Geometric-Averaged Preference Optimization for Soft Preference Labels

TL;DR

This paper introduces a geometric-averaged preference optimization method that incorporates distributional human preferences into LLM alignment, improving performance and reducing over-optimization issues.

Contribution

It proposes a novel weighted geometric average approach for soft preference labels in DPO, enhancing alignment and mitigating prior limitations.

Findings

Improved alignment performance on standard benchmarks.

More preferable responses with soft preference labels.

Significant gains with modestly-confident labels.

Abstract

Many algorithms for aligning LLMs with human preferences assume that human preferences are binary and deterministic. However, human preferences can vary across individuals, and therefore should be represented distributionally. In this work, we introduce the distributional soft preference labels and improve Direct Preference Optimization (DPO) with a weighted geometric average of the LLM output likelihood in the loss function. This approach adjusts the scale of learning loss based on the soft labels such that the loss would approach zero when the responses are closer to equally preferred. This simple modification can be easily applied to any DPO-based methods and mitigate over-optimization and objective mismatch, which prior works suffer from. Our experiments simulate the soft preference labels with AI feedback from LLMs and demonstrate that geometric averaging consistently improves performance on standard benchmarks for alignment research. In particular, we observe more preferable responses than binary labels and significant improvements where modestly-confident labels are in the majority.