ 3
366
+ ],
367
+ "continuous": true
368
+ },
369
+ "end_effector_rotation_relative": {
370
+ "absolute": true,
371
+ "rotation_type": "quaternion",
372
+ "shape": [
373
+ 4
374
+ ],
375
+ "continuous": true
376
+ },
377
+ "gripper_qpos": {
378
+ "absolute": true,
379
+ "rotation_type": null,
380
+ "shape": [
381
+ 2
382
+ ],
383
+ "continuous": true
384
+ }
385
+ },
386
+ "action": {
387
+ "base_motion": {
388
+ "absolute": true,
389
+ "rotation_type": null,
390
+ "shape": [
391
+ 4
392
+ ],
393
+ "continuous": true
394
+ },
395
+ "control_mode": {
396
+ "absolute": true,
397
+ "rotation_type": null,
398
+ "shape": [
399
+ 1
400
+ ],
401
+ "continuous": true
402
+ },
403
+ "end_effector_position": {
404
+ "absolute": true,
405
+ "rotation_type": null,
406
+ "shape": [
407
+ 3
408
+ ],
409
+ "continuous": true
410
+ },
411
+ "end_effector_rotation": {
412
+ "absolute": true,
413
+ "rotation_type": "axis_angle",
414
+ "shape": [
415
+ 3
416
+ ],
417
+ "continuous": true
418
+ },
419
+ "gripper_close": {
420
+ "absolute": true,
421
+ "rotation_type": null,
422
+ "shape": [
423
+ 1
424
+ ],
425
+ "continuous": true
426
+ }
427
+ }
428
+ },
429
+ "embodiment_tag": "new_embodiment"
430
+ }
431
+ }
rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/resolved_config.yaml ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_rkd_v2
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 2
9
+ batch_size: 32
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_rkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: action_encoder
29
+ rkd_action_encoder_projector_enabled: true
30
+ rkd_action_encoder_projector_dim: 512
31
+ rkd_action_encoder_projector_pooling: flatten
32
+ rkd_distance_loss_weight: 1.0
33
+ rkd_angle_loss_weight: 2.0
34
+ rkd_exclude_diagonal: true
35
+ - name: phase3_fm_rkd_da_fixed_0p5
36
+ max_steps: 30000
37
+ save_steps: 0
38
+ trainable:
39
+ tune_llm: false
40
+ tune_visual: false
41
+ tune_projector: true
42
+ tune_diffusion_model: true
43
+ losses:
44
+ rkd_enabled: true
45
+ rkd_fm_loss_weight: 1.0
46
+ rkd_loss_weight: 0.5
47
+ rkd_relation_mode: flatten
48
+ rkd_loss_type: distance_angle
49
+ rkd_teacher_source: action_encoder
50
+ rkd_action_encoder_projector_enabled: true
51
+ rkd_action_encoder_projector_dim: 512
52
+ rkd_action_encoder_projector_pooling: flatten
53
+ rkd_distance_loss_weight: 1.0
54
+ rkd_angle_loss_weight: 2.0
55
+ rkd_exclude_diagonal: true
56
+ resolved_sweep: {}
rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder/default/rkd_v2.2_da_flatten_action_encoder.yaml ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: rkd_v2_2_distance_angle_flatten_action_encoder_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_rkd_v2
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_action_encoder
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 2
9
+ batch_size: 32
