wehub/labmlai--annotated_deep_learning_paper_implementations
1
0
Fork 0
Code Issues Pull Requests Actions Deployments Models Agent Notes Packages Projects Releases Wiki Activity

chore: import upstream snapshot with attribution

Browse Source
This commit is contained in:
wehub-resource-sync
2026-07-13 12:19:01 +08:00
commit 3b90d1192f
2172 changed files with 594509 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/ja/optimizers/ada_belief.html
+507
View File
File diff suppressed because one or more lines are too long
docs/ja/optimizers/adam.html
+747
View File
File diff suppressed because one or more lines are too long
docs/ja/optimizers/adam_fp16.html
+502
View File
@@ -0,0 +1,502 @@
<!DOCTYPE html>
<html lang="ja">
<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="Adam オプティマイザーの簡単な PyTorch 実装/チュートリアル"/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="半精度トレーニング用の Adam オプティマイザー"/>
<meta name="twitter:description" content="Adam オプティマイザーの簡単な PyTorch 実装/チュートリアル"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/optimizers/adam_fp16.html"/>
<meta property="og:title" content="半精度トレーニング用の Adam オプティマイザー"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="半精度トレーニング用の Adam オプティマイザー"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="半精度トレーニング用の Adam オプティマイザー"/>
<meta property="og:description" content="Adam オプティマイザーの簡単な PyTorch 実装/チュートリアル"/>
<title>半精度トレーニング用の Adam オプティマイザー</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/optimizers/adam_fp16.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">optimizers</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/optimizers/adam_fp16.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1>半精度トレーニング用の Adam オプティマイザー</h1>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">10</span><span></span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Dict</span><span class="p">,</span> <span class="n">Tuple</span><span class="p">,</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">Any</span>
<span class="lineno">11</span>
<span class="lineno">12</span><span class="kn">import</span> <span class="nn">torch</span>
<span class="lineno">13</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
<span class="lineno">14</span><span class="kn">from</span> <span class="nn">torch.optim</span> <span class="kn">import</span> <span class="n">Optimizer</span>
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">torch.cuda.amp</span> <span class="kn">import</span> <span class="n">grad_scaler</span>
<span class="lineno">16</span><span class="kn">from</span> <span class="nn">collections</span> <span class="kn">import</span> <span class="n">defaultdict</span><span class="p">,</span> <span class="n">abc</span>
<span class="lineno">17</span>
<span class="lineno">18</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers</span> <span class="kn">import</span> <span class="n">WeightDecay</span>
<span class="lineno">19</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers.adam</span> <span class="kn">import</span> <span class="n">Adam</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
<h2>半精度トレーニング用の Adam オプティマイザー</h2>
<p><a href="adam.html">Adam Optimizerを拡張しましたが</a>、グラデーションとモーメントの保存にはFP32を使用しています。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">22</span><span class="k">class</span> <span class="nc">AdamFP16</span><span class="p">(</span><span class="n">Adam</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">29</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">params</span><span class="p">,</span> <span class="n">lr</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">1e-3</span><span class="p">,</span> <span class="n">betas</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">float</span><span class="p">,</span> <span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="p">(</span><span class="mf">0.9</span><span class="p">,</span> <span class="mf">0.999</span><span class="p">),</span> <span class="n">eps</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">1e-16</span><span class="p">,</span>
<span class="lineno">30</span> <span class="n">weight_decay</span><span class="p">:</span> <span class="n">WeightDecay</span> <span class="o">=</span> <span class="n">WeightDecay</span><span class="p">(),</span> <span class="n">optimized_update</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">,</span>
<span class="lineno">31</span> <span class="n">defaults</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Any</span><span class="p">]]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-3'>
<div class='docs'>
<div class='section-link'>
<a href='#section-3'>#</a>
</div>
<p>32 ビットのグラデーションを格納するパラメーター。<code class="highlight"><span></span><span class="n">GradScaler</span></code>
これには以下の定義が入力されます</p>。
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">33</span> <span class="bp">self</span><span class="o">.</span><span class="n">grad_fp32</span> <span class="o">=</span> <span class="p">{}</span></pre></div>
</div>
</div>
<div class='section' id='section-4'>
<div class='docs'>
<div class='section-link'>
<a href='#section-4'>#</a>
</div>
<p><a href="adam.html">Adam オプティマイザーイニシャライザーを呼び出す</a></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">35</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">params</span><span class="p">,</span> <span class="n">lr</span><span class="p">,</span> <span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="p">,</span> <span class="n">weight_decay</span><span class="p">,</span> <span class="n">optimized_update</span><span class="p">,</span> <span class="n">defaults</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-5'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-5'>#</a>
</div>
<h3>パラメータ状態を初期化</h3>
<ul><li><code class="highlight"><span></span><span class="n">state</span></code>
はパラメーター (テンソル) のオプティマイザー状態です</li>
<li><code class="highlight"><span></span><span class="n">group</span></code>
パラメータグループのオプティマイザ属性を格納します</li>
<li><code class="highlight"><span></span><span class="n">param</span></code>
はパラメータテンソル <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.902771em;vertical-align:-0.208331em;"></span><span class="mord coloredeq eqa" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.02778em">θ</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:-0.02778em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqe" style="">t</span></span><span class="mbin mtight" style="">−</span><span class="mord mtight" style="">1</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span></span></span></span></span></span></li></ul>
<p>すべてのステートテンソルは FP32 を使用します。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">37</span> <span class="k">def</span> <span class="nf">init_state</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">state</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">group</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">param</span><span class="p">:</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-6'>
<div class='docs'>
<div class='section-link'>
<a href='#section-6'>#</a>
</div>
<p>これは、パラメーターに対して実行されたオプティマイザーステップの数です。<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.61508em;vertical-align:0em;"></span><span class="mord coloredeq eqe" style=""><span class="mord mathnormal" style="">t</span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">49</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;step&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="mi">0</span></pre></div>
</div>
</div>
<div class='section' id='section-7'>
<div class='docs'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
<p>勾配の指数移動平均、<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.58056em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mord mathnormal" style="">m</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqe" style="">t</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">51</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;exp_avg&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">zeros_like</span><span class="p">(</span><span class="n">param</span><span class="p">,</span> <span class="n">memory_format</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">preserve_format</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">float</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-8'>
<div class='docs'>
<div class='section-link'>
<a href='#section-8'>#</a>
</div>
<p>二乗勾配値の指数移動平均、<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.58056em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqd" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.03588em">v</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqe" style="">t</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">53</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;exp_avg_sq&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">zeros_like</span><span class="p">(</span><span class="n">param</span><span class="p">,</span> <span class="n">memory_format</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">preserve_format</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">float</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-9'>
<div class='docs'>
<div class='section-link'>
<a href='#section-9'>#</a>
</div>
<p>パラメータの FP32 コピーを管理</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">55</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;fp32_copy&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="n">param</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">float</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-10'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-10'>#</a>
</div>
<h3>与えられたパラメータテンソルの更新ステップを実行する</h3>
<ul><li><code class="highlight"><span></span><span class="n">state</span></code>
はパラメーター (テンソル) のオプティマイザー状態です</li>
<li><code class="highlight"><span></span><span class="n">group</span></code>
パラメータグループのオプティマイザ属性を格納します</li>
<li><code class="highlight"><span></span><span class="n">grad</span></code>
<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.625em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.03588em;">g</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight coloredeq eqe" style=""><span class="mord mathnormal mtight" style="">t</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span>パラメータの現在の勾配テンソルです <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.902771em;vertical-align:-0.208331em;"></span><span class="mord coloredeq eqa" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.02778em">θ</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:-0.02778em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqe" style="">t</span></span><span class="mbin mtight" style="">−</span><span class="mord mtight" style="">1</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">param</span></code>
はパラメータテンソル <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.902771em;vertical-align:-0.208331em;"></span><span class="mord coloredeq eqa" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.02778em">θ</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.301108em;"><span style="top:-2.5500000000000003em;margin-left:-0.02778em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqe" style="">t</span></span><span class="mbin mtight" style="">−</span><span class="mord mtight" style="">1</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.208331em;"><span></span></span></span></span></span></span></span></span></span></span></span></li></ul>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">57</span> <span class="k">def</span> <span class="nf">step_param</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">state</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">group</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">grad</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">param</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-11'>
<div class='docs'>
<div class='section-link'>
<a href='#section-11'>#</a>
</div>
<p>FP32 パラメータを取得</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">68</span> <span class="n">param_fp32</span> <span class="o">=</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;fp32_copy&#39;</span><span class="p">]</span></pre></div>
</div>
</div>
<div class='section' id='section-12'>
<div class='docs'>
<div class='section-link'>
<a href='#section-12'>#</a>
</div>
<p>可能な場合は FP32 のグラデーションを取得</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">70</span> <span class="n">grad_fp32</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">grad_fp32</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">param</span><span class="p">,</span> <span class="kc">None</span><span class="p">)</span>
<span class="lineno">71</span> <span class="k">if</span> <span class="n">grad_fp32</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="lineno">72</span> <span class="k">del</span> <span class="bp">self</span><span class="o">.</span><span class="n">grad_fp32</span><span class="p">[</span><span class="n">param</span><span class="p">]</span>
<span class="lineno">73</span> <span class="n">grad</span> <span class="o">=</span> <span class="n">grad_fp32</span>
<span class="lineno">74</span> <span class="k">else</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-13'>
<div class='docs'>
<div class='section-link'>
<a href='#section-13'>#</a>
</div>
<p>それ以外の場合は、グラデーションを FP32 に変換します。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">76</span> <span class="n">grad</span> <span class="o">=</span> <span class="n">grad</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">float</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-14'>
<div class='docs'>
<div class='section-link'>
<a href='#section-14'>#</a>
</div>
<p>体重減少の計算</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">79</span> <span class="n">grad</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight_decay</span><span class="p">(</span><span class="n">param_fp32</span><span class="p">,</span> <span class="n">grad</span><span class="p">,</span> <span class="n">group</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-15'>
<div class='docs'>
<div class='section-link'>
<a href='#section-15'>#</a>
</div>
<p><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.58056em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mord mathnormal" style="">m</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqe" style="">t</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span>取得して <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.58056em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqd" style=""><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.03588em">v</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight coloredeq eqe" style="">t</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">82</span> <span class="n">m</span><span class="p">,</span> <span class="n">v</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">get_mv</span><span class="p">(</span><span class="n">state</span><span class="p">,</span> <span class="n">group</span><span class="p">,</span> <span class="n">grad</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-16'>
<div class='docs'>
<div class='section-link'>
<a href='#section-16'>#</a>
</div>
<p><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.61508em;vertical-align:0em;"></span><span class="mord coloredeq eqe" style=""><span class="mord mathnormal" style="">t</span></span></span></span></span></span>オプティマイザーのステップ数を増やす</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">85</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;step&#39;</span><span class="p">]</span> <span class="o">+=</span> <span class="mi">1</span></pre></div>
</div>
</div>
<div class='section' id='section-17'>
<div class='docs'>
<div class='section-link'>
<a href='#section-17'>#</a>
</div>
<p><em>Adam</em> アップデートを実行</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">88</span> <span class="bp">self</span><span class="o">.</span><span class="n">adam_update</span><span class="p">(</span><span class="n">state</span><span class="p">,</span> <span class="n">group</span><span class="p">,</span> <span class="n">param_fp32</span><span class="p">,</span> <span class="n">m</span><span class="p">,</span> <span class="n">v</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-18'>
<div class='docs'>
<div class='section-link'>
<a href='#section-18'>#</a>
</div>
<p>パラメータを設定</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">91</span> <span class="n">param</span><span class="o">.</span><span class="n">data</span> <span class="o">=</span> <span class="n">param_fp32</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">param</span><span class="o">.</span><span class="n">dtype</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-19'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-19'>#</a>
</div>
<h2>半精度グラデーションのグラデーションスケーラー</h2>
<p>PyTorch グラデーションスケーラーを FP32 グラデーションを使用するように拡張します。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">94</span><span class="k">class</span> <span class="nc">GradScalerFP16</span><span class="p">(</span><span class="n">grad_scaler</span><span class="o">.</span><span class="n">GradScaler</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-20'>
<div class='docs'>
<div class='section-link'>
<a href='#section-20'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">101</span> <span class="k">def</span> <span class="nf">_unscale_grads_</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">optimizer</span><span class="p">:</span> <span class="n">Optimizer</span><span class="p">,</span> <span class="n">inv_scale</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">found_inf</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span>
<span class="lineno">102</span> <span class="n">allow_fp16</span><span class="p">:</span> <span class="nb">bool</span><span class="p">)</span> <span class="o">-&gt;</span> <span class="n">Dict</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">,</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">]:</span>
<span class="lineno">103</span> <span class="n">per_device_inv_scale</span> <span class="o">=</span> <span class="n">grad_scaler</span><span class="o">.</span><span class="n">_MultiDeviceReplicator</span><span class="p">(</span><span class="n">inv_scale</span><span class="p">)</span>
<span class="lineno">104</span> <span class="n">per_device_found_inf</span> <span class="o">=</span> <span class="n">grad_scaler</span><span class="o">.</span><span class="n">_MultiDeviceReplicator</span><span class="p">(</span><span class="n">found_inf</span><span class="p">)</span>
<span class="lineno">105</span>
<span class="lineno">106</span> <span class="n">per_device_and_dtype_grads</span> <span class="o">=</span> <span class="n">defaultdict</span><span class="p">(</span><span class="k">lambda</span><span class="p">:</span> <span class="n">defaultdict</span><span class="p">(</span><span class="nb">list</span><span class="p">))</span> <span class="c1"># type: ignore[var-annotated]</span>
<span class="lineno">107</span>
<span class="lineno">108</span> <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span></pre></div>
</div>
</div>
