不再遗漏任何Token：解析Miles中的Token-In-Token-Out（TITO）
Blog No Token Left Behind: Demystifying Token-In-Token-Out in Miles In agentic RL, a rollout is not a single generation. It is a chain of model calls, tool outputs, harness messages, and resumed generations. Token-In-Token-Out (TITO) is a design principle that address... Miles Team: Jiajun Li, Yanbin Jiang, Mao Cheng, Shi Dong, Yusheng Su, Yueming Yuan, Zhichen Zeng, Banghua Zhu
AI 导读Miles框架提出Token-In-Token-Out（TITO）原则，解决智能体强化学习中训练-推理不匹配：确保rollout过程token序列与训练器评估序列逐位一致。TITO将多轮轨迹视为一个连续序列（每任务一个样本），节省一个数量级计算开销并维持on-policy性。三种破坏场景：反分词-再分词不匹配、聊天模板修剪推理内容、有损模板重新渲染。Miles通过推理会话服务器、三级只追加保证、可插拔TITO分词器和序列比较器实现。典型任务（如SWE-Bench）轨迹含30-50轮。
来源：LMSYS Blog · lmsys.org