Discovering Preference Optimization Algorithms with and for Large Language Models

TL;DR

This paper introduces DiscoPOP, a novel preference optimization algorithm for large language models, discovered through an LLM-driven automated search process, achieving state-of-the-art results without human-designed loss functions.

Contribution

It presents a new automated method for discovering preference optimization algorithms using LLMs, leading to the development of DiscoPOP, which outperforms existing methods.

Findings

DiscoPOP achieves state-of-the-art performance on preference optimization tasks.

The method successfully transfers to unseen tasks.

Automated discovery surpasses manually crafted loss functions.

Abstract

Offline preference optimization is a key method for enhancing and controlling the quality of Large Language Model (LLM) outputs. Typically, preference optimization is approached as an offline supervised learning task using manually-crafted convex loss functions. While these methods are based on theoretical insights, they are inherently constrained by human creativity, so the large search space of possible loss functions remains under explored. We address this by performing LLM-driven objective discovery to automatically discover new state-of-the-art preference optimization algorithms without (expert) human intervention. Specifically, we iteratively prompt an LLM to propose and implement new preference optimization loss functions based on previously-evaluated performance metrics. This process leads to the discovery of previously-unknown and performant preference optimization algorithms. The best performing of these we call Discovered Preference Optimization (DiscoPOP), a novel algorithm that adaptively blends logistic and exponential losses. Experiments demonstrate the state-of-the-art performance of DiscoPOP and its successful transfer to held-out tasks.