<div class='section' id='section-21'>
<div class='docs'>
<div class='section-link'>
<a href='#section-21'>#</a>
</div>
<p>ループスルーパラメータ</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">110</span> <span class="k">for</span> <span class="n">group</span> <span class="ow">in</span> <span class="n">optimizer</span><span class="o">.</span><span class="n">param_groups</span><span class="p">:</span>
<span class="lineno">111</span> <span class="k">for</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">group</span><span class="p">[</span><span class="s2">&quot;params&quot;</span><span class="p">]:</span></pre></div>
</div>
</div>
<div class='section' id='section-22'>
<div class='docs'>
<div class='section-link'>
<a href='#section-22'>#</a>
</div>
<p>トレーニング不可のパラメータをスキップ</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">113</span> <span class="k">if</span> <span class="n">param</span><span class="o">.</span><span class="n">grad</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="lineno">114</span> <span class="k">continue</span></pre></div>
</div>
</div>
<div class='section' id='section-23'>
<div class='docs'>
<div class='section-link'>
<a href='#section-23'>#</a>
</div>
<p>スパステンソルには実装されていません</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">116</span> <span class="k">if</span> <span class="n">param</span><span class="o">.</span><span class="n">grad</span><span class="o">.</span><span class="n">is_sparse</span><span class="p">:</span>
<span class="lineno">117</span> <span class="k">raise</span> <span class="ne">NotImplementedError</span></pre></div>
</div>
</div>
<div class='section' id='section-24'>
<div class='docs'>
<div class='section-link'>
<a href='#section-24'>#</a>
</div>
<p>FP32 <code class="highlight"><span></span><span class="n">AdamFP16</span></code>
<code class="highlight"><span></span><span class="n">optimizer</span><span class="o">.</span><span class="n">grad_fp32</span><span class="p">[</span><span class="n">param</span><span class="p">]</span></code>
グラデーションに設定されたオプティマイザーを使用している場合</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">120</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">optimizer</span><span class="p">,</span> <span class="n">AdamFP16</span><span class="p">):</span>
<span class="lineno">121</span> <span class="n">grad</span> <span class="o">=</span> <span class="n">param</span><span class="o">.</span><span class="n">grad</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">float</span><span class="p">)</span>
<span class="lineno">122</span> <span class="n">optimizer</span><span class="o">.</span><span class="n">grad_fp32</span><span class="p">[</span><span class="n">param</span><span class="p">]</span> <span class="o">=</span> <span class="n">grad</span></pre></div>
</div>
</div>
<div class='section' id='section-25'>
<div class='docs'>
<div class='section-link'>
<a href='#section-25'>#</a>
</div>
<p>それ以外の場合は、グラデーションを FP32 に変換しないでください。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">124</span> <span class="k">else</span><span class="p">:</span>
<span class="lineno">125</span> <span class="n">grad</span> <span class="o">=</span> <span class="n">param</span><span class="o">.</span><span class="n">grad</span>
<span class="lineno">126</span>
<span class="lineno">127</span> <span class="n">per_device_and_dtype_grads</span><span class="p">[</span><span class="n">grad</span><span class="o">.</span><span class="n">device</span><span class="p">][</span><span class="n">grad</span><span class="o">.</span><span class="n">dtype</span><span class="p">]</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">grad</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-26'>
<div class='docs'>
<div class='section-link'>
<a href='#section-26'>#</a>
</div>
<p>すべてのグラデーションをスケール解除</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">130</span> <span class="k">for</span> <span class="n">device</span><span class="p">,</span> <span class="n">per_dtype_grads</span> <span class="ow">in</span> <span class="n">per_device_and_dtype_grads</span><span class="o">.</span><span class="n">items</span><span class="p">():</span>
<span class="lineno">131</span> <span class="k">for</span> <span class="n">grads</span> <span class="ow">in</span> <span class="n">per_dtype_grads</span><span class="o">.</span><span class="n">values</span><span class="p">():</span>
<span class="lineno">132</span> <span class="n">torch</span><span class="o">.</span><span class="n">_amp_foreach_non_finite_check_and_unscale_</span><span class="p">(</span><span class="n">grads</span><span class="p">,</span>
<span class="lineno">133</span> <span class="n">per_device_found_inf</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">device</span><span class="p">),</span>
<span class="lineno">134</span> <span class="n">per_device_inv_scale</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">device</span><span class="p">))</span></pre></div>
</div>
</div>
<div class='section' id='section-27'>
<div class='docs'>
<div class='section-link'>
<a href='#section-27'>#</a>
</div>
<p></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">136</span> <span class="k">return</span> <span class="n">per_device_found_inf</span><span class="o">.</span><span class="n">_per_device_tensors</span></pre></div>
</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>
docs/ja/optimizers/adam_warmup.html
+243
View File
@@ -0,0 +1,243 @@
<!DOCTYPE html>
<html lang="ja">
<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="ウォームアップ機能付きの Adam オプティマイザーの簡単な PyTorch 実装/チュートリアル。"/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="ウォームアップ機能付き Adam オプティマイザー"/>
<meta name="twitter:description" content="ウォームアップ機能付きの Adam オプティマイザーの簡単な PyTorch 実装/チュートリアル。"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/optimizers/adam_warmup.html"/>
<meta property="og:title" content="ウォームアップ機能付き Adam オプティマイザー"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="ウォームアップ機能付き Adam オプティマイザー"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="ウォームアップ機能付き Adam オプティマイザー"/>
<meta property="og:description" content="ウォームアップ機能付きの Adam オプティマイザーの簡単な PyTorch 実装/チュートリアル。"/>
<title>ウォームアップ機能付き Adam オプティマイザー</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/optimizers/adam_warmup.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">optimizers</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/optimizers/adam_warmup.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1>ウォームアップ機能付き Adam オプティマイザー</h1>
<p>これにより <a href="amsgrad.html">AMSgrad オプティマイザが拡張され</a>、ウォームアップステージが追加されます。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">12</span><span></span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Dict</span>
<span class="lineno">13</span>
<span class="lineno">14</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers</span> <span class="kn">import</span> <span class="n">WeightDecay</span>
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers.amsgrad</span> <span class="kn">import</span> <span class="n">AMSGrad</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
<h2>ウォームアップ機能付き Adam オプティマイザー</h2>
<p>このクラスは、で定義されている AMSGrad オプティマイザを拡張したものです。<a href="amsgrad.html"><code class="highlight"><span></span><span class="n">amsgrad</span><span class="o">.</span><span class="n">py</span></code>
</a></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">18</span><span class="k">class</span> <span class="nc">AdamWarmup</span><span class="p">(</span><span class="n">AMSGrad</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
<h3>オプティマイザを初期化</h3>
<ul><li><code class="highlight"><span></span><span class="n">params</span></code>
はパラメータのリストです</li>
<li><code class="highlight"><span></span><span class="n">lr</span></code>
は学習率 <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqf" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">betas</span></code>
(,) <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span> のタプルです <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">2</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">eps</span></code>
<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.69444em;vertical-align:0em;"></span><span class="mord accent"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.69444em;"><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord coloredeq eqc" style=""><span class="mord mathnormal" style="">ϵ</span></span></span><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="accent-body" style="left:-0.19444em;"><span class="mord">^</span></span></span></span></span></span></span></span></span></span></span><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqc" style=""><span class="mord mathnormal" style="">ϵ</span></span></span></span></span></span>またはそれに基づいている <code class="highlight"><span></span><span class="n">optimized_update</span></code>
</li>
<li><code class="highlight"><span></span><span class="n">weight_decay</span></code>
<code class="highlight"><span></span><span class="n">WeightDecay</span></code>
で定義されているクラスのインスタンスです <a href="index.html"><code class="highlight"><span></span><span class="fm">__init__</span><span class="o">.</span><span class="n">py</span></code>
</a></li>
<li>'optimized_update'は追加後に行うことでセカンドモーメントのバイアス補正を最適化するかどうかのフラグです <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqc" style=""><span class="mord mathnormal" style="">ϵ</span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">amsgrad</span></code>
amsGradを使用するか、プレーンなAdamにフォールバックするかを示すフラグです</li>
<li><code class="highlight"><span></span><span class="n">warmup</span></code>
ウォームアップステップ数</li>
<li><code class="highlight"><span></span><span class="n">defaults</span></code>
グループ値のデフォルト辞書です。これは、クラスを拡張する場合に便利です<code class="highlight"><span></span><span class="n">AdamWarmup</span></code>
。</li></ul>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">24</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">params</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-3</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="p">(</span><span class="mf">0.9</span><span class="p">,</span> <span class="mf">0.999</span><span class="p">),</span> <span class="n">eps</span><span class="o">=</span><span class="mf">1e-16</span><span class="p">,</span>
<span class="lineno">25</span> <span class="n">weight_decay</span><span class="p">:</span> <span class="n">WeightDecay</span> <span class="o">=</span> <span class="n">WeightDecay</span><span class="p">(),</span>
<span class="lineno">26</span> <span class="n">optimized_update</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">,</span>
<span class="lineno">27</span> <span class="n">amsgrad</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">warmup</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">defaults</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-3'>
<div class='docs'>
<div class='section-link'>
<a href='#section-3'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">44</span> <span class="n">defaults</span> <span class="o">=</span> <span class="p">{}</span> <span class="k">if</span> <span class="n">defaults</span> <span class="ow">is</span> <span class="kc">None</span> <span class="k">else</span> <span class="n">defaults</span>
<span class="lineno">45</span> <span class="n">defaults</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="nb">dict</span><span class="p">(</span><span class="n">warmup</span><span class="o">=</span><span class="n">warmup</span><span class="p">))</span>
<span class="lineno">46</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">params</span><span class="p">,</span> <span class="n">lr</span><span class="p">,</span> <span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="p">,</span> <span class="n">weight_decay</span><span class="p">,</span> <span class="n">optimized_update</span><span class="p">,</span> <span class="n">amsgrad</span><span class="p">,</span> <span class="n">defaults</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-4'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-4'>#</a>
</div>
<h3>学習率を取得</h3>
<p><span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:2.40003em;vertical-align:-0.95003em;"></span><span class="mord coloredeq eqf" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mop">min</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord"><span class="delimsizing size3">(</span></span><span class="mord">1</span><span class="mpunct">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord"><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.29208em;"><span style="top:-2.314em;"><span class="pstrut" style="height:3em;"></span><span class="mord"><span class="mord coloredeq eqh" style=""><span class="mord mathnormal" style="margin-right:0.02691em">w</span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em;"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord"><span class="mord mathnormal">t</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.686em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mord"><span class="delimsizing size3">)</span></span></span></span></span></span></span><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqh" style=""><span class="mord mathnormal" style="margin-right:0.02691em">w</span></span></span></span></span></span>はウォームアップステップの数です。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">48</span> <span class="k">def</span> <span class="nf">get_lr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">state</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">group</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">]):</span></pre></div>
</div>
</div>
<div class='section' id='section-5'>
<div class='docs'>
<div class='section-link'>
<a href='#section-5'>#</a>
</div>
<p>ウォームアップ段階の場合</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">56</span> <span class="k">if</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;warmup&#39;</span><span class="p">]</span> <span class="o">&gt;</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;step&#39;</span><span class="p">]:</span></pre></div>
</div>
</div>
<div class='section' id='section-6'>
<div class='docs'>
<div class='section-link'>
<a href='#section-6'>#</a>
</div>
<p>学習率が 1 から <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord">0</span></span></span></span></span> 1 に直線的に増加している <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqf" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">58</span> <span class="k">return</span> <span class="mf">1e-8</span> <span class="o">+</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;step&#39;</span><span class="p">]</span> <span class="o">*</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;lr&#39;</span><span class="p">]</span> <span class="o">/</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;warmup&#39;</span><span class="p">]</span>
<span class="lineno">59</span> <span class="k">else</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-7'>
<div class='docs'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
<p>一定の学習率 <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqf" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">61</span> <span class="k">return</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;lr&#39;</span><span class="p">]</span></pre></div>
</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>
docs/ja/optimizers/adam_warmup_cosine_decay.html
+294
View File
@@ -0,0 +1,294 @@
<!DOCTYPE html>
<html lang="ja">
<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="GPT のウォームアップとコサインディケイを備えた Adam オプティマイザーの PyTorch 実装/チュートリアル。"/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="ウォームアップとコサインディケイを備えた Adam オプティマイザー"/>
<meta name="twitter:description" content="GPT のウォームアップとコサインディケイを備えた Adam オプティマイザーの PyTorch 実装/チュートリアル。"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/optimizers/adam_warmup_cosine_decay.html"/>
<meta property="og:title" content="ウォームアップとコサインディケイを備えた Adam オプティマイザー"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="ウォームアップとコサインディケイを備えた Adam オプティマイザー"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="ウォームアップとコサインディケイを備えた Adam オプティマイザー"/>
<meta property="og:description" content="GPT のウォームアップとコサインディケイを備えた Adam オプティマイザーの PyTorch 実装/チュートリアル。"/>
<title>ウォームアップとコサインディケイを備えた Adam オプティマイザー</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/optimizers/adam_warmup_cosine_decay.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">optimizers</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/optimizers/adam_warmup_cosine_decay.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1>ウォームアップとコサインディケイを備えた Adam オプティマイザー</h1>
<p>これにより <a href="adam.html">AMSgrad オプティマイザが拡張され</a>、ウォームアップステージが追加されます。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">11</span><span></span><span class="kn">import</span> <span class="nn">math</span>
<span class="lineno">12</span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Dict</span>
<span class="lineno">13</span>
<span class="lineno">14</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers</span> <span class="kn">import</span> <span class="n">WeightDecay</span>
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers.amsgrad</span> <span class="kn">import</span> <span class="n">AMSGrad</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
<p><a id="EmbeddingsWithPositionalEncoding"></a></p>
<h2>ウォームアップとコサインディケイを備えた Adam オプティマイザー</h2>
<p>このクラスは、で定義されている AMSGrad オプティマイザを拡張したものです。<a href="amsgrad.html"><code class="highlight"><span></span><span class="n">amsgrad</span><span class="o">.</span><span class="n">py</span></code>
</a></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">18</span><span class="k">class</span> <span class="nc">AdamWarmupCosineDecay</span><span class="p">(</span><span class="n">AMSGrad</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
<h3>オプティマイザを初期化</h3>
<ul><li><code class="highlight"><span></span><span class="n">params</span></code>
はパラメータのリストです</li>
<li><code class="highlight"><span></span><span class="n">lr</span></code>
は学習率 <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">betas</span></code>
(,) <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span> のタプルです <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">2</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">eps</span></code>
<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.69444em;vertical-align:0em;"></span><span class="mord accent"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.69444em;"><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord coloredeq eqd" style=""><span class="mord mathnormal" style="">ϵ</span></span></span><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="accent-body" style="left:-0.19444em;"><span class="mord">^</span></span></span></span></span></span></span></span></span></span></span><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqd" style=""><span class="mord mathnormal" style="">ϵ</span></span></span></span></span></span>またはそれに基づいている <code class="highlight"><span></span><span class="n">optimized_update</span></code>
</li>
<li><code class="highlight"><span></span><span class="n">weight_decay</span></code>
<code class="highlight"><span></span><span class="n">WeightDecay</span></code>
