{
    "producer": {
        "name": "modelopt",
        "version": "0.19.0"
    },
    "architecture": "LlamaForCausalLM",
    "dtype": "bfloat16",
    "logits_dtype": "float32",
    "num_hidden_layers": 88,
    "num_attention_heads": 96,
    "num_key_value_heads": 8,
    "hidden_size": 12288,
    "norm_epsilon": 1e-05,
    "vocab_size": 32768,
    "max_position_embeddings": 131072,
    "hidden_act": "silu",
    "use_parallel_embedding": true,
    "embedding_sharding_dim": 0,
    "quantization": {
        "quant_algo": "MIXED_PRECISION",
        "kv_cache_quant_algo": "FP8"
    },
    "mapping": {
        "world_size": 4,
        "tp_size": 4,
        "pp_size": 1
    },
    "head_size": 128,
    "intermediate_size": 28672,
    "position_embedding_type": "rope_gpt_neox",
    "share_embedding_table": false,
    "residual_mlp": false,
    "bias": false,
    "rotary_pct": 1.0,
    "rank": 3,
    "decoder": "llama",
    "rmsnorm": true,
    "lm_head_bias": false,
    "rotary_base": 1000000.0,
    "model_type": "llama"
}