BIue-Jay commited on
Commit
ec16f68
·
verified ·
1 Parent(s): 6a7505b

Add files using upload-large-folder tool

Browse files
a18-pour-water/epoch-020/experiment_cfg/final_processor_config.json CHANGED
@@ -22,7 +22,7 @@
22
  "random_rotation_angle": "0",
23
  "color_jitter_params": "{'brightness': 0.3, 'contrast': 0.4, 'saturation': 0.5, 'hue': 0.08}",
24
  "processor": "Qwen3VLProcessor:\n- image_processor: Qwen2VLImageProcessorFast {\n \"crop_size\": null,\n \"data_format\": \"channels_first\",\n \"default_to_square\": true,\n \"device\": null,\n \"disable_grouping\": null,\n \"do_center_crop\": null,\n \"do_convert_rgb\": true,\n \"do_normalize\": true,\n \"do_pad\": null,\n \"do_rescale\": true,\n \"do_resize\": true,\n \"image_mean\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"image_processor_type\": \"Qwen2VLImageProcessorFast\",\n \"image_std\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"input_data_format\": null,\n \"max_pixels\": null,\n \"merge_size\": 2,\n \"min_pixels\": null,\n \"pad_size\": null,\n \"patch_size\": 16,\n \"processor_class\": \"Qwen3VLProcessor\",\n \"resample\": 3,\n \"rescale_factor\": 0.00392156862745098,\n \"return_tensors\": null,\n \"size\": {\n \"longest_edge\": 16777216,\n \"shortest_edge\": 65536\n },\n \"temporal_patch_size\": 2\n}\n\n- tokenizer: Qwen2TokenizerFast(name_or_path='nvidia/Cosmos-Reason2-2B', vocab_size=151643, model_max_length=262144, is_fast=True, padding_side='left', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={\n\t151643: AddedToken(\"<|endoftext|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151644: AddedToken(\"<|im_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151645: AddedToken(\"<|im_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151646: AddedToken(\"<|object_ref_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151647: AddedToken(\"<|object_ref_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151648: AddedToken(\"<|box_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151649: AddedToken(\"<|box_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151650: AddedToken(\"<|quad_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151651: AddedToken(\"<|quad_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151652: AddedToken(\"<|vision_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151653: AddedToken(\"<|vision_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151654: AddedToken(\"<|vision_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151655: AddedToken(\"<|image_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151656: AddedToken(\"<|video_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151657: AddedToken(\"<tool_call>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151658: AddedToken(\"</tool_call>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151659: AddedToken(\"<|fim_prefix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151660: AddedToken(\"<|fim_middle|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151661: AddedToken(\"<|fim_suffix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151662: AddedToken(\"<|fim_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151663: AddedToken(\"<|repo_name|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151664: AddedToken(\"<|file_sep|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151665: AddedToken(\"<tool_response>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151666: AddedToken(\"</tool_response>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151667: AddedToken(\"<think>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151668: AddedToken(\"</think>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n}\n)\n- video_processor: Qwen3VLVideoProcessor {\n \"crop_size\": null,\n \"data_format\": \"channels_first\",\n \"default_to_square\": true,\n \"device\": null,\n \"do_center_crop\": null,\n \"do_convert_rgb\": true,\n \"do_normalize\": true,\n \"do_rescale\": true,\n \"do_resize\": true,\n \"do_sample_frames\": true,\n \"fps\": 2,\n \"image_mean\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"image_std\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"input_data_format\": null,\n \"max_frames\": 768,\n \"merge_size\": 2,\n \"min_frames\": 4,\n \"num_frames\": null,\n \"pad_size\": null,\n \"patch_size\": 16,\n \"processor_class\": \"Qwen3VLProcessor\",\n \"resample\": 3,\n \"rescale_factor\": 0.00392156862745098,\n \"return_metadata\": false,\n \"size\": {\n \"longest_edge\": 25165824,\n \"shortest_edge\": 4096\n },\n \"temporal_patch_size\": 2,\n \"video_metadata\": null,\n \"video_processor_type\": \"Qwen3VLVideoProcessor\"\n}\n\n\n{\n \"processor_class\": \"Qwen3VLProcessor\"\n}\n",
25
- "embodiment_id_mapping": "{'robocasa_panda_omron': 13, 'oxe_droid': 17, 'oxe_fractal': 18, 'oxe_language_table': 19, 'oxe_bridge': 20, 'unknown': 22, 'gr1_unified': 20, 'agibot': 26, 'sim_behavior_r1_pro': 23, 'xdof': 24, 'xdof_oss_data': 25, 'unitree_g1_full_body_with_waist_height_nav_cmd': 25, 'real_r1_pro_sharpa': 27, 'real_r1_pro_sharpa_add_view': 27, 'real_r1_pro_sharpa_relative_arm_joint': 26, 'real_r1_pro_sharpa_delta_eef': 26, 'real_r1_pro_sharpa_absolute_eef': 26, 'real_r1_pro_sharpa_meanstd': 26, 'real_r1_pro_sharpa_relative_eef': 26, 'real_r1_pro_sharpa_relative_eef_add_view': 26, 'real_r1_pro_sharpa_relative_eef_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_human': 26, 'real_r1_pro_sharpa_relative_eef_human_add_view': 26, 'real_r1_pro_sharpa_relative_eef_human_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_egodex': 26, 'real_r1_pro_sharpa_relative_eef_egodex_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_egodex_wrist_only': 26, 'real_r1_pro_sharpa_relative_eef_maxinsights': 26, 'real_r1_pro_sharpa_relative_eef_maxinsights_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_mecka': 26, 'real_r1_pro_sharpa_relative_eef_mecka_relative_hand': 26, 'real_g1_relative_eef_absolute_joints': 25, 'real_g1_relative_eef_absolute_joints_wrist_cam': 25, 'real_g1_relative_eef_relative_joints': 25, 'real_r1_pro_sharpa_relative_eef_relative_hand_relative_joint': 26, 'real_r1_pro_sharpa_relative_joint': 29, 'oxe_droid_relative_eef_relative_joint': 24, 'oxe_droid_relative_eef_relative_joint_swapped': 24, 'oxe_droid_relative_eef_relative_joint_upweight_z': 24, 'oxe_droid_relative_eef_relative_joint_upweight_z_swapped': 24, 'oxe_droid_relative_eef_relative_joint_3view': 24, 'oxe_droid_relative_eef_relative_joint_3view_swapped': 24, 'oxe_droid_relative_eef': 24, 'oxe_droid_joint_position_relative': 24, 'xdof_relative_eef_relative_joint': 27, 'xdof_relative_eef_relative_joint_subtask': 27, 'xdof_relative_eef': 27, 'xdof_relative_joint': 28, 'simpler_env_google': 0, 'simpler_env_widowx': 1, 'libero_sim': 2, 'droid_sim': 3, 'robocasa_gr1_tabletop': 10, 'new_embodiment': 10, 'unitree_g1_sonic': 11}",
26
  "shortest_image_edge": "256",
27
  "crop_fraction": "0.95",
28
  "statistics": "{}",
 