で定義されているクラスのインスタンスです <a href="index.html"><code class="highlight"><span></span><span class="fm">__init__</span><span class="o">.</span><span class="n">py</span></code>
</a></li>
<li>'optimized_update'は追加後に行うことでセカンドモーメントのバイアス補正を最適化するかどうかのフラグです <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqd" style=""><span class="mord mathnormal" style="">ϵ</span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">amsgrad</span></code>
amsGradを使用するか、プレーンなAdamにフォールバックするかを示すフラグです</li>
<li><code class="highlight"><span></span><span class="n">warmup</span></code>
ウォームアップステップ数</li>
<li><code class="highlight"><span></span><span class="n">total_steps</span></code>
ステップの総数。この時点でコサイン減衰は0に達しますが、<code class="highlight"><span></span><span class="n">lr</span></code>
取るため10％にとどまります <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.46528em;vertical-align:0em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span><span class="mspace" style="margin-right:0.2222222222222222em;"></span><span class="mbin">∗</span><span class="mspace" style="margin-right:0.2222222222222222em;"></span></span><span class="base"><span class="strut" style="height:1em;vertical-align:-0.25em;"></span><span class="mop">max</span><span class="mopen">(</span><span class="mord coloredeq eqh" style=""><span class="mord" style="">0</span></span><span class="mord">.1</span><span class="mpunct">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord mathnormal">d</span><span class="mord mathnormal">ec</span><span class="mord mathnormal">a</span><span class="mord mathnormal" style="margin-right:0.03588em;">y</span><span class="mclose">)</span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">defaults</span></code>
グループ値のデフォルト辞書です。これは、クラスを拡張する場合に便利です<code class="highlight"><span></span><span class="n">AdamWarmup</span></code>
。</li></ul>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">27</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">params</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-3</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="p">(</span><span class="mf">0.9</span><span class="p">,</span> <span class="mf">0.999</span><span class="p">),</span> <span class="n">eps</span><span class="o">=</span><span class="mf">1e-16</span><span class="p">,</span>
<span class="lineno">28</span> <span class="n">weight_decay</span><span class="p">:</span> <span class="n">WeightDecay</span> <span class="o">=</span> <span class="n">WeightDecay</span><span class="p">(),</span>
<span class="lineno">29</span> <span class="n">optimized_update</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">,</span>
<span class="lineno">30</span> <span class="n">amsgrad</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span> <span class="n">warmup</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">total_steps</span><span class="o">=</span><span class="mf">1e10</span><span class="p">,</span> <span class="n">defaults</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-3'>
<div class='docs'>
<div class='section-link'>
<a href='#section-3'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">49</span> <span class="n">defaults</span> <span class="o">=</span> <span class="p">{}</span> <span class="k">if</span> <span class="n">defaults</span> <span class="ow">is</span> <span class="kc">None</span> <span class="k">else</span> <span class="n">defaults</span>
<span class="lineno">50</span> <span class="n">defaults</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="nb">dict</span><span class="p">(</span><span class="n">warmup</span><span class="o">=</span><span class="n">warmup</span><span class="p">,</span> <span class="n">total_steps</span><span class="o">=</span><span class="n">total_steps</span><span class="p">))</span>
<span class="lineno">51</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">params</span><span class="p">,</span> <span class="n">lr</span><span class="p">,</span> <span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="p">,</span> <span class="n">weight_decay</span><span class="p">,</span> <span class="n">optimized_update</span><span class="p">,</span> <span class="n">amsgrad</span><span class="p">,</span> <span class="n">defaults</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-4'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-4'>#</a>
</div>
<h3>学習率を取得</h3>
<p><span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:2.40003em;vertical-align:-0.95003em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mop">min</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord"><span class="delimsizing size3">(</span></span><span class="mord">1</span><span class="mpunct">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord"><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.29208em;"><span style="top:-2.314em;"><span class="pstrut" style="height:3em;"></span><span class="mord"><span class="mord coloredeq eqi" style=""><span class="mord mathnormal" style="margin-right:0.02691em">w</span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em;"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord"><span class="mord mathnormal">t</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.686em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mord"><span class="delimsizing size3">)</span></span></span></span></span></span></span><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqi" style=""><span class="mord mathnormal" style="margin-right:0.02691em">w</span></span></span></span></span></span>はウォームアップステップの数です。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">53</span> <span class="k">def</span> <span class="nf">get_lr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">state</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">group</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">]):</span></pre></div>
</div>
</div>
<div class='section' id='section-5'>
<div class='docs'>
<div class='section-link'>
<a href='#section-5'>#</a>
</div>
<p>ウォームアップ段階の場合</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">61</span> <span class="k">if</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;warmup&#39;</span><span class="p">]</span> <span class="o">&gt;</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;step&#39;</span><span class="p">]:</span></pre></div>
</div>
</div>
<div class='section' id='section-6'>
<div class='docs'>
<div class='section-link'>
<a href='#section-6'>#</a>
</div>
<p>学習率が 1 から <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.64444em;vertical-align:0em;"></span><span class="mord coloredeq eqh" style=""><span class="mord" style="">0</span></span></span></span></span></span> 1 に直線的に増加している <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">63</span> <span class="k">return</span> <span class="mf">1e-8</span> <span class="o">+</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;step&#39;</span><span class="p">]</span> <span class="o">*</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;lr&#39;</span><span class="p">]</span> <span class="o">/</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;warmup&#39;</span><span class="p">]</span>
<span class="lineno">64</span> <span class="k">else</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-7'>
<div class='docs'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
<p>一定の学習率 <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">66</span> <span class="n">progress</span> <span class="o">=</span> <span class="p">(</span><span class="n">state</span><span class="p">[</span><span class="s1">&#39;step&#39;</span><span class="p">]</span> <span class="o">-</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;warmup&#39;</span><span class="p">])</span> <span class="o">/</span> <span class="nb">max</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;total_steps&#39;</span><span class="p">]</span> <span class="o">-</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;warmup&#39;</span><span class="p">])</span>
<span class="lineno">67</span> <span class="k">return</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;lr&#39;</span><span class="p">]</span> <span class="o">*</span> <span class="nb">max</span><span class="p">(</span><span class="mf">0.1</span><span class="p">,</span> <span class="mf">0.5</span> <span class="o">*</span> <span class="p">(</span><span class="mf">1.0</span> <span class="o">+</span> <span class="n">math</span><span class="o">.</span><span class="n">cos</span><span class="p">(</span><span class="n">math</span><span class="o">.</span><span class="n">pi</span> <span class="o">*</span> <span class="n">progress</span><span class="p">)))</span></pre></div>
</div>
</div>
<div class='section' id='section-8'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-8'>#</a>
</div>
<h3>さまざまなウォームアップとモデルサイズの学習率をプロット</h3>
<p><img alt="Plot of learning rate" src="noam_lr.png"></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">70</span><span class="k">def</span> <span class="nf">_test_lr</span><span class="p">():</span></pre></div>
</div>
</div>
<div class='section' id='section-9'>
<div class='docs'>
<div class='section-link'>
<a href='#section-9'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">76</span> <span class="kn">import</span> <span class="nn">matplotlib.pyplot</span> <span class="k">as</span> <span class="nn">plt</span>
<span class="lineno">77</span> <span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="lineno">78</span> <span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
<span class="lineno">79</span>
<span class="lineno">80</span> <span class="n">model</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="mi">10</span><span class="p">)</span>
<span class="lineno">81</span> <span class="n">opt</span> <span class="o">=</span> <span class="n">AdamWarmupCosineDecay</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">warmup</span><span class="o">=</span><span class="mi">5000</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-4</span><span class="p">,</span> <span class="n">total_steps</span><span class="o">=</span><span class="mf">4e6</span><span class="p">)</span>
<span class="lineno">82</span> <span class="n">steps</span> <span class="o">=</span> <span class="mi">20_000</span>
<span class="lineno">83</span> <span class="n">plt</span><span class="o">.</span><span class="n">plot</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">steps</span><span class="p">),</span> <span class="p">[</span><span class="n">opt</span><span class="o">.</span><span class="n">get_lr</span><span class="p">({</span><span class="s1">&#39;step&#39;</span><span class="p">:</span> <span class="n">i</span><span class="p">},</span> <span class="n">opt</span><span class="o">.</span><span class="n">defaults</span><span class="p">)</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">steps</span><span class="p">)])</span>
<span class="lineno">84</span> <span class="n">plt</span><span class="o">.</span><span class="n">legend</span><span class="p">([</span><span class="s2">&quot;5000:4e6&quot;</span><span class="p">,</span> <span class="s2">&quot;5000:2e6&quot;</span><span class="p">,</span> <span class="s2">&quot;5000:1e6&quot;</span><span class="p">])</span>
<span class="lineno">85</span> <span class="n">plt</span><span class="o">.</span><span class="n">title</span><span class="p">(</span><span class="s2">&quot;Learning Rate&quot;</span><span class="p">)</span>
<span class="lineno">86</span> <span class="n">plt</span><span class="o">.</span><span class="n">show</span><span class="p">()</span>
<span class="lineno">87</span>
<span class="lineno">88</span> <span class="n">steps</span> <span class="o">=</span> <span class="nb">int</span><span class="p">(</span><span class="mf">6e6</span><span class="p">)</span>
<span class="lineno">89</span> <span class="n">step_size</span> <span class="o">=</span> <span class="mi">1000</span>
<span class="lineno">90</span> <span class="n">plt</span><span class="o">.</span><span class="n">plot</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">steps</span><span class="p">,</span> <span class="n">step_size</span><span class="p">),</span> <span class="p">[</span><span class="n">opt</span><span class="o">.</span><span class="n">get_lr</span><span class="p">({</span><span class="s1">&#39;step&#39;</span><span class="p">:</span> <span class="n">i</span><span class="p">},</span> <span class="n">opt</span><span class="o">.</span><span class="n">defaults</span><span class="p">)</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">steps</span><span class="p">,</span> <span class="n">step_size</span><span class="p">)])</span>
<span class="lineno">91</span> <span class="n">plt</span><span class="o">.</span><span class="n">legend</span><span class="p">([</span><span class="s2">&quot;5000:4e6&quot;</span><span class="p">,</span> <span class="s2">&quot;5000:2e6&quot;</span><span class="p">,</span> <span class="s2">&quot;5000:1e6&quot;</span><span class="p">])</span>
<span class="lineno">92</span> <span class="n">plt</span><span class="o">.</span><span class="n">title</span><span class="p">(</span><span class="s2">&quot;Learning Rate&quot;</span><span class="p">)</span>
<span class="lineno">93</span> <span class="n">plt</span><span class="o">.</span><span class="n">show</span><span class="p">()</span>
<span class="lineno">94</span>
<span class="lineno">95</span>
<span class="lineno">96</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">&#39;__main__&#39;</span><span class="p">:</span>
<span class="lineno">97</span> <span class="n">_test_lr</span><span class="p">()</span></pre></div>
</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>
docs/ja/optimizers/amsgrad.html
+563
View File
File diff suppressed because one or more lines are too long
docs/ja/optimizers/configs.html
+443
View File
@@ -0,0 +1,443 @@
<!DOCTYPE html>
<html lang="ja">
<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="これにより、オプティマイザー用の設定可能なモジュールが実装されます。"/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="設定可能なオプティマイザモジュール"/>
<meta name="twitter:description" content="これにより、オプティマイザー用の設定可能なモジュールが実装されます。"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/optimizers/configs.html"/>
<meta property="og:title" content="設定可能なオプティマイザモジュール"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="設定可能なオプティマイザモジュール"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="設定可能なオプティマイザモジュール"/>
<meta property="og:description" content="これにより、オプティマイザー用の設定可能なモジュールが実装されます。"/>
<title>設定可能なオプティマイザモジュール</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/optimizers/configs.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">optimizers</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/optimizers/configs.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1>設定可能なオプティマイザー</h1>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">10</span><span></span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Tuple</span>
<span class="lineno">11</span>
<span class="lineno">12</span><span class="kn">import</span> <span class="nn">torch</span>
<span class="lineno">13</span>
<span class="lineno">14</span><span class="kn">from</span> <span class="nn">labml.configs</span> <span class="kn">import</span> <span class="n">BaseConfigs</span><span class="p">,</span> <span class="n">option</span><span class="p">,</span> <span class="n">meta_config</span>
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers</span> <span class="kn">import</span> <span class="n">WeightDecay</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
<p><a id="OptimizerConfigs"></a></p>
<h2>オプティマイザー構成</h2>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">18</span><span class="k">class</span> <span class="nc">OptimizerConfigs</span><span class="p">(</span><span class="n">BaseConfigs</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
<p>オプティマイザー</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">26</span> <span class="n">optimizer</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">optim</span><span class="o">.</span><span class="n">Adam</span></pre></div>
</div>
</div>
<div class='section' id='section-3'>
<div class='docs'>
<div class='section-link'>
<a href='#section-3'>#</a>
</div>
<p>体重減少</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">29</span> <span class="n">weight_decay_obj</span><span class="p">:</span> <span class="n">WeightDecay</span></pre></div>
</div>
</div>
<div class='section' id='section-4'>
<div class='docs'>
<div class='section-link'>
<a href='#section-4'>#</a>
</div>
<p>重み減衰が切り離されているかどうか、つまり重み減衰が勾配に加えられないかどうか</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">32</span> <span class="n">weight_decouple</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span></pre></div>
</div>
</div>
<div class='section' id='section-5'>
<div class='docs'>
<div class='section-link'>
<a href='#section-5'>#</a>
</div>
<p>体重減少</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">34</span> <span class="n">weight_decay</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">0.0</span></pre></div>
</div>
</div>
<div class='section' id='section-6'>
<div class='docs'>
<div class='section-link'>
<a href='#section-6'>#</a>
</div>
<p>減量が絶対体重なのか、それとも学習率で掛けるべきなのか</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">36</span> <span class="n">weight_decay_absolute</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span></pre></div>
</div>
</div>
<div class='section' id='section-7'>
<div class='docs'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
<p>adamアップデートが最適化されているかどうか (イプシロンが異なる)</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">39</span> <span class="n">optimized_adam_update</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span></pre></div>
</div>
</div>
<div class='section' id='section-8'>
<div class='docs'>
<div class='section-link'>
<a href='#section-8'>#</a>
</div>
<p>最適化するパラメーター</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">42</span> <span class="n">parameters</span><span class="p">:</span> <span class="nb">any</span></pre></div>
</div>
</div>
<div class='section' id='section-9'>
<div class='docs'>
<div class='section-link'>
<a href='#section-9'>#</a>
</div>
<p>学習率 <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord mathnormal" style="margin-right:0.0037em;">α</span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">45</span> <span class="n">learning_rate</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">0.01</span></pre></div>
</div>
</div>
<div class='section' id='section-10'>
<div class='docs'>
<div class='section-link'>
<a href='#section-10'>#</a>
</div>
<p><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1em;vertical-align:-0.25em;"></span><span class="mopen">(</span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span><span class="mpunct">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">2</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span><span class="mclose">)</span></span></span></span></span>アダムのベータ値</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">47</span> <span class="n">betas</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">float</span><span class="p">,</span> <span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="p">(</span><span class="mf">0.9</span><span class="p">,</span> <span class="mf">0.999</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-11'>
