随着算力的飞速进步，有越多越多的场景希望能够实现“算力换时间”，即通过堆砌算力来缩短模型训练时间。理想情况下，我们希望投入$n$倍的算力，那么达到同样效果的时间则缩短为$1/n$，此时总的算力成...| spaces.ac.cn
In current deep learning tasks, Adam style optimizers such as Adam, Adagrad, RMSProp, Adafactor, and Lion have been widely used as alternatives to SGD style optimizers. These optimizers typically update model parameters using the sign of gradients, resulting in more stable convergence curves. The learning rate and the batch size are the most critical hyperparameters for optimizers, which require careful tuning to enable effective convergence. Previous research has shown that the optimal learn...| papers.cool
In an increasing number of domains it has been demonstrated that deep learning models can be trained using relatively large batch sizes without sacrificing data efficiency. However the limits of this massive data parallelism seem to differ from domain to domain, ranging from batches of tens of thousands in ImageNet to batches of millions in RL agents that play the game Dota 2. To our knowledge there is limited conceptual understanding of why these limits to batch size differ or how we might c...| papers.cool
GELU，全称为Gaussian Error Linear Unit，也算是RELU的变种，是一个非初等函数形式的激活函数。它由论文《Gaussian Error Linear Units (G...| spaces.ac.cn
事实上，除了写博客内容，在这几年里，笔者是花了相当一部分时间来做科学空间的“表面功夫”，为此还专门学了一点php、css和js。虽然不敢说精益求精，但总体来说网站的浏览体验应该比前几年要好得多。...| spaces.ac.cn