通过多样化初始查询提升智能体搜索效率，显著改善多跳问答表现。
Beyond Parallel Sampling: Diverse Query Initialization for Agentic Search
- 首步查询采用多样性筛选，避免并行轨迹重复检索
- 在相同算力下，多跳问答平均提升5-7个百分点
- 无需训练，适配各类开源模型与评测基准
智能体搜索的测试时扩展通常通过增加深度（更多对话轮次和令牌）或广度（更多并行轨迹）实现。本文聚焦广度扩展，发现标准并行采样存在收益递减问题，根源在于首轮查询冗余：不同轨迹生成相似初始查询，导致检索结果重叠，后续轮次依赖共享信息。为此提出DivInit——一种无需训练的首轮干预方法：从单次调用中生成n个候选查询，从中挑选k < n个多样性的种子作为并行轨迹起点。在五个开源模型和八个基准上，DivInit持续优于标准并行采样，在匹配算力条件下多跳问答平均提升5至7个百分点。代码已开源：https://github.com/cxcscmu/diverse-query-initialization。
原文摘要 · Abstract (English)
Test-time scaling for agentic search typically increases depth (i.e., more turns and tokens per trajectory) or breadth (i.e., more parallel rollouts). Here we focus on breadth scaling, showing that standard parallel sampling yields diminishing returns, tracing this to query redundancy at the first turn. When models issue similar first queries across rollouts, the threads retrieve overlapping evidence, and subsequent turns are conditioned on this shared retrieval. We address this limitation with DivInit, a training-free intervention at the first turn. Rather than sampling k independent first queries, DivInit draws n candidates from a single call, picks k < n diverse seeds, and runs them as parallel trajectories. Across five open-weight models and eight benchmarks, DivInit consistently improves over standard parallel sampling, with average gains of five to seven points on multi-hop QA at matched compute. Code available at https://github.com/cxcscmu/diverse-query-initialization
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。