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_rkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: action_encoder
29
+ rkd_action_encoder_projector_enabled: true
30
+ rkd_action_encoder_projector_dim: 512
31
+ rkd_action_encoder_projector_pooling: flatten
32
+ rkd_distance_loss_weight: 1.0
33
+ rkd_angle_loss_weight: 2.0
34
+ rkd_exclude_diagonal: true
35
+ - name: phase3_fm_rkd_da_fixed_0p5
36
+ max_steps: 30000
37
+ save_steps: 0
38
+ trainable:
39
+ tune_llm: false
40
+ tune_visual: false
41
+ tune_projector: true
42
+ tune_diffusion_model: true
43
+ losses:
44
+ rkd_enabled: true
45
+ rkd_fm_loss_weight: 1.0
46
+ rkd_loss_weight: 0.5
47
+ rkd_relation_mode: flatten
48
+ rkd_loss_type: distance_angle
49
+ rkd_teacher_source: action_encoder
50
+ rkd_action_encoder_projector_enabled: true
51
+ rkd_action_encoder_projector_dim: 512
52
+ rkd_action_encoder_projector_pooling: flatten
53
+ rkd_distance_loss_weight: 1.0
54
+ rkd_angle_loss_weight: 2.0
55
+ rkd_exclude_diagonal: true
rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/phase2/experiment_cfg/metadata.json ADDED
@@ -0,0 +1,431 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "statistics": {
4
+ "state": {
5
+ "base_position": {
6
+ "max": [
7
+ 7.3139495849609375,
8
+ 0.4876587688922882,
9
+ 0.7196521759033203
10
+ ],
11
+ "min": [
12
+ -4.94293737411499,
13
+ -6.890198230743408,
14
+ 0.6996827125549316
15
+ ],
16
+ "mean": [
17
+ 2.749288365724937,
18
+ -2.0455216239909983,
19
+ 0.700532551020533
20
+ ],
21
+ "std": [
22
+ 1.6786295897046262,
23
+ 1.312897969434244,
24
+ 0.00133751758906258
25
+ ],
26
+ "q01": [
27
+ -1.4241907881148037,
28
+ -5.1716547935950885,
29
+ 0.6999670898768614
30
+ ],
31
+ "q99": [
32
+ 6.198111545831555,
33
+ -0.5873531334104489,
34
+ 0.7038833014242587
35
+ ]
36
+ },
37
+ "base_rotation": {
38
+ "max": [
39
+ 0.0,
40
+ 0.0,
41
+ 1.0,
42
+ 1.0
43
+ ],
44
+ "min": [
45
+ 0.0,
46
+ 0.0,
47
+ -1.0,
48
+ 0.0
49
+ ],
50
+ "mean": [
51
+ 0.0,
52
+ 0.0,
53
+ 0.23084999587450247,
54
+ 0.6238431088581847
55
+ ],
56
+ "std": [
57
+ 0.0,
58
+ 0.0,
59
+ 0.6556404371644047,
60
+ 0.3572252105732042
61
+ ],
62
+ "q01": [
63
+ 0.0,
64
+ 0.0,
65
+ -0.9999999999999999,
66
+ 1.7482354593273349e-06
67
+ ],
68
+ "q99": [
69
+ 0.0,
70
+ 0.0,
71
+ 0.9999999999999999,
72
+ 0.9999999999999999
73
+ ]
74
+ },
75
+ "end_effector_position_relative": {
76
+ "max": [
77
+ 0.9014528393745422,
78
+ 0.8003877401351929,
79
+ 0.9829942584037781
80
+ ],
81
+ "min": [
82
+ -0.37471598386764526,
83
+ -0.8472502827644348,
84
+ -0.25070279836654663
85
+ ],
86
+ "mean": [
87
+ 0.28667332601863704,
88
+ -0.038315473170876704,
89
+ 0.45974658906777444
90
+ ],
91
+ "std": [
92
+ 0.17067058828305154,
93
+ 0.23569952037784195,
94
+ 0.21950702354181487
95
+ ],
96
+ "q01": [
97
+ 0.015172948963784228,
98
+ -0.44450502063220615,
99
+ 0.23314444103329338
100
+ ],
101
+ "q99": [
102
+ 0.5609069689572412,
