Hugging Face's logo

theconstruct-ai
/
sort_package_real_2000

Safetensors
Gr00tN1d7
Model card Files
xet
 Community
theconstruct-ai commited on
Commit
659ca65
·
verified ·
1 Parent(s): 668bbb6

Upload folder using huggingface_hub

Browse files
Files changed (23) hide show
  1. config.json +88 -0
  2. embodiment_id.json +58 -0
  3. experiment_cfg/conf.yaml +251 -0
  4. experiment_cfg/config.yaml +277 -0
  5. experiment_cfg/dataset_statistics.json +907 -0
  6. experiment_cfg/final_model_config.json +55 -0
  7. experiment_cfg/final_processor_config.json +0 -0
  8. global_step2000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt +3 -0
  9. global_step2000/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt +3 -0
  10. global_step2000/mp_rank_00_model_states.pt +3 -0
  11. latest +1 -0
  12. model-00001-of-00002.safetensors +3 -0
  13. model-00002-of-00002.safetensors +3 -0
  14. model.safetensors.index.json +0 -0
  15. processor_config.json +1159 -0
  16. rng_state_0.pth +3 -0
  17. rng_state_1.pth +3 -0
  18. scheduler.pt +3 -0
  19. statistics.json +0 -0
  20. trainer_state.json +1234 -0
  21. training_args.bin +3 -0
  22. wandb_config.json +1 -0
  23. zero_to_fp32.py +760 -0
config.json ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "action_horizon": 40,
3
+ "add_pos_embed": true,
4
+ "apply_sincos_state_encoding": false,
5
+ "architectures": [
6
+ "Gr00tN1d7"
7
+ ],
8
+ "attn_dropout": 0.2,
9
+ "attn_implementation": null,
10
+ "backbone_embedding_dim": 2048,
11
+ "backbone_trainable_params_fp32": true,
12
+ "color_jitter_params": {
13
+ "brightness": 0.3,
14
+ "contrast": 0.4,
15
+ "hue": 0.08,
16
+ "saturation": 0.5
17
+ },
18
+ "crop_fraction": 0.95,
19
+ "diffusion_model_cfg": {
20
+ "attention_head_dim": 48,
21
+ "dropout": 0.2,
22
+ "final_dropout": true,
23
+ "interleave_self_attention": true,
24
+ "norm_type": "ada_norm",
25
+ "num_attention_heads": 32,
26
+ "num_layers": 32,
27
+ "output_dim": 1024,
28
+ "positional_embeddings": null
29
+ },
30
+ "dtype": "bfloat16",
31
+ "exclude_state": false,
32
+ "formalize_language": true,
33
+ "hidden_size": 1024,
34
+ "image_crop_size": [
35
+ 230,
36
+ 230
37
+ ],
38
+ "image_target_size": [
39
+ 256,
40
+ 256
41
+ ],
42
+ "letter_box_transform": false,
43
+ "load_bf16": false,
44
+ "max_action_dim": 132,
45
+ "max_num_embodiments": 32,
46
+ "max_seq_len": 1024,
47
+ "max_state_dim": 132,
48
+ "model_dtype": "bfloat16",
49
+ "model_name": "nvidia/Cosmos-Reason2-2B",
50
+ "model_type": "Gr00tN1d7",
51
+ "noise_beta_alpha": 1.5,
52
+ "noise_beta_beta": 1.0,
53
+ "noise_s": 0.999,
54
+ "num_inference_timesteps": 4,
55
+ "num_timestep_buckets": 1000,
56
+ "random_history_crop": true,
57
+ "random_rotation_angle": 0,
58
+ "reproject_vision": false,
59
+ "rtc_ramp_rate": 6.0,
60
+ "select_layer": 16,
61
+ "shortest_image_edge": 256,
62
+ "state_dropout_prob": 0.2,
63
+ "state_gaussian_noise_std": 0.0,
64
+ "transformers_version": "4.57.3",
65
+ "tune_diffusion_model": true,
66
+ "tune_linear": true,
67
+ "tune_llm": false,
68
+ "tune_projector": true,
69
+ "tune_top_llm_layers": 0,
70
+ "tune_visual": false,
71
+ "tune_vlln": true,
72
+ "use_albumentations": true,
73
+ "use_alternate_vl_dit": true,
74
+ "use_flash_attention": true,
75
+ "use_future_tokens": false,
76
+ "use_mean_std": false,
77
+ "use_percentiles": true,
78
+ "use_vl_self_attention": true,
79
+ "use_vlln": true,
80
+ "vl_self_attention_cfg": {
81
+ "attention_head_dim": 64,
82
+ "dropout": 0.2,
83
+ "final_dropout": true,
84
+ "num_attention_heads": 32,
85
+ "num_layers": 4,
86
+ "positional_embeddings": null
87
+ }
88
+ }
embodiment_id.json ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "robocasa_panda_omron": 13,
3
+ "oxe_droid": 17,
4
+ "oxe_fractal": 18,
5
+ "oxe_language_table": 19,
6
+ "oxe_bridge": 20,
7
+ "unknown": 22,
8
+ "gr1_unified": 20,
9
+ "agibot": 26,
10
+ "sim_behavior_r1_pro": 23,
11
+ "xdof": 24,
12
+ "xdof_oss_data": 25,
13
+ "unitree_g1_full_body_with_waist_height_nav_cmd": 25,
14
+ "real_r1_pro_sharpa": 27,
15
+ "real_r1_pro_sharpa_add_view": 27,
16
+ "real_r1_pro_sharpa_relative_arm_joint": 26,
17
+ "real_r1_pro_sharpa_delta_eef": 26,
18
+ "real_r1_pro_sharpa_absolute_eef": 26,
19
+ "real_r1_pro_sharpa_meanstd": 26,
20
+ "real_r1_pro_sharpa_relative_eef": 26,
21
+ "real_r1_pro_sharpa_relative_eef_add_view": 26,
22
+ "real_r1_pro_sharpa_relative_eef_relative_hand": 26,
23
+ "real_r1_pro_sharpa_relative_eef_human": 26,
24
+ "real_r1_pro_sharpa_relative_eef_human_add_view": 26,
25
+ "real_r1_pro_sharpa_relative_eef_human_relative_hand": 26,
26
+ "real_r1_pro_sharpa_relative_eef_egodex": 26,
27
+ "real_r1_pro_sharpa_relative_eef_egodex_relative_hand": 26,
28
+ "real_r1_pro_sharpa_relative_eef_egodex_wrist_only": 26,
29
+ "real_r1_pro_sharpa_relative_eef_maxinsights": 26,
30
+ "real_r1_pro_sharpa_relative_eef_maxinsights_relative_hand": 26,
31
+ "real_r1_pro_sharpa_relative_eef_mecka": 26,
32
+ "real_r1_pro_sharpa_relative_eef_mecka_relative_hand": 26,
33
+ "real_g1_relative_eef_absolute_joints": 25,
34
+ "real_g1_relative_eef_absolute_joints_wrist_cam": 25,
35
+ "real_g1_relative_eef_relative_joints": 25,
36
+ "real_r1_pro_sharpa_relative_eef_relative_hand_relative_joint": 26,
37
+ "real_r1_pro_sharpa_relative_joint": 29,
38
+ "oxe_droid_relative_eef_relative_joint": 24,
39
+ "oxe_droid_relative_eef_relative_joint_swapped": 24,
40
+ "oxe_droid_relative_eef_relative_joint_upweight_z": 24,
41
+ "oxe_droid_relative_eef_relative_joint_upweight_z_swapped": 24,
42
+ "oxe_droid_relative_eef_relative_joint_3view": 24,
43
+ "oxe_droid_relative_eef_relative_joint_3view_swapped": 24,
44
+ "oxe_droid_relative_eef": 24,
45
+ "oxe_droid_joint_position_relative": 24,
46
+ "xdof_relative_eef_relative_joint": 27,
47
+ "xdof_relative_eef_relative_joint_subtask": 27,
48
+ "xdof_relative_eef": 27,
49
+ "xdof_relative_joint": 28,
50
+ "simpler_env_google": 0,
51
+ "simpler_env_widowx": 1,
52
+ "libero_sim": 2,
53
+ "droid_sim": 3,
54
+ "g1_decoupled_wbc": 10,
55
+ "g1_29dof": 10,
56
+ "g1_23dof": 10,
57
+ "new_embodiment": 10
58
+ }
experiment_cfg/conf.yaml ADDED
@@ -0,0 +1,251 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ load_config_path: null
2
+ model:
3
+ model_type: Gr00tN1d7
4
+ model_dtype: bfloat16
5
+ model_name: nvidia/Cosmos-Reason2-2B
6
+ backbone_model_type: qwen
7
+ model_revision: null
8
+ tune_top_llm_layers: 0
9
+ backbone_embedding_dim: 2048
10
+ tune_llm: false
11
+ tune_visual: false
12
+ select_layer: 12
13
+ reproject_vision: false
14
+ use_flash_attention: true
15
+ load_bf16: false
16
+ backbone_trainable_params_fp32: true
17
+ image_crop_size:
18
+ - 230
19
+ - 230
20
+ image_target_size:
21
+ - 256
22
+ - 256
23
+ shortest_image_edge: null
24
+ crop_fraction: null
25
+ random_rotation_angle: null
26
+ color_jitter_params:
27
+ brightness: 0.3
28
+ contrast: 0.4
29
+ saturation: 0.5
30
+ hue: 0.08
31
+ use_albumentations_transforms: true
32
+ extra_augmentation_config: null
33
+ formalize_language: true
34
+ apply_sincos_state_encoding: false
35
+ use_percentiles: true
36
+ use_relative_action: true
37
+ max_state_dim: 132
38
+ max_action_dim: 132
39
+ action_horizon: 40
40
+ hidden_size: 1024
41
+ input_embedding_dim: 1536
42
+ state_history_length: 1
43
+ add_pos_embed: true
44
+ attn_dropout: 0.2
45
+ use_vlln: true
46
+ max_seq_len: 1024
47
+ use_alternate_vl_dit: true
48
+ attend_text_every_n_blocks: 2
49
+ diffusion_model_cfg:
50
+ positional_embeddings: null
51
+ num_layers: 16
52
+ num_attention_heads: 32
53
+ attention_head_dim: 48
54
+ norm_type: ada_norm
55
+ dropout: 0.2
56
+ final_dropout: true
57
+ output_dim: 1024
58
+ interleave_self_attention: true
59
+ num_inference_timesteps: 4
60
+ noise_beta_alpha: 1.5
61
+ noise_beta_beta: 1.0
62
+ noise_s: 0.999
63
+ num_timestep_buckets: 1000
64
+ tune_projector: true
65
+ tune_diffusion_model: true
66
+ tune_vlln: true
67
+ state_dropout_prob: 0.2
68
+ exclude_state: false
69
+ use_mean_std: false
70
+ max_num_embodiments: 32
71
+ data:
72
+ datasets:
73
+ - dataset_paths:
74
+ - datasets/sort_packages_real
75
+ embodiment_tag: g1_29dof
76
+ mix_ratio: 1.0
77
+ dataset_type: physical_embodiment
78
+ val_dataset_path: null
79
+ modality_configs:
80
+ g1_29dof:
81
+ video:
82
+ delta_indices:
83
+ - 0
84
+ modality_keys:
85
+ - ego_view
86
+ sin_cos_embedding_keys: null
87
+ mean_std_embedding_keys: null
88
+ action_configs: null
89
+ state:
90
+ delta_indices:
91
+ - 0
92
+ modality_keys:
93
+ - left_leg
94
+ - right_leg
95
+ - waist
96
+ - left_arm
97
+ - left_hand
98
+ - right_arm
99
+ - right_hand
100
+ - projected_gravity
101
+ sin_cos_embedding_keys: null
102
+ mean_std_embedding_keys: null
103
+ action_configs: null
104
+ action:
105
+ delta_indices:
106
+ - 0
107
+ - 1
108
+ - 2
109
+ - 3
110
+ - 4
111
+ - 5
112
+ - 6
113
+ - 7
114
+ - 8
115
+ - 9
116
+ - 10
117
+ - 11
118
+ - 12
119
+ - 13
120
+ - 14
121
+ - 15
122
+ - 16
123
+ - 17
124
+ - 18
125
+ - 19
126
+ - 20
127
+ - 21
128
+ - 22
129
+ - 23
130
+ - 24
131
+ - 25
132
+ - 26
133
+ - 27
134
+ - 28
135
+ - 29
136
+ - 30
137
+ - 31
138
+ - 32
139
+ - 33
140
+ - 34
141
+ - 35
142
+ - 36
143
+ - 37
144
+ - 38
145
+ - 39
146
+ modality_keys:
147
+ - motion_token
148
+ - left_hand_joints
149
+ - right_hand_joints
150
+ sin_cos_embedding_keys: null
151
+ mean_std_embedding_keys: null
152
+ action_configs:
153
+ - rep: ABSOLUTE
154
+ type: NON_EEF
155
+ format: DEFAULT
156
+ state_key: null
157
+ - rep: ABSOLUTE
158
+ type: NON_EEF
159
+ format: DEFAULT
160
+ state_key: null
161
+ - rep: ABSOLUTE
162
+ type: NON_EEF
163
+ format: DEFAULT
164
+ state_key: null
165
+ language:
166
+ delta_indices:
167
+ - 0
168
+ modality_keys:
169
+ - annotation.human.task_description
170
+ sin_cos_embedding_keys: null
171
+ mean_std_embedding_keys: null
172
+ action_configs: null
173
+ download_cache: false
174
+ shard_size: 1024
175
+ episode_sampling_rate: 0.1
176
+ num_shards_per_epoch: 100000
177
+ override_pretraining_statistics: true
178
+ mode: single_turn
179
+ random_chop: 0.0
180
+ mock_dataset_mode: false
181
+ shuffle: true
182
+ seed: 42
183
+ multiprocessing_context: fork
184
+ allow_padding: false
185
+ subsample_ratio: 1.0
186
+ image_crop_size:
187
+ - 244
188
+ - 244
189
+ image_target_size:
190
+ - 224
191
+ - 224
192
+ video_backend: torchcodec
193
+ training:
194
+ output_dir: ./outputs
195
+ experiment_name: null
196
+ max_steps: 5000
197
+ global_batch_size: 1536
198
+ batch_size: null
199
+ gradient_accumulation_steps: 1
200
+ learning_rate: 0.0001
201
+ lr_scheduler_type: cosine
202
+ weight_decay: 1.0e-05
203
+ warmup_ratio: 0.05
204
+ warmup_steps: 0
205
+ max_grad_norm: 1.0
206
+ optim: adamw_torch
207
+ start_from_checkpoint: nvidia/GR00T-N1.7-3B
208
+ skip_weight_loading: false
209
+ tf32: true
210
+ fp16: false
211
+ bf16: true
212
+ eval_bf16: true
213
+ logging_steps: 10
214
+ save_steps: 1000
215
+ save_total_limit: 5
216
+ save_vl_model: false
217
+ save_only_model: false
218
+ upload_checkpoints: false
219
+ upload_every: 1000
220
+ upload_last_n_checkpoints: 5
221
+ max_concurrent_uploads: 2
222
+ eval_strategy: 'no'
223
+ eval_steps: 500
224
+ eval_set_split_ratio: 0.1
225
+ eval_batch_size: 2
226
+ save_best_eval_metric_name: ''
227
+ save_best_eval_metric_greater_is_better: true
228
+ deepspeed_stage: 2
229
+ gradient_checkpointing: false
230
+ transformers_trust_remote_code: true
231
+ transformers_local_files_only: false
232
+ transformers_cache_dir: null
233
+ transformers_access_token: null
234
+ use_ddp: false
235
+ ddp_bucket_cap_mb: 100
236
+ num_gpus: 2
237
+ dataloader_num_workers: 6
238
+ remove_unused_columns: false
239
+ use_wandb: true
240
+ wandb_project: gr00t-g1-29dof-gear-sonic
241
+ enable_profiling: false
242
+ max_retries: 3
243
+ assert_loss_less_than: null
244
+ add_rl_callback: false
245
+ enable_open_loop_eval: false
246
+ open_loop_eval_traj_ids:
247
+ - 0
248
+ open_loop_eval_steps_per_traj: 100
249
+ open_loop_eval_plot_indices: null
250
+ max_steps: 5000
251
+ save_steps: 1000
experiment_cfg/config.yaml ADDED
@@ -0,0 +1,277 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ - datasets/sort_packages_real
8
+ dataset_type: physical_embodiment
9
+ embodiment_tag: g1_29dof
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
+ g1_29dof:
23
+ action: !!python/object:gr00t.data.types.ModalityConfig
24
+ action_configs:
25
+ - !!python/object:gr00t.data.types.ActionConfig
26
+ format: &id001 !!python/object/apply:gr00t.data.types.ActionFormat
27
+ - default
28
+ rep: &id002 !!python/object/apply:gr00t.data.types.ActionRepresentation
