通过分阶段调度让多个智能体有序协作，大幅减少语言模型的 token 消耗。
Phase-Scheduled Multi-Agent Systems for Token-Efficient Coordination
- 按相位顺序激活智能体，只在必要时处理上下文。
- 平均节省27.3%的 token，性能损失小于2.1个百分点。
- 适合高并发、长流程任务中的高效多智能体系统设计。
基于大语言模型的多智能体系统存在严重的 token 低效问题，源于两个因素：一是无序并行执行，所有智能体同时启动；二是不受限的上下文共享，每个智能体接收全部累积信息。现有方法将协调视为结构分配问题，忽视其时间维度。本文提出相位调度多智能体系统（PSMAS），将智能体激活建模为在圆形流形上的连续控制过程。每个智能体分配固定相位 θ_i ∈ [0, 2π]，由任务依赖拓扑决定；全局扫掠信号 φ(t) 以速度 ω 旋转，仅激活相位窗口 ε 内的智能体。空闲智能体接收压缩的上下文摘要，降低每步的 token 消耗。在 LangGraph 上实现，评估于四个结构化基准（HotPotQA-MAS、HumanEval-MAS、ALFWorld-Multi、WebArena-Coord）和两个非结构化对话场景，证明扫掠动态的稳定性、收敛性与最优性。PSMAS 实现平均 27.3% 的 token 减少（范围 21.4–34.8%），任务性能保持在全激活基线的 2.1 个百分点内（p < 0.01，n = 500/配置），相比最强学习路由基线，减少 5.6 个百分点的 token 消耗，性能下降仅 2.0 个百分点。关键发现：调度与压缩是独立增益源，仅调度即贡献 18–20 个百分点的减少，对压缩退化至 α = 0.40 仍稳健。
原文摘要 · Abstract (English)
Multi-agent systems (MAS) powered by large language models suffer from severe token inefficiency arising from two compounding sources: (i) unstructured parallel execution, where all agents activate simultaneously irrespective of input readiness; and (ii) unrestricted context sharing, where every agent receives the full accumulated context regardless of relevance. Existing mitigation strategies - static pruning, hierarchical decomposition, and learned routing - treat coordination as a structural allocation problem and fundamentally ignore its temporal dimension. We propose Phase-Scheduled Multi-Agent Systems (PSMAS), a framework that reconceptualizes agent activation as continuous control over a shared attention space modeled on a circular manifold. Each agent i is assigned a fixed angular phase theta_i in the range [0, 2*pi], derived from the task dependency topology; a global sweep signal phi(t) rotates at velocity omega, activating only agents within an angular window epsilon. Idle agents receive compressed context summaries, reducing per-step token consumption. We implement PSMAS on LangGraph, evaluate on four structured benchmarks (HotPotQA-MAS, HumanEval-MAS, ALFWorld-Multi, WebArena-Coord) and two unstructured conversational settings, and prove stability, convergence, and optimality results for the sweep dynamics. PSMAS achieves a mean token reduction of 27.3 percent (range 21.4-34.8 percent) while maintaining task performance within 2.1 percentage points of a fully activated baseline (p < 0.01, n = 500 per configuration), and outperforms the strongest learned routing baseline by 5.6 percentage points in token reduction with 2.0 percentage points less performance drop. Crucially, we show that scheduling and compression are independent sources of gain: scheduling alone accounts for 18-20 percentage points of reduction, robust to compression degradation up to alpha = 0.40.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。