{
    "type": "smolvla",
    "n_obs_steps": 1,
    "input_features": {
        "observation.images.screen": {
            "type": "VISUAL",
            "shape": [
                3,
                1080,
                1920
            ]
        },
        "observation.state": {
            "type": "STATE",
            "shape": [
                3
            ]
        }
    },
    "output_features": {
        "action": {
            "type": "ACTION",
            "shape": [
                3
            ]
        }
    },
    "device": "cuda:0",
    "use_amp": false,
    "push_to_hub": false,
    "repo_id": null,
    "private": null,
    "tags": null,
    "license": null,
    "chunk_size": 21,
    "n_action_steps": 1,
    "normalization_mapping": {
        "VISUAL": "IDENTITY",
        "STATE": "IDENTITY",
        "ACTION": "IDENTITY"
    },
    "max_state_dim": 32,
    "max_action_dim": 3,
    "resize_imgs_with_padding": [
        1024,
        576
    ],
    "empty_cameras": 0,
    "adapt_to_pi_aloha": false,
    "use_delta_joint_actions_aloha": false,
    "tokenizer_max_length": 48,
    "num_steps": 10,
    "noise_std": 0.3,
    "noise_std_dex": null,
    "time_min": 0.001,
    "state_noise_std": 0.02,
    "time_beta_dex": [
        1.5,
        1.0
    ],
    "time_beta_grounding": [
        0.5,
        1.5
    ],
    "use_cache": true,
    "freeze_vision_encoder": false,
    "train_expert_only": false,
    "train_state_proj": true,
    "optimizer_lr": 5e-05,
    "optimizer_betas": [
        0.9,
        0.95
    ],
    "optimizer_eps": 1e-08,
    "optimizer_weight_decay": 1e-10,
    "optimizer_grad_clip_norm": 10,
    "scheduler_warmup_steps": 0,
    "scheduler_decay_steps": 15000,
    "scheduler_decay_lr": 2.5e-06,
    "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
    "load_vlm_weights": true,
    "add_image_special_tokens": false,
    "attention_mode": "cross_attn",
    "vision_scale_factor": 4,
    "prefix_length": -1,
    "pad_language_to": "longest",
    "num_expert_layers": -1,
    "num_vlm_layers": 16,
    "self_attn_every_n_layers": 2,
    "expert_width_multiplier": 0.75,
    "min_period": 0.004,
    "max_period": 4.0,
    "traj_reg_indices": [
        1,
        2
    ],
    "lambda_dir": 0.0,
    "lambda_traj_acc": 0.0,
    "lambda_action_recon": 0.0,
    "time_weights": null,
    "dex_grounding_ratio": 4,
    "use_lora": false,
    "lora_rank": 16,
    "lora_alpha": 32,
    "lora_dropout": 0.05
}