通过上下文折叠实现长程LLM智能体的扩展

摘要

大型語言模型（LLM）代理在處理長期任務時，其效能根本上受制於上下文長度的限制。本文提出了一種名為“上下文折疊”（Context-Folding）的框架，該框架賦予代理主動管理其工作上下文的能力。代理能夠程序性地分支進入子軌跡以處理子任務，並在完成後將其折疊，從而壓縮中間步驟，同時保留結果的簡明摘要。為了使此行為可學習，我們開發了一種端到端的強化學習框架FoldGRPO，該框架通過特定的過程獎勵來鼓勵有效的任務分解與上下文管理。在複雜的長期任務（如深度研究與軟體工程）上，我們的折疊代理在保持活躍上下文僅為十分之一大小的情況下，與ReAct基準線持平或更優，並且顯著優於依賴於基於摘要的上下文管理模型。

English

Large language model (LLM) agents are fundamentally constrained by context length on long-horizon tasks. We introduce Context-Folding, a framework that empowers agents to actively manage their working context. An agent can procedurally branch into a sub-trajectory to handle a subtask and then fold it upon completion, collapsing the intermediate steps while retaining a concise summary of the outcome. To make this behavior learnable, we develop an end-to-end reinforcement learning framework FoldGRPO with specific process rewards to encourage effective task decomposition and context management. On complex long-horizon tasks (Deep Research and SWE), our folding agent matches or outperforms the ReAct baselines while using an active context 10times smaller and significantly outperforms models that rely on summarization-based context management.

通过上下文折叠实现长程LLM智能体的扩展

Scaling Long-Horizon LLM Agent via Context-Folding

摘要

Support