跳到主内容
@wquguru
精选72Rohan Paul论文研究

微软论文:免训练滑动窗口注意力优于线性注意力

So happy to see this new Microsoft paper.

原文
发到 X

So happy to see this new Microsoft paper.

很高兴看到这篇微软的新论文。

If lower inference memory is the goal, this paper finds training-free Sliding Window Attention beats most retrofitted linear-attention methods, making it the simpler default to try first.

如果目标是降低推理内存占用,该论文发现无需训练的滑动窗口注意力(Sliding Window Attention)优于大多数经过改造的线性注意力方法,因此它成为首选尝试的更简单默认方案。

Keep only a small recent window, plus the first 4 “sink” tokens that models rely on.

仅保留一个较小的近期窗口,以及模型所依赖的前4个“sink”(吸收/汇聚)token。

With a 64-token window, this training-free setup had the best average downstream score in 9 of 11 model comparisons and recovered 99.0% of the full-attention baseline average.

在64个token的窗口设置下,这种无需训练的方案在11组模型对比中的9组取得了最佳的平均下游任务得分,并恢复了完整注意力基线平均水平的99.0%。

Many linear-attention alternatives need additional post-training; this version of SWA needs none.

许多线性注意力的替代方案需要额外的后训练步骤;而此版本的SWA则完全不需要。

The gap grew on long-context reasoning.

在长上下文推理方面,差距进一步扩大。

At 4K context, SWA reached 17.2%–23.0% on the Needle-in-a-Haystack tasks, while LoLCATs reached at most 5.8%; on BABILong, SWA scored 15% versus 3%.

在4K上下文长度下,SWA在Haystack Needle任务上的得分为17.2%–23.0%,而LoLCATs最高仅为5.8%;在BABILong任务上,SWA得分为15%,而后者为3%。

In their speed and memory test, the 64-token SWA setup was fastest and used the least memory.

在速度与内存测试中,64-token的SWA设置速度最快且内存占用最少。

Full attention still wins badly on long context, but for fixed, low memory without retraining, the paper recommends trying SWA with attention sinks first.

在长上下文处理上,完整注意力仍然具有压倒性优势;但在不重新训练且固定低内存的情况下,论文建议首先尝试带有注意力汇聚(attention sinks)的SWA。

– arxiv. org/abs/2608.28444

– arxiv.org/abs/2608.28444

Title: "Sliding-window beats linear attention"

标题:《滑动窗口胜过线性注意力》

更进一步:量化金融体系

看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力

进入量化体系 →

相似阅读

另一事件,读法相近