22
  "random_rotation_angle": "0",
23
  "color_jitter_params": "{'brightness': 0.3, 'contrast': 0.4, 'saturation': 0.5, 'hue': 0.08}",
24
  "processor": "Qwen3VLProcessor:\n- image_processor: Qwen2VLImageProcessorFast {\n \"crop_size\": null,\n \"data_format\": \"channels_first\",\n \"default_to_square\": true,\n \"device\": null,\n \"disable_grouping\": null,\n \"do_center_crop\": null,\n \"do_convert_rgb\": true,\n \"do_normalize\": true,\n \"do_pad\": null,\n \"do_rescale\": true,\n \"do_resize\": true,\n \"image_mean\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"image_processor_type\": \"Qwen2VLImageProcessorFast\",\n \"image_std\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"input_data_format\": null,\n \"max_pixels\": null,\n \"merge_size\": 2,\n \"min_pixels\": null,\n \"pad_size\": null,\n \"patch_size\": 16,\n \"processor_class\": \"Qwen3VLProcessor\",\n \"resample\": 3,\n \"rescale_factor\": 0.00392156862745098,\n \"return_tensors\": null,\n \"size\": {\n \"longest_edge\": 16777216,\n \"shortest_edge\": 65536\n },\n \"temporal_patch_size\": 2\n}\n\n- tokenizer: Qwen2TokenizerFast(name_or_path='nvidia/Cosmos-Reason2-2B', vocab_size=151643, model_max_length=262144, is_fast=True, padding_side='left', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={\n\t151643: AddedToken(\"<|endoftext|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151644: AddedToken(\"<|im_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151645: AddedToken(\"<|im_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151646: AddedToken(\"<|object_ref_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151647: AddedToken(\"<|object_ref_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151648: AddedToken(\"<|box_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151649: AddedToken(\"<|box_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151650: AddedToken(\"<|quad_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151651: AddedToken(\"<|quad_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151652: AddedToken(\"<|vision_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151653: AddedToken(\"<|vision_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151654: AddedToken(\"<|vision_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151655: AddedToken(\"<|image_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151656: AddedToken(\"<|video_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151657: AddedToken(\"<tool_call>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151658: AddedToken(\"</tool_call>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151659: AddedToken(\"<|fim_prefix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151660: AddedToken(\"<|fim_middle|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151661: AddedToken(\"<|fim_suffix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151662: AddedToken(\"<|fim_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151663: AddedToken(\"<|repo_name|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151664: AddedToken(\"<|file_sep|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151665: AddedToken(\"<tool_response>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151666: AddedToken(\"</tool_response>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151667: AddedToken(\"<think>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151668: AddedToken(\"</think>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n}\n)\n- video_processor: Qwen3VLVideoProcessor {\n \"crop_size\": null,\n \"data_format\": \"channels_first\",\n \"default_to_square\": true,\n \"device\": null,\n \"do_center_crop\": null,\n \"do_convert_rgb\": true,\n \"do_normalize\": true,\n \"do_rescale\": true,\n \"do_resize\": true,\n \"do_sample_frames\": true,\n \"fps\": 2,\n \"image_mean\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"image_std\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"input_data_format\": null,\n \"max_frames\": 768,\n \"merge_size\": 2,\n \"min_frames\": 4,\n \"num_frames\": null,\n \"pad_size\": null,\n \"patch_size\": 16,\n \"processor_class\": \"Qwen3VLProcessor\",\n \"resample\": 3,\n \"rescale_factor\": 0.00392156862745098,\n \"return_metadata\": false,\n \"size\": {\n \"longest_edge\": 25165824,\n \"shortest_edge\": 4096\n },\n \"temporal_patch_size\": 2,\n \"video_metadata\": null,\n \"video_processor_type\": \"Qwen3VLVideoProcessor\"\n}\n\n\n{\n \"processor_class\": \"Qwen3VLProcessor\"\n}\n",
25
+ "embodiment_id_mapping": "{'robocasa_panda_omron': 13, 'oxe_droid': 17, 'oxe_fractal': 18, 'oxe_language_table': 19, 'oxe_bridge': 20, 'unknown': 22, 'gr1_unified': 20, 'agibot': 26, 'sim_behavior_r1_pro': 23, 'xdof': 24, 'xdof_oss_data': 25, 'unitree_g1_full_body_with_waist_height_nav_cmd': 25, 'real_r1_pro_sharpa': 27, 'real_r1_pro_sharpa_add_view': 27, 'real_r1_pro_sharpa_relative_arm_joint': 26, 'real_r1_pro_sharpa_delta_eef': 26, 'real_r1_pro_sharpa_absolute_eef': 26, 'real_r1_pro_sharpa_meanstd': 26, 'real_r1_pro_sharpa_relative_eef': 26, 'real_r1_pro_sharpa_relative_eef_add_view': 26, 'real_r1_pro_sharpa_relative_eef_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_human': 26, 'real_r1_pro_sharpa_relative_eef_human_add_view': 26, 'real_r1_pro_sharpa_relative_eef_human_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_egodex': 26, 'real_r1_pro_sharpa_relative_eef_egodex_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_egodex_wrist_only': 26, 'real_r1_pro_sharpa_relative_eef_maxinsights': 26, 'real_r1_pro_sharpa_relative_eef_maxinsights_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_mecka': 26, 'real_r1_pro_sharpa_relative_eef_mecka_relative_hand': 26, 'real_g1_relative_eef_absolute_joints': 25, 'real_g1_relative_eef_absolute_joints_wrist_cam': 25, 'real_g1_relative_eef_relative_joints': 25, 'real_r1_pro_sharpa_relative_eef_relative_hand_relative_joint': 26, 'real_r1_pro_sharpa_relative_joint': 29, 'oxe_droid_relative_eef_relative_joint': 24, 'oxe_droid_relative_eef_relative_joint_swapped': 24, 'oxe_droid_relative_eef_relative_joint_upweight_z': 24, 'oxe_droid_relative_eef_relative_joint_upweight_z_swapped': 24, 'oxe_droid_relative_eef_relative_joint_3view': 24, 'oxe_droid_relative_eef_relative_joint_3view_swapped': 24, 'oxe_droid_relative_eef': 24, 'oxe_droid_joint_position_relative': 24, 'xdof_relative_eef_relative_joint': 27, 'xdof_relative_eef_relative_joint_subtask': 27, 'xdof_relative_eef': 27, 'xdof_relative_joint': 28, 'simpler_env_google': 0, 'simpler_env_widowx': 1, 'libero_sim': 2, 'droid_sim': 3, 'new_embodiment': 10, 'robocasa_gr1_tabletop': 10, 'unitree_g1_sonic': 11}",
26
  "shortest_image_edge": "256",
27
  "crop_fraction": "0.95",
28
  "statistics": "{}",
a3-fridge-open/epoch-010/experiment_cfg/conf.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/a3/production
177
  experiment_name: null
178
- max_steps: 56
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 28
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
@@ -230,5 +230,5 @@ training:
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
- max_steps: 56
234
- save_steps: 28
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/a3/production
177
  experiment_name: null
178
+ max_steps: 810
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 405
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
+ max_steps: 810
234
+ save_steps: 405
a3-fridge-open/epoch-010/experiment_cfg/config.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/a3/production
177
  experiment_name: null
178
- max_steps: 56
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 28
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/a3/production
177
  experiment_name: null
178
+ max_steps: 810
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 405
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
a3-fridge-open/epoch-010/protocol.json CHANGED
@@ -4,22 +4,26 @@
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "A3 open fridge door (G1)",
6
  "task_dir": "a3-fridge-open",
 
7
  "robot": "Unitree G1",
8
  "embodiment_tag": "new_embodiment",
9
  "epoch_completed": 10,
10
- "global_step": 28,
11
  "total_epochs": 20,
12
- "total_steps": 56,
13
- "save_steps": 28,
14
- "per_gpu_batch": 320,
15
- "global_batch": 10240,
16
- "num_gpus": 32,
 
