gubam commited on
Commit
6537fa4
·
verified ·
1 Parent(s): 4877142

Upload folder using huggingface_hub

Browse files
Files changed (50) hide show
  1. checkpoint-10000/config.json +70 -0
  2. checkpoint-10000/embodiment_id.json +11 -0
  3. checkpoint-10000/experiment_cfg/conf.yaml +225 -0
  4. checkpoint-10000/experiment_cfg/config.yaml +260 -0
  5. checkpoint-10000/experiment_cfg/dataset_statistics.json +149 -0
  6. checkpoint-10000/experiment_cfg/final_model_config.json +54 -0
  7. checkpoint-10000/experiment_cfg/final_processor_config.json +0 -0
  8. checkpoint-10000/model-00001-of-00002.safetensors +3 -0
  9. checkpoint-10000/model-00002-of-00002.safetensors +3 -0
  10. checkpoint-10000/model.safetensors.index.json +0 -0
  11. checkpoint-10000/optimizer.pt +3 -0
  12. checkpoint-10000/processor_config.json +471 -0
  13. checkpoint-10000/rng_state.pth +3 -0
  14. checkpoint-10000/scheduler.pt +3 -0
  15. checkpoint-10000/statistics.json +0 -0
  16. checkpoint-10000/trainer_state.json +0 -0
  17. checkpoint-10000/training_args.bin +3 -0
  18. checkpoint-10000/wandb_config.json +1 -0
  19. checkpoint-20000/config.json +70 -0
  20. checkpoint-20000/embodiment_id.json +11 -0
  21. checkpoint-20000/experiment_cfg/conf.yaml +225 -0
  22. checkpoint-20000/experiment_cfg/config.yaml +260 -0
  23. checkpoint-20000/experiment_cfg/dataset_statistics.json +149 -0
  24. checkpoint-20000/experiment_cfg/final_model_config.json +54 -0
  25. checkpoint-20000/experiment_cfg/final_processor_config.json +0 -0
  26. checkpoint-20000/model-00001-of-00002.safetensors +3 -0
  27. checkpoint-20000/model-00002-of-00002.safetensors +3 -0
  28. checkpoint-20000/model.safetensors.index.json +0 -0
  29. checkpoint-20000/optimizer.pt +3 -0
  30. checkpoint-20000/processor_config.json +471 -0
  31. checkpoint-20000/rng_state.pth +3 -0
  32. checkpoint-20000/scheduler.pt +3 -0
  33. checkpoint-20000/statistics.json +0 -0
  34. checkpoint-20000/trainer_state.json +0 -0
  35. checkpoint-20000/training_args.bin +3 -0
  36. checkpoint-20000/wandb_config.json +1 -0
  37. config.json +70 -0
  38. experiment_cfg/conf.yaml +225 -0
  39. experiment_cfg/config.yaml +260 -0
  40. experiment_cfg/dataset_statistics.json +149 -0
  41. experiment_cfg/final_model_config.json +54 -0
  42. experiment_cfg/final_processor_config.json +0 -0
  43. model-00001-of-00002.safetensors +3 -0
  44. model-00002-of-00002.safetensors +3 -0
  45. model.safetensors.index.json +0 -0
  46. processor/embodiment_id.json +11 -0
  47. processor/processor_config.json +471 -0
  48. processor/statistics.json +0 -0
  49. training_args.bin +3 -0
  50. wandb_config.json +1 -0
checkpoint-10000/config.json ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_horizon": 50,
3
+ "add_pos_embed": true,
4
+ "apply_sincos_state_encoding": true,
5
+ "architectures": [
6
+ "Gr00tN1d6"
7
+ ],
8
+ "attn_dropout": 0.2,
9
+ "attn_implementation": null,
10
+ "backbone_embedding_dim": 2048,
11
+ "backbone_model_type": "eagle",
12
+ "backbone_trainable_params_fp32": true,
13
+ "collator_overwrite_image_inputs": false,
14
+ "color_jitter_params": {
15
+ "brightness": 0.1,
16
+ "contrast": 0.1,
17
+ "hue": 0.1,
18
+ "saturation": 0.1
19
+ },
20
+ "crop_fraction": 0.95,
21
+ "diffusion_model_cfg": {
22
+ "attention_head_dim": 48,
23
+ "dropout": 0.2,
24
+ "final_dropout": true,
25
+ "interleave_self_attention": true,
26
+ "norm_type": "ada_norm",
27
+ "num_attention_heads": 32,
28
+ "num_layers": 32,
29
+ "output_dim": 1024,
30
+ "positional_embeddings": null
31
+ },
32
+ "eagle_collator": true,
33
+ "formalize_language": true,
34
+ "gemma_collator": false,
35
+ "hidden_size": 1024,
36
+ "image_crop_size": null,
37
+ "image_target_size": null,
38
+ "input_embedding_dim": 1536,
39
+ "load_bf16": true,
40
+ "max_action_dim": 128,
41
+ "max_num_embodiments": 32,
42
+ "max_seq_len": 1024,
43
+ "max_state_dim": 128,
44
+ "model_dtype": "bfloat16",
45
+ "model_name": "nvidia/Eagle-Block2A-2B-v2",
46
+ "model_type": "Gr00tN1d6",
47
+ "noise_beta_alpha": 1.5,
48
+ "noise_beta_beta": 1.0,
49
+ "noise_s": 0.999,
50
+ "num_inference_timesteps": 4,
51
+ "num_timestep_buckets": 1000,
52
+ "random_rotation_angle": null,
53
+ "reproject_vision": false,
54
+ "select_layer": 16,
55
+ "shortest_image_edge": 256,
56
+ "state_dropout_prob": 0.0,
57
+ "torch_dtype": "bfloat16",
58
+ "transformers_version": "4.51.3",
59
+ "tune_diffusion_model": true,
60
+ "tune_llm": false,
61
+ "tune_projector": true,
62
+ "tune_top_llm_layers": 4,
63
+ "tune_visual": false,
64
+ "tune_vlln": true,
65
+ "use_albumentations_transforms": true,
66
+ "use_alternate_vl_dit": true,
67
+ "use_flash_attention": true,
68
+ "use_relative_action": true,
69
+ "use_vlln": true
70
+ }
checkpoint-10000/embodiment_id.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "robocasa_panda_omron": 13,
3
+ "gr1": 20,
4
+ "behavior_r1_pro": 24,
5
+ "unitree_g1": 8,
6
+ "oxe_google": 0,
7
+ "oxe_widowx": 1,
8
+ "libero_panda": 2,
9
+ "oxe_droid": 16,
10
+ "new_embodiment": 10
11
+ }
checkpoint-10000/experiment_cfg/conf.yaml ADDED
@@ -0,0 +1,225 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ load_config_path: null
2
+ model:
3
+ model_type: Gr00tN1d6
4
+ model_dtype: bfloat16
5
+ model_name: nvidia/Eagle-Block2A-2B-v2
6
+ backbone_model_type: eagle
7
+ model_revision: null
8
+ tune_top_llm_layers: 4
9
+ backbone_embedding_dim: 2048
10
+ tune_llm: false
11
+ tune_visual: false
12
+ select_layer: 16
13
+ reproject_vision: false
14
+ use_flash_attention: true
15
+ load_bf16: false
16
+ collator_overwrite_image_inputs: false
17
+ eagle_collator: true
18
+ backbone_trainable_params_fp32: true
19
+ image_crop_size: null
20
+ image_target_size: null
21
+ shortest_image_edge: 256
22
+ crop_fraction: 0.95
23
+ random_rotation_angle: null
24
+ color_jitter_params:
25
+ brightness: 0.3
26
+ contrast: 0.4
27
+ saturation: 0.5
28
+ hue: 0.08
29
+ use_albumentations_transforms: true
30
+ extra_augmentation_config: null
31
+ formalize_language: true
32
+ apply_sincos_state_encoding: false
33
+ use_relative_action: true
34
+ max_state_dim: 29
35
+ max_action_dim: 29
36
+ action_horizon: 16
37
+ hidden_size: 1024
38
+ input_embedding_dim: 1536
39
+ add_pos_embed: true
40
+ attn_dropout: 0.2
41
+ use_vlln: true
42
+ max_seq_len: 1024
43
+ use_alternate_vl_dit: true
44
+ attend_text_every_n_blocks: 2
45
+ diffusion_model_cfg:
46
+ positional_embeddings: null
47
+ num_layers: 32
48
+ num_attention_heads: 32
49
+ attention_head_dim: 48
50
+ norm_type: ada_norm
51
+ dropout: 0.2
52
+ final_dropout: true
53
+ output_dim: 1024
54
+ interleave_self_attention: true
55
+ num_inference_timesteps: 4
56
+ noise_beta_alpha: 1.5
57
+ noise_beta_beta: 1.0
58
+ noise_s: 0.999
59
+ num_timestep_buckets: 1000
60
+ tune_projector: true
61
+ tune_diffusion_model: true
62
+ tune_vlln: true
63
+ state_dropout_prob: 0.0
64
+ state_additive_noise_scale: 0.0
65
+ max_num_embodiments: 32
66
+ data:
67
+ datasets:
68
+ - dataset_paths:
69
+ - gubam/dataset
70
+ embodiment_tag: new_embodiment
71
+ mix_ratio: 1.0
72
+ dataset_type: physical_embodiment
73
+ val_dataset_path: null
74
+ modality_configs:
75
+ new_embodiment:
76
+ video:
77
+ delta_indices:
78
+ - 0
79
+ modality_keys:
80
+ - front
81
+ - wrist
82
+ sin_cos_embedding_keys: null
83
+ mean_std_embedding_keys: null
84
+ action_configs: null
85
+ state:
86
+ delta_indices:
87
+ - 0
88
+ modality_keys:
89
+ - arm
90
+ - gripper
91
+ sin_cos_embedding_keys:
92
+ - arm
93
+ mean_std_embedding_keys: null
94
+ action_configs: null
95
+ action:
96
+ delta_indices:
97
+ - 0
98
+ - 1
99
+ - 2
100
+ - 3
101
+ - 4
102
+ - 5
103
+ - 6
104
+ - 7
105
+ - 8
106
+ - 9
107
+ - 10
108
+ - 11
109
+ - 12
110
+ - 13
111
+ - 14
112
+ - 15
113
+ - 16
114
+ - 17
115
+ - 18
116
+ - 19
117
+ - 20
118
+ - 21
119
+ - 22
120
+ - 23
121
+ - 24
122
+ - 25
123
+ - 26
124
+ - 27
125
+ - 28
126
+ - 29
127
+ modality_keys:
128
+ - arm
129
+ - gripper
130
+ sin_cos_embedding_keys: null
131
+ mean_std_embedding_keys: null
132
+ action_configs:
133
+ - rep: ABSOLUTE
134
+ type: EEF
135
+ format: XYZ_ROTVEC
136
+ state_key: null
137
+ - rep: ABSOLUTE
138
+ type: NON_EEF
139
+ format: DEFAULT
140
+ state_key: null
141
+ language:
142
+ delta_indices:
143
+ - 0
144
+ modality_keys:
145
+ - annotation.task_description
146
+ sin_cos_embedding_keys: null
147
+ mean_std_embedding_keys: null
148
+ action_configs: null
149
+ download_cache: false
150
+ shard_size: 1024
151
+ episode_sampling_rate: 0.1
152
+ num_shards_per_epoch: 100000
153
+ override_pretraining_statistics: false
154
+ mode: single_turn
155
+ random_chop: 0.0
156
+ mock_dataset_mode: false
157
+ shuffle: true
158
+ seed: 42
159
+ multiprocessing_context: fork
160
+ allow_padding: false
161
+ subsample_ratio: 1.0
162
+ image_crop_size:
163
+ - 244
164
+ - 244
165
+ image_target_size:
166
+ - 224
167
+ - 224
168
+ video_backend: torchcodec
169
+ training:
170
+ output_dir: /tmp/finetune
171
+ experiment_name: null
172
+ max_steps: 20000
173
+ global_batch_size: 32
174
+ batch_size: null
175
+ gradient_accumulation_steps: 1
176
+ learning_rate: 0.0001
177
+ lr_scheduler_type: cosine
178
+ weight_decay: 1.0e-05
179
+ warmup_ratio: 0.05
180
+ warmup_steps: 0
181
+ max_grad_norm: 1.0
182
+ optim: adamw_torch
183
+ start_from_checkpoint: nvidia/GR00T-N1.6-3B
184
+ tf32: true
185
+ fp16: false
186
+ bf16: true
187
+ eval_bf16: true
188
+ logging_steps: 10
189
+ save_steps: 10000
190
+ save_total_limit: 5
191
+ save_vl_model: false
192
+ upload_checkpoints: false
193
+ upload_every: 1000
194
+ upload_last_n_checkpoints: 5
195
+ max_concurrent_uploads: 2
196
+ eval_strategy: 'no'
197
+ eval_steps: 500
198
+ eval_set_split_ratio: 0.1
199
+ eval_batch_size: 2
200
+ save_best_eval_metric_name: ''
201
+ save_best_eval_metric_greater_is_better: true
202
+ deepspeed_stage: 2
203
+ gradient_checkpointing: false
204
+ transformers_trust_remote_code: true
205
+ transformers_local_files_only: false
206
+ transformers_cache_dir: null
207
+ transformers_access_token: null
208
+ use_ddp: false
209
+ ddp_bucket_cap_mb: 100
210
+ num_gpus: 1
211
+ dataloader_num_workers: 4
212
+ remove_unused_columns: false
213
+ use_wandb: true
214
+ wandb_project: finetune-gr00t-n1d6
215
+ enable_profiling: false
216
+ max_retries: 3
217
+ assert_loss_less_than: null
218
+ add_rl_callback: false
219
+ enable_open_loop_eval: false
220
+ open_loop_eval_traj_ids:
221
+ - 0
222
+ open_loop_eval_steps_per_traj: 100
223
+ open_loop_eval_plot_indices: null
224
+ max_steps: 20000
225
+ save_steps: 10000
checkpoint-10000/experiment_cfg/config.yaml ADDED
@@ -0,0 +1,260 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ !!python/object:gr00t.configs.base_config.Config
2
+ data: !!python/object:gr00t.configs.data.data_config.DataConfig
3
+ allow_padding: false
4
+ datasets:
5
+ - !!python/object:gr00t.configs.data.data_config.SingleDatasetConfig
6
+ dataset_paths:
7
+ - gubam/dataset
8
+ dataset_type: physical_embodiment
9
+ embodiment_tag: new_embodiment
10
+ mix_ratio: 1.0
11
+ val_dataset_path: null
12
+ download_cache: false
13
+ episode_sampling_rate: 0.1
14
+ image_crop_size:
15
+ - 244
16
+ - 244
17
+ image_target_size:
18
+ - 224
19
+ - 224
20
+ mock_dataset_mode: false
21
+ modality_configs:
22
+ new_embodiment:
23
+ action: !!python/object:gr00t.data.types.ModalityConfig
24
+ action_configs:
25
+ - !!python/object:gr00t.data.types.ActionConfig
26
+ format: !!python/object/apply:gr00t.data.types.ActionFormat
27
+ - xyz+rotvec
28
