##### ###### ##### ### # # ### # # ###### ## ## ## ## ## ## ## # # # # # ## ##### #### ##### # # # # # # # #### ## # ## ## ## ## # # # # # ## ## # ###### ## ### # ### # ######
##### ###### ##### ### # # ### # # ###### ## ## ## ## ## ## ## # # # # # ## ##### #### ##### # # # # # # # #### ## # ## ## ## ## # # # # # ## ## # ###### ## ### # ### # ######
Supervised fine-tuning (SFT) is the most common way to adapt LLMs. You show the model examples of desired behavior, and it learns to replicate those patterns.
I'll walk you through the mechanics of SFT: how it works, what hyperparameters matter, and how to avoid common problems.
