| { | |
| "normalization_mapping": { | |
| "VISUAL": "IDENTITY", | |
| "STATE": "MIN_MAX", | |
| "ACTION": "MIN_MAX" | |
| }, | |
| "input_features": { | |
| "observation.images.cam_high": { | |
| "type": "VISUAL", | |
| "shape": [ | |
| 3, | |
| 240, | |
| 320 | |
| ] | |
| }, | |
| "observation.images.cam_left_wrist": { | |
| "type": "VISUAL", | |
| "shape": [ | |
| 3, | |
| 240, | |
| 320 | |
| ] | |
| }, | |
| "observation.images.cam_right_wrist": { | |
| "type": "VISUAL", | |
| "shape": [ | |
| 3, | |
| 240, | |
| 320 | |
| ] | |
| }, | |
| "observation.state": { | |
| "type": "STATE", | |
| "shape": [ | |
| 14 | |
| ] | |
| } | |
| }, | |
| "output_features": { | |
| "action": { | |
| "type": "ACTION", | |
| "shape": [ | |
| 14 | |
| ] | |
| } | |
| }, | |
| "device": "cuda", | |
| "backbone": "Qwen/Qwen3-VL-4B-Instruct", | |
| "chunk_size": 60, | |
| "n_action_steps": 60, | |
| "max_state_dim": 32, | |
| "max_action_dim": 32, | |
| "proj_width": 2560, | |
| "num_steps": 10, | |
| "attention_implementation": "eager", | |
| "dit_hidden_size": 1536, | |
| "dit_num_heads": 32, | |
| "dit_num_layers": 16, | |
| "dit_interleave_self_attention": true, | |
| "dit_cross_attention_dim": 2560, | |
| "num_noise_per_sample": 1, | |
| "type": "spirit_vla" | |
| } | |