# Stanford 前缀滑动法让长推理提速 3 倍

- 来源：elvis (@omarsar0)
- 发布时间：2026-08-31 01:06
- AIHOT 分数：48
- AIHOT 链接：https://aihot.virxact.com/items/cmtg2peew01korov8zdcbdiyf
- 原文链接：https://x.com/omarsar0/status/2094109398604099888

## AI 摘要

Stanford 新研究提出 Prefix Sliding，在生成时丢弃推理轨迹中间 token，仅保留含指令与工具的前缀及末尾数千 token，使内存占用封顶。无需训练即可让现有模型提速 3 倍且性能持平全注意力，并支持超过 100,000 token 的 RL  rollout。论文：arxiv.org/abs/2608.26070。

## 正文

Banger paper from Stanford on efficient test-time scaling.

If you run agents that think for a long time, this one is worth your time.

(bookmark it)

Long reasoning keeps the entire trace in memory through full attention.

This means that the hardest problems, the ones that need the most thinking, are also the ones that cost the most to run.

The authors measured what the middle of a reasoning trace is actually worth.

Intermediate tokens steadily lose importance as the model keeps going.

Their new approach, Prefix Sliding, drops those tokens. It keeps the prefix, which holds the instructions and the available tools, plus a window of the last few thousand tokens. Everything in between gets discarded during generation.

Total memory stays capped no matter how long the model reasons.

Without any training, this runs existing models 3x faster while matching full-attention performance, and it enables RL rollouts past 100,000 tokens.

Paper: https://arxiv.org/abs/2608.26070

Chat with Paper: https://academy.dair.ai/papers/prefix-sliding-for-efficient-test-time-scaling-2608.26070