I’m continuing my series of articles for the Princeton Companion to Math…

把 Transformer decoder 拆开讲透：masked self-attention、cross-attention、FFN 三块子…

If you look at “gaming” keyboards, a lot of them sell for $100 or more o…

I’ve just come back from the 48th Annual IEEE Symposium on the Foundatio…

把 Transformer encoder 从“左半边”这个模糊概念拆成可操作的结构：单层里 self-attention、FFN、残差、Lay…

Here’s a language that gives near-C performance that feels like Python o…

Today I’d like to discuss (part of) a cute and surprising theorem of Fri…

从「self-attention 是排列等变的」这件几乎被忽视的事实出发，推导出位置编码不是装饰、不是工程小技巧，而是结构性必需。原论文为什么选…

Typical server utilization is between 10% and 50%. Google has demonstrat…

This month I have been at the Institute for Advanced Study, participatin…