+ rep: &id001 !!python/object/apply:gr00t.data.types.ActionRepresentation
29
+ - absolute
30
+ state_key: null
31
+ type: !!python/object/apply:gr00t.data.types.ActionType
32
+ - eef
33
+ - !!python/object:gr00t.data.types.ActionConfig
34
+ format: !!python/object/apply:gr00t.data.types.ActionFormat
35
+ - default
36
+ rep: *id001
37
+ state_key: null
38
+ type: !!python/object/apply:gr00t.data.types.ActionType
39
+ - non_eef
40
+ delta_indices:
41
+ - 0
42
+ - 1
43
+ - 2
44
+ - 3
45
+ - 4
46
+ - 5
47
+ - 6
48
+ - 7
49
+ - 8
50
+ - 9
51
+ - 10
52
+ - 11
53
+ - 12
54
+ - 13
55
+ - 14
56
+ - 15
57
+ - 16
58
+ - 17
59
+ - 18
60
+ - 19
61
+ - 20
62
+ - 21
63
+ - 22
64
+ - 23
65
+ - 24
66
+ - 25
67
+ - 26
68
+ - 27
69
+ - 28
70
+ - 29
71
+ mean_std_embedding_keys: null
72
+ modality_keys:
73
+ - arm
74
+ - gripper
75
+ sin_cos_embedding_keys: null
76
+ language: !!python/object:gr00t.data.types.ModalityConfig
77
+ action_configs: null
78
+ delta_indices:
79
+ - 0
80
+ mean_std_embedding_keys: null
81
+ modality_keys:
82
+ - annotation.task_description
83
+ sin_cos_embedding_keys: null
84
+ state: !!python/object:gr00t.data.types.ModalityConfig
85
+ action_configs: null
86
+ delta_indices:
87
+ - 0
88
+ mean_std_embedding_keys: null
89
+ modality_keys:
90
+ - arm
91
+ - gripper
92
+ sin_cos_embedding_keys:
93
+ - arm
94
+ video: !!python/object:gr00t.data.types.ModalityConfig
95
+ action_configs: null
96
+ delta_indices:
97
+ - 0
98
+ mean_std_embedding_keys: null
99
+ modality_keys:
100
+ - front
101
+ - wrist
102
+ sin_cos_embedding_keys: null
103
+ mode: single_turn
104
+ multiprocessing_context: fork
105
+ num_shards_per_epoch: 100000
106
+ override_pretraining_statistics: false
107
+ random_chop: 0.0
108
+ seed: 42
109
+ shard_size: 1024
110
+ shuffle: true
111
+ subsample_ratio: 1.0
112
+ video_backend: torchcodec
113
+ load_config_path: null
114
+ model: !!python/object:gr00t.configs.model.gr00t_n1d6.Gr00tN1d6Config
115
+ _attn_implementation_autoset: false
116
+ _attn_implementation_internal: null
117
+ _commit_hash: null
118
+ _name_or_path: ''
119
+ add_cross_attention: false
120
+ architectures: null
121
+ backbone_model_type: eagle
122
+ backbone_trainable_params_fp32: true
123
+ bad_words_ids: null
124
+ begin_suppress_tokens: null
125
+ bos_token_id: null
126
+ chunk_size_feed_forward: 0
127
+ color_jitter_params:
128
+ brightness: 0.3
129
+ contrast: 0.4
130
+ hue: 0.08
131
+ saturation: 0.5
132
+ cross_attention_hidden_size: null
133
+ decoder_start_token_id: null
134
+ diffusion_model_cfg:
135
+ attention_head_dim: 48
136
+ dropout: 0.2
137
+ final_dropout: true
138
+ interleave_self_attention: true
139
+ norm_type: ada_norm
140
+ num_attention_heads: 32
141
+ num_layers: 32
142
+ output_dim: 1024
143
+ positional_embeddings: null
144
+ diversity_penalty: 0.0
145
+ do_sample: false
146
+ eagle_collator: true
147
+ early_stopping: false
148
+ encoder_no_repeat_ngram_size: 0
149
+ eos_token_id: null
150
+ exponential_decay_length_penalty: null
151
+ extra_augmentation_config: null
152
+ finetuning_task: null
153
+ forced_bos_token_id: null
154
+ forced_eos_token_id: null
155
+ id2label:
156
+ 0: LABEL_0
157
+ 1: LABEL_1
158
+ is_decoder: false
159
+ is_encoder_decoder: false
160
+ label2id:
161
+ LABEL_0: 0
162
+ LABEL_1: 1
163
+ length_penalty: 1.0
164
+ load_bf16: false
165
+ max_length: 20
166
+ min_length: 0
167
+ model_name: nvidia/Eagle-Block2A-2B-v2
168
+ no_repeat_ngram_size: 0
169
+ num_beam_groups: 1
170
+ num_beams: 1
171
+ num_return_sequences: 1
172
+ output_attentions: false
173
+ output_hidden_states: false
174
+ output_scores: false
175
+ pad_token_id: null
176
+ prefix: null
177
+ problem_type: null
178
+ pruned_heads: {}
179
+ random_rotation_angle: null
180
+ remove_invalid_values: false
181
+ repetition_penalty: 1.0
182
+ reproject_vision: false
183
+ return_dict: true
184
+ return_dict_in_generate: false
185
+ sep_token_id: null
186
+ state_dropout_prob: 0.0
187
+ suppress_tokens: null
188
+ task_specific_params: null
189
+ temperature: 1.0
190
+ tf_legacy_loss: false
191
+ tie_encoder_decoder: false
192
+ tie_word_embeddings: true
193
+ tokenizer_class: null
194
+ top_k: 50
195
+ top_p: 1.0
196
+ torch_dtype: null
197
+ torchscript: false
198
+ transformers_version: null
199
+ tune_diffusion_model: true
200
+ tune_llm: false
201
+ tune_projector: true
202
+ tune_visual: false
203
+ typical_p: 1.0
204
+ use_bfloat16: false
205
+ use_relative_action: true
206
+ training: !!python/object:gr00t.configs.training.training_config.TrainingConfig
207
+ add_rl_callback: false
208
+ assert_loss_less_than: null
209
+ batch_size: null
210
+ bf16: true
211
+ dataloader_num_workers: 4
212
+ ddp_bucket_cap_mb: 100
213
+ deepspeed_stage: 2
214
+ enable_open_loop_eval: false
215
+ enable_profiling: false
216
+ eval_batch_size: 2
217
+ eval_bf16: true
218
+ eval_set_split_ratio: 0.1
219
+ eval_steps: 500
220
+ eval_strategy: 'no'
221
+ experiment_name: null
222
+ fp16: false
223
+ global_batch_size: 32
224
+ gradient_accumulation_steps: 1
225
+ gradient_checkpointing: false
226
+ learning_rate: 0.0001
227
+ logging_steps: 10
228
+ lr_scheduler_type: cosine
229
+ max_concurrent_uploads: 2
230
+ max_grad_norm: 1.0
231
+ max_retries: 3
232
+ max_steps: 20000
233
+ num_gpus: 1
234
+ open_loop_eval_plot_indices: null
235
+ open_loop_eval_steps_per_traj: 100
236
+ open_loop_eval_traj_ids:
237
+ - 0
238
+ optim: adamw_torch
239
+ output_dir: /tmp/finetune
240
+ remove_unused_columns: false
241
+ save_best_eval_metric_greater_is_better: true
242
+ save_best_eval_metric_name: ''
243
+ save_steps: 10000
244
+ save_total_limit: 5
245
+ save_vl_model: false
246
+ start_from_checkpoint: nvidia/GR00T-N1.6-3B
247
+ tf32: true
248
+ transformers_access_token: null
249
+ transformers_cache_dir: null
250
+ transformers_local_files_only: false
251
+ transformers_trust_remote_code: true
252
+ upload_checkpoints: false
253
+ upload_every: 1000
254
+ upload_last_n_checkpoints: 5
255
+ use_ddp: false
256
+ use_wandb: true
257
+ wandb_project: finetune-gr00t-n1d6
258
+ warmup_ratio: 0.05
259
+ warmup_steps: 0
260
+ weight_decay: 1.0e-05
checkpoint-10000/experiment_cfg/dataset_statistics.json ADDED
@@ -0,0 +1,149 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "state": {
4
+ "arm": {
5
+ "min": [
6
+ -0.30735278129577637,
7
+ -0.5557547807693481,
8
+ 0.9460603594779968,
9
+ -3.0296072959899902,
10
+ -3.136127471923828,
11
+ -0.8582668304443359
12
+ ],
13
+ "max": [
14
+ 0.4784075915813446,
15
+ 0.536034107208252,
16
+ 1.7277131080627441,
17
+ 2.922990322113037,
18
+ 3.1376261711120605,
19
+ 0.42977502942085266
20
+ ],
21
+ "mean": [
22
+ 0.13191731012878258,
23
+ 0.031970066748100066,
24
+ 1.1558844479773447,
25
+ -1.6954683609230488,
26
+ 1.6487134783988946,
27
+ -0.14286577845817536
28
+ ],
29
+ "std": [
30
+ 0.15234599091258288,
31
+ 0.25073243709252585,
32
+ 0.19292941514856915,
33
+ 1.1962757356957574,
34
+ 1.2519753680003787,
35
+ 0.2175599434355554
36
+ ],
37
+ "q01": [
38
+ -0.04688343169734043,
39
+ -0.2370660196788625,
40
+ 0.9693865812794936,
41
+ -2.460839541332334,
42
+ -0.07001938726657715,
43
+ -0.6103960370493784
44
+ ],
45
+ "q99": [
46
+ 0.2768438247195932,
47
+ 0.267643256964274,
48
+ 1.5990514549460293,
49
+ 0.01546860347780651,
50
+ 2.483356319667699,
51
+ 0.11741250933956077
52
+ ]
53
+ },
54
+ "gripper": {
55
+ "min": [
56
+ -0.001212759525515139
57
+ ],
58
+ "max": [
59
+ 0.7762507200241089
60
+ ],
61
+ "mean": [
62
+ 0.1826333326178475
63
+ ],
64
+ "std": [
65
+ 0.20001624021635997
66
+ ],
67
+ "q01": [
68
+ 0.021986669201693607
69
+ ],
70
+ "q99": [
71
+ 0.45503166863409855
72
+ ]
73
+ }
74
+ },
75
+ "action": {
76
+ "arm": {
77
+ "min": [
78
+ -0.3141818344593048,
79
+ -0.6150317192077637,
80
+ 0.9386419653892517,
81
+ -3.101818323135376,
82
+ -3.1366591453552246,
83
+ -0.9746662378311157
84
+ ],
85
+ "max": [
86
+ 0.48730289936065674,
87
+ 0.590604305267334,
88
+ 1.7818619012832642,
89
+ 3.1010582447052,
90
+ 3.128772258758545,
91
+ 0.48776116967201233
92
+ ],
93
+ "mean": [
94
+ 0.11818711284420536,
95
+ 0.028762292790606554,
96
+ 1.1608746673321966,
97
+ -1.6448431900842613,
98
+ 1.6027947065112016,
99
+ -0.15912706390756018
100
+ ],
101
+ "std": [
102
+ 0.16134523907378098,
103
+ 0.25919821992288916,
104
+ 0.20997126294727028,
105
+ 1.2448266960019687,
106
+ 1.2914855158630338,
107
+ 0.2432569420599352
108
+ ],
109
+ "q01": [
110
+ -0.07862773065169046,
111
+ -0.26523173844012143,
112
+ 0.963678525591205,
113
+ -2.513543702118574,
114
+ -0.4443971805927962,
115
+ -0.6861969386370356
116
+ ],
117
+ "q99": [
118
+ 0.28730840579975764,
119
+ 0.2975560395103185,
120
+ 1.637628068438941,
121
+ 0.489103782574516,
122
+ 2.526707288272369,
123
+ 0.15085034587810217
124
+ ]
125
+ },
126
+ "gripper": {
127
+ "min": [
128
+ -0.8607733845710754
129
+ ],
130
+ "max": [
131
+ 0.13195456564426422
132
+ ],
133
+ "mean": [
134
+ -0.332952961079147
135
+ ],
136
+ "std": [
137
+ 0.3676414530724673
138
+ ],
139
+ "q01": [
140
+ -0.848867505241717
141
+ ],
142
+ "q99": [
143
+ -0.010392814656977918
144
+ ]
145
+ }
146
+ },
147
+ "relative_action": {}
148
+ }
149
+ }
checkpoint-10000/experiment_cfg/final_model_config.json ADDED
@@ -0,0 +1,54 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "Gr00tN1d6",
3
+ "model_dtype": "bfloat16",
4
+ "model_name": "nvidia/Eagle-Block2A-2B-v2",
5
+ "backbone_model_type": "eagle",
6
+ "model_revision": null,
7
+ "tune_top_llm_layers": 4,
8
+ "backbone_embedding_dim": 2048,
9
+ "tune_llm": false,
10
+ "tune_visual": false,
11
+ "select_layer": 16,
12
+ "reproject_vision": false,
13
+ "use_flash_attention": true,
14
+ "load_bf16": true,
15
+ "collator_overwrite_image_inputs": false,
16
+ "eagle_collator": true,
17
+ "backbone_trainable_params_fp32": true,
18
+ "extra_augmentation_config": null,
19
+ "apply_sincos_state_encoding": true,
20
+ "use_relative_action": true,
21
+ "max_state_dim": 128,
22
+ "max_action_dim": 128,
23
+ "action_horizon": 50,
24
+ "hidden_size": 1024,
25
+ "input_embedding_dim": 1536,
26
+ "add_pos_embed": true,
27
+ "attn_dropout": 0.2,
28
+ "use_vlln": true,
29
+ "max_seq_len": 1024,
30
+ "use_alternate_vl_dit": true,
31
+ "attend_text_every_n_blocks": 2,
32
+ "diffusion_model_cfg": {
33
+ "attention_head_dim": 48,
34
+ "dropout": 0.2,
35
+ "final_dropout": true,
36
+ "interleave_self_attention": true,
37
+ "norm_type": "ada_norm",
38
+ "num_attention_heads": 32,
39
+ "num_layers": 32,
40
+ "output_dim": 1024,
41
+ "positional_embeddings": null
42
+ },
43
+ "num_inference_timesteps": 4,
44
+ "noise_beta_alpha": 1.5,
45
+ "noise_beta_beta": 1.0,
46
+ "noise_s": 0.999,
47
+ "num_timestep_buckets": 1000,
48
+ "tune_projector": true,
49
+ "tune_diffusion_model": true,
50
+ "tune_vlln": true,
51
+ "state_dropout_prob": 0.0,
52
+ "state_additive_noise_scale": 0.0,
53
+ "max_num_embodiments": 32
54
+ }
checkpoint-10000/experiment_cfg/final_processor_config.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-10000/model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:80a9a1a6e1501f62f58da1aa0cf5f3a5b31b98cabd690777b1b06efe35f07e7e
3
+ size 4990120184
checkpoint-10000/model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1283d14b879402feb051e6e63d9b03364c13348658089471e101fdc48a5b7971
3
+ size 4823190320
checkpoint-10000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-10000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ecda63732134006c24aaec54e17f6ea83df2a55aa3125ada1a2ce786025ab0a5
3
+ size 12960193762
checkpoint-10000/processor_config.json ADDED
