Robotics
Transformers
Safetensors
LeRobot
English
Gr00tN1d6
vision-language-action
manipulation
gr00t
nvidia
physical-ai
humanoid
reachy2
Instructions to use ganatrask/NOVA with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use ganatrask/NOVA with Transformers:
# Load model directly from transformers import Gr00tN1d6 model = Gr00tN1d6.from_pretrained("ganatrask/NOVA", dtype="auto") - LeRobot
How to use ganatrask/NOVA with LeRobot:
- Notebooks
- Google Colab
- Kaggle
| !!python/object:gr00t.configs.base_config.Config | |
| data: !!python/object:gr00t.configs.data.data_config.DataConfig | |
| allow_padding: false | |
| datasets: | |
| - !!python/object:gr00t.configs.data.data_config.SingleDatasetConfig | |
| dataset_paths: | |
| - ./datasets/reachy2_100 | |
| dataset_type: physical_embodiment | |
| embodiment_tag: reachy2 | |
| mix_ratio: 1.0 | |
| val_dataset_path: null | |
| download_cache: false | |
| episode_sampling_rate: 0.1 | |
| image_crop_size: | |
| - 244 | |
| - 244 | |
| image_target_size: | |
| - 224 | |
| - 224 | |
| mock_dataset_mode: false | |
| modality_configs: | |
| reachy2: | |
| action: !!python/object:gr00t.data.types.ModalityConfig | |
| action_configs: | |
| - !!python/object:gr00t.data.types.ActionConfig | |
| format: &id001 !!python/object/apply:gr00t.data.types.ActionFormat | |
| - default | |
| rep: !!python/object/apply:gr00t.data.types.ActionRepresentation | |
| - relative | |
| state_key: arm_joints | |
| type: &id002 !!python/object/apply:gr00t.data.types.ActionType | |
| - non_eef | |
| - !!python/object:gr00t.data.types.ActionConfig | |
| format: | |
| rep: !!python/object/apply:gr00t.data.types.ActionRepresentation | |
| - absolute | |
| state_key: null | |
| type: | |
| delta_indices: | |
| - 0 | |
| - 1 | |
| - 2 | |
| - 3 | |
| - 4 | |
| - 5 | |
| - 6 | |
| - 7 | |
| - 8 | |
| - 9 | |
| - 10 | |
| - 11 | |
| - 12 | |
| - 13 | |
| - 14 | |
| - 15 | |
| mean_std_embedding_keys: null | |
| modality_keys: | |
| - arm_joints | |
| - gripper | |
| sin_cos_embedding_keys: null | |
| language: !!python/object:gr00t.data.types.ModalityConfig | |
| action_configs: null | |
| delta_indices: | |
| - 0 | |
| mean_std_embedding_keys: null | |
| modality_keys: | |
| - annotation.human.task_description | |
| sin_cos_embedding_keys: null | |
| state: !!python/object:gr00t.data.types.ModalityConfig | |
| action_configs: null | |
| delta_indices: | |
| - 0 | |
| mean_std_embedding_keys: null | |
| modality_keys: | |
| - arm_joints | |
| sin_cos_embedding_keys: null | |
| video: !!python/object:gr00t.data.types.ModalityConfig | |
| action_configs: null | |
| delta_indices: | |
| - 0 | |
| mean_std_embedding_keys: null | |
| modality_keys: | |
| - front_cam | |
| sin_cos_embedding_keys: null | |
| mode: single_turn | |
| multiprocessing_context: fork | |
| num_shards_per_epoch: 10000 | |
| override_pretraining_statistics: false | |
| random_chop: 0.0 | |
| seed: 42 | |
| shard_size: 1024 | |
| shuffle: true | |
| subsample_ratio: 1.0 | |
| video_backend: decord | |
| load_config_path: null | |
| model: !!python/object:gr00t.configs.model.gr00t_n1d6.Gr00tN1d6Config | |
| _attn_implementation_autoset: false | |
| _attn_implementation_internal: null | |
| _commit_hash: null | |
| _name_or_path: '' | |
| add_cross_attention: false | |
| architectures: null | |
| backbone_model_type: eagle | |
| backbone_trainable_params_fp32: true | |
| bad_words_ids: null | |
| begin_suppress_tokens: null | |
| bos_token_id: null | |
| chunk_size_feed_forward: 0 | |
| color_jitter_params: null | |
| cross_attention_hidden_size: null | |
