Rethinking Mixture-of-Agents: Is Mixing Different Large Language Models Beneficial?

TL;DR

This paper questions the benefit of mixing different large language models in ensemble methods and proposes Self-MoA, which outperforms traditional mixtures by aggregating only the top-performing model, leading to state-of-the-art results.

Contribution

The paper introduces Self-MoA, a novel ensemble method that aggregates only the best LLM, demonstrating superior performance over traditional mixture-of-agents approaches.

Findings

Self-MoA outperforms standard MoA by 6.6% on AlpacaEval 2.0.

Self-MoA achieves an average of 3.8% improvement across various benchmarks.

Mixing different LLMs can lower average output quality, making top-only aggregation more effective.

Abstract

Ensembling outputs from diverse sources is a straightforward yet effective approach to boost performance. Mixture-of-Agents (MoA) is one such popular ensemble method that aggregates outputs from multiple different Large Language Models (LLMs). This paper raises the question in the context of language models: is mixing different LLMs truly beneficial? We propose Self-MoA -- an ensemble method that aggregates outputs from only the single top-performing LLM. Our extensive experiments reveal that, surprisingly, Self-MoA outperforms standard MoA that mixes different LLMs in a large number of scenarios: Self-MoA achieves $6.6\%$ improvement over MoA on the AlpacaEval 2.0 benchmark, and an average of $3.8\%$ improvement across various benchmarks, including MMLU, CRUX, and MATH. Applying Self-MoA to one of the top-ranking models in AlpacaEval 2.0 directly achieves the new state-of-the-art performance on the leaderboard. To understand the effectiveness of Self-MoA, we systematically investigate the trade-off between diversity and quality of outputs under various MoA settings. We confirm that the MoA performance is rather sensitive to the quality, and mixing different LLMs often lowers the average quality of the models. To complement the study, we identify the scenarios where mixing different LLMs could be helpful. This paper further introduces a sequential version of Self-MoA, that is capable of aggregating a large number of LLM outputs on-the-fly over multiple rounds, and is as effective as aggregating all outputs at once.