@@ -0,0 +1,471 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "processor_class": "Gr00tN1d6Processor",
3
+ "processor_kwargs": {
4
+ "modality_configs": {
5
+ "behavior_r1_pro": {
6
+ "video": {
7
+ "delta_indices": [
8
+ 0
9
+ ],
10
+ "modality_keys": [
11
+ "observation.images.rgb.head_256_256",
12
+ "observation.images.rgb.left_wrist_256_256",
13
+ "observation.images.rgb.right_wrist_256_256"
14
+ ],
15
+ "sin_cos_embedding_keys": null,
16
+ "mean_std_embedding_keys": null,
17
+ "action_configs": null
18
+ },
19
+ "state": {
20
+ "delta_indices": [
21
+ 0
22
+ ],
23
+ "modality_keys": [
24
+ "robot_pos",
25
+ "robot_ori_cos",
26
+ "robot_ori_sin",
27
+ "robot_2d_ori",
28
+ "robot_2d_ori_cos",
29
+ "robot_2d_ori_sin",
30
+ "robot_lin_vel",
31
+ "robot_ang_vel",
32
+ "arm_left_qpos",
33
+ "arm_left_qpos_sin",
34
+ "arm_left_qpos_cos",
35
+ "eef_left_pos",
36
+ "eef_left_quat",
37
+ "gripper_left_qpos",
38
+ "arm_right_qpos",
39
+ "arm_right_qpos_sin",
40
+ "arm_right_qpos_cos",
41
+ "eef_right_pos",
42
+ "eef_right_quat",
43
+ "gripper_right_qpos",
44
+ "trunk_qpos"
45
+ ],
46
+ "sin_cos_embedding_keys": null,
47
+ "mean_std_embedding_keys": null,
48
+ "action_configs": null
49
+ },
50
+ "action": {
51
+ "delta_indices": [
52
+ 0,
53
+ 1,
54
+ 2,
55
+ 3,
56
+ 4,
57
+ 5,
58
+ 6,
59
+ 7,
60
+ 8,
61
+ 9,
62
+ 10,
63
+ 11,
64
+ 12,
65
+ 13,
66
+ 14,
67
+ 15,
68
+ 16,
69
+ 17,
70
+ 18,
71
+ 19,
72
+ 20,
73
+ 21,
74
+ 22,
75
+ 23,
76
+ 24,
77
+ 25,
78
+ 26,
79
+ 27,
80
+ 28,
81
+ 29,
82
+ 30,
83
+ 31
84
+ ],
85
+ "modality_keys": [
86
+ "base",
87
+ "torso",
88
+ "left_arm",
89
+ "left_gripper",
90
+ "right_arm",
91
+ "right_gripper"
92
+ ],
93
+ "sin_cos_embedding_keys": null,
94
+ "mean_std_embedding_keys": null,
95
+ "action_configs": [
96
+ {
97
+ "rep": "ABSOLUTE",
98
+ "type": "NON_EEF",
99
+ "format": "DEFAULT",
100
+ "state_key": null
101
+ },
102
+ {
103
+ "rep": "RELATIVE",
104
+ "type": "NON_EEF",
105
+ "format": "DEFAULT",
106
+ "state_key": "trunk_qpos"
107
+ },
108
+ {
109
+ "rep": "RELATIVE",
110
+ "type": "NON_EEF",
111
+ "format": "DEFAULT",
112
+ "state_key": "arm_left_qpos"
113
+ },
114
+ {
115
+ "rep": "ABSOLUTE",
116
+ "type": "NON_EEF",
117
+ "format": "DEFAULT",
118
+ "state_key": null
119
+ },
120
+ {
121
+ "rep": "RELATIVE",
122
+ "type": "NON_EEF",
123
+ "format": "DEFAULT",
124
+ "state_key": "arm_right_qpos"
125
+ },
126
+ {
127
+ "rep": "ABSOLUTE",
128
+ "type": "NON_EEF",
129
+ "format": "DEFAULT",
130
+ "state_key": null
131
+ }
132
+ ]
133
+ },
134
+ "language": {
135
+ "delta_indices": [
136
+ 0
137
+ ],
138
+ "modality_keys": [
139
+ "annotation.human.coarse_action"
140
+ ],
141
+ "sin_cos_embedding_keys": null,
142
+ "mean_std_embedding_keys": null,
143
+ "action_configs": null
144
+ }
145
+ },
146
+ "gr1": {
147
+ "video": {
148
+ "delta_indices": [
149
+ 0
150
+ ],
151
+ "modality_keys": [
152
+ "ego_view_bg_crop_pad_res256_freq20"
153
+ ],
154
+ "sin_cos_embedding_keys": null,
155
+ "mean_std_embedding_keys": null,
156
+ "action_configs": null
157
+ },
158
+ "state": {
159
+ "delta_indices": [
160
+ 0
161
+ ],
162
+ "modality_keys": [
163
+ "left_arm",
164
+ "right_arm",
165
+ "left_hand",
166
+ "right_hand",
167
+ "waist"
168
+ ],
169
+ "sin_cos_embedding_keys": [
170
+ "left_arm",
171
+ "right_arm",
172
+ "left_hand",
173
+ "right_hand",
174
+ "waist"
175
+ ],
176
+ "mean_std_embedding_keys": null,
177
+ "action_configs": null
178
+ },
179
+ "action": {
180
+ "delta_indices": [
181
+ 0,
182
+ 1,
183
+ 2,
184
+ 3,
185
+ 4,
186
+ 5,
187
+ 6,
188
+ 7,
189
+ 8,
190
+ 9,
191
+ 10,
192
+ 11,
193
+ 12,
194
+ 13,
195
+ 14,
196
+ 15
197
+ ],
198
+ "modality_keys": [
199
+ "left_arm",
200
+ "right_arm",
201
+ "left_hand",
202
+ "right_hand",
203
+ "waist"
204
+ ],
205
+ "sin_cos_embedding_keys": null,
206
+ "mean_std_embedding_keys": null,
207
+ "action_configs": [
208
+ {
209
+ "rep": "RELATIVE",
210
+ "type": "NON_EEF",
211
+ "format": "DEFAULT",
212
+ "state_key": null
213
+ },
214
+ {
215
+ "rep": "RELATIVE",
216
+ "type": "NON_EEF",
217
+ "format": "DEFAULT",
218
+ "state_key": null
219
+ },
220
+ {
221
+ "rep": "RELATIVE",
222
+ "type": "NON_EEF",
223
+ "format": "DEFAULT",
224
+ "state_key": null
225
+ },
226
+ {
227
+ "rep": "RELATIVE",
228
+ "type": "NON_EEF",
229
+ "format": "DEFAULT",
230
+ "state_key": null
231
+ },
232
+ {
233
+ "rep": "ABSOLUTE",
234
+ "type": "NON_EEF",
235
+ "format": "DEFAULT",
236
+ "state_key": null
237
+ }
238
+ ]
239
+ },
240
+ "language": {
241
+ "delta_indices": [
242
+ 0
243
+ ],
244
+ "modality_keys": [
245
+ "task"
246
+ ],
247
+ "sin_cos_embedding_keys": null,
248
+ "mean_std_embedding_keys": null,
249
+ "action_configs": null
250
+ }
251
+ },
252
+ "robocasa_panda_omron": {
253
+ "video": {
254
+ "delta_indices": [
255
+ 0
256
+ ],
257
+ "modality_keys": [
258
+ "res256_image_side_0",
259
+ "res256_image_side_1",
260
+ "res256_image_wrist_0"
261
+ ],
262
+ "sin_cos_embedding_keys": null,
263
+ "mean_std_embedding_keys": null,
264
+ "action_configs": null
265
+ },
266
+ "state": {
267
+ "delta_indices": [
268
+ 0
269
+ ],
270
+ "modality_keys": [
271
+ "end_effector_position_relative",
272
+ "end_effector_rotation_relative",
273
+ "gripper_qpos",
274
+ "base_position",
275
+ "base_rotation"
276
+ ],
277
+ "sin_cos_embedding_keys": null,
278
+ "mean_std_embedding_keys": null,
279
+ "action_configs": null
280
+ },
281
+ "action": {
282
+ "delta_indices": [
283
+ 0,
284
+ 1,
285
+ 2,
286
+ 3,
287
+ 4,
288
+ 5,
289
+ 6,
290
+ 7,
291
+ 8,
292
+ 9,
293
+ 10,
294
+ 11,
295
+ 12,
296
+ 13,
297
+ 14,
298
+ 15
299
+ ],
300
+ "modality_keys": [
301
+ "end_effector_position",
302
+ "end_effector_rotation",
303
+ "gripper_close",
304
+ "base_motion",
305
+ "control_mode"
306
+ ],
307
+ "sin_cos_embedding_keys": null,
308
+ "mean_std_embedding_keys": null,
309
+ "action_configs": [
310
+ {
311
+ "rep": "ABSOLUTE",
312
+ "type": "NON_EEF",
313
+ "format": "DEFAULT",
314
+ "state_key": null
315
+ },
316
+ {
317
+ "rep": "ABSOLUTE",
318
+ "type": "NON_EEF",
319
+ "format": "DEFAULT",
320
+ "state_key": null
321
+ },
322
+ {
323
+ "rep": "ABSOLUTE",
324
+ "type": "NON_EEF",
325
+ "format": "DEFAULT",
326
+ "state_key": null
327
+ },
328
+ {
329
+ "rep": "ABSOLUTE",
330
+ "type": "NON_EEF",
331
+ "format": "DEFAULT",
332
+ "state_key": null
333
+ },
334
+ {
335
+ "rep": "ABSOLUTE",
336
+ "type": "NON_EEF",
337
+ "format": "DEFAULT",
338
+ "state_key": null
339
+ }
340
+ ]
341
+ },
342
+ "language": {
343
+ "delta_indices": [
344
+ 0
345
+ ],
346
+ "modality_keys": [
347
+ "annotation.human.action.task_description"
348
+ ],
349
+ "sin_cos_embedding_keys": null,
350
+ "mean_std_embedding_keys": null,
351
+ "action_configs": null
352
+ }
353
+ },
354
+ "new_embodiment": {
355
+ "video": {
356
+ "delta_indices": [
357
+ 0
358
+ ],
359
+ "modality_keys": [
360
+ "front",
361
+ "wrist"
362
+ ],
363
+ "sin_cos_embedding_keys": null,
364
+ "mean_std_embedding_keys": null,
365
+ "action_configs": null
366
+ },
367
+ "state": {
368
+ "delta_indices": [
369
+ 0
370
+ ],
371
+ "modality_keys": [
372
+ "arm",
373
+ "gripper"
374
+ ],
375
+ "sin_cos_embedding_keys": [
376
+ "arm"
377
+ ],
378
+ "mean_std_embedding_keys": null,
379
+ "action_configs": null
380
+ },
381
+ "action": {
382
+ "delta_indices": [
383
+ 0,
384
+ 1,
385
+ 2,
386
+ 3,
387
+ 4,
388
+ 5,
389
+ 6,
390
+ 7,
391
+ 8,
392
+ 9,
393
+ 10,
394
+ 11,
395
+ 12,
396
+ 13,
397
+ 14,
398
+ 15,
399
+ 16,
400
+ 17,
401
+ 18,
402
+ 19,
403
+ 20,
404
+ 21,
405
+ 22,
406
+ 23,
407
+ 24,
408
+ 25,
409
+ 26,
410
+ 27,
411
+ 28,
412
+ 29
413
+ ],
414
+ "modality_keys": [
415
+ "arm",
416
+ "gripper"
417
+ ],
418
+ "sin_cos_embedding_keys": null,
419
+ "mean_std_embedding_keys": null,
420
+ "action_configs": [
421
+ {
422
+ "rep": "ABSOLUTE",
423
+ "type": "EEF",
424
+ "format": "XYZ_ROTVEC",
425
+ "state_key": null
426
+ },
427
+ {
428
+ "rep": "ABSOLUTE",
429
+ "type": "NON_EEF",
430
+ "format": "DEFAULT",
431
+ "state_key": null
432
+ }
433
+ ]
434
+ },
435
+ "language": {
436
+ "delta_indices": [
437
+ 0
438
+ ],
439
+ "modality_keys": [
440
+ "annotation.task_description"
441
+ ],
442
+ "sin_cos_embedding_keys": null,
443
+ "mean_std_embedding_keys": null,
444
+ "action_configs": null
445
+ }
446
+ }
447
+ },
448
+ "image_crop_size": null,
449
+ "image_target_size": null,
450
+ "use_albumentations": true,
451
+ "random_rotation_angle": null,
452
+ "color_jitter_params": {
453
+ "brightness": 0.3,
454
+ "contrast": 0.4,
455
+ "saturation": 0.5,
456
+ "hue": 0.08
457
+ },
458
+ "shortest_image_edge": 256,
459
+ "crop_fraction": 0.95,
460
+ "model_name": "nvidia/Eagle-Block2A-2B-v2",
461
+ "model_type": "eagle",
462
+ "formalize_language": true,
463
+ "max_state_dim": 128,
464
+ "max_action_dim": 128,
465
+ "max_action_horizon": 50,
466
+ "use_percentiles": false,
467
+ "clip_outliers": true,
468
+ "apply_sincos_state_encoding": true,
469
+ "use_relative_action": true
470
+ }
471
+ }
checkpoint-10000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:07c9c68f9f5a4ff7c2756642a3e4376e946cce84760795260e1e09012773add2
3
+ size 14645
checkpoint-10000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9793ce798c508501a83c7de96591494637b8ee20bbad166324a3caeddb4cdc8e
3
+ size 1465
checkpoint-10000/statistics.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-10000/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-10000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74fc2963f498bcc7aa9fb904bda507a455bd7ba7c1ed3b99b740bdb1946df479
3
+ size 5713
checkpoint-10000/wandb_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"project": "finetune-gr00t-n1d6", "run_id": "finetune"}
checkpoint-20000/config.json ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_horizon": 50,
3
+ "add_pos_embed": true,
4
+ "apply_sincos_state_encoding": true,
5
+ "architectures": [
6
+ "Gr00tN1d6"
7
+ ],
8
+ "attn_dropout": 0.2,
9
+ "attn_implementation": null,
10
+ "backbone_embedding_dim": 2048,
11
+ "backbone_model_type": "eagle",
12
+ "backbone_trainable_params_fp32": true,
13
+ "collator_overwrite_image_inputs": false,
14
+ "color_jitter_params": {
15
+ "brightness": 0.1,
16
+ "contrast": 0.1,
17
+ "hue": 0.1,
18
+ "saturation": 0.1
19
+ },
20
+ "crop_fraction": 0.95,
21
+ "diffusion_model_cfg": {
22
+ "attention_head_dim": 48,
23
+ "dropout": 0.2,
24
+ "final_dropout": true,
25
+ "interleave_self_attention": true,
26
+ "norm_type": "ada_norm",
27
+ "num_attention_heads": 32,
28
+ "num_layers": 32,
29
+ "output_dim": 1024,
30
+ "positional_embeddings": null
31
+ },
32
+ "eagle_collator": true,
33
+ "formalize_language": true,
34
+ "gemma_collator": false,
35
+ "hidden_size": 1024,
36
+ "image_crop_size": null,
37
+ "image_target_size": null,
38
+ "input_embedding_dim": 1536,
39
+ "load_bf16": true,
40
+ "max_action_dim": 128,
41
+ "max_num_embodiments": 32,
42
+ "max_seq_len": 1024,
43
+ "max_state_dim": 128,
44
+ "model_dtype": "bfloat16",
45
+ "model_name": "nvidia/Eagle-Block2A-2B-v2",
46
+ "model_type": "Gr00tN1d6",
47
+ "noise_beta_alpha": 1.5,
48
+ "noise_beta_beta": 1.0,
49
+ "noise_s": 0.999,
50
+ "num_inference_timesteps": 4,
51
+ "num_timestep_buckets": 1000,
52
+ "random_rotation_angle": null,
53
+ "reproject_vision": false,
54
+ "select_layer": 16,
55
+ "shortest_image_edge": 256,
56
+ "state_dropout_prob": 0.0,
57
+ "torch_dtype": "bfloat16",
58
+ "transformers_version": "4.51.3",
59
+ "tune_diffusion_model": true,
60
+ "tune_llm": false,
61
+ "tune_projector": true,
62
+ "tune_top_llm_layers": 4,
63
+ "tune_visual": false,
