preference tuning

Preference tuning is nowadays often performed through Reinforcement Learning From Human Feedback (or RLHF) refines the supervised fine-tuning with preferred stylistic choices.