103
+ 0.38454179128009547,
104
+ 0.7028102736473852
105
+ ]
106
+ },
107
+ "end_effector_rotation_relative": {
108
+ "max": [
109
+ 0.9999998807907104,
110
+ 0.9984455108642578,
111
+ 0.9434727430343628,
112
+ 0.9062229990959167
113
+ ],
114
+ "min": [
115
+ -0.9999930262565613,
116
+ -0.9988337755203247,
117
+ -0.9618149995803833,
118
+ 2.1872274658107926e-07
119
+ ],
120
+ "mean": [
121
+ -0.25672297401336,
122
+ 0.024425335806687216,
123
+ -0.08740977164483847,
124
+ 0.16608739644018397
125
+ ],
126
+ "std": [
127
+ 0.7932585208392383,
128
+ 0.3102260195580416,
129
+ 0.3772472576315148,
130
+ 0.17451959300158773
131
+ ],
132
+ "q01": [
133
+ -0.99379006987882,
134
+ -0.5478102050038828,
135
+ -0.6101777952971636,
136
+ 0.002274085014134748
137
+ ],
138
+ "q99": [
139
+ 0.8804856038896288,
140
+ 0.5910611092873037,
141
+ 0.5216058042993562,
142
+ 0.5231965974012255
143
+ ]
144
+ },
145
+ "gripper_qpos": {
146
+ "max": [
147
+ 0.055869169533252716,
148
+ 0.010916369967162609
149
+ ],
150
+ "min": [
151
+ -0.011436971835792065,
152
+ -0.05664053186774254
153
+ ],
154
+ "mean": [
155
+ 0.031651551516052555,
156
+ -0.03161481387920421
157
+ ],
158
+ "std": [
159
+ 0.013119769222217526,
160
+ 0.01306078225192402
161
+ ],
162
+ "q01": [
163
+ 0.0060999246446777336,
164
+ -0.04062601653964198
165
+ ],
166
+ "q99": [
167
+ 0.04054891621517283,
168
+ -0.0059163567113456345
169
+ ]
170
+ }
171
+ },
172
+ "action": {
173
+ "base_motion": {
174
+ "max": [
175
+ 1.0,
176
+ 1.0,
177
+ 1.0,
178
+ 0.0
179
+ ],
180
+ "min": [
181
+ -1.0,
182
+ -1.0,
183
+ -1.0,
184
+ 0.0
185
+ ],
186
+ "mean": [
187
+ 0.008144896753718373,
188
+ -0.00018893135146078263,
189
+ -0.0008739062727221845,
190
+ 0.0
191
+ ],
192
+ "std": [
193
+ 0.11030045424364411,
194
+ 0.10148082570313594,
195
+ 0.08975698373467506,
196
+ 0.0
197
+ ],
198
+ "q01": [
199
+ -0.07287373067581518,
200
+ -0.09446994199569948,
201
+ -0.07702818259249572,
202
+ 0.0
203
+ ],
204
+ "q99": [
205
+ 0.04485485414407898,
206
+ 0.09626941605965177,
207
+ 0.07610665906122742,
208
+ 0.0
209
+ ]
210
+ },
211
+ "control_mode": {
212
+ "max": [
213
+ 1.0
214
+ ],
215
+ "min": [
216
+ -1.0
217
+ ],
218
+ "mean": [
219
+ -0.9221974313425939
220
+ ],
221
+ "std": [
222
+ 0.38670194342612674
223
+ ],
224
+ "q01": [
225
+ -1.0
226
+ ],
227
+ "q99": [
228
+ -0.384693946883099
229
+ ]
230
+ },
231
+ "end_effector_position": {
232
+ "max": [
233
+ 1.0,
234
+ 1.0,
235
+ 1.0
236
+ ],
237
+ "min": [
238
+ -1.0,
239
+ -1.0,
240
+ -1.0
241
+ ],
242
+ "mean": [
243
+ 0.01976129923017491,
244
+ -0.020870631344490034,
245
+ -0.05993476517349181
246
+ ],
247
+ "std": [
248
+ 0.4347024990804053,
249
+ 0.4221662976569997,
250
+ 0.3845506843044066
251
+ ],
252
+ "q01": [
253
+ -0.8835792939767807,
254
+ -0.9280919251713778,
255
+ -0.783361071470956
256
+ ],
257
+ "q99": [
258
+ 0.7622084438449307,
259
+ 0.8625125252609077,