64
+ "tune_vlln": true,
65
+ "use_albumentations_transforms": true,
66
+ "use_alternate_vl_dit": true,
67
+ "use_flash_attention": true,
68
+ "use_relative_action": true,
69
+ "use_vlln": true
70
+ }
checkpoint-20000/embodiment_id.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "robocasa_panda_omron": 13,
3
+ "gr1": 20,
4
+ "behavior_r1_pro": 24,
5
+ "unitree_g1": 8,
6
+ "oxe_google": 0,
7
+ "oxe_widowx": 1,
8
+ "libero_panda": 2,
9
+ "oxe_droid": 16,
10
+ "new_embodiment": 10
11
+ }
checkpoint-20000/experiment_cfg/conf.yaml ADDED
@@ -0,0 +1,225 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ load_config_path: null
2
+ model:
3
+ model_type: Gr00tN1d6
4
+ model_dtype: bfloat16
5
+ model_name: nvidia/Eagle-Block2A-2B-v2
6
+ backbone_model_type: eagle
7
+ model_revision: null
8
+ tune_top_llm_layers: 4
9
+ backbone_embedding_dim: 2048
10
+ tune_llm: false
11
+ tune_visual: false
12
+ select_layer: 16
13
+ reproject_vision: false
14
+ use_flash_attention: true
15
+ load_bf16: false
16
+ collator_overwrite_image_inputs: false
17
+ eagle_collator: true
18
+ backbone_trainable_params_fp32: true
19
+ image_crop_size: null
20
+ image_target_size: null
21
+ shortest_image_edge: 256
22
+ crop_fraction: 0.95
23
+ random_rotation_angle: null
24
+ color_jitter_params:
25
+ brightness: 0.3
26
+ contrast: 0.4
27
+ saturation: 0.5
28
+ hue: 0.08
29
+ use_albumentations_transforms: true
30
+ extra_augmentation_config: null
31
+ formalize_language: true
32
+ apply_sincos_state_encoding: false
33
+ use_relative_action: true
34
+ max_state_dim: 29
35
+ max_action_dim: 29
36
+ action_horizon: 16
37
+ hidden_size: 1024
38
+ input_embedding_dim: 1536
39
+ add_pos_embed: true
40
+ attn_dropout: 0.2
41
+ use_vlln: true
42
+ max_seq_len: 1024
43
+ use_alternate_vl_dit: true
44
+ attend_text_every_n_blocks: 2
45
+ diffusion_model_cfg:
46
+ positional_embeddings: null
47
+ num_layers: 32
48
+ num_attention_heads: 32
49
+ attention_head_dim: 48
50
+ norm_type: ada_norm
51
+ dropout: 0.2
52
+ final_dropout: true
53
+ output_dim: 1024
54
+ interleave_self_attention: true
55
+ num_inference_timesteps: 4
56
+ noise_beta_alpha: 1.5
57
+ noise_beta_beta: 1.0
58
+ noise_s: 0.999
59
+ num_timestep_buckets: 1000
60
+ tune_projector: true
61
+ tune_diffusion_model: true
62
+ tune_vlln: true
63
+ state_dropout_prob: 0.0
64
+ state_additive_noise_scale: 0.0
65
+ max_num_embodiments: 32
66
+ data:
67
+ datasets:
68
+ - dataset_paths:
69
+ - gubam/dataset
70
+ embodiment_tag: new_embodiment
71
+ mix_ratio: 1.0
72
+ dataset_type: physical_embodiment
73
+ val_dataset_path: null
74
+ modality_configs:
75
+ new_embodiment:
76
+ video:
77
+ delta_indices:
78
+ - 0
79
+ modality_keys:
80
+ - front
81
+ - wrist
82
+ sin_cos_embedding_keys: null
83
+ mean_std_embedding_keys: null
84
+ action_configs: null
85
+ state:
86
+ delta_indices:
87
+ - 0
88
+ modality_keys:
89
+ - arm
90
+ - gripper
91
+ sin_cos_embedding_keys:
92
+ - arm
93
+ mean_std_embedding_keys: null
94
+ action_configs: null
95
+ action:
96
+ delta_indices:
97
+ - 0
98
+ - 1
99
+ - 2
100
+ - 3
101
+ - 4
102
+ - 5
103
+ - 6
104
+ - 7
105
+ - 8
106
+ - 9
107
+ - 10
108
+ - 11
109
+ - 12
110
+ - 13
111
+ - 14
112
+ - 15
113
+ - 16
114
+ - 17
115
+ - 18
116
+ - 19
117
+ - 20
118
+ - 21
119
+ - 22
120
+ - 23
121
+ - 24
122
+ - 25
123
+ - 26
124
+ - 27
125
+ - 28
126
+ - 29
127
+ modality_keys:
128
+ - arm
129
+ - gripper
130
+ sin_cos_embedding_keys: null
131
+ mean_std_embedding_keys: null
132
+ action_configs:
133
+ - rep: ABSOLUTE
134
+ type: EEF
135
+ format: XYZ_ROTVEC
136
+ state_key: null
137
+ - rep: ABSOLUTE
138
+ type: NON_EEF
139
+ format: DEFAULT
140
+ state_key: null
141
+ language:
142
+ delta_indices:
143
+ - 0
144
+ modality_keys:
145
+ - annotation.task_description
146
+ sin_cos_embedding_keys: null
147
+ mean_std_embedding_keys: null
148
+ action_configs: null
149
+ download_cache: false
150
+ shard_size: 1024
151
+ episode_sampling_rate: 0.1
152
+ num_shards_per_epoch: 100000
153
+ override_pretraining_statistics: false
154
+ mode: single_turn
155
+ random_chop: 0.0
156
+ mock_dataset_mode: false
157
+ shuffle: true
158
+ seed: 42
159
+ multiprocessing_context: fork
160
+ allow_padding: false
161
+ subsample_ratio: 1.0
162
+ image_crop_size:
163
+ - 244
164
+ - 244
165
+ image_target_size:
166
+ - 224
167
+ - 224
168
+ video_backend: torchcodec
169
+ training:
170
+ output_dir: /tmp/finetune
171
+ experiment_name: null
172
+ max_steps: 20000
173
+ global_batch_size: 32
174
+ batch_size: null
175
+ gradient_accumulation_steps: 1
176
+ learning_rate: 0.0001
177
+ lr_scheduler_type: cosine
178
+ weight_decay: 1.0e-05
179
+ warmup_ratio: 0.05
180
+ warmup_steps: 0
181
+ max_grad_norm: 1.0
182
+ optim: adamw_torch
183
+ start_from_checkpoint: nvidia/GR00T-N1.6-3B
184
+ tf32: true
185
+ fp16: false
186
+ bf16: true
187
+ eval_bf16: true
188
+ logging_steps: 10
189
+ save_steps: 10000
190
+ save_total_limit: 5
191
+ save_vl_model: false
192
+ upload_checkpoints: false
193
+ upload_every: 1000
194
+ upload_last_n_checkpoints: 5
195
+ max_concurrent_uploads: 2
196
+ eval_strategy: 'no'
197
+ eval_steps: 500
198
+ eval_set_split_ratio: 0.1
199
+ eval_batch_size: 2
200
+ save_best_eval_metric_name: ''
201
+ save_best_eval_metric_greater_is_better: true
202
+ deepspeed_stage: 2
203
+ gradient_checkpointing: false
204
+ transformers_trust_remote_code: true
205
+ transformers_local_files_only: false
206
+ transformers_cache_dir: null
207
+ transformers_access_token: null
208
+ use_ddp: false
209
+ ddp_bucket_cap_mb: 100
210
+ num_gpus: 1
211
+ dataloader_num_workers: 4
212
+ remove_unused_columns: false
213
+ use_wandb: true
214
+ wandb_project: finetune-gr00t-n1d6
215
+ enable_profiling: false
216
+ max_retries: 3
217
+ assert_loss_less_than: null
218
+ add_rl_callback: false
219
+ enable_open_loop_eval: false
220
+ open_loop_eval_traj_ids:
221
+ - 0
222
+ open_loop_eval_steps_per_traj: 100
223
+ open_loop_eval_plot_indices: null
224
+ max_steps: 20000
225
+ save_steps: 10000
checkpoint-20000/experiment_cfg/config.yaml ADDED
@@ -0,0 +1,260 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ !!python/object:gr00t.configs.base_config.Config
2
+ data: !!python/object:gr00t.configs.data.data_config.DataConfig
3
+ allow_padding: false
4
+ datasets:
5
+ - !!python/object:gr00t.configs.data.data_config.SingleDatasetConfig
6
+ dataset_paths:
7
+ - gubam/dataset
8
+ dataset_type: physical_embodiment
9
+ embodiment_tag: new_embodiment
10
+ mix_ratio: 1.0
11
+ val_dataset_path: null
12
+ download_cache: false
13
+ episode_sampling_rate: 0.1
14
+ image_crop_size:
15
+ - 244
16
+ - 244
17
+ image_target_size:
18
+ - 224
19
+ - 224
20
+ mock_dataset_mode: false
21
+ modality_configs:
22
+ new_embodiment:
23
+ action: !!python/object:gr00t.data.types.ModalityConfig
24
+ action_configs:
25
+ - !!python/object:gr00t.data.types.ActionConfig
26
+ format: !!python/object/apply:gr00t.data.types.ActionFormat
27
+ - xyz+rotvec
28
+ rep: &id001 !!python/object/apply:gr00t.data.types.ActionRepresentation
29
+ - absolute
30
+ state_key: null
31
+ type: !!python/object/apply:gr00t.data.types.ActionType
32
+ - eef
33
+ - !!python/object:gr00t.data.types.ActionConfig
34
+ format: !!python/object/apply:gr00t.data.types.ActionFormat
35
+ - default
36
+ rep: *id001
37
+ state_key: null
38
+ type: !!python/object/apply:gr00t.data.types.ActionType
39
+ - non_eef
40
+ delta_indices:
41
+ - 0
42
+ - 1
43
+ - 2
44
+ - 3
45
+ - 4
46
+ - 5
47
+ - 6
48
+ - 7
49
+ - 8
50
+ - 9
51
+ - 10
52
+ - 11
53
+ - 12
54
+ - 13
55
+ - 14
56
+ - 15
57
+ - 16
58
+ - 17
59
+ - 18
60
+ - 19
61
+ - 20
62
+ - 21
63
+ - 22
64
+ - 23
65
+ - 24
66
+ - 25
67
+ - 26
68
+ - 27
69
+ - 28
70
+ - 29
71
+ mean_std_embedding_keys: null
72
+ modality_keys:
73
+ - arm
74
+ - gripper
75
+ sin_cos_embedding_keys: null
76
+ language: !!python/object:gr00t.data.types.ModalityConfig
77
+ action_configs: null
78
+ delta_indices:
79
+ - 0
80
+ mean_std_embedding_keys: null
81
+ modality_keys:
82
+ - annotation.task_description
83
+ sin_cos_embedding_keys: null
84
+ state: !!python/object:gr00t.data.types.ModalityConfig
85
+ action_configs: null
86
+ delta_indices:
87
+ - 0
88
+ mean_std_embedding_keys: null
89
+ modality_keys:
90
+ - arm
91
+ - gripper
92
+ sin_cos_embedding_keys:
93
+ - arm
94
+ video: !!python/object:gr00t.data.types.ModalityConfig
95
+ action_configs: null
96
+ delta_indices:
97
+ - 0
98
+ mean_std_embedding_keys: null
99
+ modality_keys:
100
+ - front
101
+ - wrist
102
+ sin_cos_embedding_keys: null
103
+ mode: single_turn
104
+ multiprocessing_context: fork
105
+ num_shards_per_epoch: 100000
106
+ override_pretraining_statistics: false
107
+ random_chop: 0.0
108
+ seed: 42
109
+ shard_size: 1024
110
+ shuffle: true
111
+ subsample_ratio: 1.0
112
+ video_backend: torchcodec
113
+ load_config_path: null
114
+ model: !!python/object:gr00t.configs.model.gr00t_n1d6.Gr00tN1d6Config
115
+ _attn_implementation_autoset: false
116
+ _attn_implementation_internal: null
117
+ _commit_hash: null
118
+ _name_or_path: ''
119
+ add_cross_attention: false
120
+ architectures: null
121
+ backbone_model_type: eagle
122
+ backbone_trainable_params_fp32: true
123
+ bad_words_ids: null
124
+ begin_suppress_tokens: null
125
+ bos_token_id: null
126
+ chunk_size_feed_forward: 0
127
+ color_jitter_params:
128
+ brightness: 0.3
129
+ contrast: 0.4
130
+ hue: 0.08
131
+ saturation: 0.5
132
+ cross_attention_hidden_size: null
133
+ decoder_start_token_id: null
134
+ diffusion_model_cfg:
135
+ attention_head_dim: 48
136
+ dropout: 0.2
137
+ final_dropout: true
138
+ interleave_self_attention: true
139
+ norm_type: ada_norm
140
+ num_attention_heads: 32
141
+ num_layers: 32
142
+ output_dim: 1024
143
+ positional_embeddings: null
144
+ diversity_penalty: 0.0
145
+ do_sample: false
146
+ eagle_collator: true
147
+ early_stopping: false
148
+ encoder_no_repeat_ngram_size: 0
149
+ eos_token_id: null
150
+ exponential_decay_length_penalty: null
151
+ extra_augmentation_config: null
152
+ finetuning_task: null
153
+ forced_bos_token_id: null
154
+ forced_eos_token_id: null
155
+ id2label:
156
+ 0: LABEL_0
157
+ 1: LABEL_1
158
+ is_decoder: false
159
+ is_encoder_decoder: false
160
+ label2id:
161
+ LABEL_0: 0
162
+ LABEL_1: 1
163
+ length_penalty: 1.0
164
+ load_bf16: false
165
+ max_length: 20
166
+ min_length: 0
167
+ model_name: nvidia/Eagle-Block2A-2B-v2
168
+ no_repeat_ngram_size: 0
169
+ num_beam_groups: 1
170
+ num_beams: 1
171
+ num_return_sequences: 1
172
+ output_attentions: false
173
+ output_hidden_states: false
174
+ output_scores: false
175
+ pad_token_id: null
176
+ prefix: null
177
+ problem_type: null
178
+ pruned_heads: {}
179
+ random_rotation_angle: null
180
+ remove_invalid_values: false
181
+ repetition_penalty: 1.0
182
+ reproject_vision: false
183
+ return_dict: true
184
+ return_dict_in_generate: false
185
+ sep_token_id: null
186
+ state_dropout_prob: 0.0
187
+ suppress_tokens: null
188
+ task_specific_params: null
189
+ temperature: 1.0
190
+ tf_legacy_loss: false
191
+ tie_encoder_decoder: false
192
+ tie_word_embeddings: true
193
+ tokenizer_class: null
194
+ top_k: 50
195
+ top_p: 1.0
196
+ torch_dtype: null
197
+ torchscript: false
198
+ transformers_version: null
199
+ tune_diffusion_model: true
200
+ tune_llm: false
201
+ tune_projector: true
202
+ tune_visual: false
203
+ typical_p: 1.0
204
+ use_bfloat16: false
205
+ use_relative_action: true
206
+ training: !!python/object:gr00t.configs.training.training_config.TrainingConfig
