Hi Nikhil, my article has a section on fine-tuning pointing to a notebook. To tune for private code, you need to prepare your data in the format similar to https://huggingface.co/datasets/b-mc2/sql-create-context, e.g. question, context, answer (e.g. question: "create hello world in Python", context: <blank in this case>, answer: <hello world in Python>), and you can refer to generate_and_tokenize_prompt function to see how the training instruction is prepared