29
+ - absolute
30
+ state_key: null
31
+ type: &id003 !!python/object/apply:gr00t.data.types.ActionType
32
+ - non_eef
33
+ - !!python/object:gr00t.data.types.ActionConfig
34
+ format: *id001
35
+ rep: *id002
36
+ state_key: null
37
+ type: *id003
38
+ - !!python/object:gr00t.data.types.ActionConfig
39
+ format: *id001
40
+ rep: *id002
41
+ state_key: null
42
+ type: *id003
43
+ delta_indices:
44
+ - 0
45
+ - 1
46
+ - 2
47
+ - 3
48
+ - 4
49
+ - 5
50
+ - 6
51
+ - 7
52
+ - 8
53
+ - 9
54
+ - 10
55
+ - 11
56
+ - 12
57
+ - 13
58
+ - 14
59
+ - 15
60
+ - 16
61
+ - 17
62
+ - 18
63
+ - 19
64
+ - 20
65
+ - 21
66
+ - 22
67
+ - 23
68
+ - 24
69
+ - 25
70
+ - 26
71
+ - 27
72
+ - 28
73
+ - 29
74
+ - 30
75
+ - 31
76
+ - 32
77
+ - 33
78
+ - 34
79
+ - 35
80
+ - 36
81
+ - 37
82
+ - 38
83
+ - 39
84
+ mean_std_embedding_keys: null
85
+ modality_keys:
86
+ - motion_token
87
+ - left_hand_joints
88
+ - right_hand_joints
89
+ sin_cos_embedding_keys: null
90
+ language: !!python/object:gr00t.data.types.ModalityConfig
91
+ action_configs: null
92
+ delta_indices:
93
+ - 0
94
+ mean_std_embedding_keys: null
95
+ modality_keys:
96
+ - annotation.human.task_description
97
+ sin_cos_embedding_keys: null
98
+ state: !!python/object:gr00t.data.types.ModalityConfig
99
+ action_configs: null
100
+ delta_indices:
101
+ - 0
102
+ mean_std_embedding_keys: null
103
+ modality_keys:
104
+ - left_leg
105
+ - right_leg
106
+ - waist
107
+ - left_arm
108
+ - left_hand
109
+ - right_arm
110
+ - right_hand
111
+ - projected_gravity
112
+ sin_cos_embedding_keys: null
113
+ video: !!python/object:gr00t.data.types.ModalityConfig
114
+ action_configs: null
115
+ delta_indices:
116
+ - 0
117
+ mean_std_embedding_keys: null
118
+ modality_keys:
119
+ - ego_view
120
+ sin_cos_embedding_keys: null
121
+ mode: single_turn
122
+ multiprocessing_context: fork
123
+ num_shards_per_epoch: 100000
124
+ override_pretraining_statistics: true
125
+ random_chop: 0.0
126
+ seed: 42
127
+ shard_size: 1024
128
+ shuffle: true
129
+ subsample_ratio: 1.0
130
+ video_backend: torchcodec
131
+ load_config_path: null
132
+ model: !!python/object:gr00t.configs.model.gr00t_n1d7.Gr00tN1d7Config
133
+ _attn_implementation_internal: null
134
+ _commit_hash: null
135
+ _name_or_path: ''
136
+ _output_attentions: false
137
+ add_cross_attention: false
138
+ architectures: null
139
+ backbone_trainable_params_fp32: true
140
+ bad_words_ids: null
141
+ begin_suppress_tokens: null
142
+ bos_token_id: null
143
+ chunk_size_feed_forward: 0
144
+ color_jitter_params:
145
+ brightness: 0.3
146
+ contrast: 0.4
147
+ hue: 0.08
148
+ saturation: 0.5
149
+ cross_attention_hidden_size: null
150
+ decoder_start_token_id: null
151
+ diffusion_model_cfg:
152
+ attention_head_dim: 48
153
+ dropout: 0.2
154
+ final_dropout: true
155
+ interleave_self_attention: true
156
+ norm_type: ada_norm
157
+ num_attention_heads: 32
158
+ num_layers: 16
159
+ output_dim: 1024
160
+ positional_embeddings: null
161
+ diversity_penalty: 0.0
162
+ do_sample: false
163
+ dtype: null
164
+ early_stopping: false
165
+ encoder_no_repeat_ngram_size: 0
166
+ eos_token_id: null
167
+ exponential_decay_length_penalty: null
168
+ extra_augmentation_config: null
169
+ finetuning_task: null
170
+ forced_bos_token_id: null
171
+ forced_eos_token_id: null
172
+ id2label:
173
+ 0: LABEL_0
174
+ 1: LABEL_1
175
+ is_decoder: false
176
+ is_encoder_decoder: false
177
+ label2id:
178
+ LABEL_0: 0
179
+ LABEL_1: 1
180
+ length_penalty: 1.0
181
+ load_bf16: false
182
+ max_length: 20
183
+ min_length: 0
184
+ model_name: nvidia/Cosmos-Reason2-2B
185
+ no_repeat_ngram_size: 0
186
+ num_beam_groups: 1
187
+ num_beams: 1
188
+ num_return_sequences: 1
189
+ output_hidden_states: false
190
+ output_scores: false
191
+ pad_token_id: null
192
+ prefix: null
193
+ problem_type: null
194
+ pruned_heads: {}
195
+ random_rotation_angle: null
196
+ remove_invalid_values: false
197
+ repetition_penalty: 1.0
198
+ reproject_vision: false
199
+ return_dict: true
200
+ return_dict_in_generate: false
201
+ sep_token_id: null
202
+ state_dropout_prob: 0.2
203
+ suppress_tokens: null
204
+ task_specific_params: null
205
+ temperature: 1.0
206
+ tf_legacy_loss: false
207
+ tie_encoder_decoder: false
208
+ tie_word_embeddings: true
209
+ tokenizer_class: null
210
+ top_k: 50
211
+ top_p: 1.0
212
+ torchscript: false
213
+ transformers_version: null
214
+ tune_diffusion_model: true
215
+ tune_llm: false
216
+ tune_projector: true
217
+ tune_visual: false
218
+ typical_p: 1.0
219
+ use_bfloat16: false
220
+ use_relative_action: true
221
+ training: !!python/object:gr00t.configs.training.training_config.TrainingConfig
222
+ add_rl_callback: false
223
+ assert_loss_less_than: null
224
+ batch_size: null
225
+ bf16: true
226
+ dataloader_num_workers: 6
227
+ ddp_bucket_cap_mb: 100
228
+ deepspeed_stage: 2
229
+ enable_open_loop_eval: false
230
+ enable_profiling: false
231
+ eval_batch_size: 2
232
+ eval_bf16: true
233
+ eval_set_split_ratio: 0.1
234
+ eval_steps: 500
235
+ eval_strategy: 'no'
236
+ experiment_name: null
237
+ fp16: false
238
+ global_batch_size: 1536
239
+ gradient_accumulation_steps: 1
240
+ gradient_checkpointing: false
241
+ learning_rate: 0.0001
242
+ logging_steps: 10
243
+ lr_scheduler_type: cosine
244
+ max_concurrent_uploads: 2
245
+ max_grad_norm: 1.0
246
+ max_retries: 3
247
+ max_steps: 5000
248
+ num_gpus: 2
249
+ open_loop_eval_plot_indices: null
250
+ open_loop_eval_steps_per_traj: 100
251
+ open_loop_eval_traj_ids:
252
+ - 0
253
+ optim: adamw_torch
254
+ output_dir: ./outputs
255
+ remove_unused_columns: false
256
+ save_best_eval_metric_greater_is_better: true
257
+ save_best_eval_metric_name: ''
258
+ save_only_model: false
259
+ save_steps: 1000
260
+ save_total_limit: 5
261
+ save_vl_model: false
262
+ skip_weight_loading: false
263
+ start_from_checkpoint: nvidia/GR00T-N1.7-3B
264
+ tf32: true
265
+ transformers_access_token: null
266
+ transformers_cache_dir: null
267
+ transformers_local_files_only: false
268
+ transformers_trust_remote_code: true
269
+ upload_checkpoints: false
270
+ upload_every: 1000
271
+ upload_last_n_checkpoints: 5
272
+ use_ddp: false
273
+ use_wandb: true
274
+ wandb_project: gr00t-g1-29dof-gear-sonic
275
+ warmup_ratio: 0.05
276
+ warmup_steps: 0
277
+ weight_decay: 1.0e-05
experiment_cfg/dataset_statistics.json ADDED
@@ -0,0 +1,907 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "g1_29dof": {
3
+ "state": {
4
+ "left_leg": {
5
+ "min": [
6
+ -0.5459340214729309,
7
+ 0.020836573094129562,
8
+ 0.03307798132300377,
9
+ 0.18152958154678345,
10
+ -0.3211824595928192,
11
+ -0.12429654598236084
12
+ ],
13
+ "max": [
14
+ 0.037899065762758255,
15
+ 0.3440207540988922,
16
+ 0.8120443820953369,
17
+ 0.6794171333312988,
18
+ 0.06066618487238884,
19
+ 0.07582387328147888
20
+ ],
21
+ "mean": [
22
+ -0.20662064850330353,
23
+ 0.16517671942710876,
24
+ 0.37475070357322693,
25
+ 0.368203729391098,
26
+ -0.1281968355178833,
27
+ -0.010802671313285828
28
+ ],
29
+ "std": [
30
+ 0.1077035665512085,
31
+ 0.05811949819326401,
32
+ 0.17629078030586243,
33
+ 0.08000018447637558,
34
+ 0.054561641067266464,
35
+ 0.025703581050038338
36
+ ],
37
+ "q01": [
38
+ -0.4678820306062698,
39
+ 0.07722962506115437,
40
+ 0.14612941384315492,
41
+ 0.22429704502224923,
42
+ -0.24932930871844292,
43
+ -0.07663071535527706
44
+ ],
45
+ "q99": [
46
+ 0.0019284329609945416,
47
+ 0.30674280643463137,
48
+ 0.7542048096656799,
49
+ 0.5877020955085754,
50
+ -0.00820838207378974,
51
+ 0.05381844338029624
52
+ ]
53
+ },
54
+ "right_leg": {
55
+ "min": [
56
+ -0.540952205657959,
57
+ -0.6762639284133911,
58
+ -0.9704035520553589,
59
+ 0.17207002639770508,
60
+ -0.24158212542533875,
61
+ -0.07556027919054031
62
+ ],
63
+ "max": [
64
+ 0.0653926283121109,
65
+ -0.042815107852220535,
66
+ 0.08482426404953003,
67
+ 0.9727909564971924,
68
+ 0.22199997305870056,
69
+ 0.14264413714408875
70
+ ],
71
+ "mean": [
72
+ -0.193039670586586,
73
+ -0.21999937295913696,
74
+ -0.40548714995384216,
75
+ 0.3346921503543854,
76
+ -0.07188944518566132,
77
+ 0.04187983274459839
78
+ ],
79
+ "std": [
80
+ 0.10288046300411224,
81
+ 0.11948458105325699,
82
+ 0.22097143530845642,
83
+ 0.08815735578536987,
84
+ 0.040201835334300995,
85
+ 0.02890731394290924
86
+ ],
87
+ "q01": [
88
+ -0.4420629927515984,
89
+ -0.5392772448062897,
90
+ -0.8507518917322159,
91
+ 0.19727469518780708,
92
+ -0.18496158257126807,
93
+ -0.022557649910449982
94
+ ],
95
+ "q99": [
96
+ 0.0042635779036209335,
97
+ -0.07850794352591034,
98
+ -0.10300550691783425,
99
+ 0.6013008415699007,
100
+ 0.011049804165959387,
101
+ 0.11685092002153397
102
+ ]
103
+ },
104
+ "waist": {
105
+ "min": [
106
+ -0.4189922511577606,
107
+ 0.0,
108
+ 0.0
109
+ ],
110
+ "max": [
111
+ 0.3251686096191406,
112
+ 0.0,
113
+ 0.0
114
+ ],
115
+ "mean": [
116
+ -0.036889638751745224,
117
+ 0.0,
118
+ 0.0
119
+ ],
120
+ "std": [
121
+ 0.08375038951635361,
122
+ 0.0,
123
+ 0.0
124
+ ],
125
+ "q01": [
126
+ -0.29784566938877105,
127
+ 0.0,
128
+ 0.0
129
+ ],
130
+ "q99": [
131
+ 0.21038212612271373,
132
+ 0.0,
133
+ 0.0
134
+ ]
135
+ },
136
+ "left_arm": {
137
+ "min": [
138
+ -1.1030880212783813,
139
+ -0.1675993949174881,
140
+ -0.8565964698791504,
141
+ -0.6908906102180481,
142
+ -0.8318610191345215,
143
+ 0.0,
144
+ 0.0
145
+ ],
146
+ "max": [
147
+ 0.3145259916782379,
148
+ 0.6950730681419373,
149
+ 0.634636640548706,
150
+ 1.0954540967941284,
151
+ 0.426111102104187,
152
+ 0.0,
153
+ 0.0
154
+ ],
155
+ "mean": [
156
+ -0.5297300219535828,
157
+ 0.049589674919843674,
158
+ -0.2067471146583557,
159
+ 0.04027409106492996,
160
+ -0.16137352585792542,
161
+ 0.0,
162
+ 0.0
163
+ ],
164
+ "std": [
165
+ 0.1920243203639984,
166
+ 0.10622277855873108,
167
+ 0.17039525508880615,
168
+ 0.3023645877838135,
169
+ 0.18287615478038788,
170
+ 0.0,
171
+ 0.0
172
+ ],
173
+ "q01": [
174
+ -0.9600922465324402,
175
+ -0.1571415938436985,
176
+ -0.650602256655693,
177
+ -0.42840524643659594,
178
+ -0.6665949642658233,
179
+ 0.0,
180
+ 0.0
181
+ ],
182
+ "q99": [
183
+ 0.00679002243094253,
184
+ 0.41254399955272725,
185
+ 0.37448701351881053,
186
+ 0.8449516332149517,
187
+ 0.21284368053078764,
188
+ 0.0,
189
+ 0.0
190
+ ]
191
+ },
192
+ "left_hand": {
193
+ "min": [
194
+ 0.0,
195
+ 0.0,
196
+ 0.0,
197
+ 0.0,
198
+ 0.0,
199
+ 0.0,
200
+ 0.0
201
+ ],
202
+ "max": [
203
+ 0.0,
204
+ 0.0,
205
+ 0.0,
206
+ 0.0,
207
+ 0.0,
208
+ 0.0,
209
+ 0.0
210
+ ],
211
+ "mean": [
212
+ 0.0,
213
+ 0.0,
214
+ 0.0,
215
+ 0.0,
216
+ 0.0,
217
+ 0.0,
218
+ 0.0
219
+ ],
220
+ "std": [
221
+ 0.0,
222
+ 0.0,
223
+ 0.0,
224
+ 0.0,
225
+ 0.0,
226
+ 0.0,
227
+ 0.0
228
+ ],
229
+ "q01": [
230
+ 0.0,
231
+ 0.0,
232
+ 0.0,
233
+ 0.0,
234
+ 0.0,
235
+ 0.0,
236
+ 0.0
237
+ ],
238
+ "q99": [
239
+ 0.0,
240
+ 0.0,
241
+ 0.0,
242
+ 0.0,
243
+ 0.0,
244
+ 0.0,
245
+ 0.0
246
+ ]
247
+ },
248
+ "right_arm": {
249
+ "min": [
250
+ -1.0336154699325562,
251
+ -1.020492672920227,
252
+ -0.5443115234375,
253
+ -0.6091341972351074,
254
+ -0.09871406108140945,
255
+ 0.0,
256
+ 0.0
257
+ ],
258
+ "max": [
259
+ 0.1786608248949051,
260
+ 0.12750017642974854,
261
+ 0.8439651131629944,
262
+ 1.1767909526824951,
263
+ 1.327252984046936,
264
+ 0.0,
265
+ 0.0
266
+ ],
267
+ "mean": [
268
+ -0.46119454503059387,
269
+ -0.22351405024528503,
270
+ 0.032373715192079544,
271
+ 0.08123326301574707,
272
+ 0.42776578664779663,
273
+ 0.0,
274
+ 0.0
275
+ ],
276
+ "std": [
277
+ 0.18687885999679565,
278
+ 0.16154557466506958,
279
+ 0.18049120903015137,
280
+ 0.36587652564048767,
281
+ 0.16936077177524567,
282
+ 0.0,
283
+ 0.0
284
+ ],
285
+ "q01": [
286
+ -0.9048523896932602,
287
+ -0.7503394240140915,
288
+ -0.3831618160009384,
289
+ -0.4913097244501114,
290
+ 0.0538766410574317,
291
+ 0.0,
292
+ 0.0
293
+ ],
294
+ "q99": [
295
+ -0.1103991630673406,
296
+ 0.10943503081798575,
297
+ 0.6203017145395281,
298
+ 0.9831467711925507,
299
+ 0.9494794869422917,
300
+ 0.0,
301
+ 0.0
302
+ ]
303
+ },
304
+ "right_hand": {
305
+ "min": [
306
+ 0.0,
307
+ 0.0,
308
+ 0.0,
309
+ 0.0,
310
+ 0.0,
311
+ 0.0,
312
+ 0.0
313
+ ],
314
+ "max": [
315
+ 0.0,
316
+ 0.0,
317
+ 0.0,
318
+ 0.0,
319
+ 0.0,
320
+ 0.0,
321
+ 0.0
322
+ ],
323
+ "mean": [
324
+ 0.0,
325
+ 0.0,
326
+ 0.0,
327
+ 0.0,
328
+ 0.0,
329
+ 0.0,
330
+ 0.0
331
+ ],
332
+ "std": [
333
+ 0.0,
334
+ 0.0,
335
+ 0.0,
336
+ 0.0,
337
+ 0.0,
338
+ 0.0,
339
+ 0.0
340
+ ],
341
+ "q01": [
342
+ 0.0,
343
+ 0.0,
344
+ 0.0,
345
+ 0.0,
346
+ 0.0,
347
+ 0.0,
348
+ 0.0
349
+ ],
350
+ "q99": [
351
+ 0.0,
352
+ 0.0,
353
+ 0.0,
354
+ 0.0,
355
+ 0.0,
356
+ 0.0,
357
+ 0.0