<div class='docs'>
<div class='section-link'>
<a href='#section-11'>#</a>
</div>
<p>イプシロン・フォー・アダム <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord mathnormal">ϵ</span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">49</span> <span class="n">eps</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">1e-08</span></pre></div>
</div>
</div>
<div class='section' id='section-12'>
<div class='docs'>
<div class='section-link'>
<a href='#section-12'>#</a>
</div>
<p>シンガポールドルのモメンタム</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">52</span> <span class="n">momentum</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">0.5</span></pre></div>
</div>
</div>
<div class='section' id='section-13'>
<div class='docs'>
<div class='section-link'>
<a href='#section-13'>#</a>
</div>
<p>アムスグラードを使用するかどうか</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">54</span> <span class="n">amsgrad</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span></pre></div>
</div>
</div>
<div class='section' id='section-14'>
<div class='docs'>
<div class='section-link'>
<a href='#section-14'>#</a>
</div>
<p>ウォームアップオプティマイザーのステップ数</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">57</span> <span class="n">warmup</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">2_000</span></pre></div>
</div>
</div>
<div class='section' id='section-15'>
<div class='docs'>
<div class='section-link'>
<a href='#section-15'>#</a>
</div>
<p>オプティマイザステップの総数 (コサイン減衰用)</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">59</span> <span class="n">total_steps</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="nb">int</span><span class="p">(</span><span class="mf">1e10</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-16'>
<div class='docs'>
<div class='section-link'>
<a href='#section-16'>#</a>
</div>
<p>AdabElief で SGD に縮退させるかどうか</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">62</span> <span class="n">degenerate_to_sgd</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span></pre></div>
</div>
</div>
<div class='section' id='section-17'>
<div class='docs'>
<div class='section-link'>
<a href='#section-17'>#</a>
</div>
<p>AdabElief でレクティファイドアダムを使用するかどうか</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">65</span> <span class="n">rectify</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span></pre></div>
</div>
</div>
<div class='section' id='section-18'>
<div class='docs'>
<div class='section-link'>
<a href='#section-18'>#</a>
</div>
<p>Noam オプティマイザーのモデル埋め込みサイズ</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">68</span> <span class="n">d_model</span><span class="p">:</span> <span class="nb">int</span></pre></div>
</div>
</div>
<div class='section' id='section-19'>
<div class='docs'>
<div class='section-link'>
<a href='#section-19'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">70</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="lineno">71</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">_primary</span><span class="o">=</span><span class="s1">&#39;optimizer&#39;</span><span class="p">)</span>
<span class="lineno">72</span>
<span class="lineno">73</span>
<span class="lineno">74</span><span class="n">meta_config</span><span class="p">(</span><span class="n">OptimizerConfigs</span><span class="o">.</span><span class="n">parameters</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-20'>
<div class='docs'>
<div class='section-link'>
<a href='#section-20'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">77</span><span class="nd">@option</span><span class="p">(</span><span class="n">OptimizerConfigs</span><span class="o">.</span><span class="n">weight_decay_obj</span><span class="p">,</span> <span class="s1">&#39;L2&#39;</span><span class="p">)</span>
<span class="lineno">78</span><span class="k">def</span> <span class="nf">_weight_decay</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">OptimizerConfigs</span><span class="p">):</span>
<span class="lineno">79</span> <span class="k">return</span> <span class="n">WeightDecay</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">weight_decay</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">weight_decouple</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">weight_decay_absolute</span><span class="p">)</span>
<span class="lineno">80</span>
<span class="lineno">81</span>
<span class="lineno">82</span><span class="nd">@option</span><span class="p">(</span><span class="n">OptimizerConfigs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="s1">&#39;SGD&#39;</span><span class="p">)</span>
<span class="lineno">83</span><span class="k">def</span> <span class="nf">_sgd_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">OptimizerConfigs</span><span class="p">):</span>
<span class="lineno">84</span> <span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">optim</span><span class="o">.</span><span class="n">SGD</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">parameters</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">learning_rate</span><span class="p">,</span> <span class="n">c</span><span class="o">.</span><span class="n">momentum</span><span class="p">,</span>
<span class="lineno">85</span> <span class="n">weight_decay</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">weight_decay</span><span class="p">)</span>
<span class="lineno">86</span>
<span class="lineno">87</span>
<span class="lineno">88</span><span class="nd">@option</span><span class="p">(</span><span class="n">OptimizerConfigs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="s1">&#39;Adam&#39;</span><span class="p">)</span>
<span class="lineno">89</span><span class="k">def</span> <span class="nf">_adam_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">OptimizerConfigs</span><span class="p">):</span>
<span class="lineno">90</span> <span class="k">if</span> <span class="n">c</span><span class="o">.</span><span class="n">amsgrad</span><span class="p">:</span>
<span class="lineno">91</span> <span class="kn">from</span> <span class="nn">labml_nn.optimizers.amsgrad</span> <span class="kn">import</span> <span class="n">AMSGrad</span>
<span class="lineno">92</span> <span class="k">return</span> <span class="n">AMSGrad</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">parameters</span><span class="p">,</span>
<span class="lineno">93</span> <span class="n">lr</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">learning_rate</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">eps</span><span class="p">,</span>
<span class="lineno">94</span> <span class="n">optimized_update</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">optimized_adam_update</span><span class="p">,</span>
<span class="lineno">95</span> <span class="n">weight_decay</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">weight_decay_obj</span><span class="p">,</span> <span class="n">amsgrad</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">amsgrad</span><span class="p">)</span>
<span class="lineno">96</span> <span class="k">else</span><span class="p">:</span>
<span class="lineno">97</span> <span class="kn">from</span> <span class="nn">labml_nn.optimizers.adam</span> <span class="kn">import</span> <span class="n">Adam</span>
<span class="lineno">98</span> <span class="k">return</span> <span class="n">Adam</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">parameters</span><span class="p">,</span>
<span class="lineno">99</span> <span class="n">lr</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">learning_rate</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">eps</span><span class="p">,</span>
<span class="lineno">100</span> <span class="n">optimized_update</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">optimized_adam_update</span><span class="p">,</span>
<span class="lineno">101</span> <span class="n">weight_decay</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">weight_decay_obj</span><span class="p">)</span>
<span class="lineno">102</span>
<span class="lineno">103</span>
<span class="lineno">104</span><span class="nd">@option</span><span class="p">(</span><span class="n">OptimizerConfigs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="s1">&#39;AdamW&#39;</span><span class="p">)</span>
<span class="lineno">105</span><span class="k">def</span> <span class="nf">_adam_warmup_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">OptimizerConfigs</span><span class="p">):</span>
<span class="lineno">106</span> <span class="kn">from</span> <span class="nn">labml_nn.optimizers.adam_warmup</span> <span class="kn">import</span> <span class="n">AdamWarmup</span>
<span class="lineno">107</span> <span class="k">return</span> <span class="n">AdamWarmup</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">parameters</span><span class="p">,</span>
<span class="lineno">108</span> <span class="n">lr</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">learning_rate</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">eps</span><span class="p">,</span>
<span class="lineno">109</span> <span class="n">weight_decay</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">weight_decay_obj</span><span class="p">,</span> <span class="n">amsgrad</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">amsgrad</span><span class="p">,</span> <span class="n">warmup</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">warmup</span><span class="p">)</span>
<span class="lineno">110</span>
<span class="lineno">111</span>
<span class="lineno">112</span><span class="nd">@option</span><span class="p">(</span><span class="n">OptimizerConfigs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="s1">&#39;RAdam&#39;</span><span class="p">)</span>
<span class="lineno">113</span><span class="k">def</span> <span class="nf">_radam_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">OptimizerConfigs</span><span class="p">):</span>
<span class="lineno">114</span> <span class="kn">from</span> <span class="nn">labml_nn.optimizers.radam</span> <span class="kn">import</span> <span class="n">RAdam</span>
<span class="lineno">115</span> <span class="k">return</span> <span class="n">RAdam</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">parameters</span><span class="p">,</span>
<span class="lineno">116</span> <span class="n">lr</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">learning_rate</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">eps</span><span class="p">,</span>
<span class="lineno">117</span> <span class="n">weight_decay</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">weight_decay_obj</span><span class="p">,</span> <span class="n">amsgrad</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">amsgrad</span><span class="p">,</span>
<span class="lineno">118</span> <span class="n">degenerated_to_sgd</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">degenerate_to_sgd</span><span class="p">)</span>
<span class="lineno">119</span>
<span class="lineno">120</span>
<span class="lineno">121</span><span class="nd">@option</span><span class="p">(</span><span class="n">OptimizerConfigs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="s1">&#39;AdaBelief&#39;</span><span class="p">)</span>
<span class="lineno">122</span><span class="k">def</span> <span class="nf">_ada_belief_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">OptimizerConfigs</span><span class="p">):</span>
<span class="lineno">123</span> <span class="kn">from</span> <span class="nn">labml_nn.optimizers.ada_belief</span> <span class="kn">import</span> <span class="n">AdaBelief</span>
<span class="lineno">124</span> <span class="k">return</span> <span class="n">AdaBelief</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">parameters</span><span class="p">,</span>
<span class="lineno">125</span> <span class="n">lr</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">learning_rate</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">eps</span><span class="p">,</span>
<span class="lineno">126</span> <span class="n">weight_decay</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">weight_decay_obj</span><span class="p">,</span> <span class="n">amsgrad</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">amsgrad</span><span class="p">,</span>
<span class="lineno">127</span> <span class="n">degenerate_to_sgd</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">degenerate_to_sgd</span><span class="p">,</span>
<span class="lineno">128</span> <span class="n">rectify</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">rectify</span><span class="p">)</span>
<span class="lineno">129</span>
<span class="lineno">130</span>
<span class="lineno">131</span><span class="nd">@option</span><span class="p">(</span><span class="n">OptimizerConfigs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="s1">&#39;Noam&#39;</span><span class="p">)</span>
<span class="lineno">132</span><span class="k">def</span> <span class="nf">_noam_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">OptimizerConfigs</span><span class="p">):</span>
<span class="lineno">133</span> <span class="kn">from</span> <span class="nn">labml_nn.optimizers.noam</span> <span class="kn">import</span> <span class="n">Noam</span>
<span class="lineno">134</span> <span class="k">return</span> <span class="n">Noam</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">parameters</span><span class="p">,</span>
<span class="lineno">135</span> <span class="n">lr</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">learning_rate</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">eps</span><span class="p">,</span>
<span class="lineno">136</span> <span class="n">weight_decay</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">weight_decay_obj</span><span class="p">,</span> <span class="n">amsgrad</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">amsgrad</span><span class="p">,</span> <span class="n">warmup</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">warmup</span><span class="p">,</span>
<span class="lineno">137</span> <span class="n">d_model</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">d_model</span><span class="p">)</span>
<span class="lineno">138</span>
<span class="lineno">139</span>
<span class="lineno">140</span><span class="nd">@option</span><span class="p">(</span><span class="n">OptimizerConfigs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="s1">&#39;AdamWarmupCosineDecay&#39;</span><span class="p">)</span>
<span class="lineno">141</span><span class="k">def</span> <span class="nf">_noam_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">OptimizerConfigs</span><span class="p">):</span>
<span class="lineno">142</span> <span class="kn">from</span> <span class="nn">labml_nn.optimizers.adam_warmup_cosine_decay</span> <span class="kn">import</span> <span class="n">AdamWarmupCosineDecay</span>
<span class="lineno">143</span> <span class="k">return</span> <span class="n">AdamWarmupCosineDecay</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">parameters</span><span class="p">,</span>
<span class="lineno">144</span> <span class="n">lr</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">learning_rate</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">eps</span><span class="p">,</span>
<span class="lineno">145</span> <span class="n">weight_decay</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">weight_decay_obj</span><span class="p">,</span> <span class="n">amsgrad</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">amsgrad</span><span class="p">,</span>
<span class="lineno">146</span> <span class="n">warmup</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">warmup</span><span class="p">,</span> <span class="n">total_steps</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">total_steps</span><span class="p">)</span></pre></div>
</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>
docs/ja/optimizers/index.html
+583
View File
@@ -0,0 +1,583 @@
<!DOCTYPE html>
<html lang="ja">
<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="人気のある勾配降下ベースのオプティマイザーのPyTorch実装/チュートリアルのセット。現在、Adam、Masgrad、および Adamのオプティマイザーが含まれています。"/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="オプティマイザー"/>
<meta name="twitter:description" content="人気のある勾配降下ベースのオプティマイザーのPyTorch実装/チュートリアルのセット。現在、Adam、Masgrad、および Adamのオプティマイザーが含まれています。"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/optimizers/index.html"/>
<meta property="og:title" content="オプティマイザー"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="オプティマイザー"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="オプティマイザー"/>
<meta property="og:description" content="人気のある勾配降下ベースのオプティマイザーのPyTorch実装/チュートリアルのセット。現在、Adam、Masgrad、および Adamのオプティマイザーが含まれています。"/>
<title>オプティマイザー</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/optimizers/index.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">optimizers</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/optimizers/__init__.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1>オプティマイザー</h1>
<h2>オプティマイザーの実装</h2>
<ul><li><a href="adam.html">アダム・オプティマイザー</a></li>
<li><a href="amsgrad.html">マスグラードオプティマイザー</a></li>
<li><a href="adam_warmup.html">ウォームアップ機能付き Adam オプティマイザー</a></li>
<li><a href="noam.html">ノームオプティマイザー</a></li>
<li><a href="radam.html">修正されたアダムオプティマイザー</a></li>
<li><a href="ada_belief.html">アダブリリーフオプティマイザー</a></li></ul>
<p>この <a href="mnist_experiment.html">MNIST の例では</a>、これらのオプティマイザーを使用しています。</p>
<h2>汎用アダプティブオプティマイザー基本クラスとウェイトディケイ</h2>
<p>このファイルは、<em>Adam</em> の共通基本クラスとその拡張を定義しています。基本クラスは、再利用が可能なため、最小限のコードで他のオプティマイザを実装するのに役立ちます</p>。
<p>また、L2の重み減衰用の特別なクラスを定義しているので、各オプティマイザー内に実装する必要がなく、オプティマイザーを変更せずにL1のような他の重み減衰にも簡単に拡張できます。</p>
<p>PyTorch オプティマイザの概念は次のとおりです。</p>
<h3>パラメータグループ</h3>
<p>PyTorch オプティマイザーは、パラメーターをグループと呼ばれるセットにグループ化します。各グループには、学習率などの独自のハイパーパラメータを設定できます</p>。
<p>たいていの場合、グループが 1 つしかありません。このとき、オプティマイザを次のように初期化します</p>。
<pre class="highlight lang-python"><code><span></span><span class="n">Optimizer</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">())</span></code></pre>
<p>オプティマイザを初期化するときに、複数のパラメータグループを定義できます。</p>
<pre class="highlight lang-python"><code><span></span><span class="n">Optimizer</span><span class="p">([{</span><span class="s1">&#39;params&#39;</span><span class="p">:</span> <span class="n">model1</span><span class="o">.</span><span class="n">parameters</span><span class="p">()},</span> <span class="p">{</span><span class="s1">&#39;params&#39;</span><span class="p">:</span> <span class="n">model2</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="s1">&#39;lr&#39;</span><span class="p">:</span> <span class="mi">2</span><span class="p">}])</span></code></pre>
<p>ここにグループのリストを渡します。各グループは辞書で、パラメータは 'params' です。任意のハイパーパラメータも指定します。ハイパーパラメータが定義されていない場合は、デフォルトでオプティマイザレベルのデフォルトになります</p>。
<p>を使用してこれらのグループとそのハイパーパラメータにアクセスしたり、変更したりすることができます。<code class="highlight"><span></span><span class="n">optimizer</span><span class="o">.</span><span class="n">param_groups</span></code>
私が出会ったほとんどの学習率スケジュールの実装は、これにアクセスして「lr」を変更します</p>。
<h3>州</h3>
