Fine tuning instruct modes

Hi, can we get the ability to fine tune the latest instruct models? If that’s not on the roadmap, what’s the best way to get to similar results? Like hand crafting X number of prompts to get to a Likert score of X?