358
+ ]
359
+ },
360
+ "projected_gravity": {
361
+ "min": [
362
+ -0.11367977410554886,
363
+ -0.06603632867336273,
364
+ -1.0
365
+ ],
366
+ "max": [
367
+ 0.24466784298419952,
368
+ 0.05539962649345398,
369
+ -0.9692031741142273
370
+ ],
371
+ "mean": [
372
+ 0.03607587143778801,
373
+ -0.00853223167359829,
374
+ -0.9977879524230957
375
+ ],
376
+ "std": [
377
+ 0.05618465691804886,
378
+ 0.013086325488984585,
379
+ 0.0034188576974017518
380
+ ],
381
+ "q01": [
382
+ -0.06162475544959307,
383
+ -0.03968637753278017,
384
+ -0.9999926686286926
385
+ ],
386
+ "q99": [
387
+ 0.17451374292373673,
388
+ 0.027327275015413814,
389
+ -0.9845989102125168
390
+ ]
391
+ }
392
+ },
393
+ "action": {
394
+ "motion_token": {
395
+ "min": [
396
+ -0.25,
397
+ -0.1875,
398
+ -0.5625,
399
+ -0.5625,
400
+ -0.5625,
401
+ -0.25,
402
+ -0.375,
403
+ -0.125,
404
+ -0.375,
405
+ -0.3125,
406
+ -0.125,
407
+ -0.0625,
408
+ -0.4375,
409
+ -0.25,
410
+ -0.3125,
411
+ -0.25,
412
+ -0.25,
413
+ -0.375,
414
+ -0.375,
415
+ -0.1875,
416
+ -0.375,
417
+ -0.5625,
418
+ -0.4375,
419
+ -0.25,
420
+ -0.25,
421
+ 0.0,
422
+ -0.4375,
423
+ -0.0625,
424
+ -0.125,
425
+ -0.3125,
426
+ -0.375,
427
+ -0.1875,
428
+ -0.4375,
429
+ -0.25,
430
+ -0.5,
431
+ -0.1875,
432
+ -0.5,
433
+ -0.3125,
434
+ -0.3125,
435
+ -0.4375,
436
+ -0.3125,
437
+ -0.0625,
438
+ -0.25,
439
+ -0.5,
440
+ -0.25,
441
+ -0.25,
442
+ -0.375,
443
+ -0.25,
444
+ -0.1875,
445
+ -0.125,
446
+ -0.3125,
447
+ -0.1875,
448
+ -0.3125,
449
+ -0.4375,
450
+ -0.5,
451
+ -0.25,
452
+ -0.1875,
453
+ -0.5625,
454
+ -0.125,
455
+ 0.0,
456
+ -0.1875,
457
+ -0.1875,
458
+ -0.25,
459
+ -0.3125
460
+ ],
461
+ "max": [
462
+ 0.375,
463
+ 0.25,
464
+ 0.125,
465
+ 0.1875,
466
+ 0.25,
467
+ 0.25,
468
+ 0.125,
469
+ 0.5625,
470
+ 0.5,
471
+ 0.3125,
472
+ 0.25,
473
+ 0.4375,
474
+ 0.0625,
475
+ 0.125,
476
+ 0.3125,
477
+ 0.1875,
478
+ 0.4375,
479
+ 0.125,
480
+ 0.1875,
481
+ 0.375,
482
+ 0.125,
483
+ 0.0,
484
+ 0.3125,
485
+ 0.375,
486
+ 0.5,
487
+ 0.4375,
488
+ 0.3125,
489
+ 0.4375,
490
+ 0.4375,
491
+ 0.4375,
492
+ 0.25,
493
+ 0.25,
494
+ 0.0625,
495
+ 0.375,
496
+ 0.125,
497
+ 0.5,
498
+ 0.1875,
499
+ 0.4375,
500
+ 0.25,
501
+ 0.25,
502
+ 0.125,
503
+ 0.5,
504
+ 0.375,
505
+ 0.4375,
506
+ 0.3125,
507
+ 0.125,
508
+ 0.3125,
509
+ 0.1875,
510
+ 0.5,
511
+ 0.3125,
512
+ 0.5,
513
+ 0.375,
514
+ 0.4375,
515
+ 0.1875,
516
+ 0.1875,
517
+ 0.4375,
518
+ 0.25,
519
+ 0.125,
520
+ 0.5,
521
+ 0.4375,
522
+ 0.3125,
523
+ 0.375,
524
+ 0.4375,
525
+ 0.4375
526
+ ],
527
+ "mean": [
528
+ 0.08140476793050766,
529
+ 0.06792771071195602,
530
+ -0.22392231225967407,
531
+ -0.26640209555625916,
532
+ -0.19179262220859528,
533
+ -0.01847255602478981,
534
+ -0.13173852860927582,
535
+ 0.2070552557706833,
536
+ 0.09662047773599625,
537
+ -0.08532193303108215,
538
+ 0.11379750072956085,
539
+ 0.17242835462093353,
540
+ -0.2373971939086914,
541
+ -0.10906066745519638,
542
+ -0.022489041090011597,
543
+ -0.03505147248506546,
544
+ 0.14378361403942108,
545
+ -0.07929611206054688,
546
+ -0.14308619499206543,
547
+ 0.16388896107673645,
548
+ -0.1588498055934906,
549
+ -0.2928672432899475,
550
+ -0.07414671778678894,
551
+ 0.025249283760786057,
552
+ 0.06719863414764404,
553
+ 0.21122564375400543,
554
+ 0.01722286269068718,
555
+ 0.1611025184392929,
556
+ 0.19415012001991272,
557
+ 0.019569450989365578,
558
+ -0.030947672203183174,
559
+ 0.046191759407520294,
560
+ -0.21559466421604156,
561
+ 0.12881238758563995,
562
+ -0.2516317069530487,
563
+ 0.23479630053043365,
564
+ -0.12657931447029114,
565
+ 0.03772767260670662,
566
+ 0.047308299690485,
567
+ -0.05665426701307297,
568
+ -0.04151713848114014,
569
+ 0.2534947693347931,
570
+ 0.0656258761882782,
571
+ -0.09658336639404297,
572
+ 0.0887250304222107,
573
+ -0.06463048607110977,
574
+ -0.08554785698652267,
575
+ -0.016205644235014915,
576
+ 0.10814385861158371,
577
+ 0.06956486403942108,
578
+ 0.15535829961299896,
579
+ 0.10281001031398773,
580
+ 0.11934635788202286,
581
+ -0.09885901212692261,
582
+ -0.21955439448356628,
583
+ 0.19556352496147156,
584
+ 0.06404110789299011,
585
+ -0.2392318844795227,
586
+ 0.15785112977027893,
587
+ 0.2314062863588333,
588
+ 0.09991879761219025,
589
+ 0.09364412724971771,
590
+ 0.04072912409901619,
591
+ 0.13782545924186707
592
+ ],
593
+ "std": [
594
+ 0.09975534677505493,
595
+ 0.06287963688373566,
596
+ 0.10252486914396286,
597
+ 0.09614555537700653,
598
+ 0.09721450507640839,
599
+ 0.07409973442554474,
600
+ 0.08462771028280258,
601
+ 0.09551585465669632,
602
+ 0.14713865518569946,
603
+ 0.09029823541641235,
604
+ 0.05245383828878403,
605
+ 0.0718272402882576,
606
+ 0.0715542733669281,
607
+ 0.054989010095596313,
608
+ 0.0905834287405014,
609
+ 0.07303649187088013,
610
+ 0.09359410405158997,
611
+ 0.08907555788755417,
612
+ 0.07803433388471603,
613
+ 0.07319825887680054,
614
+ 0.0711868479847908,
615
+ 0.07639683783054352,
616
+ 0.12524324655532837,
617
+ 0.08468863368034363,
618
+ 0.10562840849161148,
619
+ 0.07204841077327728,
620
+ 0.11544526368379593,
621
+ 0.07498978823423386,
622
+ 0.07564231008291245,
623
+ 0.10132701694965363,
624
+ 0.10170868039131165,
625
+ 0.05836925283074379,
626
+ 0.0926022008061409,
627
+ 0.08838827908039093,
628
+ 0.0864035040140152,
629
+ 0.12131625413894653,
630
+ 0.09940194338560104,
631
+ 0.1156344786286354,
632
+ 0.09147944301366806,
633
+ 0.10684174299240112,
634
+ 0.08788376301527023,
635
+ 0.07355199754238129,
636
+ 0.10123647004365921,
637
+ 0.14389897882938385,
638
+ 0.10935746133327484,
639
+ 0.05960744991898537,
640
+ 0.09588974714279175,
641
+ 0.06536087393760681,
642
+ 0.08791742473840714,
643
+ 0.06530021131038666,
644
+ 0.10736631602048874,
645
+ 0.06333936750888824,
646
+ 0.10655359923839569,
647
+ 0.08971486240625381,
648
+ 0.09705425798892975,
649
+ 0.09006437659263611,
650
+ 0.06474755704402924,
651
+ 0.10175342112779617,
652
+ 0.0893244668841362,
653
+ 0.082696333527565,
654
+ 0.06853081285953522,
655
+ 0.06663510948419571,
656
+ 0.1082838848233223,
657
+ 0.13823480904102325
658
+ ],
659
+ "q01": [
660
+ -0.125,
661
+ -0.0625,
662
+ -0.4375,
663
+ -0.5,
664
+ -0.4375,
665
+ -0.1875,
666
+ -0.3125,
667
+ 0.0,
668
+ -0.25,
669
+ -0.3125,
670
+ -0.0625,
671
+ 0.0,
672
+ -0.375,
673
+ -0.1875,
674
+ -0.1875,
675
+ -0.1875,
676
+ -0.125,
677
+ -0.25,
678
+ -0.3125,
679
+ 0.0,
680
+ -0.3125,
681
+ -0.4375,
682
+ -0.375,
683
+ -0.125,
684
+ -0.1875,
685
+ 0.0,
686
+ -0.3125,
687
+ 0.0,
688
+ -0.0625,
689
+ -0.25,
690
+ -0.3125,
691
+ -0.125,
692
+ -0.375,
693
+ -0.125,
694
+ -0.4375,
695
+ -0.0625,
696
+ -0.4375,
697
+ -0.25,
698
+ -0.25,
699
+ -0.375,
700
+ -0.25,
701
+ 0.0625,
702
+ -0.1875,
703
+ -0.375,
704
+ -0.1875,
705
+ -0.25,
706
+ -0.3125,
707
+ -0.1875,
708
+ -0.0625,
709
+ -0.0625,
710
+ -0.1875,
711
+ -0.0625,
712
+ -0.125,
713
+ -0.3125,
714
+ -0.4375,
715
+ -0.0625,
716
+ -0.0625,
717
+ -0.5,
718
+ 0.0,
719
+ 0.0625,
720
+ -0.0625,
721
+ -0.0625,
722
+ -0.1875,
723
+ -0.1875
724
+ ],
725
+ "q99": [
726
+ 0.3125,
727
+ 0.1875,
728
+ 0.125,
729
+ 0.0,
730
+ 0.125,
731
+ 0.125,
732
+ 0.0625,
733
+ 0.5,
734
+ 0.4375,
735
+ 0.1875,
736
+ 0.1875,
737
+ 0.375,
738
+ -0.0625,
739
+ 0.0,
740
+ 0.1875,
741
+ 0.125,
742
+ 0.375,
743
+ 0.0625,
744
+ 0.0625,
745
+ 0.3125,
746
+ 0.0,
747
+ -0.125,
748
+ 0.1875,
749
+ 0.25,
750
+ 0.375,
751
+ 0.375,
752
+ 0.25,
753
+ 0.3125,
754
+ 0.375,
755
+ 0.3125,
756
+ 0.1875,
757
+ 0.1875,
758
+ 0.0,
759
+ 0.3125,
760
+ 0.0,
761
+ 0.5,
762
+ 0.125,
763
+ 0.375,
764
+ 0.1875,
765
+ 0.125,
766
+ 0.125,
767
+ 0.4375,
768
+ 0.3125,
769
+ 0.3125,
770
+ 0.3125,
771
+ 0.0625,
772
+ 0.1875,
773
+ 0.125,
774
+ 0.375,
775
+ 0.25,
776
+ 0.4375,
777
+ 0.25,
778
+ 0.375,
779
+ 0.125,
780
+ 0.0625,
781
+ 0.375,
782
+ 0.25,
783
+ 0.0,
784
+ 0.4375,
785
+ 0.4375,
786
+ 0.25,
787
+ 0.25,
788
+ 0.3125,
789
+ 0.375
790
+ ]
791
+ },
792
+ "left_hand_joints": {
793
+ "min": [
794
+ 0.0,
795
+ 0.0,
796
+ 0.0,
797
+ 0.0,
798
+ 0.0,
799
+ 0.0,
800
+ 0.0
801
+ ],
802
+ "max": [
803
+ 0.0,
804
+ 0.0,
805
+ 0.0,
806
+ 0.0,
807
+ 0.0,
808
+ 0.0,
809
+ 0.0
810
+ ],
811
+ "mean": [
812
+ 0.0,
813
+ 0.0,
814
+ 0.0,
815
+ 0.0,
816
+ 0.0,
817
+ 0.0,
818
+ 0.0
819
+ ],
820
+ "std": [
821
+ 0.0,
822
+ 0.0,
823
+ 0.0,
824
+ 0.0,
825
+ 0.0,
826
+ 0.0,
827
+ 0.0
828
+ ],
829
+ "q01": [
830
+ 0.0,
831
+ 0.0,
832
+ 0.0,
833
+ 0.0,
834
+ 0.0,
835
+ 0.0,
836
+ 0.0
837
+ ],
838
+ "q99": [
839
+ 0.0,
840
+ 0.0,
841
+ 0.0,
842
+ 0.0,
843
+ 0.0,
844
+ 0.0,
845
+ 0.0
846
+ ]
847
+ },
848
+ "right_hand_joints": {
849
+ "min": [
850
+ 0.0,
851
+ 0.0,
852
+ 0.0,
853
+ 0.0,
854
+ 0.0,
855
+ 0.0,
856
+ 0.0
857
+ ],
858
+ "max": [
859
+ 0.0,
860
+ 0.0,
861
+ 0.0,
862
+ 0.0,
863
+ 0.0,
864
+ 0.0,
865
+ 0.0
866
+ ],
867
+ "mean": [
868
+ 0.0,
869
+ 0.0,
870
+ 0.0,
871
+ 0.0,
872
+ 0.0,
873
+ 0.0,
874
+ 0.0
875
+ ],
876
+ "std": [
877
+ 0.0,
878
+ 0.0,
879
+ 0.0,
880
+ 0.0,
881
+ 0.0,
882
+ 0.0,
883
+ 0.0
884
+ ],
885
+ "q01": [
886
+ 0.0,
887
+ 0.0,
888
+ 0.0,
889
+ 0.0,
890
+ 0.0,
891
+ 0.0,
892
+ 0.0
893
+ ],
894
+ "q99": [
895
+ 0.0,
896
+ 0.0,
897
+ 0.0,
898
+ 0.0,
899
+ 0.0,
900
+ 0.0,
901
+ 0.0
902
+ ]
903
+ }
904
+ },
905
+ "relative_action": {}
906
+ }
907
+ }
experiment_cfg/final_model_config.json ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_type": "Gr00tN1d7",
3
+ "model_dtype": "bfloat16",
4
+ "model_name": "nvidia/Cosmos-Reason2-2B",
5
+ "backbone_model_type": "qwen",
6
+ "model_revision": null,
7
+ "tune_top_llm_layers": 0,
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": false,
15
+ "backbone_trainable_params_fp32": true,
16
+ "extra_augmentation_config": null,
17
+ "apply_sincos_state_encoding": false,
18
+ "use_percentiles": true,
19
+ "use_relative_action": false,
20
+ "max_state_dim": 132,
21
+ "max_action_dim": 132,
22
+ "action_horizon": 40,
23
+ "hidden_size": 1024,
24
+ "input_embedding_dim": 1536,
25
+ "state_history_length": 1,
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": 0.2,
52
+ "exclude_state": false,
53
+ "use_mean_std": false,
54
+ "max_num_embodiments": 32
55
+ }
experiment_cfg/final_processor_config.json ADDED
The diff for this file is too large to render. See raw diff
 
global_step2000/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e761185ce5ae9b7ed5be98f9ca61e87453f5a35e2898ea58b491658d6a402ab7
3
+ size 9723131425
global_step2000/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e9733693ce9981e45aaf01ab5d74f9ef7c4a1afaea3777db7674cf21df1e7e67
3
+ size 9723123937
global_step2000/mp_rank_00_model_states.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e068f865657f1582f9ffeadad2de34214c03db6a729e2402000cb31f23533fd
3
+ size 9335640879
latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step2000
model-00001-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e0ad99b311055c5ebe55fd84f89bc85d4189918285c4061bc0f3df88722bbc7a
3
+ size 4990519232
model-00002-of-00002.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6eb0971e425f9a5e597be2feb0130672bb348466e0d45f2c6878775a5d8823f8
3
+ size 1919980184
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
processor_config.json ADDED
@@ -0,0 +1,1159 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "processor_class": "Gr00tN1d7Processor",
3
+ "processor_kwargs": {
4
+ "modality_configs": {
5
+ "real_g1_relative_eef_relative_joints": {
6
+ "video": {
7
+ "delta_indices": [
8
+ -20,
9
+ 0
10
+ ],
11
+ "modality_keys": [
12
+ "ego_view"
13
+ ],
14
+ "sin_cos_embedding_keys": null,
15
+ "mean_std_embedding_keys": null,
16
+ "action_configs": null
17
+ },
18
+ "state": {
19
+ "delta_indices": [
20
+ 0
21
+ ],
22
+ "modality_keys": [
23
+ "left_wrist_eef_9d",
24
+ "right_wrist_eef_9d",
25
+ "left_hand",
26
+ "right_hand",
27
+ "left_arm",
28
+ "right_arm",
29
+ "waist"
30
+ ],
31
+ "sin_cos_embedding_keys": null,
32
+ "mean_std_embedding_keys": null,
33
+ "action_configs": null
34
+ },
35
+ "action": {
36
+ "delta_indices": [
37
+ 0,
38
+ 1,
39
+ 2,
40
+ 3,
41
+ 4,
42
+ 5,
43
+ 6,
44
+ 7,
45
+ 8,
46
+ 9,
47
+ 10,
48
+ 11,
49
+ 12,
50
+ 13,
51
+ 14,
52
+ 15,
53
+ 16,
54
+ 17,
55
+ 18,
56
+ 19,
57
+ 20,
58
+ 21,
59
+ 22,
60
+ 23,
61
+ 24,
62
+ 25,
63
+ 26,
64
+ 27,
65
+ 28,
66
+ 29,
67
+ 30,
68
+ 31,
69
+ 32,
70
+ 33,
71
+ 34,
72
+ 35,
73
+ 36,
74
+ 37,
75
+ 38,
76
+ 39
77
+ ],
78
+ "modality_keys": [
79
+ "left_wrist_eef_9d",
80
+ "right_wrist_eef_9d",
81
+ "left_hand",
82
+ "right_hand",
83
+ "left_arm",
84
+ "right_arm",
85
+ "waist",
86
+ "base_height_command",
87
+ "navigate_command"
88
+ ],
89
+ "sin_cos_embedding_keys": null,
90
+ "mean_std_embedding_keys": null,
91
+ "action_configs": [
92
+ {
93
