{
    "type": "vla",
    "n_obs_steps": 1,
    "input_features": {
        "observation.state": {
            "type": "STATE",
            "shape": [
                7
            ]
        },
        "observation.images.wrist": {
            "type": "VISUAL",
            "shape": [
                3,
                256,
                256
            ]
        },
        "observation.images.front": {
            "type": "VISUAL",
            "shape": [
                3,
                256,
                256
            ]
        },
        "observation.vfm_features.wrist": {
            "type": "STATE",
            "shape": [
                16,
                16,
                768
            ]
        },
        "observation.vfm_features.front": {
            "type": "STATE",
            "shape": [
                16,
                16,
                768
            ]
        },
        "observation.skill_feature": {
            "type": "STATE",
            "shape": [
                1024
            ]
        },
        "observation.object_feature": {
            "type": "STATE",
            "shape": [
                1024
            ]
        },
        "observation.vlm_features.wrist": {
            "type": "STATE",
            "shape": [
                16,
                16,
                1024
            ]
        },
        "observation.vlm_features.front": {
            "type": "STATE",
            "shape": [
                16,
                16,
                1024
            ]
        }
    },
    "output_features": {
        "action": {
            "type": "ACTION",
            "shape": [
                7
            ]
        }
    },
    "device": "cuda:0",
    "use_amp": false,
    "push_to_hub": true,
    "repo_id": "Tsagkas/vla_state_modulation_v2_chunk",
    "private": null,
    "tags": null,
    "license": null,
    "pretrained_path": null,
    "num_cameras": 2,
    "chunk_size": 100,
    "n_action_steps": 100,
    "state_dim": 7,
    "action_dim": 7,
    "state_hidden_dim": 256,
    "state_output_dim": 768,
    "num_transformer_heads": 12,
    "transformer_hidden_dim": 768,
    "transformer_expansion_factor": 4,
    "attention_probs_dropout_prob": 0.0,
    "hidden_dropout_prob": 0.1,
    "temporal_ensemble_coeff": 0.01,
    "optimizer_lr": 0.0001,
    "optimizer_weight_decay": 0.0001,
    "normalization_mapping": {
        "VISUAL": "IDENTITY",
        "STATE": "MEAN_STD",
        "ACTION": "MEAN_STD"
    }
}