组合元路由学习：智能体工作流的可执行基准
原标题：Learning Compositional Meta-Routing for Agentic Workflows: An Executable Benchmark
AI 摘要
该研究提出了一种可执行的基准测试和预算感知的元路由器，用于组合智能体工作流中的异构操作。基准包含216个训练、72个开发、108个测试和108个锁定词汇转移挑战任务，覆盖数据分析、冻结语料库研究和文档处理。学习策略在测试集上达到100%成功率，比静态工作流高6.5个百分点，成本降低43%；但在挑战集上成功率降至75.9%，落后于静态路由的93.5%，表明词汇泛化是主要限制。
正文节选
Computer Science > Machine Learning Title:Learning Compositional Meta-Routing for Agentic Workflows: An Executable Benchmark View PDF HTML (experimental) Abstract:Agentic systems must decide not only what answer to produce, but which reasoning and execution operations should precede it. A controller may answer directly, decompose a request, retrieve evidence, execute code, delegate to a specialist, or verify an intermediate result. Existing routing work largely selects model endpoint