+ "rep": "RELATIVE",
94
+ "type": "EEF",
95
+ "format": "XYZ_ROT6D",
96
+ "state_key": "left_wrist_eef_9d"
97
+ },
98
+ {
99
+ "rep": "RELATIVE",
100
+ "type": "EEF",
101
+ "format": "XYZ_ROT6D",
102
+ "state_key": "right_wrist_eef_9d"
103
+ },
104
+ {
105
+ "rep": "ABSOLUTE",
106
+ "type": "NON_EEF",
107
+ "format": "DEFAULT",
108
+ "state_key": "left_hand"
109
+ },
110
+ {
111
+ "rep": "ABSOLUTE",
112
+ "type": "NON_EEF",
113
+ "format": "DEFAULT",
114
+ "state_key": "right_hand"
115
+ },
116
+ {
117
+ "rep": "RELATIVE",
118
+ "type": "NON_EEF",
119
+ "format": "DEFAULT",
120
+ "state_key": "left_arm"
121
+ },
122
+ {
123
+ "rep": "RELATIVE",
124
+ "type": "NON_EEF",
125
+ "format": "DEFAULT",
126
+ "state_key": "right_arm"
127
+ },
128
+ {
129
+ "rep": "ABSOLUTE",
130
+ "type": "NON_EEF",
131
+ "format": "DEFAULT",
132
+ "state_key": "waist"
133
+ },
134
+ {
135
+ "rep": "ABSOLUTE",
136
+ "type": "NON_EEF",
137
+ "format": "DEFAULT",
138
+ "state_key": "base_height_command"
139
+ },
140
+ {
141
+ "rep": "ABSOLUTE",
142
+ "type": "NON_EEF",
143
+ "format": "DEFAULT",
144
+ "state_key": "navigate_command"
145
+ }
146
+ ]
147
+ },
148
+ "language": {
149
+ "delta_indices": [
150
+ 0
151
+ ],
152
+ "modality_keys": [
153
+ "annotation.human.task_description"
154
+ ],
155
+ "sin_cos_embedding_keys": null,
156
+ "mean_std_embedding_keys": null,
157
+ "action_configs": null
158
+ }
159
+ },
160
+ "real_r1_pro_sharpa_relative_eef_mecka": {
161
+ "video": {
162
+ "delta_indices": [
163
+ -30,
164
+ 0
165
+ ],
166
+ "modality_keys": [
167
+ "ego_view_cropratio_res320x240_freq30"
168
+ ],
169
+ "sin_cos_embedding_keys": null,
170
+ "mean_std_embedding_keys": null,
171
+ "action_configs": null
172
+ },
173
+ "state": {
174
+ "delta_indices": [
175
+ 0
176
+ ],
177
+ "modality_keys": [
178
+ "left_wrist_eef",
179
+ "right_wrist_eef",
180
+ "left_hand_joints",
181
+ "right_hand_joints"
182
+ ],
183
+ "sin_cos_embedding_keys": null,
184
+ "mean_std_embedding_keys": null,
185
+ "action_configs": null
186
+ },
187
+ "action": {
188
+ "delta_indices": [
189
+ 0,
190
+ 1,
191
+ 2,
192
+ 3,
193
+ 4,
194
+ 5,
195
+ 6,
196
+ 7,
197
+ 8,
198
+ 9,
199
+ 10,
200
+ 11,
201
+ 12,
202
+ 13,
203
+ 14,
204
+ 15,
205
+ 16,
206
+ 17,
207
+ 18,
208
+ 19,
209
+ 20,
210
+ 21,
211
+ 22,
212
+ 23,
213
+ 24,
214
+ 25,
215
+ 26,
216
+ 27,
217
+ 28,
218
+ 29,
219
+ 30,
220
+ 31,
221
+ 32,
222
+ 33,
223
+ 34,
224
+ 35,
225
+ 36,
226
+ 37,
227
+ 38,
228
+ 39
229
+ ],
230
+ "modality_keys": [
231
+ "left_wrist_eef",
232
+ "right_wrist_eef",
233
+ "left_hand_joints",
234
+ "right_hand_joints"
235
+ ],
236
+ "sin_cos_embedding_keys": null,
237
+ "mean_std_embedding_keys": null,
238
+ "action_configs": [
239
+ {
240
+ "rep": "RELATIVE",
241
+ "type": "EEF",
242
+ "format": "XYZ_ROT6D",
243
+ "state_key": "left_wrist_eef"
244
+ },
245
+ {
246
+ "rep": "RELATIVE",
247
+ "type": "EEF",
248
+ "format": "XYZ_ROT6D",
249
+ "state_key": "right_wrist_eef"
250
+ },
251
+ {
252
+ "rep": "ABSOLUTE",
253
+ "type": "NON_EEF",
254
+ "format": "DEFAULT",
255
+ "state_key": "left_hand_joints"
256
+ },
257
+ {
258
+ "rep": "ABSOLUTE",
259
+ "type": "NON_EEF",
260
+ "format": "DEFAULT",
261
+ "state_key": "right_hand_joints"
262
+ }
263
+ ]
264
+ },
265
+ "language": {
266
+ "delta_indices": [
267
+ 0
268
+ ],
269
+ "modality_keys": [
270
+ "annotation.human.coarse_action"
271
+ ],
272
+ "sin_cos_embedding_keys": null,
273
+ "mean_std_embedding_keys": null,
274
+ "action_configs": null
275
+ }
276
+ },
277
+ "real_r1_pro_sharpa_relative_eef_human": {
278
+ "video": {
279
+ "delta_indices": [
280
+ -20,
281
+ 0
282
+ ],
283
+ "modality_keys": [
284
+ "ego_view_res320x240_freq20",
285
+ "left_wrist_view_res320x240_freq20",
286
+ "right_wrist_view_res320x240_freq20"
287
+ ],
288
+ "sin_cos_embedding_keys": null,
289
+ "mean_std_embedding_keys": null,
290
+ "action_configs": null
291
+ },
292
+ "state": {
293
+ "delta_indices": [
294
+ 0
295
+ ],
296
+ "modality_keys": [
297
+ "left_wrist_eef",
298
+ "right_wrist_eef",
299
+ "left_hand_joints",
300
+ "right_hand_joints"
301
+ ],
302
+ "sin_cos_embedding_keys": null,
303
+ "mean_std_embedding_keys": null,
304
+ "action_configs": null
305
+ },
306
+ "action": {
307
+ "delta_indices": [
308
+ 0,
309
+ 1,
310
+ 2,
311
+ 3,
312
+ 4,
313
+ 5,
314
+ 6,
315
+ 7,
316
+ 8,
317
+ 9,
318
+ 10,
319
+ 11,
320
+ 12,
321
+ 13,
322
+ 14,
323
+ 15,
324
+ 16,
325
+ 17,
326
+ 18,
327
+ 19,
328
+ 20,
329
+ 21,
330
+ 22,
331
+ 23,
332
+ 24,
333
+ 25,
334
+ 26,
335
+ 27,
336
+ 28,
337
+ 29,
338
+ 30,
339
+ 31,
340
+ 32,
341
+ 33,
342
+ 34,
343
+ 35,
344
+ 36,
345
+ 37,
346
+ 38,
347
+ 39
348
+ ],
349
+ "modality_keys": [
350
+ "left_wrist_eef",
351
+ "right_wrist_eef",
352
+ "left_hand_joints",
353
+ "right_hand_joints"
354
+ ],
355
+ "sin_cos_embedding_keys": null,
356
+ "mean_std_embedding_keys": null,
357
+ "action_configs": [
358
+ {
359
+ "rep": "RELATIVE",
360
+ "type": "EEF",
361
+ "format": "XYZ_ROT6D",
362
+ "state_key": "left_wrist_eef"
363
+ },
364
+ {
365
+ "rep": "RELATIVE",
366
+ "type": "EEF",
367
+ "format": "XYZ_ROT6D",
368
+ "state_key": "right_wrist_eef"
369
+ },
370
+ {
371
+ "rep": "ABSOLUTE",
372
+ "type": "NON_EEF",
373
+ "format": "DEFAULT",
374
+ "state_key": "left_hand_joints"
375
+ },
376
+ {
377
+ "rep": "ABSOLUTE",
378
+ "type": "NON_EEF",
379
+ "format": "DEFAULT",
380
+ "state_key": "right_hand_joints"
381
+ }
382
+ ]
383
+ },
384
+ "language": {
385
+ "delta_indices": [
386
+ 0
387
+ ],
388
+ "modality_keys": [
389
+ "annotation.human.coarse_action"
390
+ ],
391
+ "sin_cos_embedding_keys": null,
392
+ "mean_std_embedding_keys": null,
393
+ "action_configs": null
394
+ }
395
+ },
396
+ "real_r1_pro_sharpa_relative_eef": {
397
+ "video": {
398
+ "delta_indices": [
399
+ -20,
400
+ 0
401
+ ],
402
+ "modality_keys": [
403
+ "ego_view_res320x240_freq20",
404
+ "left_wrist_view_res320x240_freq20",
405
+ "right_wrist_view_res320x240_freq20"
406
+ ],
407
+ "sin_cos_embedding_keys": null,
408
+ "mean_std_embedding_keys": null,
409
+ "action_configs": null
410
+ },
411
+ "state": {
412
+ "delta_indices": [
413
+ 0
414
+ ],
415
+ "modality_keys": [
416
+ "left_wrist_eef",
417
+ "right_wrist_eef",
418
+ "left_hand_joints",
419
+ "right_hand_joints"
420
+ ],
421
+ "sin_cos_embedding_keys": null,
422
+ "mean_std_embedding_keys": null,
423
+ "action_configs": null
424
+ },
425
+ "action": {
426
+ "delta_indices": [
427
+ 0,
428
+ 1,
429
+ 2,
430
+ 3,
431
+ 4,
432
+ 5,
433
+ 6,
434
+ 7,
435
+ 8,
436
+ 9,
437
+ 10,
438
+ 11,
439
+ 12,
440
+ 13,
441
+ 14,
442
+ 15,
443
+ 16,
444
+ 17,
445
+ 18,
446
+ 19,
447
+ 20,
448
+ 21,
449
+ 22,
450
+ 23,
451
+ 24,
452
+ 25,
453
+ 26,
454
+ 27,
455
+ 28,
456
+ 29,
457
+ 30,
458
+ 31,
459
+ 32,
460
+ 33,
461
+ 34,
462
+ 35,
463
+ 36,
464
+ 37,
465
+ 38,
466
+ 39
467
+ ],
468
+ "modality_keys": [
469
+ "left_wrist_eef",
470
+ "right_wrist_eef",
471
+ "left_hand_joints",
472
+ "right_hand_joints"
473
+ ],
474
+ "sin_cos_embedding_keys": null,
475
+ "mean_std_embedding_keys": null,
476
+ "action_configs": [
477
+ {
478
+ "rep": "RELATIVE",
479
+ "type": "EEF",
480
+ "format": "XYZ_ROT6D",
481
+ "state_key": "left_wrist_eef"
482
+ },
483
+ {
484
+ "rep": "RELATIVE",
485
+ "type": "EEF",
486
+ "format": "XYZ_ROT6D",
487
+ "state_key": "right_wrist_eef"
488
+ },
489
+ {
490
+ "rep": "ABSOLUTE",
491
+ "type": "NON_EEF",
492
+ "format": "DEFAULT",
493
+ "state_key": "left_hand_joints"
494
+ },
495
+ {
496
+ "rep": "ABSOLUTE",
497
+ "type": "NON_EEF",
498
+ "format": "DEFAULT",
499
+ "state_key": "right_hand_joints"
500
+ }
501
+ ]
502
+ },
503
+ "language": {
504
+ "delta_indices": [
505
+ 0
506
+ ],
507
+ "modality_keys": [
508
+ "annotation.human.coarse_action"
509
+ ],
510
+ "sin_cos_embedding_keys": null,
511
+ "mean_std_embedding_keys": null,
512
+ "action_configs": null
513
+ }
514
+ },
515
+ "xdof_relative_eef_relative_joint": {
516
+ "video": {
517
+ "delta_indices": [
518
+ -30,
519
+ 0
520
+ ],
521
+ "modality_keys": [
522
+ "top_camera-images-rgb_320_240",
523
+ "left_camera-images-rgb_320_240",
524
+ "right_camera-images-rgb_320_240"
525
+ ],
526
+ "sin_cos_embedding_keys": null,
527
+ "mean_std_embedding_keys": null,
528
+ "action_configs": null
529
+ },
530
+ "state": {
531
+ "delta_indices": [
532
+ 0
533
+ ],
534
+ "modality_keys": [
535
+ "left_wrist_eef",
536
+ "right_wrist_eef",
537
+ "left_gripper_pos",
538
+ "right_gripper_pos",
539
+ "left_joint_pos",
540
+ "right_joint_pos"
541
+ ],
542
+ "sin_cos_embedding_keys": null,
543
+ "mean_std_embedding_keys": null,
544
+ "action_configs": null
545
+ },
546
+ "action": {
547
+ "delta_indices": [
548
+ 0,
549
+ 1,
550
+ 2,
551
+ 3,
552
+ 4,
553
+ 5,
554
+ 6,
555
+ 7,
556
+ 8,
557
+ 9,
558
+ 10,
559
+ 11,
560
+ 12,
561
+ 13,
562
+ 14,
563
+ 15,
564
+ 16,
565
+ 17,
566
+ 18,
567
+ 19,
568
+ 20,
569
+ 21,
570
+ 22,
571
+ 23,
572
+ 24,
573
+ 25,
574
+ 26,
575
+ 27,
576
+ 28,
577
+ 29,
578
+ 30,
579
+ 31,
580
+ 32,
581
+ 33,
582
+ 34,
583
+ 35,
584
+ 36,
585
+ 37,
586
+ 38,
587
+ 39
588
+ ],
589
+ "modality_keys": [
590
+ "left_wrist_eef",
591
+ "right_wrist_eef",
592
+ "left_gripper_pos",
593
+ "right_gripper_pos",
594
+ "left_joint_pos",
595
+ "right_joint_pos"
596
+ ],
597
+ "sin_cos_embedding_keys": null,
598
+ "mean_std_embedding_keys": null,
599
+ "action_configs": [
600
+ {
601
+ "rep": "RELATIVE",
602
+ "type": "EEF",
603
+ "format": "XYZ_ROT6D",
604
+ "state_key": "left_wrist_eef"
605
+ },
606
+ {
607
+ "rep": "RELATIVE",
608
+ "type": "EEF",
609
+ "format": "XYZ_ROT6D",
610
+ "state_key": "right_wrist_eef"
611
+ },
612
+ {
613
+ "rep": "ABSOLUTE",
614
+ "type": "NON_EEF",
615
+ "format": "DEFAULT",
616
+ "state_key": "left_gripper_pos"
617
+ },
618
+ {
619
+ "rep": "ABSOLUTE",
620
+ "type": "NON_EEF",
621
+ "format": "DEFAULT",
622
+ "state_key": "right_gripper_pos"
623
+ },
624
+ {
625
+ "rep": "RELATIVE",
626
+ "type": "NON_EEF",
627
+ "format": "DEFAULT",
628
+ "state_key": "left_joint_pos"
629
+ },
630
+ {
631
+ "rep": "RELATIVE",
632
+ "type": "NON_EEF",
633
+ "format": "DEFAULT",
634
+ "state_key": "right_joint_pos"
635
+ }
636
+ ]
637
+ },
638
+ "language": {
639
+ "delta_indices": [
640
+ 0
641
+ ],
642
+ "modality_keys": [
643
+ "annotation.task"
644
+ ],
645
+ "sin_cos_embedding_keys": null,
646
+ "mean_std_embedding_keys": null,
647
+ "action_configs": null
648
+ }
649
+ },
650
+ "real_r1_pro_sharpa_relative_eef_maxinsights": {
651
+ "video": {
652
+ "delta_indices": [
653
+ -30,
654
+ 0
655
+ ],
656
+ "modality_keys": [
657
+ "ego_view_cropratio_res320x240_freq30"
658
+ ],
659
+ "sin_cos_embedding_keys": null,
660
+ "mean_std_embedding_keys": null,
661
+ "action_configs": null
662
+ },
663
+ "state": {
664
+ "delta_indices": [
665
+ 0
666
+ ],
667
+ "modality_keys": [
668
+ "left_wrist_eef",
669
+ "right_wrist_eef",
670
+ "left_hand_joints",
671
+ "right_hand_joints"
672
+ ],
673
+ "sin_cos_embedding_keys": null,
674
+ "mean_std_embedding_keys": null,
675
+ "action_configs": null
676
+ },
677
+ "action": {
678
+ "delta_indices": [
679
+ 0,
680
+ 1,
681
+ 2,
682
+ 3,
683
+ 4,
684
+ 5,
685
+ 6,
686
+ 7,
687
+ 8,
688
+ 9,
689
+ 10,
690
+ 11,
691
+ 12,
692
+ 13,
693
+ 14,
694
+ 15,
695
+ 16,
696
+ 17,
697
+ 18,
698
+ 19,
699
+ 20,
700
+ 21,
701
+ 22,
702
+ 23,
703
+ 24,
704
+ 25,
705
+ 26,
706
+ 27,
707
+ 28,
708
+ 29,
709
+ 30,
710
+ 31,
711
+ 32,
712
+ 33,
713
+ 34,
714
+ 35,
715
+ 36,
716
+ 37,
717
+ 38,
718
+ 39
719
+ ],
720
+ "modality_keys": [
721
+ "left_wrist_eef",
722
+ "right_wrist_eef",
723
+ "left_hand_joints",
724
+ "right_hand_joints"
725
+ ],
726
+ "sin_cos_embedding_keys": null,
727
+ "mean_std_embedding_keys": null,
728
+ "action_configs": [
729
+ {
730
+ "rep": "RELATIVE",
731
+ "type": "EEF",
732
+ "format": "XYZ_ROT6D",
733
+ "state_key": "left_wrist_eef"
734
+ },
735
+ {
736
+ "rep": "RELATIVE",
737
+ "type": "EEF",
738
+ "format": "XYZ_ROT6D",
739
+ "state_key": "right_wrist_eef"
740
+ },
741
+ {
742
+ "rep": "ABSOLUTE",
743
+ "type": "NON_EEF",
744
+ "format": "DEFAULT",
745
+ "state_key": "left_hand_joints"
746
+ },
747
+ {
748
+ "rep": "ABSOLUTE",
749
+ "type": "NON_EEF",
750
+ "format": "DEFAULT",
751
+ "state_key": "right_hand_joints"
752
+ }
753
+ ]
754
+ },
755
+ "language": {
756
+ "delta_indices": [
757
+ 0
758
+ ],
759
+ "modality_keys": [
760
+ "annotation.human.coarse_action"
761
+ ],
762
+ "sin_cos_embedding_keys": null,
763
+ "mean_std_embedding_keys": null,
764
+ "action_configs": null
765
+ }
766
+ },
767
+ "xdof_relative_eef_relative_joint_subtask": {
768
+ "video": {
769
+ "delta_indices": [
770
+ -30,
771
+ 0
772