17
  "lr": 0.0001,
18
  "lr_schedule": "cosine, warmup_ratio 0.05",
 
19
  "deepspeed_stage": 2,
20
  "episodes": 100,
21
  "frames": 28180,
22
- "fps": 30,
 
23
  "cameras": [
24
  "observation.images.ego_view",
25
  "observation.images.chest"
@@ -43,10 +47,10 @@
43
  "action_dim": 76,
44
  "action_chunk": 16,
45
  "model_action_horizon": 40,
46
- "state_encoding": "raw (apply_sincos_state_encoding false)",
47
  "normalization": "statistics.json (percentiles, use_percentiles true)",
48
- "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09) via code/hb_groot/hb_launch.py; GR00T native random crop / colour jitter / rotation bypassed in training (config.json still lists the native defaults, they are not used)",
49
- "config": "experiment_cfg/conf.yaml (full training config); dataset_meta/modality.json (state/action/video keys)",
50
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
51
  "resume_state_included": false,
52
  "files": {
@@ -54,8 +58,8 @@
54
  "dataset_meta/hb-dataset-receipt.json": 815,
55
  "dataset_meta/modality.json": 1537,
56
  "embodiment_id.json": 2226,
57
- "experiment_cfg/conf.yaml": 5789,
58
- "experiment_cfg/config.yaml": 5760,
59
  "experiment_cfg/dataset_statistics.json": 20379,
60
  "experiment_cfg/final_model_config.json": 1586,
61
  "experiment_cfg/final_processor_config.json": 2548329,
 
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "A3 open fridge door (G1)",
6
  "task_dir": "a3-fridge-open",
7
+ "training_data": "a3-fridge-open-expert-v1",
8
  "robot": "Unitree G1",
9
  "embodiment_tag": "new_embodiment",
10
  "epoch_completed": 10,
11
+ "global_step": 405,
12
  "total_epochs": 20,
13
+ "total_steps": 810,
14
+ "save_steps": 405,
15
+ "per_gpu_batch": 87,
16
+ "num_gpus": 8,
17
+ "num_nodes": 1,
18
+ "global_batch": 696,
19
  "lr": 0.0001,
20
  "lr_schedule": "cosine, warmup_ratio 0.05",
21
+ "weight_decay": 1e-05,
22
  "deepspeed_stage": 2,
23
  "episodes": 100,
24
  "frames": 28180,
25
+ "fps": 30.0,
26
+ "control_hz": 30.0,
27
  "cameras": [
28
  "observation.images.ego_view",
29
  "observation.images.chest"
 
47
  "action_dim": 76,
48
  "action_chunk": 16,
49
  "model_action_horizon": 40,
50
+ "state_encoding": "raw (apply_sincos_state_encoding false, GR00T N1.7 default)",
51
  "normalization": "statistics.json (percentiles, use_percentiles true)",
52
+ "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09), on from epoch 1; model-native image augmentation off (via code/hb_groot/hb_launch.py; config.json still lists GR00T's native augmentation defaults, they are bypassed in training)",
53
+ "config": "experiment_cfg/conf.yaml (full model config); dataset_meta/modality.json (state/action/video keys)",
54
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
55
  "resume_state_included": false,
56
  "files": {
 
58
  "dataset_meta/hb-dataset-receipt.json": 815,
59
  "dataset_meta/modality.json": 1537,
60
  "embodiment_id.json": 2226,
61
+ "experiment_cfg/conf.yaml": 5790,
62
+ "experiment_cfg/config.yaml": 5759,
63
  "experiment_cfg/dataset_statistics.json": 20379,
64
  "experiment_cfg/final_model_config.json": 1586,
65
  "experiment_cfg/final_processor_config.json": 2548329,
a3-fridge-open/epoch-020/experiment_cfg/conf.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/a3/production
177
  experiment_name: null
178
- max_steps: 56
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 28
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
@@ -230,5 +230,5 @@ training:
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
- max_steps: 56
234
- save_steps: 28
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/a3/production
177
  experiment_name: null
178
+ max_steps: 810
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 405
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
+ max_steps: 810
234
+ save_steps: 405
a3-fridge-open/epoch-020/experiment_cfg/config.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/a3/production
177
  experiment_name: null
178
- max_steps: 56
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 28
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/a3/production
177
  experiment_name: null
178
+ max_steps: 810
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 405
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
a3-fridge-open/epoch-020/protocol.json CHANGED
@@ -4,22 +4,26 @@
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "A3 open fridge door (G1)",
6
  "task_dir": "a3-fridge-open",
 
7
  "robot": "Unitree G1",
8
  "embodiment_tag": "new_embodiment",
9
  "epoch_completed": 20,
10
- "global_step": 56,
11
  "total_epochs": 20,
12
- "total_steps": 56,
13
- "save_steps": 28,
14
- "per_gpu_batch": 320,
15
- "global_batch": 10240,
16
- "num_gpus": 32,
 
17
  "lr": 0.0001,
18
  "lr_schedule": "cosine, warmup_ratio 0.05",
 
19
  "deepspeed_stage": 2,
20
  "episodes": 100,
21
  "frames": 28180,
22
- "fps": 30,
 
23
  "cameras": [
24
  "observation.images.ego_view",
25
  "observation.images.chest"
@@ -43,10 +47,10 @@
43
  "action_dim": 76,
44
  "action_chunk": 16,
45
  "model_action_horizon": 40,
46
- "state_encoding": "raw (apply_sincos_state_encoding false)",
47
  "normalization": "statistics.json (percentiles, use_percentiles true)",
48
- "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09) via code/hb_groot/hb_launch.py; GR00T native random crop / colour jitter / rotation bypassed in training (config.json still lists the native defaults, they are not used)",
49
- "config": "experiment_cfg/conf.yaml (full training config); dataset_meta/modality.json (state/action/video keys)",
50
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
51
  "resume_state_included": false,
52
  "files": {
@@ -54,8 +58,8 @@
54
  "dataset_meta/hb-dataset-receipt.json": 815,
55
  "dataset_meta/modality.json": 1537,
56
  "embodiment_id.json": 2226,
57
- "experiment_cfg/conf.yaml": 5789,
58
- "experiment_cfg/config.yaml": 5760,
59
  "experiment_cfg/dataset_statistics.json": 20379,
60
  "experiment_cfg/final_model_config.json": 1586,
61
  "experiment_cfg/final_processor_config.json": 2548329,
 
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "A3 open fridge door (G1)",
6
  "task_dir": "a3-fridge-open",
7
+ "training_data": "a3-fridge-open-expert-v1",
8
  "robot": "Unitree G1",
9
  "embodiment_tag": "new_embodiment",
10
  "epoch_completed": 20,
11
+ "global_step": 810,
12
  "total_epochs": 20,
13
+ "total_steps": 810,
14
+ "save_steps": 405,
15
+ "per_gpu_batch": 87,
16
+ "num_gpus": 8,
17
+ "num_nodes": 1,
18
+ "global_batch": 696,
19
  "lr": 0.0001,
20
  "lr_schedule": "cosine, warmup_ratio 0.05",
21
+ "weight_decay": 1e-05,
22
  "deepspeed_stage": 2,
23
  "episodes": 100,
24
  "frames": 28180,
25
+ "fps": 30.0,
26
+ "control_hz": 30.0,
27
  "cameras": [
28
  "observation.images.ego_view",
29
  "observation.images.chest"
 
