{
    "type": "trace_act",
    "n_obs_steps": 1,
    "input_features": {
        "observation.state": {
            "type": "STATE",
            "shape": [
                14
            ]
        },
        "observation.ee_pixel_left": {
            "type": "STATE",
            "shape": [
                2
            ]
        },
        "observation.ee_pixel_right": {
            "type": "STATE",
            "shape": [
                2
            ]
        },
        "observation.images.cam_high": {
            "type": "VISUAL",
            "shape": [
                3,
                256,
                256
            ]
        },
        "observation.images.cam_left_wrist": {
            "type": "VISUAL",
            "shape": [
                3,
                256,
                256
            ]
        },
        "observation.images.cam_right_wrist": {
            "type": "VISUAL",
            "shape": [
                3,
                256,
                256
            ]
        }
    },
    "output_features": {
        "action": {
            "type": "ACTION",
            "shape": [
                14
            ]
        }
    },
    "device": "cpu",
    "use_amp": false,
    "use_peft": false,
    "push_to_hub": true,
    "repo_id": "quincyu/trace_robotwin",
    "private": null,
    "tags": null,
    "license": null,
    "pretrained_path": null,
    "vjepa_model_id": "facebook/vjepa2-vitg-fpc64-256",
    "trace_horizon": 20,
    "trace_predictor_ckpt": null,
    "video_key": "observation.video",
    "language_key": "observation.language",
    "chunk_size": 20,
    "n_action_steps": 20,
    "drop_n_last_frames": 10,
    "ee_pixel_image_resolution": 256,
    "normalization_mapping": {
        "VISUAL": "MEAN_STD",
        "STATE": "MEAN_STD",
        "ACTION": "MIN_MAX"
    },
    "vision_backbone": "resnet18",
    "pretrained_backbone_weights": "ResNet18_Weights.IMAGENET1K_V1",
    "replace_final_stride_with_dilation": false,
    "pre_norm": false,
    "dim_model": 512,
    "n_heads": 8,
    "dim_feedforward": 3200,
    "feedforward_activation": "relu",
    "n_encoder_layers": 4,
    "n_decoder_layers": 1,
    "use_vae": true,
    "latent_dim": 32,
    "n_vae_encoder_layers": 4,
    "temporal_ensemble_coeff": null,
    "dropout": 0.1,
    "kl_weight": 10.0,
    "optimizer_lr": 1e-05,
    "optimizer_weight_decay": 0.0001,
    "optimizer_lr_backbone": 1e-05,
    "compile_model": true,
    "compile_backend": "cudagraphs"
}