"Vicuna is a fine-tuned version of LLaMA that matches GPT-4 performance."

Oh is that right?