47
  "action_dim": 76,
48
  "action_chunk": 16,
49
  "model_action_horizon": 40,
50
+ "state_encoding": "raw (apply_sincos_state_encoding false, GR00T N1.7 default)",
51
  "normalization": "statistics.json (percentiles, use_percentiles true)",
52
+ "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09), on from epoch 1; model-native image augmentation off (via code/hb_groot/hb_launch.py; config.json still lists GR00T's native augmentation defaults, they are bypassed in training)",
53
+ "config": "experiment_cfg/conf.yaml (full model config); dataset_meta/modality.json (state/action/video keys)",
54
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
55
  "resume_state_included": false,
56
  "files": {
 
58
  "dataset_meta/hb-dataset-receipt.json": 815,
59
  "dataset_meta/modality.json": 1537,
60
  "embodiment_id.json": 2226,
61
+ "experiment_cfg/conf.yaml": 5790,
62
+ "experiment_cfg/config.yaml": 5759,
63
  "experiment_cfg/dataset_statistics.json": 20379,
64
  "experiment_cfg/final_model_config.json": 1586,
65
  "experiment_cfg/final_processor_config.json": 2548329,
a7-cabinet-close/epoch-010/experiment_cfg/conf.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/a7/production
177
  experiment_name: null
178
- max_steps: 60
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 30
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
@@ -230,5 +230,5 @@ training:
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
- max_steps: 60
234
- save_steps: 30
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/a7/production
177
  experiment_name: null
178
+ max_steps: 882
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 441
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
+ max_steps: 882
234
+ save_steps: 441
a7-cabinet-close/epoch-010/experiment_cfg/config.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/a7/production
177
  experiment_name: null
178
- max_steps: 60
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 30
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/a7/production
177
  experiment_name: null
178
+ max_steps: 882
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 441
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
a7-cabinet-close/epoch-010/protocol.json CHANGED
@@ -4,22 +4,26 @@
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "A7 close cabinet door (G1)",
6
  "task_dir": "a7-cabinet-close",
 
7
  "robot": "Unitree G1",
8
  "embodiment_tag": "new_embodiment",
9
  "epoch_completed": 10,
10
- "global_step": 30,
11
  "total_epochs": 20,
12
- "total_steps": 60,
13
- "save_steps": 30,
14
- "per_gpu_batch": 320,
15
- "global_batch": 10240,
16
- "num_gpus": 32,
 
17
  "lr": 0.0001,
18
  "lr_schedule": "cosine, warmup_ratio 0.05",
 
19
  "deepspeed_stage": 2,
20
  "episodes": 100,
21
  "frames": 30664,
22
- "fps": 30,
 
23
  "cameras": [
24
  "observation.images.ego_view",
25
  "observation.images.chest"
@@ -43,10 +47,10 @@
43
  "action_dim": 76,
44
  "action_chunk": 16,
45
  "model_action_horizon": 40,
46
- "state_encoding": "raw (apply_sincos_state_encoding false)",
47
  "normalization": "statistics.json (percentiles, use_percentiles true)",
48
- "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09) via code/hb_groot/hb_launch.py; GR00T native random crop / colour jitter / rotation bypassed in training (config.json still lists the native defaults, they are not used)",
49
- "config": "experiment_cfg/conf.yaml (full training config); dataset_meta/modality.json (state/action/video keys)",
50
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
51
  "resume_state_included": false,
52
  "files": {
@@ -54,8 +58,8 @@
54
  "dataset_meta/hb-dataset-receipt.json": 817,
55
  "dataset_meta/modality.json": 1537,
56
  "embodiment_id.json": 2226,
57
- "experiment_cfg/conf.yaml": 5789,
58
- "experiment_cfg/config.yaml": 5760,
59
  "experiment_cfg/dataset_statistics.json": 20397,
60
  "experiment_cfg/final_model_config.json": 1586,
61
  "experiment_cfg/final_processor_config.json": 2548329,
 
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "A7 close cabinet door (G1)",
6
  "task_dir": "a7-cabinet-close",
7
+ "training_data": "a7-cabinet-close-expert-v1",
8
  "robot": "Unitree G1",
9
  "embodiment_tag": "new_embodiment",
10
  "epoch_completed": 10,
11
+ "global_step": 441,
12
  "total_epochs": 20,
13
+ "total_steps": 882,
14
+ "save_steps": 441,
15
+ "per_gpu_batch": 87,
16
+ "num_gpus": 8,
17
+ "num_nodes": 1,
18
+ "global_batch": 696,
19
  "lr": 0.0001,
20
  "lr_schedule": "cosine, warmup_ratio 0.05",
21
+ "weight_decay": 1e-05,
22
  "deepspeed_stage": 2,
23
  "episodes": 100,
24
  "frames": 30664,
25
+ "fps": 30.0,
26
+ "control_hz": 30.0,
27
  "cameras": [
28
  "observation.images.ego_view",
29
  "observation.images.chest"
 
47
  "action_dim": 76,
48
  "action_chunk": 16,
49
  "model_action_horizon": 40,
50
+ "state_encoding": "raw (apply_sincos_state_encoding false, GR00T N1.7 default)",
51
  "normalization": "statistics.json (percentiles, use_percentiles true)",
52
+ "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09), on from epoch 1; model-native image augmentation off (via code/hb_groot/hb_launch.py; config.json still lists GR00T's native augmentation defaults, they are bypassed in training)",
53
+ "config": "experiment_cfg/conf.yaml (full model config); dataset_meta/modality.json (state/action/video keys)",
54
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
55
  "resume_state_included": false,
56
  "files": {
 
58
  "dataset_meta/hb-dataset-receipt.json": 817,
59
  "dataset_meta/modality.json": 1537,
60
  "embodiment_id.json": 2226,
61
+ "experiment_cfg/conf.yaml": 5790,
62
+ "experiment_cfg/config.yaml": 5759,
63
  "experiment_cfg/dataset_statistics.json": 20397,
64
  "experiment_cfg/final_model_config.json": 1586,
65
  "experiment_cfg/final_processor_config.json": 2548329,
a7-cabinet-close/epoch-020/experiment_cfg/conf.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/a7/production
177
  experiment_name: null
178
- max_steps: 60
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 30
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
@@ -230,5 +230,5 @@ training:
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
- max_steps: 60
234
- save_steps: 30
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/a7/production
177
  experiment_name: null
178
+ max_steps: 882
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 441
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
+ max_steps: 882
234
+ save_steps: 441
a7-cabinet-close/epoch-020/experiment_cfg/config.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/a7/production
177
  experiment_name: null
178
- max_steps: 60
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 30
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/a7/production
177
  experiment_name: null
178
+ max_steps: 882
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 441
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
a7-cabinet-close/epoch-020/protocol.json CHANGED
@@ -4,22 +4,26 @@
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "A7 close cabinet door (G1)",
6
  "task_dir": "a7-cabinet-close",
 
7
  "robot": "Unitree G1",
8
  "embodiment_tag": "new_embodiment",
9
  "epoch_completed": 20,
10
- "global_step": 60,
11
  "total_epochs": 20,
12
- "total_steps": 60,
13
- "save_steps": 30,
14
- "per_gpu_batch": 320,
15
- "global_batch": 10240,
16
- "num_gpus": 32,
 
17
  "lr": 0.0001,
18
  "lr_schedule": "cosine, warmup_ratio 0.05",
 
19
  "deepspeed_stage": 2,
20
  "episodes": 100,
21
  "frames": 30664,
22
- "fps": 30,
 
