arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2604.01664cs.AI
收藏

ContextBudget: 长时间 horizon 搜索代理的预算感知上下文管理

ContextBudget: Budget-Aware Context Management for Long-Horizon Search Agents

Zhejiang University(浙江大学);Alibaba Group(阿里巴巴集团)

Yong Wu, YanZhao Zheng, TianZe Xu, ZhenTao Zhang, YuanQiang Yu, JiHuai Zhu, Chao Ma, BinBin Lin, BaoHua Dong, HangCheng Zhu, RuoHui Huang, Gang Yu

AI总结:

本文提出BACM方法,通过预算约束的序列决策问题管理上下文,结合BACM-RL强化学习方法,在多目标问答和长时间网页浏览任务中优于现有方法,尤其在高复杂度场景下表现突出。

摘要:

基于大语言模型的代理在长时间推理方面具有潜力,但其上下文大小受限于部署因素(如内存、延迟和成本),导致上下文预算受限。随着交互历史增长,这引发了在保留过去信息和保持上下文限制之间的权衡。为此,我们提出了预算感知上下文管理(BACM),将上下文管理视为具有上下文预算约束的序列决策问题。它使代理在纳入新观察之前能够评估可用预算,并决定何时以及压缩多少交互历史。我们进一步开发了BACM-RL,一种端到端的课程式强化学习方法,在不同上下文预算下学习压缩策略。在组合多目标问答和长时间网页浏览基准测试中,实验表明BACM-RL在模型规模和任务复杂性上均优于现有方法,在高复杂度设置中比强基线高出超过1.6倍,同时在预算减少时仍保持强劲优势,而大多数方法表现出下降趋势。

英文摘要:

LLM-based agents show strong potential for long-horizon reasoning, yet their context size is limited by deployment factors (e.g., memory, latency, and cost), yielding a constrained context budget. As interaction histories grow, this induces a trade-off between retaining past information and staying within the context limit. To address this challenge, we propose Budget-Aware Context Management (BACM), which formulates context management as a sequential decision problem with a context budget constraint. It enables agents to assess the available budget before incorporating new observations and decide when and how much of the interaction history to compress. We further develop BACM-RL, an end-to-end curriculum-based reinforcement learning approach that learns compression strategies under varying context budgets. Experiments on compositional multi-objective QA and long-horizon web browsing benchmarks show that BACM-RL consistently outperforms prior methods across model scales and task complexities, achieving over $1.6\times$ gains over strong baselines in high-complexity settings, while maintaining strong advantages as budgets shrink, where most methods exhibit a downward performance trend.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