207
+ add_rl_callback: false
208
+ assert_loss_less_than: null
209
+ batch_size: null
210
+ bf16: true
211
+ dataloader_num_workers: 4
212
+ ddp_bucket_cap_mb: 100
213
+ deepspeed_stage: 2
214
+ enable_open_loop_eval: false
215
+ enable_profiling: false
216
+ eval_batch_size: 2
217
+ eval_bf16: true
218
+ eval_set_split_ratio: 0.1
219
+ eval_steps: 500
220
+ eval_strategy: 'no'
221
+ experiment_name: null
222
+ fp16: false
223
+ global_batch_size: 32
224
+ gradient_accumulation_steps: 1
225
+ gradient_checkpointing: false
226
+ learning_rate: 0.0001
227
+ logging_steps: 10
228
+ lr_scheduler_type: cosine
229
+ max_concurrent_uploads: 2
230
+ max_grad_norm: 1.0
231
+ max_retries: 3
232
+ max_steps: 20000
233
+ num_gpus: 1
234
+ open_loop_eval_plot_indices: null
235
+ open_loop_eval_steps_per_traj: 100
236
+ open_loop_eval_traj_ids:
237
+ - 0
238
+ optim: adamw_torch
239
+ output_dir: /tmp/finetune
240
+ remove_unused_columns: false
241
+ save_best_eval_metric_greater_is_better: true
242
+ save_best_eval_metric_name: ''
243
+ save_steps: 10000
244
+ save_total_limit: 5
245
+ save_vl_model: false
246
+ start_from_checkpoint: nvidia/GR00T-N1.6-3B
247
+ tf32: true
248
+ transformers_access_token: null
249
+ transformers_cache_dir: null
250
+ transformers_local_files_only: false
251
+ transformers_trust_remote_code: true
252
+ upload_checkpoints: false
253
+ upload_every: 1000
254
+ upload_last_n_checkpoints: 5
255
+ use_ddp: false
256
+ use_wandb: true
257
+ wandb_project: finetune-gr00t-n1d6
258
+ warmup_ratio: 0.05
259
+ warmup_steps: 0
260
+ weight_decay: 1.0e-05
checkpoint-20000/experiment_cfg/dataset_statistics.json ADDED
@@ -0,0 +1,149 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "state": {
4
+ "arm": {
5
+ "min": [
6
+ -0.30735278129577637,
7
+ -0.5557547807693481,
8
+ 0.9460603594779968,
9
+ -3.0296072959899902,
10
+ -3.136127471923828,
11
+ -0.8582668304443359
12
+ ],
13
+ "max": [
14
+ 0.4784075915813446,
15
+ 0.536034107208252,
16
+ 1.7277131080627441,
17
+ 2.922990322113037,
18
+ 3.1376261711120605,
19
+ 0.42977502942085266
20
+ ],
21
+ "mean": [
22
+ 0.13191731012878258,
23
+ 0.031970066748100066,
24
+ 1.1558844479773447,
25
+ -1.6954683609230488,
26
+ 1.6487134783988946,
27
+ -0.14286577845817536
28
+ ],
29
+ "std": [
30
+ 0.15234599091258288,
31
+ 0.25073243709252585,
32
+ 0.19292941514856915,
33
+ 1.1962757356957574,
34
+ 1.2519753680003787,
35
+ 0.2175599434355554
36
+ ],
37
+ "q01": [
38
+ -0.04688343169734043,
39
+ -0.2370660196788625,
40
+ 0.9693865812794936,
41
+ -2.460839541332334,
42
+ -0.07001938726657715,
43
+ -0.6103960370493784
44
+ ],
45
+ "q99": [
46
+ 0.2768438247195932,
47
+ 0.267643256964274,
48
+ 1.5990514549460293,
49
+ 0.01546860347780651,
50
+ 2.483356319667699,
51
+ 0.11741250933956077
52
+ ]
53
+ },
54
+ "gripper": {
55
+ "min": [
56
+ -0.001212759525515139
57
+ ],
58
+ "max": [
59
+ 0.7762507200241089
60
+ ],
61
+ "mean": [
62
+ 0.1826333326178475
63
+ ],
64
+ "std": [
65
+ 0.20001624021635997
66
+ ],
67
+ "q01": [
68
+ 0.021986669201693607
69
+ ],
70
+ "q99": [
71
+ 0.45503166863409855
72
+ ]
73
+ }
74
+ },
75
+ "action": {
76
+ "arm": {
77
+ "min": [
78
+ -0.3141818344593048,
79
+ -0.6150317192077637,
80
+ 0.9386419653892517,
81
+ -3.101818323135376,
82
+ -3.1366591453552246,
83
+ -0.9746662378311157
84
+ ],
85
+ "max": [
86
+ 0.48730289936065674,
87
+ 0.590604305267334,
88
+ 1.7818619012832642,
89
+ 3.1010582447052,
90
+ 3.128772258758545,
91
+ 0.48776116967201233
92
+ ],
93
+ "mean": [
94
+ 0.11818711284420536,
95
+ 0.028762292790606554,
96
+ 1.1608746673321966,
97
+ -1.6448431900842613,
98
+ 1.6027947065112016,
99
+ -0.15912706390756018
100
+ ],
101
+ "std": [
102
+ 0.16134523907378098,
103
+ 0.25919821992288916,
104
+ 0.20997126294727028,
105
+ 1.2448266960019687,
106
+ 1.2914855158630338,
107
+ 0.2432569420599352
108
+ ],
109
+ "q01": [
110
+ -0.07862773065169046,
111
+ -0.26523173844012143,
112
+ 0.963678525591205,
113
+ -2.513543702118574,
114
+ -0.4443971805927962,
115
+ -0.6861969386370356
116
+ ],
117
+ "q99": [
118
+ 0.28730840579975764,
119
+ 0.2975560395103185,
120
+ 1.637628068438941,
121
+ 0.489103782574516,
122
+ 2.526707288272369,
123
+ 0.15085034587810217
124
+ ]
125
+ },
126
+ "gripper": {
127
+ "min": [
128
+ -0.8607733845710754
129
+ ],
130
+ "max": [
131
+ 0.13195456564426422
132
+ ],
133
+ "mean": [
134
+ -0.332952961079147
135
+ ],
136
+ "std": [
137
+ 0.3676414530724673
138
+ ],
139
+ "q01": [
140
+ -0.848867505241717
141
+ ],
142
+ "q99": [
143
+ -0.010392814656977918
144
+ ]
145
+ }
146
+ },
147
+ "relative_action": {}
148
+ }
149
+ }
checkpoint-20000/experiment_cfg/final_model_config.json ADDED
@@ -0,0 +1,54 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "Gr00tN1d6",
3
+ "model_dtype": "bfloat16",
4
+ "model_name": "nvidia/Eagle-Block2A-2B-v2",
5
+ "backbone_model_type": "eagle",
6
+ "model_revision": null,
7
+ "tune_top_llm_layers": 4,
8
+ "backbone_embedding_dim": 2048,
9
+ "tune_llm": false,
10
+ "tune_visual": false,
11
+ "select_layer": 16,
12
+ "reproject_vision": false,
13
+ "use_flash_attention": true,
14
+ "load_bf16": true,
15
+ "collator_overwrite_image_inputs": false,
16
+ "eagle_collator": true,
17
+ "backbone_trainable_params_fp32": true,
18
+ "extra_augmentation_config": null,
19
+ "apply_sincos_state_encoding": true,
20
+ "use_relative_action": true,
21
+ "max_state_dim": 128,
22
+ "max_action_dim": 128,
23
+ "action_horizon": 50,
24
+ "hidden_size": 1024,
25
+ "input_embedding_dim": 1536,
26
+ "add_pos_embed": true,
27
+ "attn_dropout": 0.2,
28
+ "use_vlln": true,
29
+ "max_seq_len": 1024,
30
+ "use_alternate_vl_dit": true,
31
+ "attend_text_every_n_blocks": 2,
32
+ "diffusion_model_cfg": {
33
+ "attention_head_dim": 48,
34
+ "dropout": 0.2,
35
+ "final_dropout": true,
36
+ "interleave_self_attention": true,
37
+ "norm_type": "ada_norm",
38
+ "num_attention_heads": 32,
39
+ "num_layers": 32,
40
+ "output_dim": 1024,
41
+ "positional_embeddings": null
42
+ },
43
+ "num_inference_timesteps": 4,
44
+ "noise_beta_alpha": 1.5,
45
+ "noise_beta_beta": 1.0,
46
+ "noise_s": 0.999,
47
+ "num_timestep_buckets": 1000,
48
+ "tune_projector": true,
49
+ "tune_diffusion_model": true,
50
+ "tune_vlln": true,
51
+ "state_dropout_prob": 0.0,
52
+ "state_additive_noise_scale": 0.0,
53
+ "max_num_embodiments": 32
54
+ }
checkpoint-20000/experiment_cfg/final_processor_config.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-20000/model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8736b74d559b4e11e3ad82ebc06897e33f828899b64d1a30d122b430e31dc69c
3
+ size 4990120184
checkpoint-20000/model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:70ff15e1834de78e43332688c1220e0347de51af9b8850542f93025af919b6b5
3
+ size 4823190320
checkpoint-20000/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-20000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d52ea41a30637cecf7962133eafb4ad840ffbabf9b1d05f6cc87fd9ec8b2482
3
+ size 12960193762
checkpoint-20000/processor_config.json ADDED
@@ -0,0 +1,471 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "processor_class": "Gr00tN1d6Processor",
3
+ "processor_kwargs": {
4
+ "modality_configs": {
5
+ "behavior_r1_pro": {
6
+ "video": {
7
+ "delta_indices": [
8
+ 0
9
+ ],
10
+ "modality_keys": [
11
+ "observation.images.rgb.head_256_256",
12
+ "observation.images.rgb.left_wrist_256_256",
13
+ "observation.images.rgb.right_wrist_256_256"
14
+ ],
15
+ "sin_cos_embedding_keys": null,
16
+ "mean_std_embedding_keys": null,
17
+ "action_configs": null
18
+ },
19
+ "state": {
20
+ "delta_indices": [
21
+ 0
22
+ ],
23
+ "modality_keys": [
24
+ "robot_pos",
25
+ "robot_ori_cos",
26
+ "robot_ori_sin",
27
+ "robot_2d_ori",
28
+ "robot_2d_ori_cos",
29
+ "robot_2d_ori_sin",
30
+ "robot_lin_vel",
31
+ "robot_ang_vel",
32
+ "arm_left_qpos",
33
+ "arm_left_qpos_sin",
34
+ "arm_left_qpos_cos",
35
+ "eef_left_pos",
36
+ "eef_left_quat",
37
+ "gripper_left_qpos",
38
+ "arm_right_qpos",
39
+ "arm_right_qpos_sin",
40
+ "arm_right_qpos_cos",
41
+ "eef_right_pos",
42
+ "eef_right_quat",
43
+ "gripper_right_qpos",
44
+ "trunk_qpos"
45
+ ],
46
+ "sin_cos_embedding_keys": null,
47
+ "mean_std_embedding_keys": null,
48
+ "action_configs": null
49
+ },
50
+ "action": {
51
+ "delta_indices": [
52
+ 0,
53
+ 1,
54
+ 2,
55
+ 3,
56
+ 4,
57
+ 5,
58
+ 6,
59
+ 7,
60
+ 8,
61
+ 9,
62
+ 10,
63
+ 11,
64
+ 12,
65
+ 13,
66
+ 14,
67
+ 15,
68
+ 16,
69
+ 17,
70
+ 18,
71
+ 19,
72
+ 20,
73
+ 21,
74
+ 22,
75
+ 23,
76
+ 24,
77
+ 25,
78
+ 26,
79
+ 27,
80
+ 28,
81
+ 29,
82
+ 30,
83
+ 31
84
+ ],
85
+ "modality_keys": [
86
+ "base",
87
+ "torso",
88
+ "left_arm",
89
+ "left_gripper",
90
+ "right_arm",
91
+ "right_gripper"
92
+ ],
93
+ "sin_cos_embedding_keys": null,
94
+ "mean_std_embedding_keys": null,
95
+ "action_configs": [
96
+ {
97
+ "rep": "ABSOLUTE",
98
+ "type": "NON_EEF",
99
+ "format": "DEFAULT",
100
+ "state_key": null
101
+ },
102
+ {
103
+ "rep": "RELATIVE",
104
+ "type": "NON_EEF",
105
+ "format": "DEFAULT",
106
+ "state_key": "trunk_qpos"
107
+ },
108
+ {
109
+ "rep": "RELATIVE",
110
+ "type": "NON_EEF",
111
+ "format": "DEFAULT",
112
+ "state_key": "arm_left_qpos"
113
+ },
114
+ {
115
+ "rep": "ABSOLUTE",
116
+ "type": "NON_EEF",
117
+ "format": "DEFAULT",
118
+ "state_key": null
119
+ },
120
+ {
121
+ "rep": "RELATIVE",
122
+ "type": "NON_EEF",
123
+ "format": "DEFAULT",
124
+ "state_key": "arm_right_qpos"
125
+ },
126
+ {
127
+ "rep": "ABSOLUTE",
128
+ "type": "NON_EEF",
129
+ "format": "DEFAULT",
130
+ "state_key": null
131
+ }
132
+ ]
133
+ },
134
+ "language": {
135
+ "delta_indices": [
136
+ 0
137
+ ],
138
+ "modality_keys": [
139
+ "annotation.human.coarse_action"
140
+ ],
141
+ "sin_cos_embedding_keys": null,
142
+ "mean_std_embedding_keys": null,
143
+ "action_configs": null
144
+ }
145
+ },
146
+ "gr1": {
147
+ "video": {
148
+ "delta_indices": [
149
+ 0
150
+ ],
151
+ "modality_keys": [
152
+ "ego_view_bg_crop_pad_res256_freq20"
153
+ ],
154
+ "sin_cos_embedding_keys": null,
155
+ "mean_std_embedding_keys": null,
156
+ "action_configs": null
157
+ },
158
+ "state": {
159
+ "delta_indices": [
160
+ 0
161
+ ],
162
+ "modality_keys": [
163
+ "left_arm",
164
+ "right_arm",
165
+ "left_hand",
166
+ "right_hand",
167
+ "waist"
168
+ ],
169
+ "sin_cos_embedding_keys": [
170
+ "left_arm",
171
+ "right_arm",
172
+ "left_hand",
173
+ "right_hand",
174
+ "waist"
175
+ ],
176
+ "mean_std_embedding_keys": null,
177
+ "action_configs": null
178
+ },
179
+ "action": {
180
+ "delta_indices": [
181
+ 0,
182
+ 1,
183
+ 2,
184
+ 3,
185
+ 4,
186
+ 5,
187
+ 6,
188
+ 7,
189
+ 8,
190
+ 9,
191
+ 10,
192
+ 11,
193
+ 12,
194
+ 13,
195
+ 14,
196
+ 15
197
+ ],
198
+ "modality_keys": [
199
+ "left_arm",
200
+ "right_arm",
201
+ "left_hand",
202
+ "right_hand",
203
+ "waist"
204
+ ],
205
+ "sin_cos_embedding_keys": null,
206
+ "mean_std_embedding_keys": null,
207
+ "action_configs": [
208
+ {
209
+ "rep": "RELATIVE",
210
+ "type": "NON_EEF",
211
+ "format": "DEFAULT",
212
+ "state_key": null
213
+ },
214
+ {
215
+ "rep": "RELATIVE",
216
+ "type": "NON_EEF",
217
+ "format": "DEFAULT",
218
+ "state_key": null
219
+ },
220
+ {
221
+ "rep": "RELATIVE",
222
+ "type": "NON_EEF",
223
+ "format": "DEFAULT",
224
+ "state_key": null
225
+ },
226
+ {
227
+ "rep": "RELATIVE",
228
+ "type": "NON_EEF",
229
+ "format": "DEFAULT",
230
+ "state_key": null
231
+ },
232
+ {
233
+ "rep": "ABSOLUTE",
234
+ "type": "NON_EEF",
235
+ "format": "DEFAULT",
236
+ "state_key": null
237
+ }
238
+ ]
239
+ },
240
+ "language": {
241
+ "delta_indices": [
242
+ 0
