Primary sourcearXiv cs.AI
View sourceEvent ·Single source
Prefix Sliding for efficient test-time scaling
Prefix Sliding 提出在推理过程中丢弃非前缀或最后几千个token的token，以减少内存需求，提高测试时间扩展效率。无需训练，Prefix Sliding 可使现有模型速度提升3倍，训练后使用强化学习可进一步提升性能。代码开源于GitHub。
Current read
Prefix 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。
2 sources37° AI heatLast updated 2026/8/26 17:37:15
Evidence
2 sourcesRelated source 1Stanford AI Lab
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.