| { | |
| "chunk_size": 50, | |
| "n_action_steps": 50, | |
| "max_state_dim": 32, | |
| "max_action_dim": 32, | |
| "action_dim": 8, | |
| "state_dim": 8, | |
| "image_keys": [ | |
| "observation.images.cam0", | |
| "observation.images.cam1" | |
| ], | |
| "resize_imgs_with_padding": [ | |
| 512, | |
| 512 | |
| ], | |
| "empty_cameras": 0, | |
| "tokenizer_max_length": 48, | |
| "pad_language_to": "max_length", | |
| "num_steps": 10, | |
| "min_period": 0.004, | |
| "max_period": 4.0, | |
| "use_cache": true, | |
| "add_image_special_tokens": false, | |
| "prefix_length": 0, | |
| "freeze_vision_encoder": true, | |
| "train_expert_only": true, | |
| "train_state_proj": true, | |
| "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct", | |
| "load_vlm_weights": false, | |
| "attention_mode": "cross_attn", | |
| "num_expert_layers": 0, | |
| "num_vlm_layers": 16, | |
| "self_attn_every_n_layers": 2, | |
| "expert_width_multiplier": 0.75, | |
| "device": "cuda" | |
| } |