config.json

{
    "vocab_size": 4096,
    "max_position_embeddings": 256,
    "hidden_size": 256,
    "num_attention_heads": 8,
    "num_hidden_layers": 8,
    "intermediate_size": 682,
    "rms_norm_eps": 1e-05,
    "model_type": "llama",
    "architectures": [
        "LlamaForCausalLM"
    ],
    "device": "cuda"
}