| { |
| "type": "vla0_smol", |
| "n_obs_steps": 1, |
| "input_features": { |
| "observation.image": { |
| "type": "VISUAL", |
| "shape": [ |
| 3, |
| 96, |
| 96 |
| ] |
| }, |
| "observation.state": { |
| "type": "STATE", |
| "shape": [ |
| 2 |
| ] |
| } |
| }, |
| "output_features": { |
| "action": { |
| "type": "ACTION", |
| "shape": [ |
| 2 |
| ] |
| } |
| }, |
| "device": "cuda", |
| "use_amp": true, |
| "amp_dtype": "bf16", |
| "push_to_hub": true, |
| "repo_id": "olegbalakhnov/pusht_mtp", |
| "private": null, |
| "tags": null, |
| "license": null, |
| "pretrained_path": null, |
| "chunk_size": 10, |
| "n_action_steps": 5, |
| "ensemble_size": 0, |
| "use_remote_client": false, |
| "use_streaming": true, |
| "vllm_url": "http://localhost:8000/", |
| "vllm_api_key": "EMPTY", |
| "normalization_mapping": { |
| "VISUAL": "IDENTITY", |
| "STATE": "MIN_MAX", |
| "ACTION": "MIN_MAX" |
| }, |
| "n_action_bins": 512, |
| "n_state_bins": 512, |
| "max_decoding_steps": 512, |
| "use_cache": true, |
| "optimizer_lr": 5e-05, |
| "optimizer_betas": [ |
| 0.9, |
| 0.95 |
| ], |
| "optimizer_eps": 1e-08, |
| "optimizer_weight_decay": 0.01, |
| "scheduler_warmup_steps": 1000, |
| "scheduler_decay_steps": 30000, |
| "scheduler_decay_lr": 5e-06, |
| "checkpoint_path": null, |
| "vlm_checkpoint": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct", |
| "precision": "float32", |
| "freeze_vision_encoder": false, |
| "do_image_splitting": false, |
| "drop_n_last_frames": 1, |
| "relative_actions": true, |
| "use_state": true, |
| "start_actions_token": 49279, |
| "grad_clip_norm": 1.0, |
| "action_mask_aug_per": 0.15, |
| "action_mask_skip_per": 0.1, |
| "num_train_mtp_heads": 5, |
| "num_inference_mtp_heads": 5, |
| "mtp_layers_ids": [ |
| 1, |
| 17, |
| 30 |
| ], |
| "crop_shape": [ |
| 84, |
| 84 |
| ] |
| } |