<p>オプティマイザーは、各パラメーター (テンソル) の状態 (辞書) を辞書に保持します。<code class="highlight"><span></span><span class="n">optimizer</span><span class="o">.</span><span class="n">state</span></code>
ここで、オプティマイザーは指数平均などを管理します</p>。
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">62</span><span></span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Dict</span><span class="p">,</span> <span class="n">Tuple</span><span class="p">,</span> <span class="n">Any</span>
<span class="lineno">63</span>
<span class="lineno">64</span><span class="kn">import</span> <span class="nn">torch</span>
<span class="lineno">65</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
<span class="lineno">66</span><span class="kn">from</span> <span class="nn">torch.optim.optimizer</span> <span class="kn">import</span> <span class="n">Optimizer</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
<h2><em>Adam</em> と拡張機能の基底クラス</h2>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">69</span><span class="k">class</span> <span class="nc">GenericAdaptiveOptimizer</span><span class="p">(</span><span class="n">Optimizer</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
<h3>[初期化]</h3>
<ul><li><code class="highlight"><span></span><span class="n">params</span></code>
パラメータのコレクションまたはパラメータグループのセットです。</li>
<li><code class="highlight"><span></span><span class="n">defaults</span></code>
デフォルトのハイパーパラメータの辞書</li>
<li><code class="highlight"><span></span><span class="n">lr</span></code>
は学習率 <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord mathnormal" style="margin-right:0.0037em;">α</span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">betas</span></code>
はタプルです <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1em;vertical-align:-0.25em;"></span><span class="mopen">(</span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span><span class="mpunct">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">2</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span><span class="mclose">)</span></span></span></span></span></li>
</ul><li><code class="highlight"><span></span><span class="n">eps</span></code>
は <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord mathnormal">ϵ</span></span></span></span></span></li>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">74</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">params</span><span class="p">,</span> <span class="n">defaults</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">Any</span><span class="p">],</span> <span class="n">lr</span><span class="p">:</span> <span class="nb">float</span><span class="p">,</span> <span class="n">betas</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">float</span><span class="p">,</span> <span class="nb">float</span><span class="p">],</span> <span class="n">eps</span><span class="p">:</span> <span class="nb">float</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-3'>
<div class='docs'>
<div class='section-link'>
<a href='#section-3'>#</a>
</div>
<p>ハイパーパラメータを確認</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">86</span> <span class="k">if</span> <span class="ow">not</span> <span class="mf">0.0</span> <span class="o">&lt;=</span> <span class="n">lr</span><span class="p">:</span>
<span class="lineno">87</span> <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Invalid learning rate: </span><span class="si">{</span><span class="n">lr</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="lineno">88</span> <span class="k">if</span> <span class="ow">not</span> <span class="mf">0.0</span> <span class="o">&lt;=</span> <span class="n">eps</span><span class="p">:</span>
<span class="lineno">89</span> <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Invalid epsilon value: </span><span class="si">{</span><span class="n">eps</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="lineno">90</span> <span class="k">if</span> <span class="ow">not</span> <span class="mf">0.0</span> <span class="o">&lt;=</span> <span class="n">betas</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">&lt;</span> <span class="mf">1.0</span><span class="p">:</span>
<span class="lineno">91</span> <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Invalid beta parameter at index 0: </span><span class="si">{</span><span class="n">betas</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="lineno">92</span> <span class="k">if</span> <span class="ow">not</span> <span class="mf">0.0</span> <span class="o">&lt;=</span> <span class="n">betas</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">&lt;</span> <span class="mf">1.0</span><span class="p">:</span>
<span class="lineno">93</span> <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Invalid beta parameter at index 1: </span><span class="si">{</span><span class="n">betas</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-4'>
<div class='docs'>
<div class='section-link'>
<a href='#section-4'>#</a>
</div>
<p>ハイパーパラメータをデフォルトに追加</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">96</span> <span class="n">defaults</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="nb">dict</span><span class="p">(</span><span class="n">lr</span><span class="o">=</span><span class="n">lr</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="o">=</span><span class="n">eps</span><span class="p">))</span></pre></div>
</div>
</div>
<div class='section' id='section-5'>
<div class='docs'>
<div class='section-link'>
<a href='#section-5'>#</a>
</div>
<p>PyTorch オプティマイザーを初期化します。これにより、デフォルトのハイパーパラメータを使用してパラメータグループが作成されます</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">99</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">params</span><span class="p">,</span> <span class="n">defaults</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-6'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-6'>#</a>
</div>
<h3>与えられたパラメータテンソルの状態を初期化</h3>
<p><code class="highlight"><span></span><span class="n">state</span></code>
これをオーバーライドしてパラメータを初期化するコードを使うべきです。<code class="highlight"><span></span><span class="n">param</span></code>
<code class="highlight"><span></span><span class="n">group</span></code>
<code class="highlight"><span></span><span class="n">param</span></code>
が属するパラメータグループディクショナリです。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">101</span> <span class="k">def</span> <span class="nf">init_state</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">state</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">group</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">param</span><span class="p">:</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-7'>
<div class='docs'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">108</span> <span class="k">pass</span></pre></div>
</div>
</div>
<div class='section' id='section-8'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-8'>#</a>
</div>
<h3>パラメーターテンソルでオプティマイザーステップを実行する</h3>
<p>これをオーバーライドして、<code class="highlight"><span></span><span class="n">param</span></code>
テンソルで最適化ステップを実行する必要があります。ここで<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.69444em;vertical-align:0em;"></span><span class="mord mathnormal" style="margin-right:0.02778em;">θ</span></span></span></span></span><code class="highlight"><span></span><span class="n">grad</span></code>
、はそのパラメーターの勾配、はそのパラメーターのオプティマイザー状態ディクショナリ、<code class="highlight"><span></span><span class="n">state</span></code>
<code class="highlight"><span></span><span class="n">group</span></code>
はディクショナリが属するパラメーターグループです。<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.625em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.03588em;">g</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.2805559999999999em;"><span style="top:-2.5500000000000003em;margin-left:-0.03588em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mathnormal mtight">t</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span> <code class="highlight"><span></span><span class="n">param</span></code>
</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">110</span> <span class="k">def</span> <span class="nf">step_param</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">state</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">group</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">grad</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">param</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-9'>
<div class='docs'>
<div class='section-link'>
<a href='#section-9'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">119</span> <span class="k">pass</span></pre></div>
</div>
</div>
<div class='section' id='section-10'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-10'>#</a>
</div>
<h3>オプティマイザーステップ</h3>
<p><em>すべてのAdamベースのオプティマイザーが必要とする一般的な処理を行うテンプレートメソッドを作成しました</em>。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">121</span> <span class="nd">@torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">()</span>
<span class="lineno">122</span> <span class="k">def</span> <span class="nf">step</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">closure</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-11'>
<div class='docs'>
<div class='section-link'>
<a href='#section-11'>#</a>
</div>
<p>損失を計算します。</p>
<p>🤔 いつこれが必要なのかわかりません。自分で呼び出すのではなく、<code class="highlight"><span></span><span class="n">loss</span><span class="o">.</span><span class="n">backward</span></code>
損失を計算して損失を出して返す関数を定義すれば、その関数を渡せると思います<code class="highlight"><span></span><span class="n">optimizer</span><span class="o">.</span><span class="n">step</span></code>
。🤷‍♂️</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">133</span> <span class="n">loss</span> <span class="o">=</span> <span class="kc">None</span>
<span class="lineno">134</span> <span class="k">if</span> <span class="n">closure</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="lineno">135</span> <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">enable_grad</span><span class="p">():</span>
<span class="lineno">136</span> <span class="n">loss</span> <span class="o">=</span> <span class="n">closure</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-12'>
<div class='docs'>
<div class='section-link'>
<a href='#section-12'>#</a>
</div>
<p>パラメータグループを繰り返し処理する</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">139</span> <span class="k">for</span> <span class="n">group</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">param_groups</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-13'>
<div class='docs'>
<div class='section-link'>
<a href='#section-13'>#</a>
</div>
<p>パラメータグループ内のパラメータを繰り返し処理します</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">141</span> <span class="k">for</span> <span class="n">param</span> <span class="ow">in</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;params&#39;</span><span class="p">]:</span></pre></div>
</div>
</div>
<div class='section' id='section-14'>
<div class='docs'>
<div class='section-link'>
<a href='#section-14'>#</a>
</div>
<p>パラメータにグラデーションがない場合はスキップ</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">143</span> <span class="k">if</span> <span class="n">param</span><span class="o">.</span><span class="n">grad</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="lineno">144</span> <span class="k">continue</span></pre></div>
</div>
</div>
<div class='section' id='section-15'>
<div class='docs'>
<div class='section-link'>
<a href='#section-15'>#</a>
</div>
<p>勾配テンソルを取得</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">146</span> <span class="n">grad</span> <span class="o">=</span> <span class="n">param</span><span class="o">.</span><span class="n">grad</span><span class="o">.</span><span class="n">data</span></pre></div>
</div>
</div>
<div class='section' id='section-16'>
<div class='docs'>
<div class='section-link'>
<a href='#section-16'>#</a>
</div>
<p>スパースグラデーションは扱いません</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">148</span> <span class="k">if</span> <span class="n">grad</span><span class="o">.</span><span class="n">is_sparse</span><span class="p">:</span>
<span class="lineno">149</span> <span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s1">&#39;GenericAdaptiveOptimizer does not support sparse gradients,&#39;</span>
<span class="lineno">150</span> <span class="s1">&#39; please consider SparseAdam instead&#39;</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-17'>
<div class='docs'>
<div class='section-link'>
<a href='#section-17'>#</a>
</div>
<p>パラメータの状態を取得</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">153</span> <span class="n">state</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">state</span><span class="p">[</span><span class="n">param</span><span class="p">]</span></pre></div>
</div>
</div>
<div class='section' id='section-18'>
<div class='docs'>
<div class='section-link'>
<a href='#section-18'>#</a>
</div>
<p>状態が初期化されていない場合は状態を初期化します</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">156</span> <span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="n">state</span><span class="p">)</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="lineno">157</span> <span class="bp">self</span><span class="o">.</span><span class="n">init_state</span><span class="p">(</span><span class="n">state</span><span class="p">,</span> <span class="n">group</span><span class="p">,</span> <span class="n">param</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-19'>
<div class='docs'>
<div class='section-link'>
<a href='#section-19'>#</a>
</div>
<p>パラメータの最適化手順を実行してください</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">160</span> <span class="bp">self</span><span class="o">.</span><span class="n">step_param</span><span class="p">(</span><span class="n">state</span><span class="p">,</span> <span class="n">group</span><span class="p">,</span> <span class="n">grad</span><span class="p">,</span> <span class="n">param</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-20'>
<div class='docs'>
<div class='section-link'>
<a href='#section-20'>#</a>
</div>
<p>決済から計算した損失額を返金</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">163</span> <span class="k">return</span> <span class="n">loss</span></pre></div>
</div>
</div>
<div class='section' id='section-21'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-21'>#</a>
</div>
<h2>L2 ウェイト・ディケイ</h2>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">166</span><span class="k">class</span> <span class="nc">WeightDecay</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-22'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-22'>#</a>
</div>
<h3>体重減衰を初期化</h3>
<ul><li><code class="highlight"><span></span><span class="n">weight_decay</span></code>
は減衰係数</li>
<li><code class="highlight"><span></span><span class="n">weight_decouple</span></code>
グラデーションにウェイトディケイを追加するか、パラメータから直接ディケイを加えるかを示すフラグです。グラデーションに追加すると、通常のオプティマイザーの更新が行われます</li>。
<li><code class="highlight"><span></span><span class="n">absolute</span></code>
このフラグは重量減衰係数が絶対値かどうかを示します。これは、ディケイをパラメータに直接適用する場合に適用できます。これが false の場合、実際の減衰は <code class="highlight"><span></span><span class="n">weight_decay</span></code>
</li>
</ul><li><code class="highlight"><span></span><span class="n">learning_rate</span></code>
。</li>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">171</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">weight_decay</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">0.</span><span class="p">,</span> <span class="n">weight_decouple</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">,</span> <span class="n">absolute</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-23'>
<div class='docs'>
<div class='section-link'>
<a href='#section-23'>#</a>
</div>
<p>ハイパーパラメータをチェック</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">184</span> <span class="k">if</span> <span class="ow">not</span> <span class="mf">0.0</span> <span class="o">&lt;=</span> <span class="n">weight_decay</span><span class="p">:</span>
<span class="lineno">185</span> <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="sa">f</span><span class="s2">&quot;Invalid weight_decay value: </span><span class="si">{</span><span class="n">weight_decay</span><span class="si">}</span><span class="s2">&quot;</span><span class="p">)</span>
<span class="lineno">186</span>
<span class="lineno">187</span> <span class="bp">self</span><span class="o">.</span><span class="n">absolute</span> <span class="o">=</span> <span class="n">absolute</span>
<span class="lineno">188</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight_decouple</span> <span class="o">=</span> <span class="n">weight_decouple</span>
<span class="lineno">189</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight_decay</span> <span class="o">=</span> <span class="n">weight_decay</span></pre></div>
</div>
</div>
<div class='section' id='section-24'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-24'>#</a>
</div>
<p>パラメータグループのデフォルト値を返す</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">191</span> <span class="k">def</span> <span class="nf">defaults</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-25'>
<div class='docs'>
<div class='section-link'>
<a href='#section-25'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">195</span> <span class="k">return</span> <span class="nb">dict</span><span class="p">(</span><span class="n">weight_decay</span><span class="o">=</span><span class="bp">self</span><span class="o">.</span><span class="n">weight_decay</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-26'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-26'>#</a>
</div>
<h3>ウェイトディケイを実行してグラデーションを戻す</h3>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">197</span> <span class="k">def</span> <span class="fm">__call__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">param</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">,</span> <span class="n">grad</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">,</span> <span class="n">group</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">]):</span></pre></div>
</div>
</div>
<div class='section' id='section-27'>
<div class='docs'>
<div class='section-link'>
<a href='#section-27'>#</a>
</div>
<p>パラメータで直接ディケイを行う場合</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">203</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">weight_decouple</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-28'>
<div class='docs'>
<div class='section-link'>
<a href='#section-28'>#</a>
</div>
<p>重量減衰係数が絶対値の場合</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">205</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">absolute</span><span class="p">:</span>
<span class="lineno">206</span> <span class="n">param</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">mul_</span><span class="p">(</span><span class="mf">1.0</span> <span class="o">-</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;weight_decay&#39;</span><span class="p">])</span></pre></div>
</div>
</div>
<div class='section' id='section-29'>
<div class='docs'>
<div class='section-link'>
<a href='#section-29'>#</a>
</div>
<p>それ以外の場合は、</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">208</span> <span class="k">else</span><span class="p">:</span>