243
+ ],
244
+ "modality_keys": [
245
+ "task"
246
+ ],
247
+ "sin_cos_embedding_keys": null,
248
+ "mean_std_embedding_keys": null,
249
+ "action_configs": null
250
+ }
251
+ },
252
+ "robocasa_panda_omron": {
253
+ "video": {
254
+ "delta_indices": [
255
+ 0
256
+ ],
257
+ "modality_keys": [
258
+ "res256_image_side_0",
259
+ "res256_image_side_1",
260
+ "res256_image_wrist_0"
261
+ ],
262
+ "sin_cos_embedding_keys": null,
263
+ "mean_std_embedding_keys": null,
264
+ "action_configs": null
265
+ },
266
+ "state": {
267
+ "delta_indices": [
268
+ 0
269
+ ],
270
+ "modality_keys": [
271
+ "end_effector_position_relative",
272
+ "end_effector_rotation_relative",
273
+ "gripper_qpos",
274
+ "base_position",
275
+ "base_rotation"
276
+ ],
277
+ "sin_cos_embedding_keys": null,
278
+ "mean_std_embedding_keys": null,
279
+ "action_configs": null
280
+ },
281
+ "action": {
282
+ "delta_indices": [
283
+ 0,
284
+ 1,
285
+ 2,
286
+ 3,
287
+ 4,
288
+ 5,
289
+ 6,
290
+ 7,
291
+ 8,
292
+ 9,
293
+ 10,
294
+ 11,
295
+ 12,
296
+ 13,
297
+ 14,
298
+ 15
299
+ ],
300
+ "modality_keys": [
301
+ "end_effector_position",
302
+ "end_effector_rotation",
303
+ "gripper_close",
304
+ "base_motion",
305
+ "control_mode"
306
+ ],
307
+ "sin_cos_embedding_keys": null,
308
+ "mean_std_embedding_keys": null,
309
+ "action_configs": [
310
+ {
311
+ "rep": "ABSOLUTE",
312
+ "type": "NON_EEF",
313
+ "format": "DEFAULT",
314
+ "state_key": null
315
+ },
316
+ {
317
+ "rep": "ABSOLUTE",
318
+ "type": "NON_EEF",
319
+ "format": "DEFAULT",
320
+ "state_key": null
321
+ },
322
+ {
323
+ "rep": "ABSOLUTE",
324
+ "type": "NON_EEF",
325
+ "format": "DEFAULT",
326
+ "state_key": null
327
+ },
328
+ {
329
+ "rep": "ABSOLUTE",
330
+ "type": "NON_EEF",
331
+ "format": "DEFAULT",
332
+ "state_key": null
333
+ },
334
+ {
335
+ "rep": "ABSOLUTE",
336
+ "type": "NON_EEF",
337
+ "format": "DEFAULT",
338
+ "state_key": null
339
+ }
340
+ ]
341
+ },
342
+ "language": {
343
+ "delta_indices": [
344
+ 0
345
+ ],
346
+ "modality_keys": [
347
+ "annotation.human.action.task_description"
348
+ ],
349
+ "sin_cos_embedding_keys": null,
350
+ "mean_std_embedding_keys": null,
351
+ "action_configs": null
352
+ }
353
+ },
354
+ "new_embodiment": {
355
+ "video": {
356
+ "delta_indices": [
357
+ 0
358
+ ],
359
+ "modality_keys": [
360
+ "front",
361
+ "wrist"
362
+ ],
363
+ "sin_cos_embedding_keys": null,
364
+ "mean_std_embedding_keys": null,
365
+ "action_configs": null
366
+ },
367
+ "state": {
368
+ "delta_indices": [
369
+ 0
370
+ ],
371
+ "modality_keys": [
372
+ "arm",
373
+ "gripper"
374
+ ],
375
+ "sin_cos_embedding_keys": [
376
+ "arm"
377
+ ],
378
+ "mean_std_embedding_keys": null,
379
+ "action_configs": null
380
+ },
381
+ "action": {
382
+ "delta_indices": [
383
+ 0,
384
+ 1,
385
+ 2,
386
+ 3,
387
+ 4,
388
+ 5,
389
+ 6,
390
+ 7,
391
+ 8,
392
+ 9,
393
+ 10,
394
+ 11,
395
+ 12,
396
+ 13,
397
+ 14,
398
+ 15,
399
+ 16,
400
+ 17,
401
+ 18,
402
+ 19,
403
+ 20,
404
+ 21,
405
+ 22,
406
+ 23,
407
+ 24,
408
+ 25,
409
+ 26,
410
+ 27,
411
+ 28,
412
+ 29
413
+ ],
414
+ "modality_keys": [
415
+ "arm",
416
+ "gripper"
417
+ ],
418
+ "sin_cos_embedding_keys": null,
419
+ "mean_std_embedding_keys": null,
420
+ "action_configs": [
421
+ {
422
+ "rep": "ABSOLUTE",
423
+ "type": "EEF",
424
+ "format": "XYZ_ROTVEC",
425
+ "state_key": null
426
+ },
427
+ {
428
+ "rep": "ABSOLUTE",
429
+ "type": "NON_EEF",
430
+ "format": "DEFAULT",
431
+ "state_key": null
432
+ }
433
+ ]
434
+ },
435
+ "language": {
436
+ "delta_indices": [
437
+ 0
438
+ ],
439
+ "modality_keys": [
440
+ "annotation.task_description"
441
+ ],
442
+ "sin_cos_embedding_keys": null,
443
+ "mean_std_embedding_keys": null,
444
+ "action_configs": null
445
+ }
446
+ }
447
+ },
448
+ "image_crop_size": null,
449
+ "image_target_size": null,
450
+ "use_albumentations": true,
451
+ "random_rotation_angle": null,
452
+ "color_jitter_params": {
453
+ "brightness": 0.3,
454
+ "contrast": 0.4,
455
+ "saturation": 0.5,
456
+ "hue": 0.08
457
+ },
458
+ "shortest_image_edge": 256,
459
+ "crop_fraction": 0.95,
460
+ "model_name": "nvidia/Eagle-Block2A-2B-v2",
461
+ "model_type": "eagle",
462
+ "formalize_language": true,
463
+ "max_state_dim": 128,
464
+ "max_action_dim": 128,
465
+ "max_action_horizon": 50,
466
+ "use_percentiles": false,
467
+ "clip_outliers": true,
468
+ "apply_sincos_state_encoding": true,
469
+ "use_relative_action": true
470
+ }
471
+ }
checkpoint-20000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d4a1de5747f9ffcdd9d4cfa8b1400047a68af64364bebc905e6429c3449b7022
3
+ size 14645
checkpoint-20000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7fa802a80def971b73ec74284a6aa44d0b2ea101bd38ed41a3b1c1a0b4001f00
3
+ size 1465
checkpoint-20000/statistics.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-20000/trainer_state.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-20000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74fc2963f498bcc7aa9fb904bda507a455bd7ba7c1ed3b99b740bdb1946df479
3
+ size 5713
checkpoint-20000/wandb_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"project": "finetune-gr00t-n1d6", "run_id": "finetune"}
config.json ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_horizon": 50,
3
+ "add_pos_embed": true,
4
+ "apply_sincos_state_encoding": true,
5
+ "architectures": [
6
+ "Gr00tN1d6"
7
+ ],
8
+ "attn_dropout": 0.2,
9
+ "attn_implementation": null,
10
+ "backbone_embedding_dim": 2048,
11
+ "backbone_model_type": "eagle",
12
+ "backbone_trainable_params_fp32": true,
13
+ "collator_overwrite_image_inputs": false,
14
+ "color_jitter_params": {
15
+ "brightness": 0.1,
16
+ "contrast": 0.1,
17
+ "hue": 0.1,
18
+ "saturation": 0.1
19
+ },
20
+ "crop_fraction": 0.95,
21
+ "diffusion_model_cfg": {
22
+ "attention_head_dim": 48,
23
+ "dropout": 0.2,
24
+ "final_dropout": true,
25
+ "interleave_self_attention": true,
26
+ "norm_type": "ada_norm",
27
+ "num_attention_heads": 32,
28
+ "num_layers": 32,
29
+ "output_dim": 1024,
30
+ "positional_embeddings": null
31
+ },
32
+ "eagle_collator": true,
33
+ "formalize_language": true,
34
+ "gemma_collator": false,
35
+ "hidden_size": 1024,
36
+ "image_crop_size": null,
37
+ "image_target_size": null,
38
+ "input_embedding_dim": 1536,
39
+ "load_bf16": true,
40
+ "max_action_dim": 128,
41
+ "max_num_embodiments": 32,
42
+ "max_seq_len": 1024,
43
+ "max_state_dim": 128,
44
+ "model_dtype": "bfloat16",
45
+ "model_name": "nvidia/Eagle-Block2A-2B-v2",
46
+ "model_type": "Gr00tN1d6",
47
+ "noise_beta_alpha": 1.5,
48
+ "noise_beta_beta": 1.0,
49
+ "noise_s": 0.999,
50
+ "num_inference_timesteps": 4,
51
+ "num_timestep_buckets": 1000,
52
+ "random_rotation_angle": null,
53
+ "reproject_vision": false,
54
+ "select_layer": 16,
55
+ "shortest_image_edge": 256,
56
+ "state_dropout_prob": 0.0,
57
+ "torch_dtype": "bfloat16",
58
+ "transformers_version": "4.51.3",
59
+ "tune_diffusion_model": true,
60
+ "tune_llm": false,
61
+ "tune_projector": true,
62
+ "tune_top_llm_layers": 4,
63
+ "tune_visual": false,
64
+ "tune_vlln": true,
65
+ "use_albumentations_transforms": true,
66
+ "use_alternate_vl_dit": true,
67
+ "use_flash_attention": true,
68
+ "use_relative_action": true,
69
+ "use_vlln": true
70
+ }
experiment_cfg/conf.yaml ADDED
@@ -0,0 +1,225 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ load_config_path: null
2
+ model:
3
+ model_type: Gr00tN1d6
4
+ model_dtype: bfloat16
5
+ model_name: nvidia/Eagle-Block2A-2B-v2
6
+ backbone_model_type: eagle
7
+ model_revision: null
8
+ tune_top_llm_layers: 4
9
+ backbone_embedding_dim: 2048
10
+ tune_llm: false
11
+ tune_visual: false
12
+ select_layer: 16
13
+ reproject_vision: false
14
+ use_flash_attention: true
15
+ load_bf16: false
16
+ collator_overwrite_image_inputs: false
17
+ eagle_collator: true
18
+ backbone_trainable_params_fp32: true
19
+ image_crop_size: null
20
+ image_target_size: null
21
+ shortest_image_edge: 256
22
+ crop_fraction: 0.95
23
+ random_rotation_angle: null
24
+ color_jitter_params:
25
+ brightness: 0.3
26
+ contrast: 0.4
27
+ saturation: 0.5
28
+ hue: 0.08
29
+ use_albumentations_transforms: true
30
+ extra_augmentation_config: null
31
+ formalize_language: true
32
+ apply_sincos_state_encoding: false
33
+ use_relative_action: true
34
+ max_state_dim: 29
35
+ max_action_dim: 29
36
+ action_horizon: 16
37
+ hidden_size: 1024
38
+ input_embedding_dim: 1536
39
+ add_pos_embed: true
40
+ attn_dropout: 0.2
41
+ use_vlln: true
42
+ max_seq_len: 1024
43
+ use_alternate_vl_dit: true
44
+ attend_text_every_n_blocks: 2
45
+ diffusion_model_cfg:
46
+ positional_embeddings: null
47
+ num_layers: 32
48
+ num_attention_heads: 32
49
+ attention_head_dim: 48
50
+ norm_type: ada_norm
51
+ dropout: 0.2
52
+ final_dropout: true
53
+ output_dim: 1024
54
+ interleave_self_attention: true
55
+ num_inference_timesteps: 4
56
+ noise_beta_alpha: 1.5
57
+ noise_beta_beta: 1.0
58
+ noise_s: 0.999
59
+ num_timestep_buckets: 1000
60
+ tune_projector: true
61
+ tune_diffusion_model: true
62
+ tune_vlln: true
63
+ state_dropout_prob: 0.0
64
+ state_additive_noise_scale: 0.0
65
+ max_num_embodiments: 32
66
+ data:
67
+ datasets:
68
+ - dataset_paths:
69
+ - gubam/dataset
70
+ embodiment_tag: new_embodiment
71
+ mix_ratio: 1.0
72
+ dataset_type: physical_embodiment
73
+ val_dataset_path: null
74
+ modality_configs:
75
+ new_embodiment:
76
+ video:
77
+ delta_indices:
78
+ - 0
79
+ modality_keys:
80
+ - front
81
+ - wrist
82
+ sin_cos_embedding_keys: null
83
+ mean_std_embedding_keys: null
84
+ action_configs: null
85
+ state:
86
+ delta_indices:
87
+ - 0
88
+ modality_keys:
89
+ - arm
90
+ - gripper
91
+ sin_cos_embedding_keys:
92
+ - arm
93
+ mean_std_embedding_keys: null
94
+ action_configs: null
95
+ action:
96
+ delta_indices:
97
+ - 0
98
+ - 1
99
+ - 2
100
+ - 3
101
+ - 4
102
+ - 5
103
+ - 6
104
+ - 7
105
+ - 8
106
+ - 9
107
+ - 10
108
+ - 11
109
+ - 12
110
+ - 13
111
+ - 14
112
+ - 15
113
+ - 16
114
+ - 17
115
+ - 18
116
+ - 19
117
+ - 20
118
+ - 21
119
+ - 22
120
+ - 23
121
+ - 24
122
+ - 25
123
+ - 26
124
+ - 27
125
+ - 28
126
+ - 29
127
+ modality_keys:
128
+ - arm
129
+ - gripper
130
+ sin_cos_embedding_keys: null
131
+ mean_std_embedding_keys: null
132
+ action_configs:
133
+ - rep: ABSOLUTE
134
+ type: EEF
135
+ format: XYZ_ROTVEC
136
+ state_key: null
137
+ - rep: ABSOLUTE
138
+ type: NON_EEF
139
+ format: DEFAULT
140
+ state_key: null
141
+ language:
142
+ delta_indices:
143
+ - 0
144
+ modality_keys:
145
+ - annotation.task_description
146
+ sin_cos_embedding_keys: null
147
+ mean_std_embedding_keys: null
148
+ action_configs: null
149
+ download_cache: false
150
+ shard_size: 1024
151
+ episode_sampling_rate: 0.1
152
+ num_shards_per_epoch: 100000
153
+ override_pretraining_statistics: false
154
+ mode: single_turn
155
+ random_chop: 0.0
156
+ mock_dataset_mode: false
157
+ shuffle: true
158
+ seed: 42
159
+ multiprocessing_context: fork
160
+ allow_padding: false
161
+ subsample_ratio: 1.0
162
+ image_crop_size:
163
+ - 244
164
+ - 244
165
+ image_target_size:
166
+ - 224
167
+ - 224
168
+ video_backend: torchcodec
169
+ training:
170
+ output_dir: /tmp/finetune
171
+ experiment_name: null
172
+ max_steps: 20000
173
+ global_batch_size: 32
174
+ batch_size: null
175
+ gradient_accumulation_steps: 1
176
+ learning_rate: 0.0001
177
+ lr_scheduler_type: cosine
178
+ weight_decay: 1.0e-05
179
+ warmup_ratio: 0.05
180
+ warmup_steps: 0
181
+ max_grad_norm: 1.0
182
+ optim: adamw_torch
183
+ start_from_checkpoint: nvidia/GR00T-N1.6-3B
184
+ tf32: true
185
+ fp16: false
186
