Fine-tuning instructions for large language models