config.json

{
  "model_type": "llama",
  "architectures": [
    "FastLanguageModel"
  ],
  "max_seq_length": 2048,
  "hidden_size": 4096,
  "num_attention_heads": 32,
  "num_hidden_layers": 32,
  "r": 16,
  "lora_alpha": 16,
  "lora_dropout": 0,
  "use_gradient_checkpointing": true
}