MMoA: An AI-Agent framework with recurrence for Memoried Mixture-of-Agents

MMoA：一种用于记忆混合智能体（Memoried Mixture-of-Agents）的循环 AI 智能体框架

Abstract: The Mixture-of-Agents (MoA) framework has shown promise in improving large language model (LLM) performance by aggregating outputs from multiple agents. However, existing MoA systems often rely on static routers that do not fully capture temporal and contextual dependencies across aggregation layers.

摘要： 混合智能体（MoA）框架在通过聚合多个智能体的输出以提升大语言模型（LLM）性能方面展现出了巨大潜力。然而，现有的 MoA 系统通常依赖于静态路由，无法充分捕捉跨聚合层的时间和上下文依赖关系。

To address this limitation, we propose MMoA, a recurrent MoA architecture that integrates LSTM-based gating into the agent selection process. The recurrence router adaptively modulates agent contributions based on both current inputs and historical routing decisions, enabling more context-aware aggregation.

为了解决这一局限性，我们提出了 MMoA，这是一种将基于 LSTM 的门控机制集成到智能体选择过程中的循环 MoA 架构。循环路由（Recurrence router）能够根据当前输入和历史路由决策自适应地调节智能体的贡献，从而实现更具上下文感知能力的聚合。

We evaluate MMoA on standard instruction-following benchmarks, including AlpacaEval 2.0, MT-Bench, and Arena-Hard. The results show that MMoA achieves comparable accuracy to traditional MoA while reducing computational overhead by dynamically activating fewer agents.

我们在 AlpacaEval 2.0、MT-Bench 和 Arena-Hard 等标准指令遵循基准测试中对 MMoA 进行了评估。结果表明，MMoA 在实现与传统 MoA 相当的准确率的同时，通过动态激活较少的智能体，降低了计算开销。

For example, on AlpacaEval 2.0, MMoA achieves a win rate of 58.0%, compared with 59.8% for MoA, while improving runtime efficiency by up to 4.6%. These results suggest that MMoA provides a scalable and efficient approach for adaptive multi-agent LLM systems.

例如，在 AlpacaEval 2.0 上，MMoA 的胜率为 58.0%（MoA 为 59.8%），同时运行效率提升了高达 4.6%。这些结果表明，MMoA 为自适应多智能体 LLM 系统提供了一种可扩展且高效的方案。