+ ],
773
+ "modality_keys": [
774
+ "top_camera-images-rgb_320_240",
775
+ "left_camera-images-rgb_320_240",
776
+ "right_camera-images-rgb_320_240"
777
+ ],
778
+ "sin_cos_embedding_keys": null,
779
+ "mean_std_embedding_keys": null,
780
+ "action_configs": null
781
+ },
782
+ "state": {
783
+ "delta_indices": [
784
+ 0
785
+ ],
786
+ "modality_keys": [
787
+ "left_wrist_eef",
788
+ "right_wrist_eef",
789
+ "left_gripper_pos",
790
+ "right_gripper_pos",
791
+ "left_joint_pos",
792
+ "right_joint_pos"
793
+ ],
794
+ "sin_cos_embedding_keys": null,
795
+ "mean_std_embedding_keys": null,
796
+ "action_configs": null
797
+ },
798
+ "action": {
799
+ "delta_indices": [
800
+ 0,
801
+ 1,
802
+ 2,
803
+ 3,
804
+ 4,
805
+ 5,
806
+ 6,
807
+ 7,
808
+ 8,
809
+ 9,
810
+ 10,
811
+ 11,
812
+ 12,
813
+ 13,
814
+ 14,
815
+ 15,
816
+ 16,
817
+ 17,
818
+ 18,
819
+ 19,
820
+ 20,
821
+ 21,
822
+ 22,
823
+ 23,
824
+ 24,
825
+ 25,
826
+ 26,
827
+ 27,
828
+ 28,
829
+ 29,
830
+ 30,
831
+ 31,
832
+ 32,
833
+ 33,
834
+ 34,
835
+ 35,
836
+ 36,
837
+ 37,
838
+ 38,
839
+ 39
840
+ ],
841
+ "modality_keys": [
842
+ "left_wrist_eef",
843
+ "right_wrist_eef",
844
+ "left_gripper_pos",
845
+ "right_gripper_pos",
846
+ "left_joint_pos",
847
+ "right_joint_pos"
848
+ ],
849
+ "sin_cos_embedding_keys": null,
850
+ "mean_std_embedding_keys": null,
851
+ "action_configs": [
852
+ {
853
+ "rep": "RELATIVE",
854
+ "type": "EEF",
855
+ "format": "XYZ_ROT6D",
856
+ "state_key": "left_wrist_eef"
857
+ },
858
+ {
859
+ "rep": "RELATIVE",
860
+ "type": "EEF",
861
+ "format": "XYZ_ROT6D",
862
+ "state_key": "right_wrist_eef"
863
+ },
864
+ {
865
+ "rep": "ABSOLUTE",
866
+ "type": "NON_EEF",
867
+ "format": "DEFAULT",
868
+ "state_key": "left_gripper_pos"
869
+ },
870
+ {
871
+ "rep": "ABSOLUTE",
872
+ "type": "NON_EEF",
873
+ "format": "DEFAULT",
874
+ "state_key": "right_gripper_pos"
875
+ },
876
+ {
877
+ "rep": "RELATIVE",
878
+ "type": "NON_EEF",
879
+ "format": "DEFAULT",
880
+ "state_key": "left_joint_pos"
881
+ },
882
+ {
883
+ "rep": "RELATIVE",
884
+ "type": "NON_EEF",
885
+ "format": "DEFAULT",
886
+ "state_key": "right_joint_pos"
887
+ }
888
+ ]
889
+ },
890
+ "language": {
891
+ "delta_indices": [
892
+ 0
893
+ ],
894
+ "modality_keys": [
895
+ "annotation.sub_task"
896
+ ],
897
+ "sin_cos_embedding_keys": null,
898
+ "mean_std_embedding_keys": null,
899
+ "action_configs": null
900
+ }
901
+ },
902
+ "oxe_droid_relative_eef_relative_joint": {
903
+ "video": {
904
+ "delta_indices": [
905
+ -15,
906
+ 0
907
+ ],
908
+ "modality_keys": [
909
+ "exterior_image_1_left",
910
+ "wrist_image_left"
911
+ ],
912
+ "sin_cos_embedding_keys": null,
913
+ "mean_std_embedding_keys": null,
914
+ "action_configs": null
915
+ },
916
+ "state": {
917
+ "delta_indices": [
918
+ 0
919
+ ],
920
+ "modality_keys": [
921
+ "eef_9d",
922
+ "gripper_position",
923
+ "joint_position"
924
+ ],
925
+ "sin_cos_embedding_keys": null,
926
+ "mean_std_embedding_keys": null,
927
+ "action_configs": null
928
+ },
929
+ "action": {
930
+ "delta_indices": [
931
+ 0,
932
+ 1,
933
+ 2,
934
+ 3,
935
+ 4,
936
+ 5,
937
+ 6,
938
+ 7,
939
+ 8,
940
+ 9,
941
+ 10,
942
+ 11,
943
+ 12,
944
+ 13,
945
+ 14,
946
+ 15,
947
+ 16,
948
+ 17,
949
+ 18,
950
+ 19,
951
+ 20,
952
+ 21,
953
+ 22,
954
+ 23,
955
+ 24,
956
+ 25,
957
+ 26,
958
+ 27,
959
+ 28,
960
+ 29,
961
+ 30,
962
+ 31,
963
+ 32,
964
+ 33,
965
+ 34,
966
+ 35,
967
+ 36,
968
+ 37,
969
+ 38,
970
+ 39
971
+ ],
972
+ "modality_keys": [
973
+ "eef_9d",
974
+ "gripper_position",
975
+ "joint_position"
976
+ ],
977
+ "sin_cos_embedding_keys": null,
978
+ "mean_std_embedding_keys": null,
979
+ "action_configs": [
980
+ {
981
+ "rep": "RELATIVE",
982
+ "type": "EEF",
983
+ "format": "XYZ_ROT6D",
984
+ "state_key": "eef_9d"
985
+ },
986
+ {
987
+ "rep": "ABSOLUTE",
988
+ "type": "NON_EEF",
989
+ "format": "DEFAULT",
990
+ "state_key": "gripper_position"
991
+ },
992
+ {
993
+ "rep": "RELATIVE",
994
+ "type": "NON_EEF",
995
+ "format": "DEFAULT",
996
+ "state_key": "joint_position"
997
+ }
998
+ ]
999
+ },
1000
+ "language": {
1001
+ "delta_indices": [
1002
+ 0
1003
+ ],
1004
+ "modality_keys": [
1005
+ "annotation.language.language_instruction"
1006
+ ],
1007
+ "sin_cos_embedding_keys": null,
1008
+ "mean_std_embedding_keys": null,
1009
+ "action_configs": null
1010
+ }
1011
+ },
1012
+ "g1_29dof": {
1013
+ "video": {
1014
+ "delta_indices": [
1015
+ 0
1016
+ ],
1017
+ "modality_keys": [
1018
+ "ego_view"
1019
+ ],
1020
+ "sin_cos_embedding_keys": null,
1021
+ "mean_std_embedding_keys": null,
1022
+ "action_configs": null
1023
+ },
1024
+ "state": {
1025
+ "delta_indices": [
1026
+ 0
1027
+ ],
1028
+ "modality_keys": [
1029
+ "left_leg",
1030
+ "right_leg",
1031
+ "waist",
1032
+ "left_arm",
1033
+ "left_hand",
1034
+ "right_arm",
1035
+ "right_hand",
1036
+ "projected_gravity"
1037
+ ],
1038
+ "sin_cos_embedding_keys": null,
1039
+ "mean_std_embedding_keys": null,
1040
+ "action_configs": null
1041
+ },
1042
+ "action": {
1043
+ "delta_indices": [
1044
+ 0,
1045
+ 1,
1046
+ 2,
1047
+ 3,
1048
+ 4,
1049
+ 5,
1050
+ 6,
1051
+ 7,
1052
+ 8,
1053
+ 9,
1054
+ 10,
1055
+ 11,
1056
+ 12,
1057
+ 13,
1058
+ 14,
1059
+ 15,
1060
+ 16,
1061
+ 17,
1062
+ 18,
1063
+ 19,
1064
+ 20,
1065
+ 21,
1066
+ 22,
1067
+ 23,
1068
+ 24,
1069
+ 25,
1070
+ 26,
1071
+ 27,
1072
+ 28,
1073
+ 29,
1074
+ 30,
1075
+ 31,
1076
+ 32,
1077
+ 33,
1078
+ 34,
1079
+ 35,
1080
+ 36,
1081
+ 37,
1082
+ 38,
1083
+ 39
1084
+ ],
1085
+ "modality_keys": [
1086
+ "motion_token",
1087
+ "left_hand_joints",
1088
+ "right_hand_joints"
1089
+ ],
1090
+ "sin_cos_embedding_keys": null,
1091
+ "mean_std_embedding_keys": null,
1092
+ "action_configs": [
1093
+ {
1094
+ "rep": "ABSOLUTE",
1095
+ "type": "NON_EEF",
1096
+ "format": "DEFAULT",
1097
+ "state_key": null
1098
+ },
1099
+ {
1100
+ "rep": "ABSOLUTE",
1101
+ "type": "NON_EEF",
1102
+ "format": "DEFAULT",
1103
+ "state_key": null
1104
+ },
1105
+ {
1106
+ "rep": "ABSOLUTE",
1107
+ "type": "NON_EEF",
1108
+ "format": "DEFAULT",
1109
+ "state_key": null
1110
+ }
1111
+ ]
1112
+ },
1113
+ "language": {
1114
+ "delta_indices": [
1115
+ 0
1116
+ ],
1117
+ "modality_keys": [
1118
+ "annotation.human.task_description"
1119
+ ],
1120
+ "sin_cos_embedding_keys": null,
1121
+ "mean_std_embedding_keys": null,
1122
+ "action_configs": null
1123
+ }
1124
+ }
1125
+ },
1126
+ "image_crop_size": [
1127
+ 230,
1128
+ 230
1129
+ ],
1130
+ "image_target_size": [
1131
+ 256,
1132
+ 256
1133
+ ],
1134
+ "use_albumentations": true,
1135
+ "random_rotation_angle": 0,
1136
+ "color_jitter_params": {
1137
+ "brightness": 0.3,
1138
+ "contrast": 0.4,
1139
+ "saturation": 0.5,
1140
+ "hue": 0.08
1141
+ },
1142
+ "shortest_image_edge": 256,
1143
+ "crop_fraction": 0.95,
1144
+ "letter_box_transform": false,
1145
+ "model_name": "nvidia/Cosmos-Reason2-2B",
1146
+ "model_type": "qwen",
1147
+ "formalize_language": true,
1148
+ "max_state_dim": 132,
1149
+ "max_action_dim": 132,
1150
+ "max_action_horizon": 40,
1151
+ "use_percentiles": true,
1152
+ "use_mean_std": false,
1153
+ "clip_outliers": true,
1154
+ "apply_sincos_state_encoding": false,
1155
+ "use_relative_action": true,
1156
+ "exclude_state": false,
1157
+ "state_dropout_prob": 0.2
1158
+ }
1159
+ }
rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:15b44942714a3996ffc43833a460457cf1073f8b7749cd17d5f6708236b29b37
3
+ size 14917
rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aab70e504f8d1c00b61c138a7f4a2da6de87ea9148f4fe20a775139bc4888b43
3
+ size 14917
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c4691914aee4c32e4f8aab72151088c2579d2e5f36d40120e3845612c8dae1f0
3
+ size 1465
statistics.json ADDED
The diff for this file is too large to render. See raw diff
 
trainer_state.json ADDED
@@ -0,0 +1,1234 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 0.4,
6
+ "eval_steps": 500,
7
+ "global_step": 2000,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "grad_norm": 0.12638650834560394,
14
+ "learning_rate": 3.6e-06,
15
+ "loss": 1.1641,
16
+ "step": 10
17
+ },
18
+ {
19
+ "grad_norm": 0.1471242904663086,
20
+ "learning_rate": 7.6e-06,
21
+ "loss": 1.1535,
22
+ "step": 20
23
+ },
24
+ {
25
+ "grad_norm": 0.18666379153728485,
26
+ "learning_rate": 1.16e-05,
27
+ "loss": 1.1457,
28
+ "step": 30
29
+ },
30
+ {
31
+ "grad_norm": 0.14849230647087097,
32
+ "learning_rate": 1.56e-05,
33
+ "loss": 1.1223,
34
+ "step": 40
35
+ },
36
+ {
37
+ "grad_norm": 0.27684518694877625,
38
+ "learning_rate": 1.9600000000000002e-05,
39
+ "loss": 1.0957,
40
+ "step": 50
41
+ },
42
+ {
43
+ "grad_norm": 0.2600708305835724,
44
+ "learning_rate": 2.36e-05,
45
+ "loss": 1.0719,
46
+ "step": 60
47
+ },
48
+ {
49
+ "grad_norm": 0.5172604322433472,
50
+ "learning_rate": 2.7600000000000003e-05,
51
+ "loss": 1.0266,
52
+ "step": 70
53
+ },
54
+ {
55
+ "grad_norm": 0.44987690448760986,
56
+ "learning_rate": 3.16e-05,
57
+ "loss": 0.9682,
58
+ "step": 80
59
+ },
60
+ {
61
+ "grad_norm": 0.7133139967918396,
62
+ "learning_rate": 3.56e-05,
63
+ "loss": 0.9043,
64
+ "step": 90
65
+ },
66
+ {
67
+ "grad_norm": 0.652862548828125,
68
+ "learning_rate": 3.960000000000001e-05,
69
+ "loss": 0.8439,
70
+ "step": 100
71
+ },
72
+ {
73
+ "grad_norm": 0.6717215180397034,
74
+ "learning_rate": 4.36e-05,
75
+ "loss": 0.7908,
76
+ "step": 110
77
+ },
78
+ {
79
+ "grad_norm": 0.8351187109947205,
80
+ "learning_rate": 4.76e-05,
81
+ "loss": 0.7508,
82
+ "step": 120
83
+ },
84
+ {
85
+ "grad_norm": 0.951558530330658,
86
+ "learning_rate": 5.16e-05,
87
+ "loss": 0.7117,
88
+ "step": 130
89
+ },
90
+ {
91
+ "grad_norm": 0.7069472074508667,
92
+ "learning_rate": 5.560000000000001e-05,
93
+ "loss": 0.6824,
94
+ "step": 140
95
+ },
96
+ {
97
+ "grad_norm": 0.8377196788787842,
98
+ "learning_rate": 5.96e-05,
99
+ "loss": 0.6523,
100
+ "step": 150
101
+ },
102
+ {
103
+ "grad_norm": 0.7963162064552307,
104
+ "learning_rate": 6.36e-05,
105
+ "loss": 0.6193,
106
+ "step": 160
107
+ },
108
+ {
109
+ "grad_norm": 1.0922210216522217,
110
+ "learning_rate": 6.76e-05,
111
+ "loss": 0.5893,
112
+ "step": 170
113
+ },
114
+ {
115
+ "grad_norm": 1.2284756898880005,
116
+ "learning_rate": 7.16e-05,
117
+ "loss": 0.5648,
118
+ "step": 180
119
+ },
120
+ {
121
+ "grad_norm": 0.8342697024345398,
122
+ "learning_rate": 7.560000000000001e-05,
123
+ "loss": 0.5453,
124
+ "step": 190
125
+ },
126
+ {
127
+ "grad_norm": 1.121726155281067,
128
+ "learning_rate": 7.960000000000001e-05,
129
+ "loss": 0.5277,
130
+ "step": 200
131
+ },
132
+ {
133
+ "grad_norm": 0.8587872982025146,
134
+ "learning_rate": 8.36e-05,
135
+ "loss": 0.5082,
136
+ "step": 210
137
+ },
138
+ {
139
+ "grad_norm": 1.2177629470825195,
140
+ "learning_rate": 8.76e-05,
141
+ "loss": 0.4881,
142
+ "step": 220
143
+ },
144
+ {
145
+ "grad_norm": 1.0331931114196777,
146
+ "learning_rate": 9.16e-05,
147
+ "loss": 0.467,
148
+ "step": 230
149
+ },
150
+ {
151
+ "grad_norm": 1.178455114364624,
152
+ "learning_rate": 9.56e-05,
153
+ "loss": 0.4453,
154
+ "step": 240
155
+ },
156
+ {
157
+ "grad_norm": 1.0000393390655518,
158
+ "learning_rate": 9.960000000000001e-05,
159
+ "loss": 0.4235,
160
+ "step": 250
161
+ },
162
+ {
163
+ "grad_norm": 1.2453455924987793,
164
+ "learning_rate": 9.999911419878559e-05,
165
+ "loss": 0.4034,
166
+ "step": 260
167
+ },
168
+ {
169
+ "grad_norm": 1.474942684173584,
170
+ "learning_rate": 9.999605221019081e-05,
171
+ "loss": 0.3801,
172
+ "step": 270
173
+ },
174
+ {
175
+ "grad_norm": 0.9545984864234924,
176
+ "learning_rate": 9.999080323230761e-05,
177
+ "loss": 0.3551,
178
+ "step": 280
179
+ },
180
+ {
181
+ "grad_norm": 1.0123811960220337,
182
+ "learning_rate": 9.998336749474329e-05,
183
+ "loss": 0.329,
184
+ "step": 290
185
+ },
186
+ {
187
+ "grad_norm": 1.214898705482483,
188
+ "learning_rate": 9.997374532276107e-05,
189
+ "loss": 0.3039,
190
+ "step": 300
191
+ },
192
+ {
193
+ "grad_norm": 1.2700718641281128,
194
+ "learning_rate": 9.996193713726596e-05,
195
+ "loss": 0.2804,
196
+ "step": 310
197
+ },
198
+ {
199
+ "grad_norm": 1.0745455026626587,
200
+ "learning_rate": 9.994794345478624e-05,
201
+ "loss": 0.2604,
202
+ "step": 320
203
+ },
204
+ {
205
+ "grad_norm": 1.0220052003860474,
206
+ "learning_rate": 9.99317648874509e-05,
207
+ "loss": 0.2303,
208
+ "step": 330
209
+ },
210
+ {
211
+ "grad_norm": 0.9410139918327332,
212
+ "learning_rate": 9.991340214296292e-05,
213
+ "loss": 0.2094,
214
+ "step": 340
215
+ },
216
+ {
217
+ "grad_norm": 1.046053171157837,
218
+ "learning_rate": 9.989285602456819e-05,
219
+ "loss": 0.1971,
220
+ "step": 350
221
+ },
222
+ {
223
+ "grad_norm": 1.174741268157959,
