RAG, fine-tuning, API calling and gptscript for Llama 3 running locally