23
  "cameras": [
24
  "observation.images.ego_view",
25
  "observation.images.chest"
@@ -43,10 +47,10 @@
43
  "action_dim": 76,
44
  "action_chunk": 16,
45
  "model_action_horizon": 40,
46
- "state_encoding": "raw (apply_sincos_state_encoding false)",
47
  "normalization": "statistics.json (percentiles, use_percentiles true)",
48
- "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09) via code/hb_groot/hb_launch.py; GR00T native random crop / colour jitter / rotation bypassed in training (config.json still lists the native defaults, they are not used)",
49
- "config": "experiment_cfg/conf.yaml (full training config); dataset_meta/modality.json (state/action/video keys)",
50
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
51
  "resume_state_included": false,
52
  "files": {
@@ -54,8 +58,8 @@
54
  "dataset_meta/hb-dataset-receipt.json": 817,
55
  "dataset_meta/modality.json": 1537,
56
  "embodiment_id.json": 2226,
57
- "experiment_cfg/conf.yaml": 5789,
58
- "experiment_cfg/config.yaml": 5760,
59
  "experiment_cfg/dataset_statistics.json": 20397,
60
  "experiment_cfg/final_model_config.json": 1586,
61
  "experiment_cfg/final_processor_config.json": 2548329,
 
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "A7 close cabinet door (G1)",
6
  "task_dir": "a7-cabinet-close",
7
+ "training_data": "a7-cabinet-close-expert-v1",
8
  "robot": "Unitree G1",
9
  "embodiment_tag": "new_embodiment",
10
  "epoch_completed": 20,
11
+ "global_step": 882,
12
  "total_epochs": 20,
13
+ "total_steps": 882,
14
+ "save_steps": 441,
15
+ "per_gpu_batch": 87,
16
+ "num_gpus": 8,
17
+ "num_nodes": 1,
18
+ "global_batch": 696,
19
  "lr": 0.0001,
20
  "lr_schedule": "cosine, warmup_ratio 0.05",
21
+ "weight_decay": 1e-05,
22
  "deepspeed_stage": 2,
23
  "episodes": 100,
24
  "frames": 30664,
25
+ "fps": 30.0,
26
+ "control_hz": 30.0,
27
  "cameras": [
28
  "observation.images.ego_view",
29
  "observation.images.chest"
 
47
  "action_dim": 76,
48
  "action_chunk": 16,
49
  "model_action_horizon": 40,
50
+ "state_encoding": "raw (apply_sincos_state_encoding false, GR00T N1.7 default)",
51
  "normalization": "statistics.json (percentiles, use_percentiles true)",
52
+ "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09), on from epoch 1; model-native image augmentation off (via code/hb_groot/hb_launch.py; config.json still lists GR00T's native augmentation defaults, they are bypassed in training)",
53
+ "config": "experiment_cfg/conf.yaml (full model config); dataset_meta/modality.json (state/action/video keys)",
54
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
55
  "resume_state_included": false,
56
  "files": {
 
58
  "dataset_meta/hb-dataset-receipt.json": 817,
59
  "dataset_meta/modality.json": 1537,
60
  "embodiment_id.json": 2226,
61
+ "experiment_cfg/conf.yaml": 5790,
62
+ "experiment_cfg/config.yaml": 5759,
63
  "experiment_cfg/dataset_statistics.json": 20397,
64
  "experiment_cfg/final_model_config.json": 1586,
65
  "experiment_cfg/final_processor_config.json": 2548329,
b6-pedal-bin/epoch-010/embodiment_id.json CHANGED
@@ -51,7 +51,7 @@
51
  "simpler_env_widowx": 1,
52
  "libero_sim": 2,
53
  "droid_sim": 3,
54
- "robocasa_gr1_tabletop": 10,
55
  "new_embodiment": 10,
 
56
  "unitree_g1_sonic": 11
57
  }
 
51
  "simpler_env_widowx": 1,
52
  "libero_sim": 2,
53
  "droid_sim": 3,
 
54
  "new_embodiment": 10,
55
+ "robocasa_gr1_tabletop": 10,
56
  "unitree_g1_sonic": 11
57
  }
b6-pedal-bin/epoch-010/experiment_cfg/conf.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/b6/production
177
  experiment_name: null
178
- max_steps: 128
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 64
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
@@ -230,5 +230,5 @@ training:
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
- max_steps: 128
234
- save_steps: 64
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/b6/production
177
  experiment_name: null
178
+ max_steps: 1872
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 936
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
230
  - 0
231
  open_loop_eval_steps_per_traj: 100
232
  open_loop_eval_plot_indices: null
233
+ max_steps: 1872
234
+ save_steps: 936
b6-pedal-bin/epoch-010/experiment_cfg/config.yaml CHANGED
@@ -173,10 +173,10 @@ data:
173
  - 224
174
  - 224
175
  training:
176
- output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-bmax/b6/production
177
  experiment_name: null
178
- max_steps: 128
179
- global_batch_size: 10240
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
@@ -193,7 +193,7 @@ training:
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
- save_steps: 64
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
@@ -216,7 +216,7 @@ training:
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
- num_gpus: 32
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
 
173
  - 224
174
  - 224
175
  training:
176
+ output_dir: /mnt/nas/cross/Team/RobotAgent/Personal/baijiaqi/humanoid-benchmark-work/preflight-20261104T1648Z/policies-20261109/groot-runs-b696/b6/production
177
  experiment_name: null
178
+ max_steps: 1872
179
+ global_batch_size: 696
180
  per_gpu_batch_size: null
181
  gradient_accumulation_steps: 1
182
  learning_rate: 0.0001
 
193
  bf16: true
194
  eval_bf16: true
195
  logging_steps: 10
196
+ save_steps: 936
197
  save_total_limit: 2
198
  save_vl_model: false
199
  save_only_model: false
 