224
+ "learning_rate": 9.98701274310205e-05,
225
+ "loss": 0.1773,
226
+ "step": 360
227
+ },
228
+ {
229
+ "grad_norm": 1.0372276306152344,
230
+ "learning_rate": 9.984521735654218e-05,
231
+ "loss": 0.1615,
232
+ "step": 370
233
+ },
234
+ {
235
+ "grad_norm": 1.0307488441467285,
236
+ "learning_rate": 9.981812689078057e-05,
237
+ "loss": 0.1429,
238
+ "step": 380
239
+ },
240
+ {
241
+ "grad_norm": 2.1818134784698486,
242
+ "learning_rate": 9.978885721876041e-05,
243
+ "loss": 0.1284,
244
+ "step": 390
245
+ },
246
+ {
247
+ "grad_norm": 0.9964090585708618,
248
+ "learning_rate": 9.975740962083198e-05,
249
+ "loss": 0.1258,
250
+ "step": 400
251
+ },
252
+ {
253
+ "grad_norm": 0.9651419520378113,
254
+ "learning_rate": 9.972378547261504e-05,
255
+ "loss": 0.1093,
256
+ "step": 410
257
+ },
258
+ {
259
+ "grad_norm": 1.0263780355453491,
260
+ "learning_rate": 9.968798624493885e-05,
261
+ "loss": 0.1062,
262
+ "step": 420
263
+ },
264
+ {
265
+ "grad_norm": 0.9092751741409302,
266
+ "learning_rate": 9.965001350377753e-05,
267
+ "loss": 0.1009,
268
+ "step": 430
269
+ },
270
+ {
271
+ "grad_norm": 1.1471643447875977,
272
+ "learning_rate": 9.960986891018183e-05,
273
+ "loss": 0.0967,
274
+ "step": 440
275
+ },
276
+ {
277
+ "grad_norm": 0.95041823387146,
278
+ "learning_rate": 9.95675542202063e-05,
279
+ "loss": 0.0941,
280
+ "step": 450
281
+ },
282
+ {
283
+ "grad_norm": 0.9787719249725342,
284
+ "learning_rate": 9.952307128483256e-05,
285
+ "loss": 0.0942,
286
+ "step": 460
287
+ },
288
+ {
289
+ "grad_norm": 0.9564359188079834,
290
+ "learning_rate": 9.947642204988835e-05,
291
+ "loss": 0.096,
292
+ "step": 470
293
+ },
294
+ {
295
+ "grad_norm": 0.9926328659057617,
296
+ "learning_rate": 9.942760855596226e-05,
297
+ "loss": 0.0931,
298
+ "step": 480
299
+ },
300
+ {
301
+ "grad_norm": 0.8731317520141602,
302
+ "learning_rate": 9.937663293831471e-05,
303
+ "loss": 0.0901,
304
+ "step": 490
305
+ },
306
+ {
307
+ "grad_norm": 1.1485458612442017,
308
+ "learning_rate": 9.932349742678433e-05,
309
+ "loss": 0.0875,
310
+ "step": 500
311
+ },
312
+ {
313
+ "grad_norm": 0.9461956024169922,
314
+ "learning_rate": 9.926820434569051e-05,
315
+ "loss": 0.0892,
316
+ "step": 510
317
+ },
318
+ {
319
+ "grad_norm": 0.8998910784721375,
320
+ "learning_rate": 9.921075611373179e-05,
321
+ "loss": 0.0853,
322
+ "step": 520
323
+ },
324
+ {
325
+ "grad_norm": 1.0842655897140503,
326
+ "learning_rate": 9.915115524387988e-05,
327
+ "loss": 0.0841,
328
+ "step": 530
329
+ },
330
+ {
331
+ "grad_norm": 1.0050004720687866,
332
+ "learning_rate": 9.908940434326997e-05,
333
+ "loss": 0.0864,
334
+ "step": 540
335
+ },
336
+ {
337
+ "grad_norm": 1.1228138208389282,
338
+ "learning_rate": 9.902550611308645e-05,
339
+ "loss": 0.0839,
340
+ "step": 550
341
+ },
342
+ {
343
+ "grad_norm": 0.8371630907058716,
344
+ "learning_rate": 9.895946334844494e-05,
345
+ "loss": 0.087,
346
+ "step": 560
347
+ },
348
+ {
349
+ "grad_norm": 0.9920855760574341,
350
+ "learning_rate": 9.889127893826989e-05,
351
+ "loss": 0.0833,
352
+ "step": 570
353
+ },
354
+ {
355
+ "grad_norm": 0.8540536165237427,
356
+ "learning_rate": 9.882095586516831e-05,
357
+ "loss": 0.0808,
358
+ "step": 580
359
+ },
360
+ {
361
+ "grad_norm": 1.0284672975540161,
362
+ "learning_rate": 9.874849720529921e-05,
363
+ "loss": 0.0778,
364
+ "step": 590
365
+ },
366
+ {
367
+ "grad_norm": 0.7409518361091614,
368
+ "learning_rate": 9.867390612823914e-05,
369
+ "loss": 0.0776,
370
+ "step": 600
371
+ },
372
+ {
373
+ "grad_norm": 0.7265949249267578,
374
+ "learning_rate": 9.859718589684344e-05,
375
+ "loss": 0.0775,
376
+ "step": 610
377
+ },
378
+ {
379
+ "grad_norm": 0.8303801417350769,
380
+ "learning_rate": 9.851833986710353e-05,
381
+ "loss": 0.0766,
382
+ "step": 620
383
+ },
384
+ {
385
+ "grad_norm": 0.846651017665863,
386
+ "learning_rate": 9.843737148800023e-05,
387
+ "loss": 0.0737,
388
+ "step": 630
389
+ },
390
+ {
391
+ "grad_norm": 0.8616825342178345,
392
+ "learning_rate": 9.835428430135271e-05,
393
+ "loss": 0.0761,
394
+ "step": 640
395
+ },
396
+ {
397
+ "grad_norm": 0.6212893724441528,
398
+ "learning_rate": 9.82690819416637e-05,
399
+ "loss": 0.0734,
400
+ "step": 650
401
+ },
402
+ {
403
+ "grad_norm": 0.950407862663269,
404
+ "learning_rate": 9.818176813596041e-05,
405
+ "loss": 0.0733,
406
+ "step": 660
407
+ },
408
+ {
409
+ "grad_norm": 0.8584088683128357,
410
+ "learning_rate": 9.809234670363159e-05,
411
+ "loss": 0.0729,
412
+ "step": 670
413
+ },
414
+ {
415
+ "grad_norm": 0.8692848682403564,
416
+ "learning_rate": 9.800082155626034e-05,
417
+ "loss": 0.0715,
418
+ "step": 680
419
+ },
420
+ {
421
+ "grad_norm": 0.813220202922821,
422
+ "learning_rate": 9.790719669745312e-05,
423
+ "loss": 0.0705,
424
+ "step": 690
425
+ },
426
+ {
427
+ "grad_norm": 0.9423041939735413,
428
+ "learning_rate": 9.781147622266455e-05,
429
+ "loss": 0.0696,
430
+ "step": 700
431
+ },
432
+ {
433
+ "grad_norm": 0.8096789121627808,
434
+ "learning_rate": 9.771366431901831e-05,
435
+ "loss": 0.0729,
436
+ "step": 710
437
+ },
438
+ {
439
+ "grad_norm": 0.949595034122467,
440
+ "learning_rate": 9.761376526512394e-05,
441
+ "loss": 0.0704,
442
+ "step": 720
443
+ },
444
+ {
445
+ "grad_norm": 0.8399985432624817,
446
+ "learning_rate": 9.751178343088963e-05,
447
+ "loss": 0.07,
448
+ "step": 730
449
+ },
450
+ {
451
+ "grad_norm": 0.708450198173523,
452
+ "learning_rate": 9.740772327733123e-05,
453
+ "loss": 0.0701,
454
+ "step": 740
455
+ },
456
+ {
457
+ "grad_norm": 0.8553755879402161,
458
+ "learning_rate": 9.730158935637697e-05,
459
+ "loss": 0.0696,
460
+ "step": 750
461
+ },
462
+ {
463
+ "grad_norm": 0.7043846845626831,
464
+ "learning_rate": 9.719338631066834e-05,
465
+ "loss": 0.0675,
466
+ "step": 760
467
+ },
468
+ {
469
+ "grad_norm": 0.8093163371086121,
470
+ "learning_rate": 9.708311887335713e-05,
471
+ "loss": 0.0656,
472
+ "step": 770
473
+ },
474
+ {
475
+ "grad_norm": 0.7058002352714539,
476
+ "learning_rate": 9.697079186789823e-05,
477
+ "loss": 0.0665,
478
+ "step": 780
479
+ },
480
+ {
481
+ "grad_norm": 0.6831558346748352,
482
+ "learning_rate": 9.685641020783876e-05,
483
+ "loss": 0.0641,
484
+ "step": 790
485
+ },
486
+ {
487
+ "grad_norm": 0.799716055393219,
488
+ "learning_rate": 9.67399788966031e-05,
489
+ "loss": 0.0627,
490
+ "step": 800
491
+ },
492
+ {
493
+ "grad_norm": 0.982138454914093,
494
+ "learning_rate": 9.662150302727395e-05,
495
+ "loss": 0.0637,
496
+ "step": 810
497
+ },
498
+ {
499
+ "grad_norm": 0.761420726776123,
500
+ "learning_rate": 9.650098778236968e-05,
501
+ "loss": 0.0653,
502
+ "step": 820
503
+ },
504
+ {
505
+ "grad_norm": 0.7208728790283203,
506
+ "learning_rate": 9.637843843361749e-05,
507
+ "loss": 0.0636,
508
+ "step": 830
509
+ },
510
+ {
511
+ "grad_norm": 0.8718505501747131,
512
+ "learning_rate": 9.62538603417229e-05,
513
+ "loss": 0.0619,
514
+ "step": 840
515
+ },
516
+ {
517
+ "grad_norm": 0.969348132610321,
518
+ "learning_rate": 9.612725895613526e-05,
519
+ "loss": 0.0615,
520
+ "step": 850
521
+ },
522
+ {
523
+ "grad_norm": 0.8718401789665222,
524
+ "learning_rate": 9.599863981480926e-05,
525
+ "loss": 0.0623,
526
+ "step": 860
527
+ },
528
+ {
529
+ "grad_norm": 0.6654447317123413,
530
+ "learning_rate": 9.586800854396283e-05,
531
+ "loss": 0.0608,
532
+ "step": 870
533
+ },
534
+ {
535
+ "grad_norm": 0.7654517889022827,
536
+ "learning_rate": 9.573537085783095e-05,
537
+ "loss": 0.0602,
538
+ "step": 880
539
+ },
540
+ {
541
+ "grad_norm": 0.8219175934791565,
542
+ "learning_rate": 9.560073255841571e-05,
543
+ "loss": 0.0594,
544
+ "step": 890
545
+ },
546
+ {
547
+ "grad_norm": 0.6851817965507507,
548
+ "learning_rate": 9.546409953523247e-05,
549
+ "loss": 0.0591,
550
+ "step": 900
551
+ },
552
+ {
553
+ "grad_norm": 0.7167078256607056,
554
+ "learning_rate": 9.532547776505229e-05,
555
+ "loss": 0.0574,
556
+ "step": 910
557
+ },
558
+ {
559
+ "grad_norm": 0.8798813819885254,
560
+ "learning_rate": 9.518487331164048e-05,
561
+ "loss": 0.0591,
562
+ "step": 920
563
+ },
564
+ {
565
+ "grad_norm": 0.5818827748298645,
566
+ "learning_rate": 9.504229232549134e-05,
567
+ "loss": 0.0574,
568
+ "step": 930
569
+ },
570
+ {
571
+ "grad_norm": 0.749271810054779,
572
+ "learning_rate": 9.489774104355909e-05,
573
+ "loss": 0.0574,
574
+ "step": 940
575
+ },
576
+ {
577
+ "grad_norm": 0.6833415031433105,
578
+ "learning_rate": 9.475122578898507e-05,
579
+ "loss": 0.0572,
580
+ "step": 950
581
+ },
582
+ {
583
+ "grad_norm": 0.6324511766433716,
584
+ "learning_rate": 9.460275297082119e-05,
585
+ "loss": 0.0554,
586
+ "step": 960
587
+ },
588
+ {
589
+ "grad_norm": 0.7919363975524902,
590
+ "learning_rate": 9.445232908374948e-05,
591
+ "loss": 0.0586,
592
+ "step": 970
593
+ },
594
+ {
595
+ "grad_norm": 0.6597112417221069,
596
+ "learning_rate": 9.429996070779808e-05,
597
+ "loss": 0.0561,
598
+ "step": 980
599
+ },
600
+ {
601
+ "grad_norm": 0.6228717565536499,
602
+ "learning_rate": 9.414565450805333e-05,
603
+ "loss": 0.0558,
604
+ "step": 990
605
+ },
606
+ {
607
+ "grad_norm": 0.7159019708633423,
608
+ "learning_rate": 9.398941723436831e-05,
609
+ "loss": 0.0575,
610
+ "step": 1000
611
+ },
612
+ {
613
+ "grad_norm": 0.8466086387634277,
614
+ "learning_rate": 9.383125572106752e-05,
615
+ "loss": 0.056,
616
+ "step": 1010
617
+ },
618
+ {
619
+ "grad_norm": 0.6202797293663025,
620
+ "learning_rate": 9.367117688664791e-05,
621
+ "loss": 0.0562,
622
+ "step": 1020
623
+ },
624
+ {
625
+ "grad_norm": 0.7398084402084351,
626
+ "learning_rate": 9.35091877334763e-05,
627
+ "loss": 0.0543,
628
+ "step": 1030
629
+ },
630
+ {
631
+ "grad_norm": 0.8202595114707947,
632
+ "learning_rate": 9.334529534748297e-05,
633
+ "loss": 0.0523,
634
+ "step": 1040
635
+ },
636
+ {
637
+ "grad_norm": 0.7353408932685852,
638
+ "learning_rate": 9.317950689785188e-05,
639
+ "loss": 0.0557,
640
+ "step": 1050
641
+ },
642
+ {
643
+ "grad_norm": 0.6730363368988037,
644
+ "learning_rate": 9.301182963670688e-05,
645
+ "loss": 0.0542,
646
+ "step": 1060
647
+ },
648
+ {
649
+ "grad_norm": 0.7827518582344055,
650
+ "learning_rate": 9.284227089879456e-05,
651
+ "loss": 0.0566,
652
+ "step": 1070
653
+ },
654
+ {
655
+ "grad_norm": 0.5198912024497986,
656
+ "learning_rate": 9.26708381011634e-05,
657
+ "loss": 0.0538,
658
+ "step": 1080
659
+ },
660
+ {
661
+ "grad_norm": 0.5192142128944397,
662
+ "learning_rate": 9.249753874283937e-05,
663
+ "loss": 0.0521,
664
+ "step": 1090
665
+ },
666
+ {
667
+ "grad_norm": 0.6771655678749084,
668
+ "learning_rate": 9.232238040449779e-05,
669
+ "loss": 0.0524,
670
+ "step": 1100
671
+ },
672
+ {
673
+ "grad_norm": 0.6441646218299866,
674
+ "learning_rate": 9.214537074813181e-05,
675
+ "loss": 0.0531,
676
+ "step": 1110
677
+ },
678
+ {
679
+ "grad_norm": 0.7039899826049805,
680
+ "learning_rate": 9.196651751671724e-05,
681
+ "loss": 0.054,
682
+ "step": 1120
683
+ },
684
+ {
685
+ "grad_norm": 0.7219718098640442,
686
+ "learning_rate": 9.178582853387384e-05,
687
+ "loss": 0.0522,
688
+ "step": 1130
689
+ },
690
+ {
691
+ "grad_norm": 0.5394071340560913,
692
+ "learning_rate": 9.160331170352304e-05,
693
+ "loss": 0.0522,
694
+ "step": 1140
695
+ },
696
+ {
697
+ "grad_norm": 0.6993072032928467,
698
+ "learning_rate": 9.141897500954229e-05,
699
+ "loss": 0.0525,
700
+ "step": 1150
701
+ },
702
+ {
703
+ "grad_norm": 0.694438636302948,
704
+ "learning_rate": 9.123282651541576e-05,
705
+ "loss": 0.0527,
706
+ "step": 1160
707
+ },
708
+ {
709
+ "grad_norm": 0.5264086127281189,
710
+ "learning_rate": 9.104487436388161e-05,
711
+ "loss": 0.0507,
712
+ "step": 1170
713
+ },
714
+ {
715
+ "grad_norm": 0.6317558288574219,
716
+ "learning_rate": 9.085512677657582e-05,
717
+ "loss": 0.0492,
718
+ "step": 1180
719
+ },
720
+ {
721
+ "grad_norm": 0.5922719836235046,
722
+ "learning_rate": 9.066359205367258e-05,
723
+ "loss": 0.0507,
724
+ "step": 1190
725
+ },
726
+ {
727
+ "grad_norm": 0.5283981561660767,
728
+ "learning_rate": 9.047027857352112e-05,
729
+ "loss": 0.0498,
730
+ "step": 1200
731
+ },
732
+ {
733
+ "grad_norm": 0.6506936550140381,
734
+ "learning_rate": 9.027519479227935e-05,
735
+ "loss": 0.0477,
736
+ "step": 1210
737
+ },
738
+ {
739
+ "grad_norm": 0.8326100707054138,
740
+ "learning_rate": 9.007834924354383e-05,
741
+ "loss": 0.0492,
742
+ "step": 1220
743
+ },
744
+ {
745
+ "grad_norm": 0.6938901543617249,
746
+ "learning_rate": 8.987975053797655e-05,
747
+ "loss": 0.05,
748
+ "step": 1230
749
+ },
750
+ {
751
+ "grad_norm": 0.6680724024772644,
752
+ "learning_rate": 8.967940736292825e-05,
753
+ "loss": 0.0504,
