Who is this book for?| www.math.uci.edu
29| spaces.ac.cn
前言：去年写过一篇WGAN-GP的入门读物《互怼的艺术：从零直达WGAN-GP》，提到通过梯度惩罚来为WGAN的判别器增加Lipschitz约束（下面简称“L约束”）。前几天遐想时再次想到了WG...| spaces.ac.cn
众所周知，完整训练一次大型LLM的成本是昂贵的，这就决定了我们不可能直接在大型LLM上反复测试超参数。一个很自然的想法是希望可以在同结构的小模型上仔细搜索超参数，找到最优组合后直接迁移到大模型上...| spaces.ac.cn
The push to train ever larger neural networks has motivated the study of initialization and training at large network width. A key challenge is to scale training so that a network's internal representations evolve nontrivially at all widths, a process known as feature learning. Here, we show that feature learning is achieved by scaling the spectral norm of weight matrices and their updates like $\sqrt{\texttt{fan-out}/\texttt{fan-in}}$, in contrast to widely used but heuristic scalings based ...| papers.cool
Recently, the Muon optimizer based on matrix orthogonalization has demonstrated strong results in training small-scale language models, but the scalability to larger models has not been proven. We identify two crucial techniques for scaling up Muon: (1) adding weight decay and (2) carefully adjusting the per-parameter update scale. These techniques allow Muon to work out-of-the-box on large-scale training without the need of hyper-parameter tuning. Scaling law experiments indicate that Muon a...| papers.cool
随着LLM时代的到来，学术界对于优化器的研究热情似乎有所减退。这主要是因为目前主流的AdamW已经能够满足大多数需求，而如果对优化器“大动干戈”，那么需要巨大的验证成本。因此，当前优化器的变化，...| spaces.ac.cn
在文章《初探muP：超参数的跨模型尺度迁移规律》中，我们基于前向传播、反向传播、损失增量和特征变化的尺度不变性推导了muP（Maximal Update Parametrization）。可能对...| spaces.ac.cn
上一篇文章中我们介绍了“伪逆”，它关系到给定矩阵$\boldsymbol{M}$和$\boldsymbol{A}$（或$\boldsymbol{B}$）时优化目标$\Vert \boldsymb...| 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