216
  transformers_access_token: null
217
  use_ddp: false
218
  ddp_bucket_cap_mb: 100
219
+ num_gpus: 8
220
  dataloader_num_workers: 8
221
  remove_unused_columns: false
222
  use_wandb: false
b6-pedal-bin/epoch-010/experiment_cfg/final_processor_config.json CHANGED
@@ -22,7 +22,7 @@
22
  "random_rotation_angle": "0",
23
  "color_jitter_params": "{'brightness': 0.3, 'contrast': 0.4, 'saturation': 0.5, 'hue': 0.08}",
24
  "processor": "Qwen3VLProcessor:\n- image_processor: Qwen2VLImageProcessorFast {\n \"crop_size\": null,\n \"data_format\": \"channels_first\",\n \"default_to_square\": true,\n \"device\": null,\n \"disable_grouping\": null,\n \"do_center_crop\": null,\n \"do_convert_rgb\": true,\n \"do_normalize\": true,\n \"do_pad\": null,\n \"do_rescale\": true,\n \"do_resize\": true,\n \"image_mean\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"image_processor_type\": \"Qwen2VLImageProcessorFast\",\n \"image_std\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"input_data_format\": null,\n \"max_pixels\": null,\n \"merge_size\": 2,\n \"min_pixels\": null,\n \"pad_size\": null,\n \"patch_size\": 16,\n \"processor_class\": \"Qwen3VLProcessor\",\n \"resample\": 3,\n \"rescale_factor\": 0.00392156862745098,\n \"return_tensors\": null,\n \"size\": {\n \"longest_edge\": 16777216,\n \"shortest_edge\": 65536\n },\n \"temporal_patch_size\": 2\n}\n\n- tokenizer: Qwen2TokenizerFast(name_or_path='nvidia/Cosmos-Reason2-2B', vocab_size=151643, model_max_length=262144, is_fast=True, padding_side='left', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={\n\t151643: AddedToken(\"<|endoftext|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151644: AddedToken(\"<|im_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151645: AddedToken(\"<|im_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151646: AddedToken(\"<|object_ref_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151647: AddedToken(\"<|object_ref_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151648: AddedToken(\"<|box_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151649: AddedToken(\"<|box_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151650: AddedToken(\"<|quad_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151651: AddedToken(\"<|quad_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151652: AddedToken(\"<|vision_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151653: AddedToken(\"<|vision_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151654: AddedToken(\"<|vision_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151655: AddedToken(\"<|image_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151656: AddedToken(\"<|video_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151657: AddedToken(\"<tool_call>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151658: AddedToken(\"</tool_call>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151659: AddedToken(\"<|fim_prefix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151660: AddedToken(\"<|fim_middle|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151661: AddedToken(\"<|fim_suffix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151662: AddedToken(\"<|fim_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151663: AddedToken(\"<|repo_name|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151664: AddedToken(\"<|file_sep|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151665: AddedToken(\"<tool_response>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151666: AddedToken(\"</tool_response>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151667: AddedToken(\"<think>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151668: AddedToken(\"</think>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n}\n)\n- video_processor: Qwen3VLVideoProcessor {\n \"crop_size\": null,\n \"data_format\": \"channels_first\",\n \"default_to_square\": true,\n \"device\": null,\n \"do_center_crop\": null,\n \"do_convert_rgb\": true,\n \"do_normalize\": true,\n \"do_rescale\": true,\n \"do_resize\": true,\n \"do_sample_frames\": true,\n \"fps\": 2,\n \"image_mean\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"image_std\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"input_data_format\": null,\n \"max_frames\": 768,\n \"merge_size\": 2,\n \"min_frames\": 4,\n \"num_frames\": null,\n \"pad_size\": null,\n \"patch_size\": 16,\n \"processor_class\": \"Qwen3VLProcessor\",\n \"resample\": 3,\n \"rescale_factor\": 0.00392156862745098,\n \"return_metadata\": false,\n \"size\": {\n \"longest_edge\": 25165824,\n \"shortest_edge\": 4096\n },\n \"temporal_patch_size\": 2,\n \"video_metadata\": null,\n \"video_processor_type\": \"Qwen3VLVideoProcessor\"\n}\n\n\n{\n \"processor_class\": \"Qwen3VLProcessor\"\n}\n",
25
- "embodiment_id_mapping": "{'robocasa_panda_omron': 13, 'oxe_droid': 17, 'oxe_fractal': 18, 'oxe_language_table': 19, 'oxe_bridge': 20, 'unknown': 22, 'gr1_unified': 20, 'agibot': 26, 'sim_behavior_r1_pro': 23, 'xdof': 24, 'xdof_oss_data': 25, 'unitree_g1_full_body_with_waist_height_nav_cmd': 25, 'real_r1_pro_sharpa': 27, 'real_r1_pro_sharpa_add_view': 27, 'real_r1_pro_sharpa_relative_arm_joint': 26, 'real_r1_pro_sharpa_delta_eef': 26, 'real_r1_pro_sharpa_absolute_eef': 26, 'real_r1_pro_sharpa_meanstd': 26, 'real_r1_pro_sharpa_relative_eef': 26, 'real_r1_pro_sharpa_relative_eef_add_view': 26, 'real_r1_pro_sharpa_relative_eef_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_human': 26, 'real_r1_pro_sharpa_relative_eef_human_add_view': 26, 'real_r1_pro_sharpa_relative_eef_human_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_egodex': 26, 'real_r1_pro_sharpa_relative_eef_egodex_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_egodex_wrist_only': 26, 'real_r1_pro_sharpa_relative_eef_maxinsights': 26, 'real_r1_pro_sharpa_relative_eef_maxinsights_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_mecka': 26, 'real_r1_pro_sharpa_relative_eef_mecka_relative_hand': 26, 'real_g1_relative_eef_absolute_joints': 25, 'real_g1_relative_eef_absolute_joints_wrist_cam': 25, 'real_g1_relative_eef_relative_joints': 25, 'real_r1_pro_sharpa_relative_eef_relative_hand_relative_joint': 26, 'real_r1_pro_sharpa_relative_joint': 29, 'oxe_droid_relative_eef_relative_joint': 24, 'oxe_droid_relative_eef_relative_joint_swapped': 24, 'oxe_droid_relative_eef_relative_joint_upweight_z': 24, 'oxe_droid_relative_eef_relative_joint_upweight_z_swapped': 24, 'oxe_droid_relative_eef_relative_joint_3view': 24, 'oxe_droid_relative_eef_relative_joint_3view_swapped': 24, 'oxe_droid_relative_eef': 24, 'oxe_droid_joint_position_relative': 24, 'xdof_relative_eef_relative_joint': 27, 'xdof_relative_eef_relative_joint_subtask': 27, 'xdof_relative_eef': 27, 'xdof_relative_joint': 28, 'simpler_env_google': 0, 'simpler_env_widowx': 1, 'libero_sim': 2, 'droid_sim': 3, 'robocasa_gr1_tabletop': 10, 'new_embodiment': 10, 'unitree_g1_sonic': 11}",
26
  "shortest_image_edge": "256",
27
  "crop_fraction": "0.95",
28
  "statistics": "{}",
 
