跳到正文
arXiv · AI· Peiwen Li (Yale University), Harshit Verma (Yale University), Sophia Zhou (Yale University), Haiwen Wang (Yale University), Ngoc Bui (Yale University), Heng Chang (Tsinghua University), Fei Fei (Goodvision AI), David van Dijk (Yale University), Rex Ying (Yale University)·· 7 天前AI 评分30

CoReMem:通过前瞻性估值实现成本内化记忆的编码智能体框架

CoReMem: Cost-Internalizing Memory for Successive Coding Tasks with Prospective Valuation

AI 导读

CoReMem 提出成本内化记忆框架,通过前瞻性估值指导编码智能体在未知任务中平衡记忆效用与读写成本。 在 ChainSWE 基准测试中,该框架将 27B 模型的解决率从 21.0% 提升至 25.7%,并减少 17% 步骤与 11% 总缓存感知 token。

来源:arXiv · AI · arxiv.org