config.json

{
    "type": "smolvla",
    "n_obs_steps": 1,
    "normalization_mapping": {
        "VISUAL": "IDENTITY",
        "STATE": "MEAN_STD",
        "ACTION": "MEAN_STD"
    },
    "input_features": {
        "observation.images.scene_image": {
            "type": "VISUAL",
            "shape": [
                3,
                480,
                640
            ]
        },
        "observation.images.wrist_image": {
            "type": "VISUAL",
            "shape": [
                3,
                480,
                640
            ]
        },
        "observation.state": {
            "type": "STATE",
            "shape": [
                8
            ]
        }
    },
    "output_features": {
        "action": {
            "type": "ACTION",
            "shape": [
                8
            ]
        }
    },
    "device": "cuda",
    "use_amp": false,
    "scanner_state_pos_weight_mode": "mean",
    "scanner_state_pos_weight": 1.0,
    "chunk_size": 50,
    "n_action_steps": 50,
    "mean_std_normalization_stats": {
        "observation.state": {
            "mean": [
                0.27873939275741577,
                -0.29261285066604614,
                0.4602305293083191,
                0.08445943892002106,
                2.2974722385406494,
                -0.2794015109539032,
                0.0,
                0.0
            ],
            "std": [
                0.10405326634645462,
                0.1369001269340515,
                0.12905506789684296,
                1.0609674453735352,
                1.6164617538452148,
                0.2593933939933777,
                1.0,
                1.0
            ]
        },
        "action": {
            "mean": [
                0.27874284982681274,
                -0.2925970256328583,
                0.4602431356906891,
                0.08444322645664215,
                2.2977328300476074,
                -0.27943673729896545,
                0.0,
                0.0
            ],
            "std": [
                0.10405024886131287,
                0.1369035840034485,
                0.12906143069267273,
                1.060977578163147,
                1.6160578727722168,
                0.25939232110977173,
                1.0,
                1.0
            ]
        }
    },
    "max_state_dim": 32,
    "max_action_dim": 32,
    "resize_imgs_with_padding": [
        512,
        512
    ],
    "empty_cameras": 0,
    "adapt_to_pi_aloha": false,
    "use_delta_joint_actions_aloha": false,
    "tokenizer_max_length": 48,
    "num_steps": 10,
    "use_cache": true,
    "freeze_vision_encoder": false,
    "train_expert_only": true,
    "train_state_proj": true,
    "optimizer_lr": 0.0001,
    "optimizer_betas": [
        0.9,
        0.95
    ],
    "optimizer_eps": 1e-08,
    "optimizer_weight_decay": 1e-10,
    "optimizer_grad_clip_norm": 10,
    "scheduler_warmup_steps": 1000,
    "scheduler_decay_steps": 30000,
    "scheduler_decay_lr": 2.5e-06,
    "vlm_model_name": "HuggingFaceTB/SmolVLM2-500M-Video-Instruct",
    "load_vlm_weights": false,
    "add_image_special_tokens": false,
    "attention_mode": "cross_attn",
    "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
}