train_config.json

{
    "learning_rate": 0.0006,
    "max_iters": 57980,
    "weight_decay": 0.1,
    "beta1": 0.9,
    "beta2": 0.95,
    "grad_clip": 1.0,
    "decay_lr": true,
    "warmup_iters": 3000,
    "lr_decay_iters": 57980,
    "min_lr": 6e-05,
    "device_type": "cpu",
    "n_sample": 1932697,
    "batch_size": 5,
    "n_epoch": 3,
    "n_iters_for_estimate_loss": 1500,
    "gradient_accumulation_steps": 20
}