mirror of
https://github.com/labmlai/annotated_deep_learning_paper_implementations.git
synced 2025-08-15 10:11:39 +08:00
<!DOCTYPE html>
<html lang="zh">
<head>
<meta http-equiv="content-type" content="text/html;charset=utf-8"/>
<meta name="viewport" content="width=device-width, initial-scale=1.0"/>
<meta name="description" content=""/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
<meta name="twitter:title" content="开关变压器"/>
<meta name="twitter:description" content=""/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/transformers/switch/readme.html"/>
<meta property="og:title" content="开关变压器"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
<meta property="og:site_name" content="开关变压器"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="开关变压器"/>
<meta property="og:description" content=""/>
<title>开关变压器</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/transformers/switch/readme.html"/>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-4V3HC8HBLH');
</script>
</head>
<body>
<div id='container'>
<div id="background"></div>
<div class='section'>
<div class='docs'>
<p>
<a class="parent" href="/">home</a>
<a class="parent" href="../index.html">transformers</a>
<a class="parent" href="index.html">switch</a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
<img alt="Github"
src="https://img.shields.io/github/stars/labmlai/annotated_deep_learning_paper_implementations?style=social"
style="max-width:100%;"/></a>
<a href="https://twitter.com/labmlai" rel="nofollow" target="_blank">
<img alt="Twitter"
src="https://img.shields.io/twitter/follow/labmlai?style=social"
style="max-width:100%;"/></a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/transformers/switch/readme.md" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1><a href="https://nn.labml.ai/transformers/switch/index.html">开关变压器</a></h1>
<p>这是纸质《<a href="https://papers.labml.ai/paper/2101.03961">开关变形金刚：以简单高效的稀疏度扩展到万亿个参数模型》的</a>微型 <a href="https://pytorch.org">PyTorch</a> 实现。我们的实现只有几百万个参数，不对并行分布式训练进行建模。它进行单个 GPU 训练，但我们实现了论文中描述的切换概念。</p>
<p>Switch Transformer 通过根据令牌在参数之间切换，为每个令牌使用不同的参数。因此，只为每个代币选择了一小部分参数。因此，您可以拥有更多参数，但计算成本更低。</p>
<p>切换发生在每个变压器模块的位置前馈网络 (FFN) 上。位置前馈网络由两个按顺序完全连接的层组成。在交换机变压器中，我们有多个 FFN（多位专家），我们根据路由器选择使用哪一个。输出是一组用于选择 FFN 的概率，我们选择概率最高的概率，然后仅对其进行评估。因此，从本质上讲，计算成本与拥有单个 FFN 相同。在我们的实现中，当你有许多或大型 FFN 时，这种并行化效果不佳，因为这一切都发生在单个 GPU 上。在分布式设置中，你会将每个 FFN（每个都很大）放在不同的设备上。</p>
<p>本文引入了另一个损失术语来平衡专家（FFN）之间的负载，并讨论了路由不平衡时丢弃代币的问题。</p>
<p>这是<a href="experiment.html">训练代码和一本</a>用于在 Tiny Shakespeare 数据集上训练开关变压器的笔记本。</p>
</div>
<div class='code'>
</div>
</div>
<div class='footer'>
<a href="https://papers.labml.ai">Trending Research Papers</a>
<a href="https://labml.ai">labml.ai</a>
</div>
</div>
<script src=../../interactive.js?v=1"></script>
<script>
function handleImages() {
var images = document.querySelectorAll('p>img')
for (var i = 0; i < images.length; ++i) {
handleImage(images[i])
}
}
function handleImage(img) {
img.parentElement.style.textAlign = 'center'
var modal = document.createElement('div')
modal.id = 'modal'
var modalContent = document.createElement('div')
modal.appendChild(modalContent)
var modalImage = document.createElement('img')
modalContent.appendChild(modalImage)
var span = document.createElement('span')
span.classList.add('close')
span.textContent = 'x'
modal.appendChild(span)
img.onclick = function () {
console.log('clicked')
document.body.appendChild(modal)
modalImage.src = img.src
}
span.onclick = function () {
document.body.removeChild(modal)
}
}
handleImages()
</script>
</body>
</html>