260
+ 0.7470974753250706
261
+ ]
262
+ },
263
+ "end_effector_rotation": {
264
+ "max": [
265
+ 1.0,
266
+ 1.0,
267
+ 1.0
268
+ ],
269
+ "min": [
270
+ -1.0,
271
+ -1.0,
272
+ -1.0
273
+ ],
274
+ "mean": [
275
+ 0.008089673068544335,
276
+ -0.026498254627927348,
277
+ 0.0029083551743059005
278
+ ],
279
+ "std": [
280
+ 0.11216734503733773,
281
+ 0.13206002613798629,
282
+ 0.12615870412692864
283
+ ],
284
+ "q01": [
285
+ -0.2814627972846672,
286
+ -0.4342386516115439,
287
+ -0.34489755628340574
288
+ ],
289
+ "q99": [
290
+ 0.31879353266939386,
291
+ 0.28411797683220563,
292
+ 0.3597718671669894
293
+ ]
294
+ },
295
+ "gripper_close": {
296
+ "max": [
297
+ 1.0
298
+ ],
299
+ "min": [
300
+ -1.0
301
+ ],
302
+ "mean": [
303
+ -0.3824228874769045
304
+ ],
305
+ "std": [
306
+ 0.9240015096469786
307
+ ],
308
+ "q01": [
309
+ -1.0
310
+ ],
311
+ "q99": [
312
+ 0.5597272305813592
313
+ ]
314
+ }
315
+ }
316
+ },
317
+ "modalities": {
318
+ "video": {
319
+ "robot0_eye_in_hand": {
320
+ "resolution": [
321
+ 256,
322
+ 256
323
+ ],
324
+ "channels": 3,
325
+ "fps": 20.0
326
+ },
327
+ "robot0_agentview_left": {
328
+ "resolution": [
329
+ 256,
330
+ 256
331
+ ],
332
+ "channels": 3,
333
+ "fps": 20.0
334
+ },
335
+ "robot0_agentview_right": {
336
+ "resolution": [
337
+ 256,
338
+ 256
339
+ ],
340
+ "channels": 3,
341
+ "fps": 20.0
342
+ }
343
+ },
344
+ "state": {
345
+ "base_position": {
346
+ "absolute": true,
347
+ "rotation_type": null,
348
+ "shape": [
349
+ 3
350
+ ],
351
+ "continuous": true
352
+ },
353
+ "base_rotation": {
354
+ "absolute": true,
355
+ "rotation_type": "quaternion",
356
+ "shape": [
357
+ 4
358
+ ],
359
+ "continuous": true
360
+ },
361
+ "end_effector_position_relative": {
362
+ "absolute": true,
363
+ "rotation_type": null,
364
+ "shape": [
365
+ 3
366
+ ],
367
+ "continuous": true
368
+ },
369
+ "end_effector_rotation_relative": {
370
+ "absolute": true,
371
+ "rotation_type": "quaternion",
372
+ "shape": [
373
+ 4
374
+ ],
375
+ "continuous": true
376
+ },
377
+ "gripper_qpos": {
378
+ "absolute": true,
379
+ "rotation_type": null,
380
+ "shape": [
381
+ 2
382
+ ],
383
+ "continuous": true
384
+ }
385
+ },
386
+ "action": {
387
+ "base_motion": {
388
+ "absolute": true,
389
+ "rotation_type": null,
390
+ "shape": [
391
+ 4
392
+ ],
393
+ "continuous": true
394
+ },
395
+ "control_mode": {
396
+ "absolute": true,
397
+ "rotation_type": null,
398
+ "shape": [
399
+ 1
400
+ ],
401
+ "continuous": true
402
+ },
403
+ "end_effector_position": {
404
+ "absolute": true,
405
+ "rotation_type": null,
406
+ "shape": [
407
+ 3
408
+ ],
409
+ "continuous": true
410
+ },
411
+ "end_effector_rotation": {
412
+ "absolute": true,
413
+ "rotation_type": "axis_angle",
414
+ "shape": [
415
+ 3
416
+ ],
417
+ "continuous": true
418
+ },
419
+ "gripper_close": {
420
+ "absolute": true,
421
+ "rotation_type": null,
422
+ "shape": [
