Reward-Augmented Data Enhances Direct Preference Alignment of LLMs

TL;DR

This paper introduces reward-conditioned policies and a data relabeling method to improve preference alignment in LLMs, effectively utilizing reward scores to enhance response quality and generalization.

Contribution

It presents a novel reward-augmented dataset construction method and reward-conditioned policies that better leverage preference data for LLM alignment.

Findings

Consistently improves DPO performance across benchmarks.

Mitigates unlearning and overfitting issues in preference alignment.

Enhances generalization to optimal responses in LLMs.

Abstract

Preference alignment in Large Language Models (LLMs) has significantly improved their ability to adhere to human instructions and intentions. However, existing direct alignment algorithms primarily focus on relative preferences and often overlook the qualitative aspects of responses, despite having access to preference data that includes reward scores from judge models during AI feedback. Striving to maximize the implicit reward gap between the chosen and the slightly inferior rejected responses can cause overfitting and unnecessary unlearning of the high-quality rejected responses. The unawareness of the reward scores also drives the LLM to indiscriminately favor the low-quality chosen responses and fail to generalize to optimal responses that are sparse in data. To overcome these shortcomings, our study introduces reward-conditioned LLM policies that discern and learn from the entire spectrum of response quality within the dataset, helping extrapolate to more optimal regions. We propose an effective yet simple data relabeling method that conditions the preference pairs on quality scores to construct a reward-augmented dataset. The experiments across various benchmarks and diverse models demonstrate that our approach consistently boosts DPO by a considerable margin. Through comprehensive ablation studies, we demonstrate that our method not only maximizes the utility of preference data but also mitigates the issue of unlearning, demonstrating its broad effectiveness beyond mere data expansion. Our code is available at https://github.com/shenao-zhang/reward-augmented-preference.