Weak-to-Strong Preference Optimization: Stealing Reward from Weak Aligned Model
This work addresses the challenge of efficiently aligning strong language models for better user interaction, representing an incremental advancement by adapting weak-to-strong generalization to alignment tasks.
The paper tackles the problem of aligning language models with human preferences by proposing Weak-to-Strong Preference Optimization (WSPO), which transfers alignment behavior from weaker to stronger models, resulting in improved performance such as increasing Qwen2-7B-Instruct's win rate on Arena-Hard from 39.70 to 49.60.
Aligning language models (LMs) with human preferences has become a key area of research, enabling these models to meet diverse user needs better. Inspired by weak-to-strong generalization, where a strong LM fine-tuned on labels generated by a weaker model can consistently outperform its weak supervisor, we extend this idea to model alignment. In this work, we observe that the alignment behavior in weaker models can be effectively transferred to stronger models and even exhibit an amplification effect. Based on this insight, we propose a method called Weak-to-Strong Preference Optimization (WSPO), which achieves strong model alignment by learning the distribution differences before and after the alignment of the weak model. Experiments demonstrate that WSPO delivers outstanding performance, improving the win rate of Qwen2-7B-Instruct on Arena-Hard from 39.70 to 49.60, achieving a remarkable 47.04 length-controlled win rate on AlpacaEval 2, and scoring 7.33 on MT-bench. Our results suggest that using the weak model to elicit a strong model with a high alignment ability is feasible.