model.json

{
    "model_id": "llama3.1_8b_8gen4",
    "model_arch": "llama",
    "version": 20241202,
    "llm_config": {
        "embed_dim": 4096,
        "ffn_dim": 14336,
        "head_size": 128,
        "kv_dim": 1024,
        "n_attn_heads": 32,
        "n_attn_kv_heads": 8,
        "n_ctx": 131072,
        "n_layers": 32,
        "norm_eps": 9.999999747378752e-06,
        "vocab_size": 128256,
        "rope_config": {
            "n_rope_ctx_orig": 131072,
            "rope_attn_factor": 1.0,
            "rope_dim": 128,
            "rope_freq_base": 500000.0,
            "rope_freq_scale": 1.0,
            "rope_scale_type": "linear",
            "rope_type": 0
        }
    },
    "vision": {}
}