Self-supervised Preference Optimization: Enhance Your Language Model with Preference Degree Awareness

TL;DR

This paper introduces a Self-supervised Preference Optimization framework that enhances Large Language Models' ability to understand preference degrees, leading to improved performance over existing methods in preference-based training.

Contribution

The paper proposes a novel self-supervised preference degree loss that, when combined with alignment loss, improves LLMs' understanding of human preference intensities.

Findings

SPO significantly boosts preference optimization performance.

SPO achieves state-of-the-art results on multiple datasets.

The framework is compatible with existing preference methods.

Abstract

Recently, there has been significant interest in replacing the reward model in Reinforcement Learning with Human Feedback (RLHF) methods for Large Language Models (LLMs), such as Direct Preference Optimization (DPO) and its variants. These approaches commonly use a binary cross-entropy mechanism on pairwise samples, i.e., minimizing and maximizing the loss based on preferred or dis-preferred responses, respectively. However, while this training strategy omits the reward model, it also overlooks the varying preference degrees within different responses. We hypothesize that this is a key factor hindering LLMs from sufficiently understanding human preferences. To address this problem, we propose a novel Self-supervised Preference Optimization (SPO) framework, which constructs a self-supervised preference degree loss combined with the alignment loss, thereby helping LLMs improve their ability to understand the degree of preference. Extensive experiments are conducted on two widely used datasets of different tasks. The results demonstrate that SPO can be seamlessly integrated with existing preference optimization methods and significantly boost their performance to achieve state-of-the-art performance. We also conduct detailed analyses to offer comprehensive insights into SPO, which verifies its effectiveness. The code is available at https://github.com/lijian16/SPO.