Yep! The linked notebook includes an example of exactly that (fine-tuning a 7b model to match the syntax of GPT-4 function call responses): https://github.com/OpenPipe/OpenPipe/blob/main/examples/clas...