22
  "random_rotation_angle": "0",
23
  "color_jitter_params": "{'brightness': 0.3, 'contrast': 0.4, 'saturation': 0.5, 'hue': 0.08}",
24
  "processor": "Qwen3VLProcessor:\n- image_processor: Qwen2VLImageProcessorFast {\n \"crop_size\": null,\n \"data_format\": \"channels_first\",\n \"default_to_square\": true,\n \"device\": null,\n \"disable_grouping\": null,\n \"do_center_crop\": null,\n \"do_convert_rgb\": true,\n \"do_normalize\": true,\n \"do_pad\": null,\n \"do_rescale\": true,\n \"do_resize\": true,\n \"image_mean\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"image_processor_type\": \"Qwen2VLImageProcessorFast\",\n \"image_std\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"input_data_format\": null,\n \"max_pixels\": null,\n \"merge_size\": 2,\n \"min_pixels\": null,\n \"pad_size\": null,\n \"patch_size\": 16,\n \"processor_class\": \"Qwen3VLProcessor\",\n \"resample\": 3,\n \"rescale_factor\": 0.00392156862745098,\n \"return_tensors\": null,\n \"size\": {\n \"longest_edge\": 16777216,\n \"shortest_edge\": 65536\n },\n \"temporal_patch_size\": 2\n}\n\n- tokenizer: Qwen2TokenizerFast(name_or_path='nvidia/Cosmos-Reason2-2B', vocab_size=151643, model_max_length=262144, is_fast=True, padding_side='left', truncation_side='right', special_tokens={'eos_token': '<|im_end|>', 'pad_token': '<|endoftext|>', 'additional_special_tokens': ['<|im_start|>', '<|im_end|>', '<|object_ref_start|>', '<|object_ref_end|>', '<|box_start|>', '<|box_end|>', '<|quad_start|>', '<|quad_end|>', '<|vision_start|>', '<|vision_end|>', '<|vision_pad|>', '<|image_pad|>', '<|video_pad|>']}, clean_up_tokenization_spaces=False, added_tokens_decoder={\n\t151643: AddedToken(\"<|endoftext|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151644: AddedToken(\"<|im_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151645: AddedToken(\"<|im_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151646: AddedToken(\"<|object_ref_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151647: AddedToken(\"<|object_ref_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151648: AddedToken(\"<|box_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151649: AddedToken(\"<|box_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151650: AddedToken(\"<|quad_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151651: AddedToken(\"<|quad_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151652: AddedToken(\"<|vision_start|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151653: AddedToken(\"<|vision_end|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151654: AddedToken(\"<|vision_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151655: AddedToken(\"<|image_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151656: AddedToken(\"<|video_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=True),\n\t151657: AddedToken(\"<tool_call>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151658: AddedToken(\"</tool_call>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151659: AddedToken(\"<|fim_prefix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151660: AddedToken(\"<|fim_middle|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151661: AddedToken(\"<|fim_suffix|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151662: AddedToken(\"<|fim_pad|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151663: AddedToken(\"<|repo_name|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151664: AddedToken(\"<|file_sep|>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151665: AddedToken(\"<tool_response>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151666: AddedToken(\"</tool_response>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151667: AddedToken(\"<think>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n\t151668: AddedToken(\"</think>\", rstrip=False, lstrip=False, single_word=False, normalized=False, special=False),\n}\n)\n- video_processor: Qwen3VLVideoProcessor {\n \"crop_size\": null,\n \"data_format\": \"channels_first\",\n \"default_to_square\": true,\n \"device\": null,\n \"do_center_crop\": null,\n \"do_convert_rgb\": true,\n \"do_normalize\": true,\n \"do_rescale\": true,\n \"do_resize\": true,\n \"do_sample_frames\": true,\n \"fps\": 2,\n \"image_mean\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"image_std\": [\n 0.5,\n 0.5,\n 0.5\n ],\n \"input_data_format\": null,\n \"max_frames\": 768,\n \"merge_size\": 2,\n \"min_frames\": 4,\n \"num_frames\": null,\n \"pad_size\": null,\n \"patch_size\": 16,\n \"processor_class\": \"Qwen3VLProcessor\",\n \"resample\": 3,\n \"rescale_factor\": 0.00392156862745098,\n \"return_metadata\": false,\n \"size\": {\n \"longest_edge\": 25165824,\n \"shortest_edge\": 4096\n },\n \"temporal_patch_size\": 2,\n \"video_metadata\": null,\n \"video_processor_type\": \"Qwen3VLVideoProcessor\"\n}\n\n\n{\n \"processor_class\": \"Qwen3VLProcessor\"\n}\n",
25
+ "embodiment_id_mapping": "{'robocasa_panda_omron': 13, 'oxe_droid': 17, 'oxe_fractal': 18, 'oxe_language_table': 19, 'oxe_bridge': 20, 'unknown': 22, 'gr1_unified': 20, 'agibot': 26, 'sim_behavior_r1_pro': 23, 'xdof': 24, 'xdof_oss_data': 25, 'unitree_g1_full_body_with_waist_height_nav_cmd': 25, 'real_r1_pro_sharpa': 27, 'real_r1_pro_sharpa_add_view': 27, 'real_r1_pro_sharpa_relative_arm_joint': 26, 'real_r1_pro_sharpa_delta_eef': 26, 'real_r1_pro_sharpa_absolute_eef': 26, 'real_r1_pro_sharpa_meanstd': 26, 'real_r1_pro_sharpa_relative_eef': 26, 'real_r1_pro_sharpa_relative_eef_add_view': 26, 'real_r1_pro_sharpa_relative_eef_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_human': 26, 'real_r1_pro_sharpa_relative_eef_human_add_view': 26, 'real_r1_pro_sharpa_relative_eef_human_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_egodex': 26, 'real_r1_pro_sharpa_relative_eef_egodex_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_egodex_wrist_only': 26, 'real_r1_pro_sharpa_relative_eef_maxinsights': 26, 'real_r1_pro_sharpa_relative_eef_maxinsights_relative_hand': 26, 'real_r1_pro_sharpa_relative_eef_mecka': 26, 'real_r1_pro_sharpa_relative_eef_mecka_relative_hand': 26, 'real_g1_relative_eef_absolute_joints': 25, 'real_g1_relative_eef_absolute_joints_wrist_cam': 25, 'real_g1_relative_eef_relative_joints': 25, 'real_r1_pro_sharpa_relative_eef_relative_hand_relative_joint': 26, 'real_r1_pro_sharpa_relative_joint': 29, 'oxe_droid_relative_eef_relative_joint': 24, 'oxe_droid_relative_eef_relative_joint_swapped': 24, 'oxe_droid_relative_eef_relative_joint_upweight_z': 24, 'oxe_droid_relative_eef_relative_joint_upweight_z_swapped': 24, 'oxe_droid_relative_eef_relative_joint_3view': 24, 'oxe_droid_relative_eef_relative_joint_3view_swapped': 24, 'oxe_droid_relative_eef': 24, 'oxe_droid_joint_position_relative': 24, 'xdof_relative_eef_relative_joint': 27, 'xdof_relative_eef_relative_joint_subtask': 27, 'xdof_relative_eef': 27, 'xdof_relative_joint': 28, 'simpler_env_google': 0, 'simpler_env_widowx': 1, 'libero_sim': 2, 'droid_sim': 3, 'new_embodiment': 10, 'robocasa_gr1_tabletop': 10, 'unitree_g1_sonic': 11}",
26
  "shortest_image_edge": "256",
27
  "crop_fraction": "0.95",
28
  "statistics": "{}",
b6-pedal-bin/epoch-010/protocol.json CHANGED
@@ -4,22 +4,26 @@
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "B6 pedal-bin trash disposal (G1)",
6
  "task_dir": "b6-pedal-bin",
 
7
  "robot": "Unitree G1",
8
  "embodiment_tag": "new_embodiment",
9
  "epoch_completed": 10,
10
- "global_step": 64,
11
  "total_epochs": 20,
12
- "total_steps": 128,
13
- "save_steps": 64,
14
- "per_gpu_batch": 320,
15
- "global_batch": 10240,
16
- "num_gpus": 32,
 
17
  "lr": 0.0001,
18
  "lr_schedule": "cosine, warmup_ratio 0.05",
 
19
  "deepspeed_stage": 2,
20
  "episodes": 100,
21
  "frames": 65127,
22
- "fps": 30,
 
23
  "cameras": [
24
  "observation.images.ego_view",
25
  "observation.images.chest"
@@ -43,10 +47,10 @@
43
  "action_dim": 76,
44
  "action_chunk": 16,
45
  "model_action_horizon": 40,
46
- "state_encoding": "raw (apply_sincos_state_encoding false)",
47
  "normalization": "statistics.json (percentiles, use_percentiles true)",
48
- "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09) via code/hb_groot/hb_launch.py; GR00T native random crop / colour jitter / rotation bypassed in training (config.json still lists the native defaults, they are not used)",
49
- "config": "experiment_cfg/conf.yaml (full training config); dataset_meta/modality.json (state/action/video keys)",
50
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
51
  "resume_state_included": false,
52
  "files": {
@@ -54,8 +58,8 @@
54
  "dataset_meta/hb-dataset-receipt.json": 816,
55
  "dataset_meta/modality.json": 1537,
56
  "embodiment_id.json": 2226,
57
- "experiment_cfg/conf.yaml": 5791,
58
- "experiment_cfg/config.yaml": 5761,
59
  "experiment_cfg/dataset_statistics.json": 20262,
60
  "experiment_cfg/final_model_config.json": 1586,
61
  "experiment_cfg/final_processor_config.json": 2548329,
 