754
+ "step": 1240
755
+ },
756
+ {
757
+ "grad_norm": 0.6676413416862488,
758
+ "learning_rate": 8.947732848205846e-05,
759
+ "loss": 0.0505,
760
+ "step": 1250
761
+ },
762
+ {
763
+ "grad_norm": 0.6352572441101074,
764
+ "learning_rate": 8.927352273495204e-05,
765
+ "loss": 0.0516,
766
+ "step": 1260
767
+ },
768
+ {
769
+ "grad_norm": 0.5139537453651428,
770
+ "learning_rate": 8.906799903673265e-05,
771
+ "loss": 0.0481,
772
+ "step": 1270
773
+ },
774
+ {
775
+ "grad_norm": 0.5869550704956055,
776
+ "learning_rate": 8.88607663776726e-05,
777
+ "loss": 0.0472,
778
+ "step": 1280
779
+ },
780
+ {
781
+ "grad_norm": 0.7531103491783142,
782
+ "learning_rate": 8.865183382279978e-05,
783
+ "loss": 0.048,
784
+ "step": 1290
785
+ },
786
+ {
787
+ "grad_norm": 0.6038063764572144,
788
+ "learning_rate": 8.844121051150096e-05,
789
+ "loss": 0.0504,
790
+ "step": 1300
791
+ },
792
+ {
793
+ "grad_norm": 0.5663856267929077,
794
+ "learning_rate": 8.822890565712211e-05,
795
+ "loss": 0.0464,
796
+ "step": 1310
797
+ },
798
+ {
799
+ "grad_norm": 0.6821518540382385,
800
+ "learning_rate": 8.801492854656536e-05,
801
+ "loss": 0.0487,
802
+ "step": 1320
803
+ },
804
+ {
805
+ "grad_norm": 0.6302056908607483,
806
+ "learning_rate": 8.779928853988268e-05,
807
+ "loss": 0.0487,
808
+ "step": 1330
809
+ },
810
+ {
811
+ "grad_norm": 0.6849905848503113,
812
+ "learning_rate": 8.758199506986655e-05,
813
+ "loss": 0.0486,
814
+ "step": 1340
815
+ },
816
+ {
817
+ "grad_norm": 0.5426021218299866,
818
+ "learning_rate": 8.73630576416373e-05,
819
+ "loss": 0.049,
820
+ "step": 1350
821
+ },
822
+ {
823
+ "grad_norm": 0.6531778573989868,
824
+ "learning_rate": 8.714248583222726e-05,
825
+ "loss": 0.0497,
826
+ "step": 1360
827
+ },
828
+ {
829
+ "grad_norm": 0.7023406624794006,
830
+ "learning_rate": 8.692028929016196e-05,
831
+ "loss": 0.0472,
832
+ "step": 1370
833
+ },
834
+ {
835
+ "grad_norm": 0.7523959279060364,
836
+ "learning_rate": 8.669647773503797e-05,
837
+ "loss": 0.0465,
838
+ "step": 1380
839
+ },
840
+ {
841
+ "grad_norm": 0.6905308961868286,
842
+ "learning_rate": 8.647106095709773e-05,
843
+ "loss": 0.0496,
844
+ "step": 1390
845
+ },
846
+ {
847
+ "grad_norm": 0.48531508445739746,
848
+ "learning_rate": 8.624404881680139e-05,
849
+ "loss": 0.0466,
850
+ "step": 1400
851
+ },
852
+ {
853
+ "grad_norm": 0.4626646041870117,
854
+ "learning_rate": 8.601545124439535e-05,
855
+ "loss": 0.045,
856
+ "step": 1410
857
+ },
858
+ {
859
+ "grad_norm": 0.5854551196098328,
860
+ "learning_rate": 8.5785278239478e-05,
861
+ "loss": 0.0446,
862
+ "step": 1420
863
+ },
864
+ {
865
+ "grad_norm": 0.515155017375946,
866
+ "learning_rate": 8.555353987056224e-05,
867
+ "loss": 0.0465,
868
+ "step": 1430
869
+ },
870
+ {
871
+ "grad_norm": 0.6705062985420227,
872
+ "learning_rate": 8.532024627463505e-05,
873
+ "loss": 0.0473,
874
+ "step": 1440
875
+ },
876
+ {
877
+ "grad_norm": 0.6408502459526062,
878
+ "learning_rate": 8.508540765671407e-05,
879
+ "loss": 0.046,
880
+ "step": 1450
881
+ },
882
+ {
883
+ "grad_norm": 0.7304362058639526,
884
+ "learning_rate": 8.484903428940121e-05,
885
+ "loss": 0.0461,
886
+ "step": 1460
887
+ },
888
+ {
889
+ "grad_norm": 0.48894795775413513,
890
+ "learning_rate": 8.461113651243334e-05,
891
+ "loss": 0.0475,
892
+ "step": 1470
893
+ },
894
+ {
895
+ "grad_norm": 0.5249834060668945,
896
+ "learning_rate": 8.437172473222987e-05,
897
+ "loss": 0.0439,
898
+ "step": 1480
899
+ },
900
+ {
901
+ "grad_norm": 0.4411742091178894,
902
+ "learning_rate": 8.413080942143767e-05,
903
+ "loss": 0.0438,
904
+ "step": 1490
905
+ },
906
+ {
907
+ "grad_norm": 0.44417959451675415,
908
+ "learning_rate": 8.388840111847288e-05,
909
+ "loss": 0.0427,
910
+ "step": 1500
911
+ },
912
+ {
913
+ "grad_norm": 0.5642029047012329,
914
+ "learning_rate": 8.364451042705998e-05,
915
+ "loss": 0.0427,
916
+ "step": 1510
917
+ },
918
+ {
919
+ "grad_norm": 0.4869621694087982,
920
+ "learning_rate": 8.33991480157679e-05,
921
+ "loss": 0.0434,
922
+ "step": 1520
923
+ },
924
+ {
925
+ "grad_norm": 0.6327221393585205,
926
+ "learning_rate": 8.315232461754338e-05,
927
+ "loss": 0.0445,
928
+ "step": 1530
929
+ },
930
+ {
931
+ "grad_norm": 0.6919836401939392,
932
+ "learning_rate": 8.290405102924144e-05,
933
+ "loss": 0.0451,
934
+ "step": 1540
935
+ },
936
+ {
937
+ "grad_norm": 0.5408800840377808,
938
+ "learning_rate": 8.265433811115316e-05,
939
+ "loss": 0.0446,
940
+ "step": 1550
941
+ },
942
+ {
943
+ "grad_norm": 0.5201606750488281,
944
+ "learning_rate": 8.240319678653049e-05,
945
+ "loss": 0.0423,
946
+ "step": 1560
947
+ },
948
+ {
949
+ "grad_norm": 0.5895570516586304,
950
+ "learning_rate": 8.215063804110857e-05,
951
+ "loss": 0.0428,
952
+ "step": 1570
953
+ },
954
+ {
955
+ "grad_norm": 0.4627968966960907,
956
+ "learning_rate": 8.189667292262512e-05,
957
+ "loss": 0.042,
958
+ "step": 1580
959
+ },
960
+ {
961
+ "grad_norm": 0.7023901343345642,
962
+ "learning_rate": 8.164131254033716e-05,
963
+ "loss": 0.044,
964
+ "step": 1590
965
+ },
966
+ {
967
+ "grad_norm": 0.5879528522491455,
968
+ "learning_rate": 8.138456806453503e-05,
969
+ "loss": 0.0458,
970
+ "step": 1600
971
+ },
972
+ {
973
+ "grad_norm": 0.5253772735595703,
974
+ "learning_rate": 8.112645072605386e-05,
975
+ "loss": 0.0444,
976
+ "step": 1610
977
+ },
978
+ {
979
+ "grad_norm": 0.5280351638793945,
980
+ "learning_rate": 8.086697181578222e-05,
981
+ "loss": 0.0423,
982
+ "step": 1620
983
+ },
984
+ {
985
+ "grad_norm": 0.6314783096313477,
986
+ "learning_rate": 8.060614268416823e-05,
987
+ "loss": 0.0417,
988
+ "step": 1630
989
+ },
990
+ {
991
+ "grad_norm": 0.8157122731208801,
992
+ "learning_rate": 8.034397474072309e-05,
993
+ "loss": 0.0433,
994
+ "step": 1640
995
+ },
996
+ {
997
+ "grad_norm": 0.43822190165519714,
998
+ "learning_rate": 8.008047945352193e-05,
999
+ "loss": 0.0419,
1000
+ "step": 1650
1001
+ },
1002
+ {
1003
+ "grad_norm": 0.5351688265800476,
1004
+ "learning_rate": 7.981566834870225e-05,
1005
+ "loss": 0.0437,
1006
+ "step": 1660
1007
+ },
1008
+ {
1009
+ "grad_norm": 0.48337793350219727,
1010
+ "learning_rate": 7.954955300995961e-05,
1011
+ "loss": 0.042,
1012
+ "step": 1670
1013
+ },
1014
+ {
1015
+ "grad_norm": 0.5100006461143494,
1016
+ "learning_rate": 7.928214507804104e-05,
1017
+ "loss": 0.04,
1018
+ "step": 1680
1019
+ },
1020
+ {
1021
+ "grad_norm": 0.5312990546226501,
1022
+ "learning_rate": 7.901345625023576e-05,
1023
+ "loss": 0.0419,
1024
+ "step": 1690
1025
+ },
1026
+ {
1027
+ "grad_norm": 0.5234403610229492,
1028
+ "learning_rate": 7.874349827986354e-05,
1029
+ "loss": 0.0413,
1030
+ "step": 1700
1031
+ },
1032
+ {
1033
+ "grad_norm": 0.5953489542007446,
1034
+ "learning_rate": 7.847228297576053e-05,
1035
+ "loss": 0.0402,
1036
+ "step": 1710
1037
+ },
1038
+ {
1039
+ "grad_norm": 0.5609169006347656,
1040
+ "learning_rate": 7.819982220176276e-05,
1041
+ "loss": 0.0424,
1042
+ "step": 1720
1043
+ },
1044
+ {
1045
+ "grad_norm": 0.5080845355987549,
1046
+ "learning_rate": 7.792612787618714e-05,
1047
+ "loss": 0.0409,
1048
+ "step": 1730
1049
+ },
1050
+ {
1051
+ "grad_norm": 0.6321108341217041,
1052
+ "learning_rate": 7.765121197131009e-05,
1053
+ "loss": 0.0401,
1054
+ "step": 1740
1055
+ },
1056
+ {
1057
+ "grad_norm": 0.5982347130775452,
1058
+ "learning_rate": 7.737508651284391e-05,
1059
+ "loss": 0.0423,
1060
+ "step": 1750
1061
+ },
1062
+ {
1063
+ "grad_norm": 0.749245822429657,
1064
+ "learning_rate": 7.709776357941069e-05,
1065
+ "loss": 0.0414,
1066
+ "step": 1760
1067
+ },
1068
+ {
1069
+ "grad_norm": 0.46884486079216003,
1070
+ "learning_rate": 7.681925530201392e-05,
1071
+ "loss": 0.0419,
1072
+ "step": 1770
1073
+ },
1074
+ {
1075
+ "grad_norm": 0.5173981189727783,
1076
+ "learning_rate": 7.65395738635079e-05,
1077
+ "loss": 0.0405,
1078
+ "step": 1780
1079
+ },
1080
+ {
1081
+ "grad_norm": 0.4504048228263855,
1082
+ "learning_rate": 7.62587314980648e-05,
1083
+ "loss": 0.0397,
1084
+ "step": 1790
1085
+ },
1086
+ {
1087
+ "grad_norm": 0.4966204762458801,
1088
+ "learning_rate": 7.597674049063947e-05,
1089
+ "loss": 0.0396,
1090
+ "step": 1800
1091
+ },
1092
+ {
1093
+ "grad_norm": 0.37520983815193176,
1094
+ "learning_rate": 7.569361317643211e-05,
1095
+ "loss": 0.0391,
1096
+ "step": 1810
1097
+ },
1098
+ {
1099
+ "grad_norm": 0.4573592245578766,
1100
+ "learning_rate": 7.540936194034865e-05,
1101
+ "loss": 0.0396,
1102
+ "step": 1820
1103
+ },
1104
+ {
1105
+ "grad_norm": 0.39196085929870605,
1106
+ "learning_rate": 7.512399921645901e-05,
1107
+ "loss": 0.0394,
1108
+ "step": 1830
1109
+ },
1110
+ {
1111
+ "grad_norm": 0.5877107381820679,
1112
+ "learning_rate": 7.483753748745317e-05,
1113
+ "loss": 0.038,
1114
+ "step": 1840
1115
+ },
1116
+ {
1117
+ "grad_norm": 0.5072145462036133,
1118
+ "learning_rate": 7.454998928409516e-05,
1119
+ "loss": 0.0386,
1120
+ "step": 1850
1121
+ },
1122
+ {
1123
+ "grad_norm": 0.42455703020095825,
1124
+ "learning_rate": 7.426136718467493e-05,
1125
+ "loss": 0.0381,
1126
+ "step": 1860
1127
+ },
1128
+ {
1129
+ "grad_norm": 0.5039349794387817,
1130
+ "learning_rate": 7.397168381445812e-05,
1131
+ "loss": 0.04,
1132
+ "step": 1870
1133
+ },
1134
+ {
1135
+ "grad_norm": 0.5629942417144775,
1136
+ "learning_rate": 7.368095184513377e-05,
1137
+ "loss": 0.0391,
1138
+ "step": 1880
1139
+ },
1140
+ {
1141
+ "grad_norm": 0.7176268696784973,
1142
+ "learning_rate": 7.338918399426005e-05,
1143
+ "loss": 0.0411,
1144
+ "step": 1890
1145
+ },
1146
+ {
1147
+ "grad_norm": 0.4593715965747833,
1148
+ "learning_rate": 7.309639302470801e-05,
1149
+ "loss": 0.0413,
1150
+ "step": 1900
1151
+ },
1152
+ {
1153
+ "grad_norm": 0.3955693542957306,
1154
+ "learning_rate": 7.280259174410312e-05,
1155
+ "loss": 0.0387,
1156
+ "step": 1910
1157
+ },
1158
+ {
1159
+ "grad_norm": 0.5040004849433899,
1160
+ "learning_rate": 7.250779300426517e-05,
1161
+ "loss": 0.0385,
1162
+ "step": 1920
1163
+ },
1164
+ {
1165
+ "grad_norm": 0.5110053420066833,
1166
+ "learning_rate": 7.22120097006461e-05,
1167
+ "loss": 0.0393,
1168
+ "step": 1930
1169
+ },
1170
+ {
1171
+ "grad_norm": 0.48912274837493896,
1172
+ "learning_rate": 7.191525477176577e-05,
1173
+ "loss": 0.0378,
1174
+ "step": 1940
1175
+ },
1176
+ {
1177
+ "grad_norm": 0.549274742603302,
1178
+ "learning_rate": 7.161754119864616e-05,
1179
+ "loss": 0.0381,
1180
+ "step": 1950
1181
+ },
1182
+ {
1183
+ "grad_norm": 0.5900108218193054,
1184
+ "learning_rate": 7.131888200424339e-05,
1185
+ "loss": 0.0393,
1186
+ "step": 1960
1187
+ },
1188
+ {
1189
+ "grad_norm": 0.525826632976532,
1190
+ "learning_rate": 7.101929025287816e-05,
1191
+ "loss": 0.0389,
1192
+ "step": 1970
1193
+ },
1194
+ {
1195
+ "grad_norm": 0.5286293625831604,
1196
+ "learning_rate": 7.071877904966423e-05,
1197
+ "loss": 0.0385,
1198
+ "step": 1980
1199
+ },
1200
+ {
1201
+ "grad_norm": 0.5297350287437439,
1202
+ "learning_rate": 7.04173615399351e-05,
1203
+ "loss": 0.0384,
1204
+ "step": 1990
1205
+ },
1206
+ {
1207
+ "grad_norm": 0.47022709250450134,
1208
+ "learning_rate": 7.011505090866913e-05,
1209
+ "loss": 0.0383,
1210
+ "step": 2000
1211
+ }
1212
+ ],
1213
+ "logging_steps": 10,
1214
+ "max_steps": 5000,
1215
+ "num_input_tokens_seen": 0,
1216
+ "num_train_epochs": 9223372036854775807,
1217
+ "save_steps": 1000,
1218
+ "stateful_callbacks": {
1219
+ "TrainerControl": {
1220
+ "args": {
1221
+ "should_epoch_stop": false,
1222
+ "should_evaluate": false,
1223
+ "should_log": false,
1224
+ "should_save": true,
1225
+ "should_training_stop": false
1226
+ },
1227
+ "attributes": {}
1228
+ }
1229
+ },
1230
+ "total_flos": 0.0,
1231
+ "train_batch_size": 768,
1232
+ "trial_name": null,
1233
+ "trial_params": null
1234
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:115dd3fd62ce1dc0fa69c817bb62b4072e8f293fb6717f54cff40de48c7de270
3
+ size 7761
wandb_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"project": "gr00t-g1-29dof-gear-sonic", "run_id": "outputs"}
zero_to_fp32.py ADDED
@@ -0,0 +1,760 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example:
14
+ # python zero_to_fp32.py . output_dir/
15
+ # or
16
+ # python zero_to_fp32.py . output_dir/ --safe_serialization
17
+
18
+ import argparse
19
+ import torch
20
+ import glob
21
+ import math
22
+ import os
23
+ import re
24
+ import gc
25
+ import json
26
+ import numpy as np
27
+ from tqdm import tqdm
28
+ from collections import OrderedDict
29
+ from dataclasses import dataclass
30
+
31
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
32
+ # DeepSpeed data structures it has to be available in the current python environment.
