用轻量路由器动态选择语言或图结构，让多智能体系统更省 token 又不丢性能。
Routed Graph Handoff: Adaptive Format Selection for Multi-Agent LLM Delegation
- 用 155 tokens 的小模型路由器，按任务自动选语言或图结构通信。
- 在多个任务上压缩 2~3 倍 token 用量，性能反而提升 8.7~12.7 个百分点。
- 适合需要高效推理的多智能体系统，尤其关注降本增效的工程落地者。
多智能体大模型系统通过自然语言消息协作，占用了其 40%–60% 的 token 预算。用结构化图代替语言可降低成本，但在需要自适应推理的任务上表现不佳。本文提出「路由图移交」（Routed Graph Handoff），由一个轻量级语言模型路由器（155 个 token，0.15% 开销）为每次任务委派动态选择使用类型化依赖图或自然语言。在四个基准测试（超过 1,050 条轨迹）上，该系统在所有任务中表现匹配或优于纯语言方案：在 τ-retail 上实现 +12.7 个百分点的提升，压缩比达 3.2×（p<0.01）；在 BrowseComp 上 +8.7 个百分点，压缩比 2.2×（p<0.05）；在 BFCL 与 AppWorld 上达到持平。若无路由器，纯图方案在 AppWorld 上退化 14.6 个百分点，而路由器以近零成本消除此问题。实验表明，需配合图感知执行提示；同一图模式无解释引导则无收益。析因分析揭示尚有 8.6 个百分点的提升空间，为未来运行时自适应路由留出方向。
原文摘要 · Abstract (English)
Multi-agent LLM systems coordinate through natural-language messages that consume 40--60\% of their token budget. Replacing these with structured graphs reduces cost but fails on tasks requiring adaptive reasoning. We propose \textbf{Routed Graph Handoff}, where a lightweight LLM router (155 tokens, 0.15\% overhead) selects between a typed dependency graph and natural language for each delegation. On four benchmarks (1,050+ trajectories), the routed system matches or exceeds NL-only on every task: \textbf{+12.7\,pp} on $τ$-retail at 3.2$\times$ compression ($p{<}0.01$), \textbf{+8.7\,pp} on BrowseComp at 2.2$\times$ compression ($p{<}0.05$), and parity on BFCL and AppWorld. Without the router, graph-only delegation regresses 14.6\,pp on AppWorld; the router eliminates this at near-zero cost. A graph-aware executor prompt is required: the same schema without interpretation guidance yields no gain. An oracle analysis reveals 8.6\,pp of additional headroom, motivating execution-time adaptive routing as future work.
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。