+ bf16: true
187
+ eval_bf16: true
188
+ logging_steps: 10
189
+ save_steps: 10000
190
+ save_total_limit: 5
191
+ save_vl_model: false
192
+ upload_checkpoints: false
193
+ upload_every: 1000
194
+ upload_last_n_checkpoints: 5
195
+ max_concurrent_uploads: 2
196
+ eval_strategy: 'no'
197
+ eval_steps: 500
198
+ eval_set_split_ratio: 0.1
199
+ eval_batch_size: 2
200
+ save_best_eval_metric_name: ''
201
+ save_best_eval_metric_greater_is_better: true
202
+ deepspeed_stage: 2
203
+ gradient_checkpointing: false
204
+ transformers_trust_remote_code: true
205
+ transformers_local_files_only: false
206
+ transformers_cache_dir: null
207
+ transformers_access_token: null
208
+ use_ddp: false
209
+ ddp_bucket_cap_mb: 100
210
+ num_gpus: 1
211
+ dataloader_num_workers: 4
212
+ remove_unused_columns: false
213
+ use_wandb: true
214
+ wandb_project: finetune-gr00t-n1d6
215
+ enable_profiling: false
216
+ max_retries: 3
217
+ assert_loss_less_than: null
218
+ add_rl_callback: false
219
+ enable_open_loop_eval: false
220
+ open_loop_eval_traj_ids:
221
+ - 0
222
+ open_loop_eval_steps_per_traj: 100
223
+ open_loop_eval_plot_indices: null
224
+ max_steps: 20000
225
+ save_steps: 10000
experiment_cfg/config.yaml ADDED
@@ -0,0 +1,260 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ !!python/object:gr00t.configs.base_config.Config
2
+ data: !!python/object:gr00t.configs.data.data_config.DataConfig
3
+ allow_padding: false
4
+ datasets:
5
+ - !!python/object:gr00t.configs.data.data_config.SingleDatasetConfig
6
+ dataset_paths:
7
+ - gubam/dataset
8
+ dataset_type: physical_embodiment
9
+ embodiment_tag: new_embodiment
10
+ mix_ratio: 1.0
11
+ val_dataset_path: null
12
+ download_cache: false
13
+ episode_sampling_rate: 0.1
14
+ image_crop_size:
15
+ - 244
16
+ - 244
17
+ image_target_size:
18
+ - 224
19
+ - 224
20
+ mock_dataset_mode: false
21
+ modality_configs:
22
+ new_embodiment:
23
+ action: !!python/object:gr00t.data.types.ModalityConfig
24
+ action_configs:
25
+ - !!python/object:gr00t.data.types.ActionConfig
26
+ format: !!python/object/apply:gr00t.data.types.ActionFormat
27
+ - xyz+rotvec
28
+ rep: &id001 !!python/object/apply:gr00t.data.types.ActionRepresentation
29
+ - absolute
30
+ state_key: null
31
+ type: !!python/object/apply:gr00t.data.types.ActionType
32
+ - eef
33
+ - !!python/object:gr00t.data.types.ActionConfig
34
+ format: !!python/object/apply:gr00t.data.types.ActionFormat
35
+ - default
36
+ rep: *id001
37
+ state_key: null
38
+ type: !!python/object/apply:gr00t.data.types.ActionType
39
+ - non_eef
40
+ delta_indices:
41
+ - 0
42
+ - 1
43
+ - 2
44
+ - 3
45
+ - 4
46
+ - 5
47
+ - 6
48
+ - 7
49
+ - 8
50
+ - 9
51
+ - 10
52
+ - 11
53
+ - 12
54
+ - 13
55
+ - 14
56
+ - 15
57
+ - 16
58
+ - 17
59
+ - 18
60
+ - 19
61
+ - 20
62
+ - 21
63
+ - 22
64
+ - 23
65
+ - 24
66
+ - 25
67
+ - 26
68
+ - 27
69
+ - 28
70
+ - 29
71
+ mean_std_embedding_keys: null
72
+ modality_keys:
73
+ - arm
74
+ - gripper
75
+ sin_cos_embedding_keys: null
76
+ language: !!python/object:gr00t.data.types.ModalityConfig
77
+ action_configs: null
78
+ delta_indices:
79
+ - 0
80
+ mean_std_embedding_keys: null
81
+ modality_keys:
82
+ - annotation.task_description
83
+ sin_cos_embedding_keys: null
84
+ state: !!python/object:gr00t.data.types.ModalityConfig
85
+ action_configs: null
86
+ delta_indices:
87
+ - 0
88
+ mean_std_embedding_keys: null
89
+ modality_keys:
90
+ - arm
91
+ - gripper
92
+ sin_cos_embedding_keys:
93
+ - arm
94
+ video: !!python/object:gr00t.data.types.ModalityConfig
95
+ action_configs: null
96
+ delta_indices:
97
+ - 0
98
+ mean_std_embedding_keys: null
99
+ modality_keys:
100
+ - front
101
+ - wrist
102
+ sin_cos_embedding_keys: null
103
+ mode: single_turn
104
+ multiprocessing_context: fork
105
+ num_shards_per_epoch: 100000
106
+ override_pretraining_statistics: false
107
+ random_chop: 0.0
108
+ seed: 42
109
+ shard_size: 1024
110
+ shuffle: true
111
+ subsample_ratio: 1.0
112
+ video_backend: torchcodec
113
+ load_config_path: null
114
+ model: !!python/object:gr00t.configs.model.gr00t_n1d6.Gr00tN1d6Config
115
+ _attn_implementation_autoset: false
116
+ _attn_implementation_internal: null
117
+ _commit_hash: null
118
+ _name_or_path: ''
119
+ add_cross_attention: false
120
+ architectures: null
121
+ backbone_model_type: eagle
122
+ backbone_trainable_params_fp32: true
123
+ bad_words_ids: null
124
+ begin_suppress_tokens: null
125
+ bos_token_id: null
126
+ chunk_size_feed_forward: 0
127
+ color_jitter_params:
128
+ brightness: 0.3
129
+ contrast: 0.4
130
+ hue: 0.08
131
+ saturation: 0.5
132
+ cross_attention_hidden_size: null
133
+ decoder_start_token_id: null
134
+ diffusion_model_cfg:
135
+ attention_head_dim: 48
136
+ dropout: 0.2
137
+ final_dropout: true
138
+ interleave_self_attention: true
139
+ norm_type: ada_norm
140
+ num_attention_heads: 32
141
+ num_layers: 32
142
+ output_dim: 1024
143
+ positional_embeddings: null
144
+ diversity_penalty: 0.0
145
+ do_sample: false
146
+ eagle_collator: true
147
+ early_stopping: false
148
+ encoder_no_repeat_ngram_size: 0
149
+ eos_token_id: null
150
+ exponential_decay_length_penalty: null
151
+ extra_augmentation_config: null
152
+ finetuning_task: null
153
+ forced_bos_token_id: null
154
+ forced_eos_token_id: null
155
+ id2label:
156
+ 0: LABEL_0
157
+ 1: LABEL_1
158
+ is_decoder: false
159
+ is_encoder_decoder: false
160
+ label2id:
161
+ LABEL_0: 0
162
+ LABEL_1: 1
163
+ length_penalty: 1.0
164
+ load_bf16: false
165
+ max_length: 20
166
+ min_length: 0
167
+ model_name: nvidia/Eagle-Block2A-2B-v2
168
+ no_repeat_ngram_size: 0
169
+ num_beam_groups: 1
170
+ num_beams: 1
171
+ num_return_sequences: 1
172
+ output_attentions: false
173
+ output_hidden_states: false
174
+ output_scores: false
175
+ pad_token_id: null
176
+ prefix: null
177
+ problem_type: null
178
+ pruned_heads: {}
179
+ random_rotation_angle: null
180
+ remove_invalid_values: false
181
+ repetition_penalty: 1.0
182
+ reproject_vision: false
183
+ return_dict: true
184
+ return_dict_in_generate: false
185
+ sep_token_id: null
186
+ state_dropout_prob: 0.0
187
+ suppress_tokens: null
188
+ task_specific_params: null
189
+ temperature: 1.0
190
+ tf_legacy_loss: false
191
+ tie_encoder_decoder: false
192
+ tie_word_embeddings: true
193
+ tokenizer_class: null
194
+ top_k: 50
195
+ top_p: 1.0
196
+ torch_dtype: null
197
+ torchscript: false
198
+ transformers_version: null
199
+ tune_diffusion_model: true
200
+ tune_llm: false
201
+ tune_projector: true
202
+ tune_visual: false
203
+ typical_p: 1.0
204
+ use_bfloat16: false
205
+ use_relative_action: true
206
+ training: !!python/object:gr00t.configs.training.training_config.TrainingConfig
207
+ add_rl_callback: false
208
+ assert_loss_less_than: null
209
+ batch_size: null
210
+ bf16: true
211
+ dataloader_num_workers: 4
212
+ ddp_bucket_cap_mb: 100
213
+ deepspeed_stage: 2
214
+ enable_open_loop_eval: false
215
+ enable_profiling: false
216
+ eval_batch_size: 2
217
+ eval_bf16: true
218
+ eval_set_split_ratio: 0.1
219
+ eval_steps: 500
220
+ eval_strategy: 'no'
221
+ experiment_name: null
222
+ fp16: false
223
+ global_batch_size: 32
224
+ gradient_accumulation_steps: 1
225
+ gradient_checkpointing: false
226
+ learning_rate: 0.0001
227
+ logging_steps: 10
228
+ lr_scheduler_type: cosine
229
+ max_concurrent_uploads: 2
230
+ max_grad_norm: 1.0
231
+ max_retries: 3
232
+ max_steps: 20000
233
+ num_gpus: 1
234
+ open_loop_eval_plot_indices: null
235
+ open_loop_eval_steps_per_traj: 100
236
+ open_loop_eval_traj_ids:
237
+ - 0
238
+ optim: adamw_torch
239
+ output_dir: /tmp/finetune
240
+ remove_unused_columns: false
241
+ save_best_eval_metric_greater_is_better: true
242
+ save_best_eval_metric_name: ''
243
+ save_steps: 10000
244
+ save_total_limit: 5
245
+ save_vl_model: false
246
+ start_from_checkpoint: nvidia/GR00T-N1.6-3B
247
+ tf32: true
248
+ transformers_access_token: null
249
+ transformers_cache_dir: null
250
+ transformers_local_files_only: false
251
+ transformers_trust_remote_code: true
252
+ upload_checkpoints: false
253
+ upload_every: 1000
254
+ upload_last_n_checkpoints: 5
255
+ use_ddp: false
256
+ use_wandb: true
257
+ wandb_project: finetune-gr00t-n1d6
258
+ warmup_ratio: 0.05
259
+ warmup_steps: 0
260
+ weight_decay: 1.0e-05
experiment_cfg/dataset_statistics.json ADDED
@@ -0,0 +1,149 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "new_embodiment": {
3
+ "state": {
4
+ "arm": {
5
+ "min": [
6
+ -0.30735278129577637,
7
+ -0.5557547807693481,
8
+ 0.9460603594779968,
9
+ -3.0296072959899902,
10
+ -3.136127471923828,
11
+ -0.8582668304443359
12
+ ],
13
+ "max": [
14
+ 0.4784075915813446,
15
+ 0.536034107208252,
16
+ 1.7277131080627441,
17
+ 2.922990322113037,
18
+ 3.1376261711120605,
19
+ 0.42977502942085266
20
+ ],
21
+ "mean": [
22
+ 0.13191731012878258,
23
+ 0.031970066748100066,
24
+ 1.1558844479773447,
25
+ -1.6954683609230488,
26
+ 1.6487134783988946,
27
+ -0.14286577845817536
28
+ ],
29
+ "std": [
30
+ 0.15234599091258288,
31
+ 0.25073243709252585,
32
+ 0.19292941514856915,
33
+ 1.1962757356957574,
34
+ 1.2519753680003787,
35
+ 0.2175599434355554
36
+ ],
37
+ "q01": [
38
+ -0.04688343169734043,
39
+ -0.2370660196788625,
40
+ 0.9693865812794936,
41
+ -2.460839541332334,
42
+ -0.07001938726657715,
43
+ -0.6103960370493784
44
+ ],
45
+ "q99": [
46
+ 0.2768438247195932,
47
+ 0.267643256964274,
48
+ 1.5990514549460293,
49
+ 0.01546860347780651,
50
+ 2.483356319667699,
51
+ 0.11741250933956077
52
+ ]
53
+ },
54
+ "gripper": {
55
+ "min": [
56
+ -0.001212759525515139
57
+ ],
58
+ "max": [
59
+ 0.7762507200241089
60
+ ],
61
+ "mean": [
62
+ 0.1826333326178475
63
+ ],
64
+ "std": [
65
+ 0.20001624021635997
66
+ ],
67
+ "q01": [
68
+ 0.021986669201693607
69
+ ],
70
+ "q99": [
71
+ 0.45503166863409855
72
+ ]
73
+ }
74
+ },
75
+ "action": {
76
+ "arm": {
77
+ "min": [
78
+ -0.3141818344593048,
79
+ -0.6150317192077637,
80
+ 0.9386419653892517,
81
+ -3.101818323135376,
82
+ -3.1366591453552246,
83
+ -0.9746662378311157
84
+ ],
85
+ "max": [
86
+ 0.48730289936065674,
87
+ 0.590604305267334,
88
+ 1.7818619012832642,
89
+ 3.1010582447052,
90
+ 3.128772258758545,
91
+ 0.48776116967201233
92
+ ],
93
+ "mean": [
94
+ 0.11818711284420536,
95
+ 0.028762292790606554,
96
+ 1.1608746673321966,
97
+ -1.6448431900842613,
98
+ 1.6027947065112016,
99
+ -0.15912706390756018
100
+ ],
101
+ "std": [
102
+ 0.16134523907378098,
103
+ 0.25919821992288916,
104
+ 0.20997126294727028,
105
+ 1.2448266960019687,
106
+ 1.2914855158630338,
107
+ 0.2432569420599352
108
+ ],
109
+ "q01": [
110
+ -0.07862773065169046,
111
+ -0.26523173844012143,
112
+ 0.963678525591205,
113
+ -2.513543702118574,
114
+ -0.4443971805927962,
115
+ -0.6861969386370356
116
+ ],
117
+ "q99": [
118
+ 0.28730840579975764,
119
+ 0.2975560395103185,
120
+ 1.637628068438941,
121
+ 0.489103782574516,
122
+ 2.526707288272369,
123
+ 0.15085034587810217
124
+ ]
125
+ },
126
+ "gripper": {
127
+ "min": [
128
+ -0.8607733845710754
129
+ ],
130
+ "max": [
131
+ 0.13195456564426422
132
+ ],
133
+ "mean": [
134
+ -0.332952961079147
135
+ ],
136
+ "std": [
137
+ 0.3676414530724673
138
+ ],
139
+ "q01": [
140
+ -0.848867505241717
141
+ ],
142
+ "q99": [
143
+ -0.010392814656977918
144
+ ]
145
+ }
146
+ },
147
+ "relative_action": {}
148
+ }
149
+ }
experiment_cfg/final_model_config.json ADDED
@@ -0,0 +1,54 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "Gr00tN1d6",
3
+ "model_dtype": "bfloat16",
4
+ "model_name": "nvidia/Eagle-Block2A-2B-v2",
5
+ "backbone_model_type": "eagle",
6
