| load_config_path: null |
| model: |
| model_type: Gr00tN1d7 |
| model_dtype: bfloat16 |
| model_name: nvidia/Cosmos-Reason2-2B |
| backbone_model_type: qwen |
| model_revision: null |
| tune_top_llm_layers: 0 |
| backbone_embedding_dim: 2048 |
| tune_llm: false |
| tune_visual: false |
| select_layer: 12 |
| reproject_vision: false |
| use_flash_attention: true |
| load_bf16: false |
| backbone_trainable_params_fp32: true |
| vision_weights_path: /workspace/checkpoints/vision/paired-grounding-20260911/vision.safetensors |
| attn_supervision_layers: null |
| attn_supervision_weight: 0.5 |
| attn_supervision_mode: vision |
| attn_supervision_loss: kl |
| attn_supervision_margin: 0.9 |
| image_crop_size: |
| - 230 |
| - 230 |
| image_target_size: |
| - 256 |
| - 256 |
| shortest_image_edge: null |
| crop_fraction: null |
| random_rotation_angle: null |
| color_jitter_params: null |
| use_albumentations_transforms: true |
| extra_augmentation_config: null |
| formalize_language: true |
| apply_sincos_state_encoding: false |
| use_percentiles: true |
| use_relative_action: true |
| max_state_dim: 132 |
| max_action_dim: 132 |
| action_horizon: 40 |
| hidden_size: 1024 |
| input_embedding_dim: 1536 |
| state_history_length: 1 |
| add_pos_embed: true |
| attn_dropout: 0.2 |
| use_vlln: true |
| max_seq_len: 1024 |
| use_alternate_vl_dit: true |
| attend_text_every_n_blocks: 2 |
| diffusion_model_cfg: |
| positional_embeddings: null |
| num_layers: 16 |
| num_attention_heads: 32 |
| attention_head_dim: 48 |
| norm_type: ada_norm |
| dropout: 0.2 |
| final_dropout: true |
| output_dim: 1024 |
| interleave_self_attention: true |
| num_inference_timesteps: 4 |
| noise_beta_alpha: 1.5 |
| noise_beta_beta: 1.0 |
| noise_s: 0.999 |
| num_timestep_buckets: 1000 |
| tune_projector: true |
| tune_diffusion_model: true |
| tune_vlln: true |
| state_dropout_prob: 0.2 |
| exclude_state: false |
| use_mean_std: false |
| max_num_embodiments: 32 |
| data: |
| datasets: |
| - dataset_paths: |
| - /workspace/data/Task_000609_000612_KPI_PickPlace_CoffeeCan_PlasticBottle_KJM_WJW_lerobot |
| embodiment_tag: new_embodiment |
| mix_ratio: 1.0 |
| dataset_type: physical_embodiment |
| val_dataset_path: null |
| modality_configs: |
| new_embodiment: |
| video: |
| delta_indices: |
| - 0 |
| modality_keys: |
| - cam_left_head |
| sin_cos_embedding_keys: null |
| mean_std_embedding_keys: null |
| action_configs: null |
| state: |
| delta_indices: |
| - 0 |
| modality_keys: |
| - arm_left |
| - arm_right |
| sin_cos_embedding_keys: null |
| mean_std_embedding_keys: null |
| action_configs: null |
| language: |
| delta_indices: |
| - 0 |
| modality_keys: |
| - sub_task |
| sin_cos_embedding_keys: null |
| mean_std_embedding_keys: null |
| action_configs: null |
| action: |
| delta_indices: |
| - 0 |
| - 1 |
| - 2 |
| - 3 |
| - 4 |
| - 5 |
| - 6 |
| - 7 |
| - 8 |
| - 9 |
| - 10 |
| - 11 |
| - 12 |
| - 13 |
| - 14 |
| - 15 |
| modality_keys: |
| - arm_left |
| - arm_right |
| sin_cos_embedding_keys: null |
| mean_std_embedding_keys: null |
| action_configs: |
| - rep: ABSOLUTE |
| type: NON_EEF |
| format: DEFAULT |
| state_key: null |
| - rep: ABSOLUTE |
| type: NON_EEF |
| format: DEFAULT |
| state_key: null |
| download_cache: false |
| shard_size: 1024 |
| episode_sampling_rate: 0.1 |
| num_shards_per_epoch: 100000 |
| override_pretraining_statistics: true |
| mode: single_turn |
| random_chop: 0.0 |
| mock_dataset_mode: false |
| shuffle: true |
| seed: 42 |
| multiprocessing_context: fork |
| allow_padding: false |
| subsample_ratio: 1.0 |
| image_crop_size: |
| - 244 |
| - 244 |
| image_target_size: |
| - 224 |
| - 224 |
| video_backend: torchcodec |
| training: |
| output_dir: /workspace/checkpoints/groot/groot-paired-20260911 |
| experiment_name: null |
| max_steps: 100000 |
| global_batch_size: 24 |
| batch_size: null |
| gradient_accumulation_steps: 1 |
| learning_rate: 0.0001 |
| lr_scheduler_type: cosine |
| weight_decay: 1.0e-05 |
| warmup_ratio: 0.05 |
| warmup_steps: 0 |
| max_grad_norm: 1.0 |
| optim: adamw_torch |
| start_from_checkpoint: nvidia/GR00T-N1.7-3B |
| skip_weight_loading: false |
| tf32: true |
| fp16: false |
| bf16: true |
| eval_bf16: true |
| logging_steps: 10 |
| save_steps: 10000 |
| save_total_limit: 2 |
| save_vl_model: false |
| save_only_model: false |
| upload_checkpoints: false |
| upload_every: 1000 |
| upload_last_n_checkpoints: 5 |
| max_concurrent_uploads: 2 |
| eval_strategy: 'no' |
| eval_steps: 500 |
| eval_set_split_ratio: 0.1 |
| eval_batch_size: 2 |
| save_best_eval_metric_name: '' |
| save_best_eval_metric_greater_is_better: true |
| deepspeed_stage: 2 |
| gradient_checkpointing: false |
| transformers_trust_remote_code: true |
| transformers_local_files_only: false |
| transformers_cache_dir: null |
| transformers_access_token: null |
| use_ddp: false |
| ddp_bucket_cap_mb: 100 |
| num_gpus: 1 |
| dataloader_num_workers: 8 |
| remove_unused_columns: false |
| use_wandb: false |
| wandb_project: finetune-gr00t-n1d7 |
| enable_profiling: false |
| max_retries: 3 |
| assert_loss_less_than: null |
| add_rl_callback: false |
| enable_open_loop_eval: false |
| open_loop_eval_traj_ids: |
| - 0 |
| open_loop_eval_steps_per_traj: 100 |
| open_loop_eval_plot_indices: null |
| max_steps: 100000 |
| save_steps: 10000 |
|
|
