RLAE: Reinforcement Learning-Assisted Ensemble for LLMs

TL;DR

This paper introduces RLAE, a reinforcement learning-based framework that dynamically adjusts ensemble weights of LLMs considering context and intermediate states, significantly improving performance and generalization over fixed-weight methods.

Contribution

RLAE reformulates LLM ensembling as an MDP and employs RL agents to adaptively optimize ensemble weights based on input and output quality, a novel approach in this domain.

Findings

RLAE improves accuracy by up to 3.3% over traditional methods.

The framework generalizes well across diverse tasks without retraining.

RLAE achieves lower latency compared to fixed-weight ensemble methods.

Abstract

Ensembling large language models (LLMs) can effectively combine diverse strengths of different models, offering a promising approach to enhance performance across various tasks. However, existing methods typically rely on fixed weighting strategies that fail to adapt to the dynamic, context-dependent characteristics of LLM capabilities. In this work, we propose Reinforcement Learning-Assisted Ensemble for LLMs (RLAE), a novel framework that reformulates LLM ensemble through the lens of a Markov Decision Process (MDP). Our approach introduces a RL agent that dynamically adjusts ensemble weights by considering both input context and intermediate generation states, with the agent being trained using rewards that directly correspond to the quality of final outputs. We implement RLAE using both single-agent and multi-agent reinforcement learning algorithms ($\text{RLAE}_\text{PPO}$ and $\text{RLAE}_\text{MAPPO}$ ), demonstrating substantial improvements over conventional ensemble methods. Extensive evaluations on a diverse set of tasks show that RLAE outperforms existing approaches by up to $3.3\%$ accuracy points, offering a more effective framework for LLM ensembling. Furthermore, our method exhibits superior generalization capabilities across different tasks without the need for retraining, while simultaneously achieving lower time latency.