+ "model_revision": null,
7
+ "tune_top_llm_layers": 4,
8
+ "backbone_embedding_dim": 2048,
9
+ "tune_llm": false,
10
+ "tune_visual": false,
11
+ "select_layer": 16,
12
+ "reproject_vision": false,
13
+ "use_flash_attention": true,
14
+ "load_bf16": true,
15
+ "collator_overwrite_image_inputs": false,
16
+ "eagle_collator": true,
17
+ "backbone_trainable_params_fp32": true,
18
+ "extra_augmentation_config": null,
19
+ "apply_sincos_state_encoding": true,
20
+ "use_relative_action": true,
21
+ "max_state_dim": 128,
22
+ "max_action_dim": 128,
23
+ "action_horizon": 50,
24
+ "hidden_size": 1024,
25
+ "input_embedding_dim": 1536,
26
+ "add_pos_embed": true,
27
+ "attn_dropout": 0.2,
28
+ "use_vlln": true,
29
+ "max_seq_len": 1024,
30
+ "use_alternate_vl_dit": true,
31
+ "attend_text_every_n_blocks": 2,
32
+ "diffusion_model_cfg": {
33
+ "attention_head_dim": 48,
34
+ "dropout": 0.2,
35
+ "final_dropout": true,
36
+ "interleave_self_attention": true,
37
+ "norm_type": "ada_norm",
38
+ "num_attention_heads": 32,
39
+ "num_layers": 32,
40
+ "output_dim": 1024,
41
+ "positional_embeddings": null
42
+ },
43
+ "num_inference_timesteps": 4,
44
+ "noise_beta_alpha": 1.5,
45
+ "noise_beta_beta": 1.0,
46
+ "noise_s": 0.999,
47
+ "num_timestep_buckets": 1000,
48
+ "tune_projector": true,
49
+ "tune_diffusion_model": true,
50
+ "tune_vlln": true,
51
+ "state_dropout_prob": 0.0,
52
+ "state_additive_noise_scale": 0.0,
53
+ "max_num_embodiments": 32
54
+ }
experiment_cfg/final_processor_config.json ADDED
The diff for this file is too large to render. See raw diff
 
model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8736b74d559b4e11e3ad82ebc06897e33f828899b64d1a30d122b430e31dc69c
3
+ size 4990120184
model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:70ff15e1834de78e43332688c1220e0347de51af9b8850542f93025af919b6b5
3
+ size 4823190320
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
processor/embodiment_id.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "robocasa_panda_omron": 13,
3
+ "gr1": 20,
4
+ "behavior_r1_pro": 24,
5
+ "unitree_g1": 8,
6
+ "oxe_google": 0,
7
+ "oxe_widowx": 1,
8
+ "libero_panda": 2,
9
+ "oxe_droid": 16,
10
+ "new_embodiment": 10
11
+ }
processor/processor_config.json ADDED
@@ -0,0 +1,471 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "processor_class": "Gr00tN1d6Processor",
3
+ "processor_kwargs": {
4
+ "modality_configs": {
5
+ "behavior_r1_pro": {
6
+ "video": {
7
+ "delta_indices": [
8
+ 0
9
+ ],
10
+ "modality_keys": [
11
+ "observation.images.rgb.head_256_256",
12
+ "observation.images.rgb.left_wrist_256_256",
13
+ "observation.images.rgb.right_wrist_256_256"
14
+ ],
15
+ "sin_cos_embedding_keys": null,
16
+ "mean_std_embedding_keys": null,
17
+ "action_configs": null
18
+ },
19
+ "state": {
20
+ "delta_indices": [
21
+ 0
22
+ ],
23
+ "modality_keys": [
24
+ "robot_pos",
25
+ "robot_ori_cos",
26
+ "robot_ori_sin",
27
+ "robot_2d_ori",
28
+ "robot_2d_ori_cos",
29
+ "robot_2d_ori_sin",
30
+ "robot_lin_vel",
31
+ "robot_ang_vel",
32
+ "arm_left_qpos",
33
+ "arm_left_qpos_sin",
34
+ "arm_left_qpos_cos",
35
+ "eef_left_pos",
36
+ "eef_left_quat",
37
+ "gripper_left_qpos",
38
+ "arm_right_qpos",
39
+ "arm_right_qpos_sin",
40
+ "arm_right_qpos_cos",
41
+ "eef_right_pos",
42
+ "eef_right_quat",
43
+ "gripper_right_qpos",
44
+ "trunk_qpos"
45
+ ],
46
+ "sin_cos_embedding_keys": null,
47
+ "mean_std_embedding_keys": null,
48
+ "action_configs": null
49
+ },
50
+ "action": {
51
+ "delta_indices": [
52
+ 0,
53
+ 1,
54
+ 2,
55
+ 3,
56
+ 4,
57
+ 5,
58
+ 6,
59
+ 7,
60
+ 8,
61
+ 9,
62
+ 10,
63
+ 11,
64
+ 12,
65
+ 13,
66
+ 14,
67
+ 15,
68
+ 16,
69
+ 17,
70
+ 18,
71
+ 19,
72
+ 20,
73
+ 21,
74
+ 22,
75
+ 23,
76
+ 24,
77
+ 25,
78
+ 26,
79
+ 27,
80
+ 28,
81
+ 29,
82
+ 30,
83
+ 31
84
+ ],
85
+ "modality_keys": [
86
+ "base",
87
+ "torso",
88
+ "left_arm",
89
+ "left_gripper",
90
+ "right_arm",
91
+ "right_gripper"
92
+ ],
93
+ "sin_cos_embedding_keys": null,
94
+ "mean_std_embedding_keys": null,
95
+ "action_configs": [
96
+ {
97
+ "rep": "ABSOLUTE",
98
+ "type": "NON_EEF",
99
+ "format": "DEFAULT",
100
+ "state_key": null
101
+ },
102
+ {
103
+ "rep": "RELATIVE",
104
+ "type": "NON_EEF",
105
+ "format": "DEFAULT",
106
+ "state_key": "trunk_qpos"
107
+ },
108
+ {
109
+ "rep": "RELATIVE",
110
+ "type": "NON_EEF",
111
+ "format": "DEFAULT",
112
+ "state_key": "arm_left_qpos"
113
+ },
114
+ {
115
+ "rep": "ABSOLUTE",
116
+ "type": "NON_EEF",
117
+ "format": "DEFAULT",
118
+ "state_key": null
119
+ },
120
+ {
121
+ "rep": "RELATIVE",
122
+ "type": "NON_EEF",
123
+ "format": "DEFAULT",
124
+ "state_key": "arm_right_qpos"
125
+ },
126
+ {
127
+ "rep": "ABSOLUTE",
128
+ "type": "NON_EEF",
129
+ "format": "DEFAULT",
130
+ "state_key": null
131
+ }
132
+ ]
133
+ },
134
+ "language": {
135
+ "delta_indices": [
136
+ 0
137
+ ],
138
+ "modality_keys": [
139
+ "annotation.human.coarse_action"
140
+ ],
141
+ "sin_cos_embedding_keys": null,
142
+ "mean_std_embedding_keys": null,
143
+ "action_configs": null
144
+ }
145
+ },
146
+ "gr1": {
147
+ "video": {
148
+ "delta_indices": [
149
+ 0
150
+ ],
151
+ "modality_keys": [
152
+ "ego_view_bg_crop_pad_res256_freq20"
153
+ ],
154
+ "sin_cos_embedding_keys": null,
155
+ "mean_std_embedding_keys": null,
156
+ "action_configs": null
157
+ },
158
+ "state": {
159
+ "delta_indices": [
160
+ 0
161
+ ],
162
+ "modality_keys": [
163
+ "left_arm",
164
+ "right_arm",
165
+ "left_hand",
166
+ "right_hand",
167
+ "waist"
168
+ ],
169
+ "sin_cos_embedding_keys": [
170
+ "left_arm",
171
+ "right_arm",
172
+ "left_hand",
173
+ "right_hand",
174
+ "waist"
175
+ ],
176
+ "mean_std_embedding_keys": null,
177
+ "action_configs": null
178
+ },
179
+ "action": {
180
+ "delta_indices": [
181
+ 0,
182
+ 1,
183
+ 2,
184
+ 3,
185
+ 4,
186
+ 5,
187
+ 6,
188
+ 7,
189
+ 8,
190
+ 9,
191
+ 10,
192
+ 11,
193
+ 12,
194
+ 13,
195
+ 14,
196
+ 15
197
+ ],
198
+ "modality_keys": [
199
+ "left_arm",
200
+ "right_arm",
201
+ "left_hand",
202
+ "right_hand",
203
+ "waist"
204
+ ],
205
+ "sin_cos_embedding_keys": null,
206
+ "mean_std_embedding_keys": null,
207
+ "action_configs": [
208
+ {
209
+ "rep": "RELATIVE",
210
+ "type": "NON_EEF",
211
+ "format": "DEFAULT",
212
+ "state_key": null
213
+ },
214
+ {
215
+ "rep": "RELATIVE",
216
+ "type": "NON_EEF",
217
+ "format": "DEFAULT",
218
+ "state_key": null
219
+ },
220
+ {
221
+ "rep": "RELATIVE",
222
+ "type": "NON_EEF",
223
+ "format": "DEFAULT",
224
+ "state_key": null
225
+ },
226
+ {
227
+ "rep": "RELATIVE",
228
+ "type": "NON_EEF",
229
+ "format": "DEFAULT",
230
+ "state_key": null
231
+ },
232
+ {
233
+ "rep": "ABSOLUTE",
234
+ "type": "NON_EEF",
235
+ "format": "DEFAULT",
236
+ "state_key": null
237
+ }
238
+ ]
239
+ },
240
+ "language": {
241
+ "delta_indices": [
242
+ 0
243
+ ],
244
+ "modality_keys": [
245
+ "task"
246
+ ],
247
+ "sin_cos_embedding_keys": null,
248
+ "mean_std_embedding_keys": null,
249
+ "action_configs": null
250
+ }
251
+ },
252
+ "robocasa_panda_omron": {
253
+ "video": {
254
+ "delta_indices": [
255
+ 0
256
+ ],
257
+ "modality_keys": [
258
+ "res256_image_side_0",
259
+ "res256_image_side_1",
260
+ "res256_image_wrist_0"
261
+ ],
262
+ "sin_cos_embedding_keys": null,
263
+ "mean_std_embedding_keys": null,
264
+ "action_configs": null
265
+ },
266
+ "state": {
267
+ "delta_indices": [
268
+ 0
269
+ ],
270
+ "modality_keys": [
271
+ "end_effector_position_relative",
272
+ "end_effector_rotation_relative",
273
+ "gripper_qpos",
274
+ "base_position",
275
+ "base_rotation"
276
+ ],
277
+ "sin_cos_embedding_keys": null,
278
+ "mean_std_embedding_keys": null,
279
+ "action_configs": null
280
+ },
281
+ "action": {
282
+ "delta_indices": [
283
+ 0,
284
+ 1,
285
+ 2,
286
+ 3,
287
+ 4,
288
+ 5,
289
+ 6,
290
+ 7,
291
+ 8,
292
+ 9,
293
+ 10,
294
+ 11,
295
+ 12,
296
+ 13,
297
+ 14,
298
+ 15
299
+ ],
300
+ "modality_keys": [
301
+ "end_effector_position",
302
+ "end_effector_rotation",
303
+ "gripper_close",
304
+ "base_motion",
305
+ "control_mode"
306
+ ],
307
+ "sin_cos_embedding_keys": null,
308
+ "mean_std_embedding_keys": null,
309
+ "action_configs": [
310
+ {
311
+ "rep": "ABSOLUTE",
312
+ "type": "NON_EEF",
313
+ "format": "DEFAULT",
314
+ "state_key": null
315
+ },
316
+ {
317
+ "rep": "ABSOLUTE",
318
+ "type": "NON_EEF",
319
+ "format": "DEFAULT",
320
+ "state_key": null
321
+ },
322
+ {
323
+ "rep": "ABSOLUTE",
324
+ "type": "NON_EEF",
325
+ "format": "DEFAULT",
326
+ "state_key": null
327
+ },
328
+ {
329
+ "rep": "ABSOLUTE",
330
+ "type": "NON_EEF",
331
+ "format": "DEFAULT",
332
+ "state_key": null
333
+ },
334
+ {
335
+ "rep": "ABSOLUTE",
336
+ "type": "NON_EEF",
337
+ "format": "DEFAULT",
338
+ "state_key": null
339
+ }
340
+ ]
341
+ },
342
+ "language": {
343
+ "delta_indices": [
344
+ 0
345
+ ],
346
+ "modality_keys": [
347
+ "annotation.human.action.task_description"
348
+ ],
349
+ "sin_cos_embedding_keys": null,
350
+ "mean_std_embedding_keys": null,
351
+ "action_configs": null
352
+ }
353
+ },
354
+ "new_embodiment": {
355
+ "video": {
356
+ "delta_indices": [
357
+ 0
358
+ ],
359
+ "modality_keys": [
360
+ "front",
361
+ "wrist"
362
+ ],
363
+ "sin_cos_embedding_keys": null,
364
+ "mean_std_embedding_keys": null,
365
+ "action_configs": null
366
+ },
367
+ "state": {
368
+ "delta_indices": [
369
+ 0
370
+ ],
371
+ "modality_keys": [
372
+ "arm",
373
+ "gripper"
374
+ ],
375
+ "sin_cos_embedding_keys": [
376
+ "arm"
377
+ ],
378
+ "mean_std_embedding_keys": null,
379
+ "action_configs": null
380
+ },
381
+ "action": {
382
+ "delta_indices": [
383
+ 0,
384
+ 1,
385
+ 2,
386
+ 3,
387
+ 4,
388
+ 5,
389
+ 6,
390
+ 7,
391
+ 8,
392
+ 9,
393
+ 10,
394
+ 11,
395
+ 12,
396
+ 13,
397
+ 14,
398
+ 15,
399
+ 16,
400
+ 17,
401
+ 18,
402
+ 19,
403
+ 20,
404
+ 21,
405
+ 22,
406
+ 23,
407
+ 24,
408
+ 25,
409
+ 26,
410
+ 27,
411
+ 28,
412
+ 29
413
+ ],
414
+ "modality_keys": [
415
+ "arm",
416
+ "gripper"
417
+ ],
418
+ "sin_cos_embedding_keys": null,
419
+ "mean_std_embedding_keys": null,
420
+ "action_configs": [
421
+ {
422
+ "rep": "ABSOLUTE",
423
+ "type": "EEF",
424
+ "format": "XYZ_ROTVEC",
425
+ "state_key": null
426
+ },
427
+ {
428
+ "rep": "ABSOLUTE",
429
+ "type": "NON_EEF",
430
+ "format": "DEFAULT",
431
+ "state_key": null
432
+ }
433
+ ]
434
+ },
435
+ "language": {
436
+ "delta_indices": [
437
+ 0
438
+ ],
439
+ "modality_keys": [
440
+ "annotation.task_description"
441
+ ],
442
+ "sin_cos_embedding_keys": null,
443
+ "mean_std_embedding_keys": null,
444
+ "action_configs": null
445
+ }
446
+ }
447
+ },
448
+ "image_crop_size": null,
449
+ "image_target_size": null,
450
+ "use_albumentations": true,
451
+ "random_rotation_angle": null,
452
+ "color_jitter_params": {
453
+ "brightness": 0.3,
454
+ "contrast": 0.4,
455
+ "saturation": 0.5,
456
+ "hue": 0.08
457
+ },
458
+ "shortest_image_edge": 256,
459
+ "crop_fraction": 0.95,
460
+ "model_name": "nvidia/Eagle-Block2A-2B-v2",
461
+ "model_type": "eagle",
462
+ "formalize_language": true,
463
+ "max_state_dim": 128,
464
+ "max_action_dim": 128,
465
+ "max_action_horizon": 50,
466
+ "use_percentiles": false,
467
+ "clip_outliers": true,
468
+ "apply_sincos_state_encoding": true,
469
+ "use_relative_action": true
470
+ }
471
+ }
processor/statistics.json ADDED
The diff for this file is too large to render. See raw diff
 
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74fc2963f498bcc7aa9fb904bda507a455bd7ba7c1ed3b99b740bdb1946df479
3
+ size 5713
wandb_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"project": "finetune-gr00t-n1d6", "run_id": "finetune"}