Instructions to use maximellerbach/omx_multicubes_vla_jepa with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use maximellerbach/omx_multicubes_vla_jepa with LeRobot:
- Notebooks
- Google Colab
- Kaggle
| { | |
| "type": "vla_jepa", | |
| "n_obs_steps": 1, | |
| "input_features": { | |
| "observation.images.exterior_1_left": { | |
| "type": "VISUAL", | |
| "shape": [ | |
| 3, | |
| 224, | |
| 224 | |
| ] | |
| }, | |
| "observation.images.exterior_2_left": { | |
| "type": "VISUAL", | |
| "shape": [ | |
| 3, | |
| 224, | |
| 224 | |
| ] | |
| } | |
| }, | |
| "output_features": { | |
| "action": { | |
| "type": "ACTION", | |
| "shape": [ | |
| 6 | |
| ] | |
| } | |
| }, | |
| "device": "cuda", | |
| "use_amp": false, | |
| "use_peft": false, | |
| "push_to_hub": true, | |
| "repo_id": "maximellerbach/omx_multicubes_vla_jepa", | |
| "private": null, | |
| "tags": null, | |
| "license": null, | |
| "pretrained_path": "maximellerbach/VLA-JEPA-Pretrain", | |
| "chunk_size": 20, | |
| "n_action_steps": 20, | |
| "normalization_mapping": { | |
| "VISUAL": "IDENTITY", | |
| "STATE": "MEAN_STD", | |
| "ACTION": "MEAN_STD" | |
| }, | |
| "qwen_model_name": "Qwen/Qwen3-VL-2B-Instruct", | |
| "jepa_encoder_name": "facebook/vjepa2-vitl-fpc64-256", | |
| "freeze_qwen": false, | |
| "enable_world_model": true, | |
| "reinit_action_head": true, | |
| "tokenizer_padding_side": "left", | |
| "prompt_template": "{instruction}\n\nPredict {actions} and condition future prediction with {e_actions}.", | |
| "special_action_token": "<|action_{}|>", | |
| "embodied_action_token": "<|embodied_action|>", | |
| "action_dim": 6, | |
| "state_dim": 6, | |
| "num_action_tokens_per_timestep": 8, | |
| "num_embodied_action_tokens_per_instruction": 32, | |
| "num_inference_timesteps": 4, | |
| "action_hidden_size": 1024, | |
| "action_model_type": "DiT-B", | |
| "action_num_layers": 16, | |
| "action_dropout": 0.2, | |
| "action_num_timestep_buckets": 1000, | |
| "action_noise_beta_alpha": 1.5, | |
| "action_noise_beta_beta": 1.0, | |
| "action_noise_s": 0.999, | |
| "num_target_vision_tokens": 32, | |
| "action_max_seq_len": 1024, | |
| "num_video_frames": 8, | |
| "predictor_depth": 12, | |
| "predictor_num_heads": 8, | |
| "predictor_mlp_ratio": 4.0, | |
| "predictor_dropout": 0.0, | |
| "world_model_loss_weight": 0.1, | |
| "jepa_tubelet_size": 2, | |
| "repeated_diffusion_steps": 8, | |
| "resize_images_to": null, | |
| "torch_dtype": "bfloat16", | |
| "optimizer_lr": 0.0001, | |
| "optimizer_betas": [ | |
| 0.9, | |
| 0.95 | |
| ], | |
| "optimizer_eps": 1e-08, | |
| "optimizer_weight_decay": 1e-10, | |
| "optimizer_grad_clip_norm": 10.0, | |
| "scheduler_warmup_steps": 1000, | |
| "scheduler_decay_steps": 5000, | |
| "scheduler_decay_lr": 2.5e-06 | |
| } |