Reward-Augmented Data Enhances Direct Preference Alignment of LLMs

2024-10-10

arXiv.org

AI Summary

Existing direct preference optimization (DPO) methods rely solely on binary preference labels, ignoring the continuous gradient of response quality—leading to overfitting, erroneous rejection of high-quality responses, and poor convergence toward optimal outputs. To address this, we propose a reward-augmented preference alignment framework: for the first time, we explicitly incorporate continuous reward scores into DPO training via reward-conditioned policy modeling and preference pair relabeling, thereby constructing a quality-aware augmented dataset. This enables the policy model to perceive the full quality spectrum and extrapolate beyond observed preferences toward superior responses, effectively mitigating the “unlearning” problem. Our method achieves state-of-the-art performance on AlpacaEval, MT-Bench, and Arena-Hard-Auto. Under on-policy evaluation, it improves AlpacaEval win rates by 3.2% and boosts average accuracy on academic benchmarks.

Technology Category

• Humans and AI: Learning Human Values and Preferences
• Machine Learning: Learning Preferences or Rankings
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

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. 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 responses with the highest rewards, which 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. This dataset is easily integrated with existing direct alignment algorithms and is applicable to any preference dataset. The experimental results across instruction-following benchmarks including AlpacaEval, MT-Bench, and Arena-Hard-Auto demonstrate that our approach consistently boosts the performance of DPO by a considerable margin across diverse models. Additionally, our method improves the average accuracy on various academic benchmarks. When applying our method to on-policy data, the resulting DPO model achieves SOTA results on AlpacaEval. Through 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 dataset expansion. Our code is available at https://github.com/shenao-zhang/reward-augmented-preference.

Problem

• Language Model Optimization
• Response Quality Assessment
• Preference Alignment

Innovation

• Quality-aware Learning
• Preference Alignment
• Data Labeling Innovation