arXiv Artificial Intelligence· Yuval Kansal, Karen Ullrich, Polina Kirichenko, Ilia Kulikov, Niraj K. Jha, Mohammad Pezeshki, Mark Ibrahim·· 2 小时前AI 评分55
Self-Direction:让智能体把失误蒸馏成低秩适配器经验
Don't Forget What You Just Learned: Self-Directed Repair Turns Agent Mistakes Into Lasting Lessons
AI 导读
论文提出 Self-Direction 机制,智能体任务失败后反思自身轨迹提取经验,并立即蒸馏进一个低秩适配器,基础模型权重保持冻结,每个适配器训练的新增参数约为基座模型参数量的 0.02%。
来源:arXiv Artificial Intelligence · arxiv.org