Discovering Preference Optimization Algorithms with and for Large Language Models
This work addresses the challenge of enhancing LLM output quality for AI developers by automating algorithm discovery, though it is incremental as it builds on existing preference optimization methods.
The paper tackles the problem of offline preference optimization for Large Language Models by automatically discovering new algorithms through LLM-driven objective discovery, resulting in DiscoPOP, which achieves state-of-the-art performance and successfully transfers to held-out tasks.
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.