config.json

{
    "model_type": "vjepa2_ac",
    "architecture": {
        "encoder": {
            "model_name": "vit_large",
            "embed_dim": 1024,
            "num_patches": 196,
            "patch_size": 16,
            "img_size": 256,
            "weights": "vision_encoder.pt"
        },
        "predictor": {
            "action_dim": 6,
            "depth": 4,
            "heads": 8,
            "embed_dim": 384,
            "dropout": 0.1,
            "weights": "pytorch_model.bin"
        }
    },
    "data_config": {
        "camera": "observation.images.phone",
        "clip_len": 8,
        "frame_step": 2,
        "context_frames": 6,
        "target_frames": 2
    },
    "training_params": {
        "epochs": 300,
        "batch_size": 2,
        "lr": 0.0001,
        "weight_decay": 0.05,
        "warmup_epochs": 20
    },
    "dataset": "SO-100 Ball-Cup Robotics Dataset"
}