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "B6 pedal-bin trash disposal (G1)",
6
  "task_dir": "b6-pedal-bin",
7
+ "training_data": "b6 v5t7 ready-start (first 100 episodes)",
8
  "robot": "Unitree G1",
9
  "embodiment_tag": "new_embodiment",
10
  "epoch_completed": 10,
11
+ "global_step": 936,
12
  "total_epochs": 20,
13
+ "total_steps": 1872,
14
+ "save_steps": 936,
15
+ "per_gpu_batch": 87,
16
+ "num_gpus": 8,
17
+ "num_nodes": 1,
18
+ "global_batch": 696,
19
  "lr": 0.0001,
20
  "lr_schedule": "cosine, warmup_ratio 0.05",
21
+ "weight_decay": 1e-05,
22
  "deepspeed_stage": 2,
23
  "episodes": 100,
24
  "frames": 65127,
25
+ "fps": 30.0,
26
+ "control_hz": 50.0,
27
  "cameras": [
28
  "observation.images.ego_view",
29
  "observation.images.chest"
 
47
  "action_dim": 76,
48
  "action_chunk": 16,
49
  "model_action_horizon": 40,
50
+ "state_encoding": "raw (apply_sincos_state_encoding false, GR00T N1.7 default)",
51
  "normalization": "statistics.json (percentiles, use_percentiles true)",
52
+ "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09), on from epoch 1; model-native image augmentation off (via code/hb_groot/hb_launch.py; config.json still lists GR00T's native augmentation defaults, they are bypassed in training)",
53
+ "config": "experiment_cfg/conf.yaml (full model config); dataset_meta/modality.json (state/action/video keys)",
54
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
55
  "resume_state_included": false,
56
  "files": {
 
58
  "dataset_meta/hb-dataset-receipt.json": 816,
59
  "dataset_meta/modality.json": 1537,
60
  "embodiment_id.json": 2226,
61
+ "experiment_cfg/conf.yaml": 5792,
62
+ "experiment_cfg/config.yaml": 5760,
63
  "experiment_cfg/dataset_statistics.json": 20262,
64
  "experiment_cfg/final_model_config.json": 1586,
65
  "experiment_cfg/final_processor_config.json": 2548329,
b6-pedal-bin/epoch-020/embodiment_id.json CHANGED
@@ -51,7 +51,7 @@
51
  "simpler_env_widowx": 1,
52
  "libero_sim": 2,
53
  "droid_sim": 3,
54
- "robocasa_gr1_tabletop": 10,
55
  "new_embodiment": 10,
 
56
  "unitree_g1_sonic": 11
57
  }
 
51
  "simpler_env_widowx": 1,
52
  "libero_sim": 2,
53
  "droid_sim": 3,
 
54
  "new_embodiment": 10,
55
+ "robocasa_gr1_tabletop": 10,
56
  "unitree_g1_sonic": 11
57
  }
b6-pedal-bin/epoch-020/protocol.json CHANGED
@@ -4,22 +4,26 @@
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "B6 pedal-bin trash disposal (G1)",
6
  "task_dir": "b6-pedal-bin",
 
7
  "robot": "Unitree G1",
8
  "embodiment_tag": "new_embodiment",
9
  "epoch_completed": 20,
10
- "global_step": 128,
11
  "total_epochs": 20,
12
- "total_steps": 128,
13
- "save_steps": 64,
14
- "per_gpu_batch": 320,
15
- "global_batch": 10240,
16
- "num_gpus": 32,
 
17
  "lr": 0.0001,
18
  "lr_schedule": "cosine, warmup_ratio 0.05",
 
19
  "deepspeed_stage": 2,
20
  "episodes": 100,
21
  "frames": 65127,
22
- "fps": 30,
 
23
  "cameras": [
24
  "observation.images.ego_view",
25
  "observation.images.chest"
@@ -43,10 +47,10 @@
43
  "action_dim": 76,
44
  "action_chunk": 16,
45
  "model_action_horizon": 40,
46
- "state_encoding": "raw (apply_sincos_state_encoding false)",
47
  "normalization": "statistics.json (percentiles, use_percentiles true)",
48
- "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09) via code/hb_groot/hb_launch.py; GR00T native random crop / colour jitter / rotation bypassed in training (config.json still lists the native defaults, they are not used)",
49
- "config": "experiment_cfg/conf.yaml (full training config); dataset_meta/modality.json (state/action/video keys)",
50
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
51
  "resume_state_included": false,
52
  "files": {
@@ -54,8 +58,8 @@
54
  "dataset_meta/hb-dataset-receipt.json": 816,
55
  "dataset_meta/modality.json": 1537,
56
  "embodiment_id.json": 2226,
57
- "experiment_cfg/conf.yaml": 5791,
58
- "experiment_cfg/config.yaml": 5761,
59
  "experiment_cfg/dataset_statistics.json": 20262,
60
  "experiment_cfg/final_model_config.json": 1586,
61
  "experiment_cfg/final_processor_config.json": 2548329,
 
4
  "vlm_backbone": "nvidia/Cosmos-Reason2-2B (gated on HF; needed at inference)",
5
  "task": "B6 pedal-bin trash disposal (G1)",
6
  "task_dir": "b6-pedal-bin",
7
+ "training_data": "b6 v5t7 ready-start (first 100 episodes)",
8
  "robot": "Unitree G1",
9
  "embodiment_tag": "new_embodiment",
10
  "epoch_completed": 20,
11
+ "global_step": 1872,
12
  "total_epochs": 20,
13
+ "total_steps": 1872,
14
+ "save_steps": 936,
15
+ "per_gpu_batch": 87,
16
+ "num_gpus": 8,
17
+ "num_nodes": 1,
18
+ "global_batch": 696,
19
  "lr": 0.0001,
20
  "lr_schedule": "cosine, warmup_ratio 0.05",
21
+ "weight_decay": 1e-05,
22
  "deepspeed_stage": 2,
23
  "episodes": 100,
24
  "frames": 65127,
25
+ "fps": 30.0,
26
+ "control_hz": 50.0,
27
  "cameras": [
28
  "observation.images.ego_view",
29
  "observation.images.chest"
 
47
  "action_dim": 76,
48
  "action_chunk": 16,
49
  "model_action_horizon": 40,
50
+ "state_encoding": "raw (apply_sincos_state_encoding false, GR00T N1.7 default)",
51
  "normalization": "statistics.json (percentiles, use_percentiles true)",
52
+ "augmentation": "Humanoid Benchmark online augmentation spec v2 (2026-11-09), on from epoch 1; model-native image augmentation off (via code/hb_groot/hb_launch.py; config.json still lists GR00T's native augmentation defaults, they are bypassed in training)",
53
+ "config": "experiment_cfg/conf.yaml (full model config); dataset_meta/modality.json (state/action/video keys)",
54
  "inference": "NVIDIA Isaac-GR00T (N1.7) policy loader on this folder with embodiment_tag new_embodiment and the modality config in experiment_cfg/conf.yaml; import code/hb_groot/hb_offline_fix.py when HF_HUB_OFFLINE=1. Augmentation is training-only.",
55
  "resume_state_included": false,
56
  "files": {
 
58
  "dataset_meta/hb-dataset-receipt.json": 816,
59
  "dataset_meta/modality.json": 1537,
60
  "embodiment_id.json": 2226,
61
+ "experiment_cfg/conf.yaml": 5792,
62
+ "experiment_cfg/config.yaml": 5760,
63
  "experiment_cfg/dataset_statistics.json": 20262,
64
  "experiment_cfg/final_model_config.json": 1586,
65
  "experiment_cfg/final_processor_config.json": 2548329,