| decoder_start_token_id: null | |
| diffusion_model_cfg: | |
| attention_head_dim: 48 | |
| dropout: 0.2 | |
| final_dropout: true | |
| interleave_self_attention: true | |
| norm_type: ada_norm | |
| num_attention_heads: 32 | |
| num_layers: 32 | |
| output_dim: 1024 | |
| positional_embeddings: null | |
| diversity_penalty: 0.0 | |
| do_sample: false | |
| eagle_collator: true | |
| early_stopping: false | |
| encoder_no_repeat_ngram_size: 0 | |
| eos_token_id: null | |
| exponential_decay_length_penalty: null | |
| finetuning_task: null | |
| forced_bos_token_id: null | |
| forced_eos_token_id: null | |
| id2label: | |
| 0: LABEL_0 | |
| 1: LABEL_1 | |
| is_decoder: false | |
| is_encoder_decoder: false | |
| label2id: | |
| LABEL_0: 0 | |
| LABEL_1: 1 | |
| length_penalty: 1.0 | |
| load_bf16: false | |
| max_length: 20 | |
| min_length: 0 | |
| model_name: nvidia/Eagle-Block2A-2B-v2 | |
| no_repeat_ngram_size: 0 | |
| num_beam_groups: 1 | |
| num_beams: 1 | |
| num_return_sequences: 1 | |
| output_attentions: false | |
| output_hidden_states: false | |
| output_scores: false | |
| pad_token_id: null | |
| prefix: null | |
| problem_type: null | |
| pruned_heads: {} | |
| random_rotation_angle: null | |
| remove_invalid_values: false | |
| repetition_penalty: 1.0 | |
| reproject_vision: false | |
| return_dict: true | |
| return_dict_in_generate: false | |
| sep_token_id: null | |
| state_dropout_prob: 0.0 | |
| suppress_tokens: null | |
| task_specific_params: null | |
| temperature: 1.0 | |
| tf_legacy_loss: false | |
| tie_encoder_decoder: false | |
| tie_word_embeddings: true | |
| tokenizer_class: null | |
| top_k: 50 | |
| top_p: 1.0 | |
| torch_dtype: null | |
| torchscript: false | |
| transformers_version: null | |
| tune_diffusion_model: true | |
| tune_llm: false | |
| tune_projector: true | |
| tune_visual: false | |
| typical_p: 1.0 | |
| use_bfloat16: false | |
| use_relative_action: true | |
| training: !!python/object:gr00t.configs.training.training_config.TrainingConfig | |
| add_rl_callback: false | |
| assert_loss_less_than: null | |
| batch_size: null | |
| bf16: true | |
| dataloader_num_workers: 8 | |
| ddp_bucket_cap_mb: 100 | |
| deepspeed_stage: 2 | |
| enable_open_loop_eval: false | |
| enable_profiling: false | |
| eval_batch_size: 2 | |
| eval_bf16: true | |
| eval_set_split_ratio: 0.1 | |
| eval_steps: 500 | |
| eval_strategy: 'no' | |
| experiment_name: null | |
| fp16: false | |
| global_batch_size: 64 | |
| gradient_accumulation_steps: 1 | |
| gradient_checkpointing: false | |
| learning_rate: 0.0001 | |
| logging_steps: 10 | |
| lr_scheduler_type: cosine | |
| max_concurrent_uploads: 2 | |
| max_grad_norm: 1.0 | |
| max_retries: 3 | |
| max_steps: 30000 | |
| num_gpus: 1 | |
| open_loop_eval_plot_indices: null | |
| open_loop_eval_steps_per_traj: 100 | |
| open_loop_eval_traj_ids: | |
| - 0 | |
| optim: adamw_torch | |
| output_dir: /tmp/groot_output/reachy2 | |
| remove_unused_columns: false | |
| save_best_eval_metric_greater_is_better: true | |
| save_best_eval_metric_name: '' | |
| save_steps: 3000 | |
| save_total_limit: 5 | |
| save_vl_model: false | |
| start_from_checkpoint: nvidia/GR00T-N1.6-3B | |
| tf32: true | |
| transformers_access_token: null | |
| transformers_cache_dir: null | |
| transformers_local_files_only: false | |
| transformers_trust_remote_code: true | |
| upload_checkpoints: false | |
| upload_every: 1000 | |
| upload_last_n_checkpoints: 5 | |
| use_ddp: false | |
| use_wandb: false | |
| wandb_project: finetune-gr00t-n1d6 | |
| warmup_ratio: 0.05 | |
| warmup_steps: 0 | |
| weight_decay: 1.0e-05 | |