CarryOn:为长程 LLM 智能体学习上下文保留策略
CarryOn: Learning What to Keep in Context for Long-Horizon LLM Agents
智能体跑长任务时历史记录越长越容易乱,摘要又会丢细节。这篇论文的 CarryOn 用学习型选择器挑证据保留,不加 LLM 调用,三个基准上超基线 6.7 个百分点。做智能体开发、被上下文成本折磨的人可以看看思路。
arXiv 论文提出 CarryOn,一个轻量级上下文管理器,用于长程 LLM 智能体的执行历史管理。在 AppWorld、OfficeBench 和 Tau2 Telecom 三个基准上配合三种 LLM Actor,CarryOn 达到 75.7% 的宏平均任务成功率,超过最强基线 6.7 个百分点。
其核心机制是把紧凑的执行状态与细粒度、带来源链接的证据配对:一个学习型选择器加上自适应保留策略决定哪些证据留在上下文中,且不需要额外调用 LLM 来做上下文管理,避免了传统摘要或剪枝方法丢失后续行动所需细节的问题。
论文还提出两个评估指标:S_α 在衡量任务成功的同时计入资源消耗,NCS(normalized cost per success)度量每个成功任务花费的资源,CarryOn 在这两个指标上均取得最佳平均分。
CarryOn: Learning What to Keep in Context for Long-Horizon LLM Agents
Long-horizon LLM agents can coordinate tools and applications to complete complex, multi-step tasks. As their interaction histories grow, effective context management becomes essential for reliable and efficient execution. Existing methods reduce context through summarization, pruning, or history selection. However, shorter contexts do not necessarily support better decisions: summaries can omit exact details needed for later actions, while retaining entire interactions can carry substantial irrelevant information. We propose CarryOn, a lightweight context manager that pairs compact execution state with fine-grained, source-linked evidence. A learned selector and adaptive retention policy keep relevant evidence available without additional LLM calls for context management. We introduce two complementary evaluation metrics: $S_α$ prioritizes task success while accounting for resource use, and normalized cost per success (NCS) measures the resources spent per successful task. Across AppWorld, OfficeBench, and Tau2 Telecom with three LLM Actors, CarryOn achieves 75.7% macro-averaged task success, exceeding the strongest baseline by 6.7 percentage points. It also achieves the best average scores on both metrics.
本条情报归属 AITOP「论文」核心栏目,重点聚焦北美前沿实验室与硅谷科技巨头。在跨区域产业观察坐标中,该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:CarryOn、智能体、上下文管理、AppWorld,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明:本文原始线索采编自「arXiv cs.AI」(@Lihuan Li, Ruiyi Yang, Jiru Wang, Yifan Duan, Xiachong Lin, Hao Xue, Yang Song, Flora Salim),经 AITOP 多维评分管线结构化提炼。