Improving LLM General Preference Alignment via Optimistic Online Mirror Descent
This addresses the challenge of modeling complex human preferences for LLM alignment, representing a significant methodological advance rather than an incremental improvement.
The paper tackles the problem of aligning large language models with human preferences without relying on restrictive Bradley-Terry model assumptions, achieving an improved O(T^{-1}) theoretical bound and outperforming state-of-the-art RLHF algorithms across multiple benchmarks.
Reinforcement learning from human feedback (RLHF) has demonstrated remarkable effectiveness in aligning large language models (LLMs) with human preferences. Many existing alignment approaches rely on the Bradley-Terry (BT) model assumption, which assumes the existence of a ground-truth reward for each prompt-response pair. However, this assumption can be overly restrictive when modeling complex human preferences. In this paper, we drop the BT model assumption and study LLM alignment under general preferences, formulated as a two-player game. Drawing on theoretical insights from learning in games, we integrate optimistic online mirror descent into our alignment framework to approximate the Nash policy. Theoretically, we demonstrate that our approach achieves an $O(T^{-1})$ bound on the duality gap, improving upon the previous $O(T^{-1/2})$ result. More importantly, we implement our method and show through experiments that it outperforms state-of-the-art RLHF algorithms across multiple representative benchmarks.