智能AI None 大模型Scaling撞上成本墙：三星提出TrOPD新方法，把前沿智能塞进数亿终端 标题：Trust Region On-Policy Distillation 文章地址：https://arxiv.org/abs/2606.01249 项目地址：https://github.com/Xingrun-Xing2/TrOPD OPD token Policy 2026-09-18 机器之心
智能AI None 政策蒸馏的多面性：陷阱、机制和修复 arXiv:2605.11182v1 Announce Type: new Abstract: On-policy distillation (OPD) and on-policy self-distillation (OPSD) have and OPSD policy 2026-05-13 Siqi Zhu, Xuyan Ye, Hongyu Lu, Weiye Shi, Ge Liu