config.json

{
  "architectures": ["LlamaForCausalLM"],
  "model_type": "llama",
  "vocab_size": 32000,
  "hidden_size": 4096,
  "num_attention_heads": 32,
  "num_hidden_layers": 24,
  "max_position_embeddings": 2048,
  "layer_norm_eps": 1e-6,
  "initializer_range": 0.02,
  "do_sample": true,
  "top_k": 50,
  "top_p": 1.0
}