Weighted-Reward Preference Optimization for Implicit Model Fusion

TL;DR

This paper introduces WRPO, a novel implicit fusion method for combining heterogeneous LLMs that avoids complex alignment procedures and demonstrates superior performance on multiple benchmarks.

Contribution

WRPO is a new preference optimization approach that effectively fuses diverse LLMs without vocabulary or matrix alignment, scalable and adaptable to various models.

Findings

WRPO outperforms existing fusion methods and fine-tuning baselines.

Achieves 55.9% win rate against GPT-4-Preview-1106 on AlpacaEval-2.

Achieves 46.2% win rate against GPT-4-0314 on Arena-Hard.

Abstract

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.