Event ·Single source

HOLA:为线性注意力引入海马体互补存储,提升长程召回

HOLA论文提出为线性注意力添加海马体互补存储,保留delta-rule状态的压缩记忆并增加一个有限精确KV缓存,形成半参数测试时记忆。在340M参数、15B SlimPajama tokens训练下,WikiText困惑度从27.32降至22.92,低于全注意力Transformer++的26.88。在RULER needle recall测试中,HOLA在32k tokens(16倍训练长度)上保持稳健。

Current read

线性注意力一直有长程遗忘问题,HOLA用一个小缓存就解决了,在32k长度上效果惊人。

2 sources52° AI heatLast updated 2026/7/3 15:38:01

Evidence

2 sources
Related source 1arXiv cs.AI
View source

Conflict 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.