33
+ from deepspeed.utils import logger
34
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
35
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
36
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
37
+
38
+
39
+ @dataclass
40
+ class zero_model_state:
41
+ buffers: dict()
42
+ param_shapes: dict()
43
+ shared_params: list
44
+ ds_version: int
45
+ frozen_param_shapes: dict()
46
+ frozen_param_fragments: dict()
47
+
48
+
49
+ debug = 0
50
+
51
+ # load to cpu
52
+ device = torch.device('cpu')
53
+
54
+
55
+ def atoi(text):
56
+ return int(text) if text.isdigit() else text
57
+
58
+
59
+ def natural_keys(text):
60
+ '''
61
+ alist.sort(key=natural_keys) sorts in human order
62
+ http://nedbatchelder.com/blog/200712/human_sorting.html
63
+ (See Toothy's implementation in the comments)
64
+ '''
65
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
66
+
67
+
68
+ def get_model_state_file(checkpoint_dir, zero_stage):
69
+ if not os.path.isdir(checkpoint_dir):
70
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
71
+
72
+ # there should be only one file
73
+ if zero_stage <= 2:
74
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
75
+ elif zero_stage == 3:
76
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
77
+
78
+ if not os.path.exists(file):
79
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
80
+
81
+ return file
82
+
83
+
84
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
85
+ # XXX: need to test that this simple glob rule works for multi-node setup too
86
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
87
+
88
+ if len(ckpt_files) == 0:
89
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
90
+
91
+ return ckpt_files
92
+
93
+
94
+ def get_optim_files(checkpoint_dir):
95
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
96
+
97
+
98
+ def get_model_state_files(checkpoint_dir):
99
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
100
+
101
+
102
+ def parse_model_states(files):
103
+ zero_model_states = []
104
+ for file in files:
105
+ state_dict = torch.load(file, map_location=device, weights_only=False)
106
+
107
+ if BUFFER_NAMES not in state_dict:
108
+ raise ValueError(f"{file} is not a model state checkpoint")
109
+ buffer_names = state_dict[BUFFER_NAMES]
110
+ if debug:
111
+ print("Found buffers:", buffer_names)
112
+
113
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
114
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
115
+ param_shapes = state_dict[PARAM_SHAPES]
116
+
117
+ # collect parameters that are included in param_shapes
118
+ param_names = []
119
+ for s in param_shapes:
120
+ for name in s.keys():
121
+ param_names.append(name)
122
+
123
+ # update with frozen parameters
124
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
125
+ if frozen_param_shapes is not None:
126
+ if debug:
127
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
128
+ param_names += list(frozen_param_shapes.keys())
129
+
130
+ # handle shared params
131
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
132
+
133
+ ds_version = state_dict.get(DS_VERSION, None)
134
+
135
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
136
+
137
+ z_model_state = zero_model_state(buffers=buffers,
138
+ param_shapes=param_shapes,
139
+ shared_params=shared_params,
140
+ ds_version=ds_version,
141
+ frozen_param_shapes=frozen_param_shapes,
142
+ frozen_param_fragments=frozen_param_fragments)
143
+ zero_model_states.append(z_model_state)
144
+
145
+ return zero_model_states
146
+
147
+
148
+ def parse_optim_states(files, ds_checkpoint_dir):
149
+ total_files = len(files)
150
+ state_dicts = []
151
+ for f in tqdm(files, desc='Loading checkpoint shards'):
152
+ state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
153
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
154
+ # and also handle the case where it was already removed by another helper script
155
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
156
+ state_dicts.append(state_dict)
157
+
158
+ if ZERO_STAGE not in state_dicts[0][OPTIMIZER_STATE_DICT]:
159
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
160
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
161
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
162
+
163
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
164
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
165
+ # use the max of the partition_count to get the dp world_size.
166
+
167
+ if type(world_size) is list:
168
+ world_size = max(world_size)
169
+
170
+ if world_size != total_files:
171
+ raise ValueError(
172
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
173
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
174
+ )
175
+
176
+ # the groups are named differently in each stage
177
+ if zero_stage <= 2:
178
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
179
+ elif zero_stage == 3:
180
+ fp32_groups_key = FP32_FLAT_GROUPS
181
+ else:
182
+ raise ValueError(f"unknown zero stage {zero_stage}")
183
+
184
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
185
+ return zero_stage, world_size, fp32_flat_groups
186
+
187
+
188
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
189
+ """
190
+ Returns fp32 state_dict reconstructed from ds checkpoint
191
+
192
+ Args:
193
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
194
+
195
+ """
196
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
197
+
198
+ optim_files = get_optim_files(ds_checkpoint_dir)
199
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
200
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
201
+
202
+ model_files = get_model_state_files(ds_checkpoint_dir)
203
+
204
+ zero_model_states = parse_model_states(model_files)
205
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
206
+
207
+ if zero_stage <= 2:
208
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
209
+ exclude_frozen_parameters)
210
+ elif zero_stage == 3:
211
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
212
+ exclude_frozen_parameters)
213
+
214
+
215
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
216
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
217
+ return
218
+
219
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
220
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
221
+
222
+ if debug:
223
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
224
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
225
+
226
+ wanted_params = len(frozen_param_shapes)
227
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
228
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
229
+ print(f'Frozen params: Have {avail_numel} numels to process.')
230
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
231
+
232
+ total_params = 0
233
+ total_numel = 0
234
+ for name, shape in frozen_param_shapes.items():
235
+ total_params += 1
236
+ unpartitioned_numel = shape.numel()
237
+ total_numel += unpartitioned_numel
238
+
239
+ state_dict[name] = frozen_param_fragments[name]
240
+
241
+ if debug:
242
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
243
+
244
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
245
+
246
+
247
+ def _has_callable(obj, fn):
248
+ attr = getattr(obj, fn, None)
249
+ return callable(attr)
250
+
251
+
252
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
253
+ param_shapes = zero_model_states[0].param_shapes
254
+
255
+ # Reconstruction protocol:
256
+ #
257
+ # XXX: document this
258
+
259
+ if debug:
260
+ for i in range(world_size):
261
+ for j in range(len(fp32_flat_groups[0])):
262
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
263
+
264
+ # XXX: memory usage doubles here (zero2)
265
+ num_param_groups = len(fp32_flat_groups[0])
266
+ merged_single_partition_of_fp32_groups = []
267
+ for i in range(num_param_groups):
268
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
269
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
270
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
271
+ avail_numel = sum(
272
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
273
+
274
+ if debug:
275
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
276
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
277
+ # not asserting if there is a mismatch due to possible padding
278
+ print(f"Have {avail_numel} numels to process.")
279
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
280
+
281
+ # params
282
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
283
+ # out-of-core computing solution
284
+ total_numel = 0
285
+ total_params = 0
286
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
287
+ offset = 0
288
+ avail_numel = full_single_fp32_vector.numel()
289
+ for name, shape in shapes.items():
290
+
291
+ unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
292
+ total_numel += unpartitioned_numel
293
+ total_params += 1
294
+
295
+ if debug:
296
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
297
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
298
+ offset += unpartitioned_numel
299
+
300
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
301
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
302
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
303
+ # live optimizer object, so we are checking that the numbers are within the right range
304
+ align_to = 2 * world_size
305
+
306
+ def zero2_align(x):
307
+ return align_to * math.ceil(x / align_to)
308
+
309
+ if debug:
310
+ print(f"original offset={offset}, avail_numel={avail_numel}")
311
+
312
+ offset = zero2_align(offset)
313
+ avail_numel = zero2_align(avail_numel)
314
+
315
+ if debug:
316
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
317
+
318
+ # Sanity check
319
+ if offset != avail_numel:
320
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
321
+
322
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
323
+
324
+
325
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
326
+ exclude_frozen_parameters):
327
+ state_dict = OrderedDict()
328
+
329
+ # buffers
330
+ buffers = zero_model_states[0].buffers
331
+ state_dict.update(buffers)
332
+ if debug:
333
+ print(f"added {len(buffers)} buffers")
334
+
335
+ if not exclude_frozen_parameters:
336
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
337
+
338
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
339
+
340
+ # recover shared parameters
341
+ for pair in zero_model_states[0].shared_params:
342
+ if pair[1] in state_dict:
343
+ state_dict[pair[0]] = state_dict[pair[1]]
344
+
345
+ return state_dict
346
+
347
+
348
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
349
+ remainder = unpartitioned_numel % world_size
350
+ padding_numel = (world_size - remainder) if remainder else 0
351
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
352
+ return partitioned_numel, padding_numel
353
+
354
+
355
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
356
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
357
+ return
358
+
359
+ if debug:
360
+ for i in range(world_size):
361
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
362
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
363
+
364
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
365
+ wanted_params = len(frozen_param_shapes)
366
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
367
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
368
+ print(f'Frozen params: Have {avail_numel} numels to process.')
369
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
370
+
371
+ total_params = 0
372
+ total_numel = 0
373
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
374
+ total_params += 1
375
+ unpartitioned_numel = shape.numel()
376
+ total_numel += unpartitioned_numel
377
+
378
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
379
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
380
+
381
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
382
+
383
+ if debug:
384
+ print(
385
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
386
+ )
387
+
388
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
389
+
390
+
391
+ class GatheredTensor:
392
+ """
393
+ A pseudo tensor that collects partitioned weights.
394
+ It is more memory efficient when there are multiple groups.
395
+ """
396
+
397
+ def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
398
+ self.flat_groups = flat_groups
399
+ self.flat_groups_offset = flat_groups_offset
400
+ self.offset = offset
401
+ self.partitioned_numel = partitioned_numel
402
+ self.shape = shape
403
+ self.dtype = self.flat_groups[0][0].dtype
404
+
405
+ def contiguous(self):
406
+ """
407
+ Merge partitioned weights from flat_groups into a single tensor.
408
+ """
409
+ end_idx = self.offset + self.partitioned_numel
410
+ world_size = len(self.flat_groups)
411
+ pad_flat_param_chunks = []
412
+
413
+ for rank_i in range(world_size):
414
+ # for each rank, we need to collect weights from related group/groups
415
+ flat_groups_at_rank_i = self.flat_groups[rank_i]
416
+ start_group_id = None
417
+ end_group_id = None
418
+ for group_id in range(len(self.flat_groups_offset)):
419
+ if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
420
+ start_group_id = group_id
421
+ if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
422
+ end_group_id = group_id
423
+ break
424
+ # collect weights from related group/groups
425
+ for group_id in range(start_group_id, end_group_id + 1):
426
+ flat_tensor = flat_groups_at_rank_i[group_id]
427
+ start_offset = self.offset - self.flat_groups_offset[group_id]
428
+ end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
429
+ pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
430
+
431
+ # collect weights from all ranks
432
+ pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
433
+ param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
434
+ return param
435
+
436
+
437
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
438
+ param_shapes = zero_model_states[0].param_shapes
439
+ avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
440
+
441
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
442
+ # param, re-consolidating each param, while dealing with padding if any
443
+
444
+ # merge list of dicts, preserving order
445
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
446
+
447
+ if debug:
448
+ for i in range(world_size):
449
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
450
+
451
+ wanted_params = len(param_shapes)
452
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
453
+ # not asserting if there is a mismatch due to possible padding
454
+ avail_numel = fp32_flat_groups[0].numel() * world_size
455
+ print(f"Trainable params: Have {avail_numel} numels to process.")
456
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
457
+
458
+ # params
459
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
460
+ # out-of-core computing solution
461
+ offset = 0
462
+ total_numel = 0
463
+ total_params = 0
464
+ flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
465
+ for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
466
+ unpartitioned_numel = shape.numel()
467
+ total_numel += unpartitioned_numel
468
+ total_params += 1
469
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
470
+
471
+ if debug:
472
+ print(
473
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
474
+ )
475
+
476
+ # memory efficient tensor
477
+ tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
478
+ state_dict[name] = tensor
479
+ offset += partitioned_numel
480
+
481
+ offset *= world_size
482
+
483
+ # Sanity check
484
+ if offset != avail_numel:
485
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
486
+
487
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
488
+
489
+
490
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
491
+ exclude_frozen_parameters):
492
+ state_dict = OrderedDict()
493
+
494
+ # buffers
495
+ buffers = zero_model_states[0].buffers
496
+ state_dict.update(buffers)
497
+ if debug:
498
+ print(f"added {len(buffers)} buffers")
499
+
500
+ if not exclude_frozen_parameters:
501
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
502
+
503
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
504
+
505
+ # recover shared parameters
506
+ for pair in zero_model_states[0].shared_params:
507
+ if pair[1] in state_dict:
508
+ state_dict[pair[0]] = state_dict[pair[1]]
509
+
510
+ return state_dict
511
+
512
+
513
+ def to_torch_tensor(state_dict, return_empty_tensor=False):
514
+ """
515
+ Convert state_dict of GatheredTensor to torch tensor
516
+ """
517
+ torch_state_dict = {}
518
+ converted_tensors = {}
519
+ for name, tensor in state_dict.items():
520
+ tensor_id = id(tensor)
521
+ if tensor_id in converted_tensors: # shared tensors
522
+ shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
523
+ torch_state_dict[name] = shared_tensor
524
+ else:
525
+ converted_tensors[tensor_id] = name
526
+ if return_empty_tensor:
527
+ torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
528
+ else:
529
+ torch_state_dict[name] = tensor.contiguous()
530
+ return torch_state_dict
531
+
532
+
533
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
534
+ tag=None,
535
+ exclude_frozen_parameters=False,
536
+ lazy_mode=False):
537
+ """
538
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
539
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
540
+ via a model hub.
541
+
542
+ Args:
543
+ - ``checkpoint_dir``: path to the desired checkpoint folder
544
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
545
+ - ``exclude_frozen_parameters``: exclude frozen parameters
546
+ - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
547
+ Convert the pesduo tensor to torch tensor by ``.contiguous()``
548
+
549
+ Returns:
550
+ - pytorch ``state_dict``
551
+
552
+ A typical usage might be ::
553
+
554
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
555
+ # do the training and checkpoint saving
556
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
557
+ model = model.cpu() # move to cpu
558
+ model.load_state_dict(state_dict)
559
+ # submit to model hub or save the model to share with others
560
+
561
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
562
+ application. i.e. you will need to re-initialize the deepspeed engine, since
563
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
564
+
565
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
566
+
567
+ Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
568
+ You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
569
+ the checkpoint. Or you can load state_dict in lazy mode ::
570
+
571
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
572
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
573
+ for name, lazy_tensor in state_dict.item():
574
+ tensor = lazy_tensor.contiguous() # to cpu
575
+ print(name, tensor)
576
+ # del tensor to release memory if it no longer in use
577
+ """
578
+ if tag is None:
579
+ latest_path = os.path.join(checkpoint_dir, 'latest')
580
+ if os.path.isfile(latest_path):
581
+ with open(latest_path, 'r') as fd:
582
+ tag = fd.read().strip()
583
+ else:
584
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
585
+
586
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
587
+
588
+ if not os.path.isdir(ds_checkpoint_dir):
589
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
590
+
591
+ state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
592
+ if lazy_mode:
593
+ return state_dict
594
+ else:
595
+ return to_torch_tensor(state_dict)
596
+
597
+
598
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
599
+ output_dir,
600
+ max_shard_size="5GB",
601
+ safe_serialization=False,
602
+ tag=None,
603
+ exclude_frozen_parameters=False):
604
+ """
605
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
606
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
607
+
608
+ Args:
609
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
610
+ - ``output_dir``: directory to the pytorch fp32 state_dict output files
611
+ - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
612
+ - ``safe_serialization``: whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
613
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
614
+ - ``exclude_frozen_parameters``: exclude frozen parameters
615
+ """
616
+
617
+ # Dependency pre-check
618
+ if safe_serialization:
619
+ try:
620
+ from safetensors.torch import save_file
621
+ except ImportError:
622
+ print('If you want to use `safe_serialization`, please `pip install safetensors`')
623
+ raise
624
+ if max_shard_size is not None:
625
+ try:
626
+ from huggingface_hub import split_torch_state_dict_into_shards
627
+ except ImportError:
628
+ print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
629
+ raise
630
+
631
+ # Convert zero checkpoint to state_dict
632
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
633
+ tag,
634
+ exclude_frozen_parameters,
635
+ lazy_mode=True)
636
+
637
+ # Shard the model if it is too big.
638
+ weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
639
+ if max_shard_size is not None:
640
+ filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
641
+ # an memory-efficient approach for sharding
642
+ empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
643
+ state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
644
+ filename_pattern=filename_pattern,
645
+ max_shard_size=max_shard_size)
646
+ else:
647
+ from collections import namedtuple
648
+ StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
649
+ state_dict_split = StateDictSplit(is_sharded=False,
650
+ filename_to_tensors={weights_name: list(state_dict.keys())})
651
+
652
+ # Save the model by shard
653
+ os.makedirs(output_dir, exist_ok=True)
654
+ filename_to_tensors = state_dict_split.filename_to_tensors.items()
655
+ for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
656
+ shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
657
+ shard_state_dict = to_torch_tensor(shard_state_dict)
658
+ output_path = os.path.join(output_dir, shard_file)
659
+ if safe_serialization:
660
+ save_file(shard_state_dict, output_path, metadata={"format": "pt"})
661
+ else:
662
+ torch.save(shard_state_dict, output_path)
663
+ # release the memory of current shard
664
+ for tensor_name in list(shard_state_dict.keys()):
665
+ del state_dict[tensor_name]
666
+ del shard_state_dict[tensor_name]
667
+ del shard_state_dict
668
+ gc.collect()
669
+
670
+ # Save index if sharded
671
+ if state_dict_split.is_sharded:
672
+ index = {
673
+ "metadata": state_dict_split.metadata,
674
+ "weight_map": state_dict_split.tensor_to_filename,
675
+ }
676
+ save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
677
+ save_index_file = os.path.join(output_dir, save_index_file)
678
+ with open(save_index_file, "w", encoding="utf-8") as f:
679
+ content = json.dumps(index, indent=2, sort_keys=True) + "\n"
680
+ f.write(content)
681
+
682
+
683
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
684
+ """
685
+ 1. Put the provided model to cpu
686
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
687
+ 3. Load it into the provided model
688
+
689
+ Args:
690
+ - ``model``: the model object to update
691
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
692
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
693
+
694
+ Returns:
695
+ - ``model`: modified model
696
+
697
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
698
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
699
+ conveniently placed for you in the checkpoint folder.
700
+
701
+ A typical usage might be ::
702
+
703
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
704
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
705
+ # submit to model hub or save the model to share with others
706
+
707
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
708
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
709
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
710
+
711
+ """
712
+ logger.info("Extracting fp32 weights")
713
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
714
+
715
+ logger.info("Overwriting model with fp32 weights")
716
+ model = model.cpu()
717
+ model.load_state_dict(state_dict, strict=False)
718
+
719
+ return model
720
+
721
+
722
+ if __name__ == "__main__":
723
+ parser = argparse.ArgumentParser()
724
+ parser.add_argument("checkpoint_dir",
725
+ type=str,
726
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
727
+ parser.add_argument("output_dir",
728
+ type=str,
729
+ help="directory to the pytorch fp32 state_dict output files"
730
+ "(e.g. path/checkpoint-12-output/)")
731
+ parser.add_argument(
732
+ "--max_shard_size",
733
+ type=str,
734
+ default="5GB",
735
+ help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
736
+ "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
737
+ "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
738
+ "without CPU OOM issues.")
739
+ parser.add_argument(
740
+ "--safe_serialization",
741
+ default=False,
742
+ action='store_true',
743
+ help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
744
+ parser.add_argument("-t",
745
+ "--tag",
746
+ type=str,
747
+ default=None,
748
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
749
+ parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
750
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
751
+ args = parser.parse_args()
752
+
753
+ debug = args.debug
754
+
755
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
756
+ args.output_dir,
757
+ max_shard_size=args.max_shard_size,
758
+ safe_serialization=args.safe_serialization,
759
+ tag=args.tag,
760
+ exclude_frozen_parameters=args.exclude_frozen_parameters)