llama-3.2-1B/spinquant/config.json

{
    "dim": 2048,
    "n_layers": 16,
    "n_heads": 32,
    "n_kv_heads": 8,
    "vocab_size": 128256,
    "ffn_dim_multiplier": 1.5,
    "multiple_of": 256,
    "norm_eps": 1e-05,
    "rope_theta": 500000.0,
    "use_scaled_rope": true,
    "quantization_args": {
        "group_size": 32
    }
}