config.json

{
    "model_type": "llama",
    "block_size": 512,
    "vocab_size": 4096,
    "n_layer": 6,
    "n_head": 4,
    "n_embd": 256,
    "n_intermediate": 682,
    "rotary_dim": 42,
    "n_ctx": 512,
    "n_key_value_heads": 2,
    "flash_attention": true
  }