论文10:31Prefix Sliding for efficient test-time scalingPrefix Sliding 提升模型推理速度,无需额外训练,开源代码可查。与总结中间token或普通滑动窗口相比,效果更佳。#Prefix Sliding#test-time scaling#model efficiency#memory reduction1 个信源在谈事件专题aarXiv cs.AI@Niklas Muennighoff 等 18 人2 个信源在谈原文稍后读已读值得跟进有用关注 Prefix Sliding