423
+ 1
424
+ ],
425
+ "continuous": true
426
+ }
427
+ }
428
+ },
429
+ "embodiment_tag": "new_embodiment"
430
+ }
431
+ }
rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/resolved_config.yaml ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_rkd_v2
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 2
9
+ batch_size: 32
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_rkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: raw_action
29
+ rkd_action_encoder_projector_enabled: false
30
+ rkd_distance_loss_weight: 1.0
31
+ rkd_angle_loss_weight: 2.0
32
+ rkd_exclude_diagonal: true
33
+ - name: phase3_fm_rkd_da_fixed_0p5
34
+ max_steps: 30000
35
+ save_steps: 0
36
+ trainable:
37
+ tune_llm: false
38
+ tune_visual: false
39
+ tune_projector: true
40
+ tune_diffusion_model: true
41
+ losses:
42
+ rkd_enabled: true
43
+ rkd_fm_loss_weight: 1.0
44
+ rkd_loss_weight: 0.5
45
+ rkd_relation_mode: flatten
46
+ rkd_loss_type: distance_angle
47
+ rkd_teacher_source: raw_action
48
+ rkd_action_encoder_projector_enabled: false
49
+ rkd_distance_loss_weight: 1.0
50
+ rkd_angle_loss_weight: 2.0
51
+ rkd_exclude_diagonal: true
52
+ resolved_sweep: {}
rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action/default/rkd_v2.2_da_flatten_raw_action.yaml ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ experiment_name: rkd_v2_2_distance_angle_flatten_raw_action_phase2_phase3_aux_fixed_0p5
2
+ policy_type: groot_rkd_v2
3
+ base_model_path: ${HOME}/groot_robocasa/Atomic26_baseline/checkpoint-60000
4
+ output_root: ${HOME}/groot_robocasa/robocasa_v2/rkd_v2_2/rkd_v2_2_distance_angle_flatten_raw_action
5
+ dataset:
6
+ dataset_soup: my_atomic26_human
7
+ training:
8
+ num_gpus: 2
9
+ batch_size: 32
10
+ seed: 42
11
+ model: null
12
+ phases:
13
+ - name: phase2_rkd_da_only
14
+ max_steps: 30000
15
+ save_steps: 0
16
+ trainable:
17
+ preset: processing_line_only
18
+ tune_llm: false
19
+ tune_visual: false
20
+ tune_projector: false
21
+ tune_diffusion_model: false
22
+ losses:
23
+ rkd_enabled: true
24
+ rkd_fm_loss_weight: 0.0
25
+ rkd_loss_weight: 1.0
26
+ rkd_relation_mode: flatten
27
+ rkd_loss_type: distance_angle
28
+ rkd_teacher_source: raw_action
29
+ rkd_action_encoder_projector_enabled: false
30
+ rkd_distance_loss_weight: 1.0
31
+ rkd_angle_loss_weight: 2.0
32
+ rkd_exclude_diagonal: true
33
+ - name: phase3_fm_rkd_da_fixed_0p5
34
+ max_steps: 30000
35
+ save_steps: 0
36
+ trainable:
37
+ tune_llm: false
38
+ tune_visual: false
39
+ tune_projector: true
40
+ tune_diffusion_model: true
41
+ losses:
42
+ rkd_enabled: true
43
+ rkd_fm_loss_weight: 1.0
44
+ rkd_loss_weight: 0.5
45
+ rkd_relation_mode: flatten
46
+ rkd_loss_type: distance_angle
47
+ rkd_teacher_source: raw_action
48
+ rkd_action_encoder_projector_enabled: false
49
+ rkd_distance_loss_weight: 1.0
50
+ rkd_angle_loss_weight: 2.0
51
+ rkd_exclude_diagonal: true