<span class="lineno">209</span> <span class="n">param</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">mul_</span><span class="p">(</span><span class="mf">1.0</span> <span class="o">-</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;lr&#39;</span><span class="p">]</span> <span class="o">*</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;weight_decay&#39;</span><span class="p">])</span></pre></div>
</div>
</div>
<div class='section' id='section-30'>
<div class='docs'>
<div class='section-link'>
<a href='#section-30'>#</a>
</div>
<p>変更されていないグラデーションを返す</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">211</span> <span class="k">return</span> <span class="n">grad</span>
<span class="lineno">212</span> <span class="k">else</span><span class="p">:</span>
<span class="lineno">213</span> <span class="k">if</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;weight_decay&#39;</span><span class="p">]</span> <span class="o">!=</span> <span class="mi">0</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-31'>
<div class='docs'>
<div class='section-link'>
<a href='#section-31'>#</a>
</div>
<p>グラデーションにウェイトディケイを追加し、変更したグラデーションを返します。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">215</span> <span class="k">return</span> <span class="n">grad</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="n">param</span><span class="o">.</span><span class="n">data</span><span class="p">,</span> <span class="n">alpha</span><span class="o">=</span><span class="n">group</span><span class="p">[</span><span class="s1">&#39;weight_decay&#39;</span><span class="p">])</span>
<span class="lineno">216</span> <span class="k">else</span><span class="p">:</span>
<span class="lineno">217</span> <span class="k">return</span> <span class="n">grad</span></pre></div>
</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>
docs/ja/optimizers/mnist_experiment.html
+460
View File
@@ -0,0 +1,460 @@
<!DOCTYPE html>
<html lang="ja">
<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="これは、CNN モデルを使用してオプティマイザーをテストする簡単な MNIST の例です。"/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="オプティマイザをテストするための MNIST の例"/>
<meta name="twitter:description" content="これは、CNN モデルを使用してオプティマイザーをテストする簡単な MNIST の例です。"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/optimizers/mnist_experiment.html"/>
<meta property="og:title" content="オプティマイザをテストするための MNIST の例"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="オプティマイザをテストするための MNIST の例"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="オプティマイザをテストするための MNIST の例"/>
<meta property="og:description" content="これは、CNN モデルを使用してオプティマイザーをテストする簡単な MNIST の例です。"/>
<title>オプティマイザをテストするための MNIST の例</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/optimizers/mnist_experiment.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">optimizers</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/optimizers/mnist_experiment.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1>オプティマイザをテストするための MNIST の例</h1>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">9</span><span></span><span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span>
<span class="lineno">10</span><span class="kn">import</span> <span class="nn">torch.utils.data</span>
<span class="lineno">11</span><span class="kn">from</span> <span class="nn">labml_helpers.module</span> <span class="kn">import</span> <span class="n">Module</span>
<span class="lineno">12</span>
<span class="lineno">13</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">experiment</span><span class="p">,</span> <span class="n">tracker</span>
<span class="lineno">14</span><span class="kn">from</span> <span class="nn">labml.configs</span> <span class="kn">import</span> <span class="n">option</span>
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">labml_helpers.datasets.mnist</span> <span class="kn">import</span> <span class="n">MNISTConfigs</span>
<span class="lineno">16</span><span class="kn">from</span> <span class="nn">labml_helpers.device</span> <span class="kn">import</span> <span class="n">DeviceConfigs</span>
<span class="lineno">17</span><span class="kn">from</span> <span class="nn">labml_helpers.metrics.accuracy</span> <span class="kn">import</span> <span class="n">Accuracy</span>
<span class="lineno">18</span><span class="kn">from</span> <span class="nn">labml_helpers.seed</span> <span class="kn">import</span> <span class="n">SeedConfigs</span>
<span class="lineno">19</span><span class="kn">from</span> <span class="nn">labml_helpers.train_valid</span> <span class="kn">import</span> <span class="n">TrainValidConfigs</span><span class="p">,</span> <span class="n">BatchIndex</span><span class="p">,</span> <span class="n">hook_model_outputs</span>
<span class="lineno">20</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers.configs</span> <span class="kn">import</span> <span class="n">OptimizerConfigs</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
<h2>モデル</h2>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">23</span><span class="k">class</span> <span class="nc">Model</span><span class="p">(</span><span class="n">Module</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">27</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="lineno">28</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="lineno">29</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv1</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">20</span><span class="p">,</span> <span class="mi">5</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
<span class="lineno">30</span> <span class="bp">self</span><span class="o">.</span><span class="n">pool1</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">MaxPool2d</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span>
<span class="lineno">31</span> <span class="bp">self</span><span class="o">.</span><span class="n">conv2</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">(</span><span class="mi">20</span><span class="p">,</span> <span class="mi">50</span><span class="p">,</span> <span class="mi">5</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
<span class="lineno">32</span> <span class="bp">self</span><span class="o">.</span><span class="n">pool2</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">MaxPool2d</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span>
<span class="lineno">33</span> <span class="bp">self</span><span class="o">.</span><span class="n">fc1</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="mi">16</span> <span class="o">*</span> <span class="mi">50</span><span class="p">,</span> <span class="mi">500</span><span class="p">)</span>
<span class="lineno">34</span> <span class="bp">self</span><span class="o">.</span><span class="n">fc2</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="mi">500</span><span class="p">,</span> <span class="mi">10</span><span class="p">)</span>
<span class="lineno">35</span> <span class="bp">self</span><span class="o">.</span><span class="n">activation</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-3'>
<div class='docs'>
<div class='section-link'>
<a href='#section-3'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">37</span> <span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">x</span><span class="p">):</span>
<span class="lineno">38</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">activation</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv1</span><span class="p">(</span><span class="n">x</span><span class="p">))</span>
<span class="lineno">39</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">pool1</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="lineno">40</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">activation</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">conv2</span><span class="p">(</span><span class="n">x</span><span class="p">))</span>
<span class="lineno">41</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">pool2</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="lineno">42</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">activation</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">fc1</span><span class="p">(</span><span class="n">x</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">16</span> <span class="o">*</span> <span class="mi">50</span><span class="p">)))</span>
<span class="lineno">43</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">fc2</span><span class="p">(</span><span class="n">x</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-4'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-4'>#</a>
</div>
<h2>設定可能な実験定義</h2>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">46</span><span class="k">class</span> <span class="nc">Configs</span><span class="p">(</span><span class="n">MNISTConfigs</span><span class="p">,</span> <span class="n">TrainValidConfigs</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-5'>
<div class='docs'>
<div class='section-link'>
<a href='#section-5'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">50</span> <span class="n">optimizer</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">optim</span><span class="o">.</span><span class="n">Adam</span>
<span class="lineno">51</span> <span class="n">model</span><span class="p">:</span> <span class="n">nn</span><span class="o">.</span><span class="n">Module</span>
<span class="lineno">52</span> <span class="n">set_seed</span> <span class="o">=</span> <span class="n">SeedConfigs</span><span class="p">()</span>
<span class="lineno">53</span> <span class="n">device</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span> <span class="o">=</span> <span class="n">DeviceConfigs</span><span class="p">()</span>
<span class="lineno">54</span> <span class="n">epochs</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">10</span>
<span class="lineno">55</span>
<span class="lineno">56</span> <span class="n">is_save_models</span> <span class="o">=</span> <span class="kc">True</span>
<span class="lineno">57</span> <span class="n">model</span><span class="p">:</span> <span class="n">nn</span><span class="o">.</span><span class="n">Module</span>
<span class="lineno">58</span> <span class="n">inner_iterations</span> <span class="o">=</span> <span class="mi">10</span>
<span class="lineno">59</span>
<span class="lineno">60</span> <span class="n">accuracy_func</span> <span class="o">=</span> <span class="n">Accuracy</span><span class="p">()</span>
<span class="lineno">61</span> <span class="n">loss_func</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">CrossEntropyLoss</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-6'>
<div class='docs'>
<div class='section-link'>
<a href='#section-6'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">63</span> <span class="k">def</span> <span class="nf">init</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="lineno">64</span> <span class="n">tracker</span><span class="o">.</span><span class="n">set_queue</span><span class="p">(</span><span class="s2">&quot;loss.*&quot;</span><span class="p">,</span> <span class="mi">20</span><span class="p">,</span> <span class="kc">True</span><span class="p">)</span>
<span class="lineno">65</span> <span class="n">tracker</span><span class="o">.</span><span class="n">set_scalar</span><span class="p">(</span><span class="s2">&quot;accuracy.*&quot;</span><span class="p">,</span> <span class="kc">True</span><span class="p">)</span>
<span class="lineno">66</span> <span class="n">hook_model_outputs</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">,</span> <span class="s1">&#39;model&#39;</span><span class="p">)</span>
<span class="lineno">67</span> <span class="bp">self</span><span class="o">.</span><span class="n">state_modules</span> <span class="o">=</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">accuracy_func</span><span class="p">]</span></pre></div>
</div>
</div>
<div class='section' id='section-7'>
<div class='docs'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">69</span> <span class="k">def</span> <span class="nf">step</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">batch</span><span class="p">:</span> <span class="nb">any</span><span class="p">,</span> <span class="n">batch_idx</span><span class="p">:</span> <span class="n">BatchIndex</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-8'>
<div class='docs'>
<div class='section-link'>
<a href='#section-8'>#</a>
</div>
<p>バッチを入手</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">71</span> <span class="n">data</span><span class="p">,</span> <span class="n">target</span> <span class="o">=</span> <span class="n">batch</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">),</span> <span class="n">batch</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">device</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-9'>
<div class='docs'>
<div class='section-link'>
<a href='#section-9'>#</a>
</div>
<p>トレーニングモードの場合はグローバルステップを追加</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">74</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">is_train</span><span class="p">:</span>
<span class="lineno">75</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add_global_step</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">data</span><span class="p">))</span></pre></div>
</div>
</div>
<div class='section' id='section-10'>
<div class='docs'>
<div class='section-link'>
<a href='#section-10'>#</a>
</div>
<p>モデルを実行し、アクティベーションをログに記録するかどうかを指定します</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">78</span> <span class="k">with</span> <span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="n">is_log_activations</span><span class="o">=</span><span class="n">batch_idx</span><span class="o">.</span><span class="n">is_last</span><span class="p">):</span>
<span class="lineno">79</span> <span class="n">output</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">(</span><span class="n">data</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-11'>
<div class='docs'>
<div class='section-link'>
<a href='#section-11'>#</a>
</div>
<p>損失の計算</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">82</span> <span class="n">loss</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">loss_func</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-12'>
<div class='docs'>
<div class='section-link'>
<a href='#section-12'>#</a>
</div>
<p>精度の計算</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">84</span> <span class="bp">self</span><span class="o">.</span><span class="n">accuracy_func</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-13'>
<div class='docs'>
<div class='section-link'>
<a href='#section-13'>#</a>
</div>
<p>損失を記録する</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">86</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="s2">&quot;loss.&quot;</span><span class="p">,</span> <span class="n">loss</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-14'>
<div class='docs'>
<div class='section-link'>
<a href='#section-14'>#</a>
</div>
<p>トレーニングモードの場合は最適化</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">89</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">mode</span><span class="o">.</span><span class="n">is_train</span><span class="p">:</span></pre></div>
</div>
</div>
<div class='section' id='section-15'>
<div class='docs'>
<div class='section-link'>
<a href='#section-15'>#</a>
</div>
<p>勾配の計算</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">91</span> <span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-16'>
<div class='docs'>
<div class='section-link'>
<a href='#section-16'>#</a>
</div>
<p>最適化の一歩を踏み出す</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">94</span> <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-17'>
<div class='docs'>
<div class='section-link'>
<a href='#section-17'>#</a>
</div>
<p>パラメーターと勾配の L2 ノルムをエポックごとに 1 回記録します。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">96</span> <span class="k">if</span> <span class="n">batch_idx</span><span class="o">.</span><span class="n">is_last</span><span class="p">:</span>
<span class="lineno">97</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="s1">&#39;model&#39;</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">)</span>
<span class="lineno">98</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="s1">&#39;optimizer&#39;</span><span class="p">,</span> <span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="p">{</span><span class="s1">&#39;model&#39;</span><span class="p">:</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">}))</span></pre></div>
</div>
</div>
<div class='section' id='section-18'>
<div class='docs'>
<div class='section-link'>
<a href='#section-18'>#</a>
</div>
<p>グラデーションをクリア</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">100</span> <span class="bp">self</span><span class="o">.</span><span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-19'>
<div class='docs'>
<div class='section-link'>
<a href='#section-19'>#</a>
</div>
<p>ログを保存</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">103</span> <span class="n">tracker</span><span class="o">.</span><span class="n">save</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-20'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-20'>#</a>
</div>
<p>設定可能なオプティマイザーを作成します。構成を使用してオプティマイザのタイプとハイパーパラメータを変更できます</p>。
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">106</span><span class="nd">@option</span><span class="p">(</span><span class="n">Configs</span><span class="o">.</span><span class="n">model</span><span class="p">)</span>
<span class="lineno">107</span><span class="k">def</span> <span class="nf">model</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">Configs</span><span class="p">):</span>
<span class="lineno">108</span> <span class="k">return</span> <span class="n">Model</span><span class="p">()</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
<span class="lineno">109</span>
<span class="lineno">110</span>
<span class="lineno">111</span><span class="nd">@option</span><span class="p">(</span><span class="n">Configs</span><span class="o">.</span><span class="n">optimizer</span><span class="p">)</span>
<span class="lineno">112</span><span class="k">def</span> <span class="nf">_optimizer</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">Configs</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-21'>
<div class='docs'>
<div class='section-link'>
<a href='#section-21'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">117</span> <span class="n">opt_conf</span> <span class="o">=</span> <span class="n">OptimizerConfigs</span><span class="p">()</span>
<span class="lineno">118</span> <span class="n">opt_conf</span><span class="o">.</span><span class="n">parameters</span> <span class="o">=</span> <span class="n">c</span><span class="o">.</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">()</span>
<span class="lineno">119</span> <span class="k">return</span> <span class="n">opt_conf</span></pre></div>
</div>
</div>
<div class='section' id='section-22'>
<div class='docs'>
<div class='section-link'>
<a href='#section-22'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">122</span><span class="k">def</span> <span class="nf">main</span><span class="p">():</span>
