Weighted-Reward Preference Optimization for Implicit Model Fusion
This addresses the challenge of efficiently integrating diverse LLMs for researchers and practitioners, though it appears incremental as it builds on preference optimization techniques.
The paper tackles the problem of fusing heterogeneous open-source LLMs with varying architectures and sizes by proposing Weighted-Reward Preference Optimization (WRPO), an implicit fusion method that eliminates complex vocabulary alignment and matrix merging. The result shows that WRPO consistently outperforms existing methods, achieving a 55.9% win rate against GPT-4-Preview-1106 on AlpacaEval-2 and 46.2% against GPT-4-0314 on Arena-Hard when applied to LLaMA3-8B-Instruct.
While fusing heterogeneous open-source LLMs with varying architectures and sizes can potentially integrate the strengths of different models, existing fusion methods face significant challenges, such as vocabulary alignment and merging distribution matrices. These procedures are not only complex but also prone to introducing noise and errors. In this paper, we propose an implicit fusion method, Weighted-Reward Preference Optimization (WRPO), which leverages preference optimization between the source LLMs and the target LLM to transfer their capabilities effectively. WRPO eliminates the need for vocabulary alignment and matrix fusion and can be efficiently scaled to accommodate various LLMs. To address distributional deviations between the source and target LLMs, WRPO introduces a progressive adaptation strategy that gradually shifts reliance on preferred examples from the target LLM to the source LLMs. Extensive experiments on the MT-Bench, AlpacaEval-2, and Arena-Hard benchmarks demonstrate that WRPO consistently outperforms existing knowledge fusion methods and various fine-tuning baselines. When applied to LLaMA3-8B-Instruct as the target model, WRPO achieves a length-controlled win rate of 55.9% against GPT-4-Preview-1106 on AlpacaEval-2 and a win rate of 46.2% against GPT-4-0314 on Arena-Hard. Our code is available at https://github.com/SLIT-AI/WRPO.