特点：通过门控机制控制信息流，增强非线性表达。 优点： 适合序列建模、控制性强。 常用于： Transformer FFN、语言模型。
In the 250-plus years between the invention of the water-powered spinning jenny and artificial intelligence, we have developed technology and technique with the primary aim of reducing the number of people necessary to employ for a given amount of output. On a finite planet, the amount of output must eventually stabilise. We cannot maintain for ever the notion that everyone must have a job in order to be allowed to have a life.
，详情可参考服务器推荐
最便宜的 iPhone 又来了，只卖三千块？
a separate allocator for each of your data structures — or rather, for