<span class="lineno">123</span> <span class="n">conf</span> <span class="o">=</span> <span class="n">Configs</span><span class="p">()</span>
<span class="lineno">124</span> <span class="n">conf</span><span class="o">.</span><span class="n">inner_iterations</span> <span class="o">=</span> <span class="mi">10</span>
<span class="lineno">125</span> <span class="n">experiment</span><span class="o">.</span><span class="n">create</span><span class="p">(</span><span class="n">name</span><span class="o">=</span><span class="s1">&#39;mnist_ada_belief&#39;</span><span class="p">)</span>
<span class="lineno">126</span> <span class="n">experiment</span><span class="o">.</span><span class="n">configs</span><span class="p">(</span><span class="n">conf</span><span class="p">,</span> <span class="p">{</span><span class="s1">&#39;inner_iterations&#39;</span><span class="p">:</span> <span class="mi">10</span><span class="p">,</span></pre></div>
</div>
</div>
<div class='section' id='section-23'>
<div class='docs'>
<div class='section-link'>
<a href='#section-23'>#</a>
</div>
<p>オプティマイザーを指定</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">128</span> <span class="s1">&#39;optimizer.optimizer&#39;</span><span class="p">:</span> <span class="s1">&#39;Adam&#39;</span><span class="p">,</span>
<span class="lineno">129</span> <span class="s1">&#39;optimizer.learning_rate&#39;</span><span class="p">:</span> <span class="mf">1.5e-4</span><span class="p">})</span>
<span class="lineno">130</span> <span class="n">conf</span><span class="o">.</span><span class="n">set_seed</span><span class="o">.</span><span class="n">set</span><span class="p">()</span>
<span class="lineno">131</span> <span class="n">experiment</span><span class="o">.</span><span class="n">add_pytorch_models</span><span class="p">(</span><span class="nb">dict</span><span class="p">(</span><span class="n">model</span><span class="o">=</span><span class="n">conf</span><span class="o">.</span><span class="n">model</span><span class="p">))</span>
<span class="lineno">132</span> <span class="k">with</span> <span class="n">experiment</span><span class="o">.</span><span class="n">start</span><span class="p">():</span>
<span class="lineno">133</span> <span class="n">conf</span><span class="o">.</span><span class="n">run</span><span class="p">()</span>
<span class="lineno">134</span>
<span class="lineno">135</span>
<span class="lineno">136</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">&#39;__main__&#39;</span><span class="p">:</span>
<span class="lineno">137</span> <span class="n">main</span><span class="p">()</span></pre></div>
</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>
docs/ja/optimizers/noam.html
+328
View File
@@ -0,0 +1,328 @@
<!DOCTYPE html>
<html lang="ja">
<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="これはNoamオプティマイザーのチュートリアル/実装です。Noam Optimizerにはウォームアップ期間があり、その後は学習率が指数関数的に低下します。"/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="アテンション・イズ・オール・ユー・ニード論文の Noam オプティマイザー"/>
<meta name="twitter:description" content="これはNoamオプティマイザーのチュートリアル/実装です。Noam Optimizerにはウォームアップ期間があり、その後は学習率が指数関数的に低下します。"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/optimizers/noam.html"/>
<meta property="og:title" content="アテンション・イズ・オール・ユー・ニード論文の Noam オプティマイザー"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="アテンション・イズ・オール・ユー・ニード論文の Noam オプティマイザー"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="アテンション・イズ・オール・ユー・ニード論文の Noam オプティマイザー"/>
<meta property="og:description" content="これはNoamオプティマイザーのチュートリアル/実装です。Noam Optimizerにはウォームアップ期間があり、その後は学習率が指数関数的に低下します。"/>
<title>アテンション・イズ・オール・ユー・ニード論文の Noam オプティマイザー</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/optimizers/noam.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">optimizers</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/optimizers/noam.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1>ノームオプティマイザー</h1>
<p>これは、論文「<a href="https://papers.labml.ai/paper/1706.03762">必要なのは注意だけ」<a href="https://pytorch.org">で紹介されているオプティマイザーのPyTorch実装です</a></a>。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">14</span><span></span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Dict</span>
<span class="lineno">15</span>
<span class="lineno">16</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers</span> <span class="kn">import</span> <span class="n">WeightDecay</span>
<span class="lineno">17</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers.amsgrad</span> <span class="kn">import</span> <span class="n">AMSGrad</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
<h2>ノームオプティマイザー</h2>
<p>このクラスは、で定義されている Adam オプティマイザを拡張したものです。<a href="adam.html"><code class="highlight"><span></span><span class="n">adam</span><span class="o">.</span><span class="n">py</span></code>
</a></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">20</span><span class="k">class</span> <span class="nc">Noam</span><span class="p">(</span><span class="n">AMSGrad</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-2'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-2'>#</a>
</div>
<h3>オプティマイザを初期化</h3>
<ul><li><code class="highlight"><span></span><span class="n">params</span></code>
はパラメータのリストです</li>
<li><code class="highlight"><span></span><span class="n">lr</span></code>
は学習率 <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="margin-right:0.0037em">α</span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">betas</span></code>
(,) <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span> のタプルです <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.8888799999999999em;vertical-align:-0.19444em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05278em;">β</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.30110799999999993em;"><span style="top:-2.5500000000000003em;margin-left:-0.05278em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight">2</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">eps</span></code>
<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.69444em;vertical-align:0em;"></span><span class="mord accent"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.69444em;"><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord coloredeq eqd" style=""><span class="mord mathnormal" style="">ϵ</span></span></span><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="accent-body" style="left:-0.19444em;"><span class="mord">^</span></span></span></span></span></span></span></span></span></span></span><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqd" style=""><span class="mord mathnormal" style="">ϵ</span></span></span></span></span></span>またはそれに基づいている <code class="highlight"><span></span><span class="n">optimized_update</span></code>
</li>
<li><code class="highlight"><span></span><span class="n">weight_decay</span></code>
<code class="highlight"><span></span><span class="n">WeightDecay</span></code>
で定義されているクラスのインスタンスです <a href="index.html"><code class="highlight"><span></span><span class="fm">__init__</span><span class="o">.</span><span class="n">py</span></code>
</a></li>
<li>'optimized_update'は追加後に行うことでセカンドモーメントのバイアス補正を最適化するかどうかのフラグです <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqd" style=""><span class="mord mathnormal" style="">ϵ</span></span></span></span></span></span></li>
<li><code class="highlight"><span></span><span class="n">amsgrad</span></code>
amsGradを使用するか、プレーンなAdamにフォールバックするかを示すフラグです</li>
<li><code class="highlight"><span></span><span class="n">warmup</span></code>
ウォームアップステップ数</li>
<li><code class="highlight"><span></span><span class="n">d_model</span></code>
モデルサイズ、つまり変圧器の次元数</li>
<li><code class="highlight"><span></span><span class="n">defaults</span></code>
グループ値のデフォルト辞書です。これは、クラスを拡張する場合に便利です<code class="highlight"><span></span><span class="n">AdamWarmup</span></code>
。</li></ul>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">27</span> <span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">params</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-3</span><span class="p">,</span> <span class="n">betas</span><span class="o">=</span><span class="p">(</span><span class="mf">0.9</span><span class="p">,</span> <span class="mf">0.999</span><span class="p">),</span> <span class="n">eps</span><span class="o">=</span><span class="mf">1e-16</span><span class="p">,</span>
<span class="lineno">28</span> <span class="n">weight_decay</span><span class="p">:</span> <span class="n">WeightDecay</span> <span class="o">=</span> <span class="n">WeightDecay</span><span class="p">(),</span>
<span class="lineno">29</span> <span class="n">optimized_update</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">,</span>
<span class="lineno">30</span> <span class="n">amsgrad</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="lineno">31</span> <span class="n">warmup</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">d_model</span><span class="o">=</span><span class="mi">512</span><span class="p">,</span> <span class="n">defaults</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-3'>
<div class='docs'>
<div class='section-link'>
<a href='#section-3'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">49</span> <span class="n">defaults</span> <span class="o">=</span> <span class="p">{}</span> <span class="k">if</span> <span class="n">defaults</span> <span class="ow">is</span> <span class="kc">None</span> <span class="k">else</span> <span class="n">defaults</span>
<span class="lineno">50</span> <span class="n">defaults</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="nb">dict</span><span class="p">(</span><span class="n">warmup</span><span class="o">=</span><span class="n">warmup</span><span class="p">))</span>
<span class="lineno">51</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">params</span><span class="p">,</span> <span class="n">lr</span><span class="p">,</span> <span class="n">betas</span><span class="p">,</span> <span class="n">eps</span><span class="p">,</span> <span class="n">weight_decay</span><span class="p">,</span> <span class="n">optimized_update</span><span class="p">,</span> <span class="n">amsgrad</span><span class="p">,</span> <span class="n">defaults</span><span class="p">)</span>
<span class="lineno">52</span> <span class="bp">self</span><span class="o">.</span><span class="n">d_model</span> <span class="o">=</span> <span class="n">d_model</span></pre></div>
</div>
</div>
<div class='section' id='section-4'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-4'>#</a>
</div>
<h3>学習率を取得</h3>
<p><span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:2.40003em;vertical-align:-0.95003em;"></span><span class="mord coloredeq eqa" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqg" style="margin-right:0.0037em">α</span></span><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.32144em;"><span style="top:-2.25278em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord sqrt" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.85722em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style="padding-left:0.833em"><span class="mord" style=""><span class="mord mathnormal" style="">d</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.33610799999999996em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">m</span><span class="mord mathnormal mtight" style="">o</span><span class="mord mathnormal mtight" style="">d</span><span class="mord mathnormal mtight" style="">e</span><span class="mord mathnormal mtight" style="margin-right:0.01968em">l</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span><span style="top:-2.81722em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail" style="min-width:0.853em;height:1.08em"><svg height="1.08em" preserveaspectratio="xMinYMin slice" viewbox="0 0 400000 1080" width="400em" xmlns="http://www.w3.org/2000/svg"><path d="M95,702
c-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14
c0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54
c44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10
s173,378,173,378c0.7,0,35.3,-71,104,-213c68.7,-142,137.5,-285,206.5,-429
c69,-144,104.5,-217.7,106.5,-221
l0 -0
c5.3,-9.3,12,-14,20,-14
H400000v40H845.2724
s-225.272,467,-225.272,467s-235,486,-235,486c-2.7,4.7,-9,7,-19,7
c-6,0,-10,-1,-12,-3s-194,-422,-194,-422s-65,47,-65,47z
M834 80h400000v40h-400000z"></path></svg></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.18278000000000005em;"><span></span></span></span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord" style="">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.93em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mord" style=""><span class="mop coloredeq eqb" style=""><span style="">m</span><span style="">i</span><span style="">n</span></span><span class="mspace" style="margin-right:0.16666666666666666em"></span><span class="mord coloredeq eqb" style=""><span class="delimsizing size3" style=""><span style="">(</span></span></span><span class="mord coloredeq eqb" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.32144em;"><span style="top:-2.21746em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord sqrt" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.89254em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style="padding-left:0.833em"><span class="mord mathnormal" style="">t</span></span></span><span style="top:-2.85254em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail" style="min-width:0.853em;height:1.08em"><svg height="1.08em" preserveaspectratio="xMinYMin slice" viewbox="0 0 400000 1080" width="400em" xmlns="http://www.w3.org/2000/svg"><path d="M95,702
c-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14
c0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54
c44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10
s173,378,173,378c0.7,0,35.3,-71,104,-213c68.7,-142,137.5,-285,206.5,-429
c69,-144,104.5,-217.7,106.5,-221
l0 -0
c5.3,-9.3,12,-14,20,-14
H400000v40H845.2724
s-225.272,467,-225.272,467s-235,486,-235,486c-2.7,4.7,-9,7,-19,7
c-6,0,-10,-1,-12,-3s-194,-422,-194,-422s-65,47,-65,47z
M834 80h400000v40h-400000z"></path></svg></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.14746000000000004em;"><span></span></span></span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord" style="">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.93em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mpunct coloredeq eqb" style="">,</span><span class="mspace" style="margin-right:0.16666666666666666em"></span><span class="mord coloredeq eqb" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.29208em;"><span style="top:-2.2960000000000003em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqh" style="margin-right:0.02691em">w</span></span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.814em;"><span style="top:-2.989em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style="">3/2</span></span></span></span></span></span></span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord mathnormal" style="">t</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.704em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mord coloredeq eqb" style=""><span class="delimsizing size3" style=""><span style="">)</span></span></span></span></span></span></span></span></span></span><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.43056em;vertical-align:0em;"></span><span class="mord coloredeq eqh" style=""><span class="mord mathnormal" style="margin-right:0.02691em">w</span></span></span></span></span></span>はウォームアップステップの数です。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">54</span> <span class="k">def</span> <span class="nf">get_lr</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">state</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">],</span> <span class="n">group</span><span class="p">:</span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">any</span><span class="p">]):</span></pre></div>
</div>
</div>
<div class='section' id='section-5'>
<div class='docs'>
<div class='section-link'>
<a href='#section-5'>#</a>
</div>
<p><span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:2.40003em;vertical-align:-0.95003em;"></span><span class="mord coloredeq eqb" style=""><span class="mop" style=""><span style="">m</span><span style="">i</span><span style="">n</span></span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord" style=""><span class="delimsizing size3" style=""><span style="">(</span></span></span><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.32144em;"><span style="top:-2.21746em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord sqrt" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.89254em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style="padding-left:0.833em"><span class="mord mathnormal" style="">t</span></span></span><span style="top:-2.85254em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail" style="min-width:0.853em;height:1.08em"><svg height="1.08em" preserveaspectratio="xMinYMin slice" viewbox="0 0 400000 1080" width="400em" xmlns="http://www.w3.org/2000/svg"><path d="M95,702
c-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14
c0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54
c44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10
s173,378,173,378c0.7,0,35.3,-71,104,-213c68.7,-142,137.5,-285,206.5,-429
c69,-144,104.5,-217.7,106.5,-221
l0 -0
c5.3,-9.3,12,-14,20,-14
H400000v40H845.2724
s-225.272,467,-225.272,467s-235,486,-235,486c-2.7,4.7,-9,7,-19,7
c-6,0,-10,-1,-12,-3s-194,-422,-194,-422s-65,47,-65,47z
M834 80h400000v40h-400000z"></path></svg></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.14746000000000004em;"><span></span></span></span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord" style="">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.93em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mpunct" style="">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.29208em;"><span style="top:-2.2960000000000003em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqh" style="margin-right:0.02691em">w</span></span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.814em;"><span style="top:-2.989em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style="">3/2</span></span></span></span></span></span></span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord mathnormal" style="">t</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.704em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mord" style=""><span class="delimsizing size3" style=""><span style="">)</span></span></span></span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">62</span> <span class="n">factor</span> <span class="o">=</span> <span class="nb">min</span><span class="p">(</span><span class="n">state</span><span class="p">[</span><span class="s1">&#39;step&#39;</span><span class="p">]</span> <span class="o">**</span> <span class="p">(</span><span class="o">-</span><span class="mf">0.5</span><span class="p">),</span> <span class="n">state</span><span class="p">[</span><span class="s1">&#39;step&#39;</span><span class="p">]</span> <span class="o">*</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;warmup&#39;</span><span class="p">]</span> <span class="o">**</span> <span class="p">(</span><span class="o">-</span><span class="mf">1.5</span><span class="p">))</span></pre></div>
