optimal-transportation-based loss

This paper introduces the Distributional Preference Reward Model (DPRM) to align large language models with a diverse set of human preferences. The researchers used a beta distribution to characterize preferences,…