training.json

{
    "batch_size": 256,
    "model": {
        "dropout": 0,
        "batch_norm": true,
        "gnn_n_layers": 8,
        "d": 512,
        "n_bars": 2,
        "resolution": 8
    },
    "optimizer": {
        "betas": [
            0.9,
            0.98
        ],
        "eps": 1e-09,
        "lr": 5e-06
    },
    "lr_scheduler": {
        "peak_lr": 0.0001,
        "final_lr_scale": 0.01,
        "warmup_steps": 8000,
        "decay_steps": 800000
    },
    "beta_scheduler": {
        "anneal_start": 40000,
        "beta_max": 0.01,
        "step_size": 0.001,
        "anneal_end": 500000
    }
}