</div>
</div>
<div class='section' id='section-6'>
<div class='docs'>
<div class='section-link'>
<a href='#section-6'>#</a>
</div>
<p><span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:2.40003em;vertical-align:-0.95003em;"></span><span class="mord coloredeq eqa" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqg" style="margin-right:0.0037em">α</span></span><span class="mord" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.32144em;"><span style="top:-2.25278em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord sqrt" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.85722em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style="padding-left:0.833em"><span class="mord" style=""><span class="mord mathnormal" style="">d</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.33610799999999996em;"><span style="top:-2.5500000000000003em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">m</span><span class="mord mathnormal mtight" style="">o</span><span class="mord mathnormal mtight" style="">d</span><span class="mord mathnormal mtight" style="">e</span><span class="mord mathnormal mtight" style="margin-right:0.01968em">l</span></span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.15em;"><span></span></span></span></span></span></span></span></span><span style="top:-2.81722em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail" style="min-width:0.853em;height:1.08em"><svg height="1.08em" preserveaspectratio="xMinYMin slice" viewbox="0 0 400000 1080" width="400em" xmlns="http://www.w3.org/2000/svg"><path d="M95,702
c-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14
c0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54
c44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10
s173,378,173,378c0.7,0,35.3,-71,104,-213c68.7,-142,137.5,-285,206.5,-429
c69,-144,104.5,-217.7,106.5,-221
l0 -0
c5.3,-9.3,12,-14,20,-14
H400000v40H845.2724
s-225.272,467,-225.272,467s-235,486,-235,486c-2.7,4.7,-9,7,-19,7
c-6,0,-10,-1,-12,-3s-194,-422,-194,-422s-65,47,-65,47z
M834 80h400000v40h-400000z"></path></svg></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.18278000000000005em;"><span></span></span></span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord" style="">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.93em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mord" style=""><span class="mop coloredeq eqb" style=""><span style="">m</span><span style="">i</span><span style="">n</span></span><span class="mspace" style="margin-right:0.16666666666666666em"></span><span class="mord coloredeq eqb" style=""><span class="delimsizing size3" style=""><span style="">(</span></span></span><span class="mord coloredeq eqb" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.32144em;"><span style="top:-2.21746em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord sqrt" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.89254em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style="padding-left:0.833em"><span class="mord mathnormal" style="">t</span></span></span><span style="top:-2.85254em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail" style="min-width:0.853em;height:1.08em"><svg height="1.08em" preserveaspectratio="xMinYMin slice" viewbox="0 0 400000 1080" width="400em" xmlns="http://www.w3.org/2000/svg"><path d="M95,702
c-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14
c0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54
c44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10
s173,378,173,378c0.7,0,35.3,-71,104,-213c68.7,-142,137.5,-285,206.5,-429
c69,-144,104.5,-217.7,106.5,-221
l0 -0
c5.3,-9.3,12,-14,20,-14
H400000v40H845.2724
s-225.272,467,-225.272,467s-235,486,-235,486c-2.7,4.7,-9,7,-19,7
c-6,0,-10,-1,-12,-3s-194,-422,-194,-422s-65,47,-65,47z
M834 80h400000v40h-400000z"></path></svg></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.14746000000000004em;"><span></span></span></span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord" style="">1</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.93em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mpunct coloredeq eqb" style="">,</span><span class="mspace" style="margin-right:0.16666666666666666em"></span><span class="mord coloredeq eqb" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.29208em;"><span style="top:-2.2960000000000003em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqh" style="margin-right:0.02691em">w</span></span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.814em;"><span style="top:-2.989em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mtight" style="">3/2</span></span></span></span></span></span></span></span></span></span></span><span style="top:-3.23em;"><span class="pstrut" style="height:3em;"></span><span class="frac-line" style="border-bottom-width:0.04em"></span></span><span style="top:-3.677em;"><span class="pstrut" style="height:3em;"></span><span class="mord" style=""><span class="mord mathnormal" style="">t</span></span></span></span><span class="vlist-s">​</span></span><span class="vlist-r"><span class="vlist" style="height:0.704em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span><span class="mord coloredeq eqb" style=""><span class="delimsizing size3" style=""><span style="">)</span></span></span></span></span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">64</span> <span class="k">return</span> <span class="n">group</span><span class="p">[</span><span class="s1">&#39;lr&#39;</span><span class="p">]</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">d_model</span> <span class="o">**</span> <span class="p">(</span><span class="o">-</span><span class="mf">0.5</span><span class="p">)</span> <span class="o">*</span> <span class="n">factor</span></pre></div>
</div>
</div>
<div class='section' id='section-7'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
<h3>さまざまなウォームアップとモデルサイズの学習率をプロット</h3>
<p><img alt="Plot of learning rate" src="noam_lr.png"></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">67</span><span class="k">def</span> <span class="nf">_test_noam_lr</span><span class="p">():</span></pre></div>
</div>
</div>
<div class='section' id='section-8'>
<div class='docs'>
<div class='section-link'>
<a href='#section-8'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">73</span> <span class="kn">import</span> <span class="nn">matplotlib.pyplot</span> <span class="k">as</span> <span class="nn">plt</span>
<span class="lineno">74</span> <span class="kn">import</span> <span class="nn">numpy</span> <span class="k">as</span> <span class="nn">np</span>
<span class="lineno">75</span> <span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
<span class="lineno">76</span>
<span class="lineno">77</span> <span class="n">model</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="mi">10</span><span class="p">)</span>
<span class="lineno">78</span> <span class="n">opts</span> <span class="o">=</span> <span class="p">[</span><span class="n">Noam</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">d_model</span><span class="o">=</span><span class="mi">512</span><span class="p">,</span> <span class="n">warmup</span><span class="o">=</span><span class="mi">4000</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mi">1</span><span class="p">),</span>
<span class="lineno">79</span> <span class="n">Noam</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">d_model</span><span class="o">=</span><span class="mi">512</span><span class="p">,</span> <span class="n">warmup</span><span class="o">=</span><span class="mi">8000</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mi">1</span><span class="p">),</span>
<span class="lineno">80</span> <span class="n">Noam</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">(),</span> <span class="n">d_model</span><span class="o">=</span><span class="mi">2048</span><span class="p">,</span> <span class="n">warmup</span><span class="o">=</span><span class="mi">2000</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mi">1</span><span class="p">)]</span>
<span class="lineno">81</span> <span class="n">plt</span><span class="o">.</span><span class="n">plot</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">20000</span><span class="p">),</span> <span class="p">[[</span><span class="n">opt</span><span class="o">.</span><span class="n">get_lr</span><span class="p">({</span><span class="s1">&#39;step&#39;</span><span class="p">:</span> <span class="n">i</span><span class="p">},</span> <span class="n">opt</span><span class="o">.</span><span class="n">defaults</span><span class="p">)</span> <span class="k">for</span> <span class="n">opt</span> <span class="ow">in</span> <span class="n">opts</span><span class="p">]</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">20000</span><span class="p">)])</span>
<span class="lineno">82</span> <span class="n">plt</span><span class="o">.</span><span class="n">legend</span><span class="p">([</span><span class="s2">&quot;512:4000&quot;</span><span class="p">,</span> <span class="s2">&quot;512:8000&quot;</span><span class="p">,</span> <span class="s2">&quot;2048:2000&quot;</span><span class="p">])</span>
<span class="lineno">83</span> <span class="n">plt</span><span class="o">.</span><span class="n">title</span><span class="p">(</span><span class="s2">&quot;Learning Rate&quot;</span><span class="p">)</span>
<span class="lineno">84</span> <span class="n">plt</span><span class="o">.</span><span class="n">show</span><span class="p">()</span>
<span class="lineno">85</span>
<span class="lineno">86</span>
<span class="lineno">87</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">&#39;__main__&#39;</span><span class="p">:</span>
<span class="lineno">88</span> <span class="n">_test_noam_lr</span><span class="p">()</span></pre></div>
</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>
docs/ja/optimizers/noam_lr.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 35 KiB

docs/ja/optimizers/performance_test.html
+175
View File
@@ -0,0 +1,175 @@
<!DOCTYPE html>
<html lang="ja">
<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="この実験では、Adam 実装のパフォーマンスを比較します。"/>
<meta name="twitter:card" content="summary"/>
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta name="twitter:title" content="Adam 実装のテストパフォーマンス"/>
<meta name="twitter:description" content="この実験では、Adam 実装のパフォーマンスを比較します。"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/optimizers/performance_test.html"/>
<meta property="og:title" content="Adam 実装のテストパフォーマンス"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="Adam 実装のテストパフォーマンス"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="Adam 実装のテストパフォーマンス"/>
<meta property="og:description" content="この実験では、Adam 実装のパフォーマンスを比較します。"/>
<title>Adam 実装のテストパフォーマンス</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/optimizers/performance_test.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">optimizers</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/optimizers/performance_test.py" target="_blank">
View code on Github</a>
</p>
</div>
</div>
<div class='section' id='section-0'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-0'>#</a>
</div>
<h1>パフォーマンステスト Adam</h1>
<pre class="highlight lang-text"><code><span></span>TorchAdam warmup...[DONE] 222.59ms
TorchAdam...[DONE] 1,356.01ms
MyAdam warmup...[DONE] 119.15ms
MyAdam...[DONE] 1,192.89ms</code></pre>
<p><a href="https://colab.research.google.com/drive/1ngowaAsADj8VdZfBifu_6L6rtjGoEeoR?usp=sharing"><img alt="Open In Colab" src="https://colab.research.google.com/assets/colab-badge.svg"></a></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">19</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
<span class="lineno">20</span><span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span>
<span class="lineno">21</span><span class="kn">from</span> <span class="nn">labml_helpers.device</span> <span class="kn">import</span> <span class="n">DeviceInfo</span>
<span class="lineno">22</span><span class="kn">from</span> <span class="nn">torch.optim</span> <span class="kn">import</span> <span class="n">Adam</span> <span class="k">as</span> <span class="n">TorchAdam</span>
<span class="lineno">23</span>
<span class="lineno">24</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">monit</span>
<span class="lineno">25</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers.adam</span> <span class="kn">import</span> <span class="n">Adam</span> <span class="k">as</span> <span class="n">MyAdam</span>
<span class="lineno">26</span><span class="kn">from</span> <span class="nn">labml_nn.optimizers.mnist_experiment</span> <span class="kn">import</span> <span class="n">Model</span></pre></div>
</div>
</div>
<div class='section' id='section-1'>
<div class='docs'>
<div class='section-link'>
<a href='#section-1'>#</a>
</div>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">29</span><span class="k">def</span> <span class="nf">test</span><span class="p">():</span>
<span class="lineno">30</span> <span class="n">device_info</span> <span class="o">=</span> <span class="n">DeviceInfo</span><span class="p">(</span><span class="n">use_cuda</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">cuda_device</span><span class="o">=</span><span class="mi">0</span><span class="p">)</span>
<span class="lineno">31</span> <span class="nb">print</span><span class="p">(</span><span class="n">device_info</span><span class="p">)</span>
<span class="lineno">32</span> <span class="n">inp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randn</span><span class="p">((</span><span class="mi">64</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">28</span><span class="p">,</span> <span class="mi">28</span><span class="p">),</span> <span class="n">device</span><span class="o">=</span><span class="n">device_info</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
<span class="lineno">33</span> <span class="n">target</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">(</span><span class="mi">64</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">long</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">device_info</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
<span class="lineno">34</span> <span class="n">loss_func</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">CrossEntropyLoss</span><span class="p">()</span>
<span class="lineno">35</span> <span class="n">model</span> <span class="o">=</span> <span class="n">Model</span><span class="p">()</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device_info</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
<span class="lineno">36</span> <span class="n">my_adam</span> <span class="o">=</span> <span class="n">MyAdam</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">())</span>
<span class="lineno">37</span> <span class="n">torch_adam</span> <span class="o">=</span> <span class="n">TorchAdam</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">())</span>
<span class="lineno">38</span> <span class="n">loss</span> <span class="o">=</span> <span class="n">loss_func</span><span class="p">(</span><span class="n">model</span><span class="p">(</span><span class="n">inp</span><span class="p">),</span> <span class="n">target</span><span class="p">)</span>
<span class="lineno">39</span> <span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="lineno">40</span> <span class="k">with</span> <span class="n">monit</span><span class="o">.</span><span class="n">section</span><span class="p">(</span><span class="s1">&#39;MyAdam warmup&#39;</span><span class="p">):</span>
<span class="lineno">41</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">100</span><span class="p">):</span>
<span class="lineno">42</span> <span class="n">my_adam</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="lineno">43</span> <span class="k">with</span> <span class="n">monit</span><span class="o">.</span><span class="n">section</span><span class="p">(</span><span class="s1">&#39;MyAdam&#39;</span><span class="p">):</span>
<span class="lineno">44</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">1000</span><span class="p">):</span>
<span class="lineno">45</span> <span class="n">my_adam</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="lineno">46</span> <span class="k">with</span> <span class="n">monit</span><span class="o">.</span><span class="n">section</span><span class="p">(</span><span class="s1">&#39;TorchAdam warmup&#39;</span><span class="p">):</span>
<span class="lineno">47</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">100</span><span class="p">):</span>
<span class="lineno">48</span> <span class="n">torch_adam</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="lineno">49</span> <span class="k">with</span> <span class="n">monit</span><span class="o">.</span><span class="n">section</span><span class="p">(</span><span class="s1">&#39;TorchAdam&#39;</span><span class="p">):</span>
<span class="lineno">50</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">1000</span><span class="p">):</span>
<span class="lineno">51</span> <span class="n">torch_adam</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="lineno">52</span>
<span class="lineno">53</span>
<span class="lineno">54</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">&#39;__main__&#39;</span><span class="p">:</span>
<span class="lineno">55</span> <span class="n">test</span><span class="p">()</span></pre></div>
</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>
docs/ja/optimizers/radam.html
+782
View File
File diff suppressed because one or more lines are too long
docs/ja/optimizers/radam_r_t.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 30 KiB

docs/ja/optimizers/readme.html
+133
View File
@@ -0,0 +1,133 @@
<!DOCTYPE html>
<html lang="ja">
<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&amp;v=4"/>
<meta name="twitter:title" content="オプティマイザー"/>
<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/optimizers/readme.html"/>
<meta property="og:title" content="オプティマイザー"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="オプティマイザー"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="オプティマイザー"/>
<meta property="og:description" content=""/>
<title>オプティマイザー</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/optimizers/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">optimizers</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/optimizers/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/optimizers/index.html">オプティマイザー</a></h1>
<h2>オプティマイザーの実装</h2>
<ul><li><a href="https://nn.labml.ai/optimizers/adam.html">アダム・オプティマイザー</a></li>
<li><a href="https://nn.labml.ai/optimizers/amsgrad.html">マスグラードオプティマイザー</a></li>
<li><a href="https://nn.labml.ai/optimizers/adam_warmup.html">ウォームアップ機能付き Adam オプティマイザー</a></li>
<li><a href="https://nn.labml.ai/optimizers/noam.html">ノームオプティマイザー</a></li>
<li><a href="https://nn.labml.ai/optimizers/radam.html">レクティファイド・アダム・オプティマイザー</a></li>
<li><a href="https://nn.labml.ai/optimizers/ada_belief.html">アダブリリーフオプティマイザー</a></li></ul>
</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>