{
    "d_model": 128,
    "n_heads": 4,
    "d_head": 32,
    "n_layers": 8,
    "vocab_size": 10000,
    "mode": "standard"
}