Files
Varuna Jayasiri 2038b11d29 ja translation
2023-05-10 17:00:29 -04:00

731 lines
73 KiB
HTML
Raw Permalink Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

<!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="PyTorchを使ったトランスフォーマーで使われるマルチヘッド・アテンションを説明付きで実装しています。"/>
<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="マルチヘッド・アテンション (MHA)"/>
<meta name="twitter:description" content="PyTorchを使ったトランスフォーマーで使われるマルチヘッド・アテンションを説明付きで実装しています。"/>
<meta name="twitter:site" content="@labmlai"/>
<meta name="twitter:creator" content="@labmlai"/>
<meta property="og:url" content="https://nn.labml.ai/transformers/mha.html"/>
<meta property="og:title" content="マルチヘッド・アテンション (MHA)"/>
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&amp;v=4"/>
<meta property="og:site_name" content="マルチヘッド・アテンション (MHA)"/>
<meta property="og:type" content="object"/>
<meta property="og:title" content="マルチヘッド・アテンション (MHA)"/>
<meta property="og:description" content="PyTorchを使ったトランスフォーマーで使われるマルチヘッド・アテンションを説明付きで実装しています。"/>
<title>マルチヘッド・アテンション (MHA)</title>
<link rel="shortcut icon" href="/icon.png"/>
<link rel="stylesheet" href="../pylit.css?v=1">
<link rel="canonical" href="https://nn.labml.ai/transformers/mha.html"/>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.13.18/dist/katex.min.css" integrity="sha384-zTROYFVGOfTw7JV7KUu8udsvW2fx4lWOsCEDqhBreBwlHI4ioVRtmIvEThzJHGET" crossorigin="anonymous">
<!-- Global site tag (gtag.js) - Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-4V3HC8HBLH"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() {
dataLayer.push(arguments);
}
gtag('js', new Date());
gtag('config', 'G-4V3HC8HBLH');
</script>
</head>
<body>
<div id='container'>
<div id="background"></div>
<div class='section'>
<div class='docs'>
<p>
<a class="parent" href="/">home</a>
<a class="parent" href="index.html">transformers</a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations" target="_blank">
<img alt="Github"
src="https://img.shields.io/github/stars/labmlai/annotated_deep_learning_paper_implementations?style=social"
style="max-width:100%;"/></a>
<a href="https://twitter.com/labmlai" rel="nofollow" target="_blank">
<img alt="Twitter"
src="https://img.shields.io/twitter/follow/labmlai?style=social"
style="max-width:100%;"/></a>
</p>
<p>
<a href="https://github.com/labmlai/annotated_deep_learning_paper_implementations/tree/master/labml_nn/transformers/mha.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>マルチヘッド・アテンション (MHA)</h1>
<p><a href="https://colab.research.google.com/github/labmlai/annotated_deep_learning_paper_implementations/blob/master/labml_nn/transformers/basic/autoregressive_experiment.ipynb"><img alt="Open In Colab" src="https://colab.research.google.com/assets/colab-badge.svg"></a></p>
</a><p>これは、PyTorchの論文「<a href="https://papers.labml.ai/paper/1706.03762">注意さえあれば十分」の「多面的な注意」のチュートリアル/実装です</a><a href="https://pytorch.org/"><a href="https://nlp.seas.harvard.edu/2018/04/03/attention.html">実装は注釈付きトランスフォーマーから着想を得ています</a></p>
<p>これは、<a href="basic/autoregressive_experiment.html">NLP自己回帰用のMHAを備えた基本的なトランスフォーマーを使用するトレーニングコードです</a></p>
<p><a href="basic/autoregressive_experiment.html">これは簡単な変圧器をトレーニングする実験実装です</a></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">24</span><span></span><span class="kn">import</span> <span class="nn">math</span>
<span class="lineno">25</span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">List</span>
<span class="lineno">26</span>
<span class="lineno">27</span><span class="kn">import</span> <span class="nn">torch</span>
<span class="lineno">28</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
<span class="lineno">29</span>
<span class="lineno">30</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">tracker</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="PrepareMHA"></a></p>
<h2>マルチヘッド・アテンションに備えましょう</h2>
<p>このモジュールは線形変換を行い、ベクトルを指定された数のヘッドに分割してマルチヘッドアテンションを行います。これは、<strong>キー</strong><strong>クエリ</strong><strong>および値のベクトルを変換するために使用されます</strong></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">33</span><span class="k">class</span> <span class="nc">PrepareForMultiHeadAttention</span><span class="p">(</span><span class="n">nn</span><span class="o">.</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">44</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">d_model</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">heads</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">d_k</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">bias</span><span class="p">:</span> <span class="nb">bool</span><span class="p">):</span>
<span class="lineno">45</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</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>線形変換用の線形層</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">47</span> <span class="bp">self</span><span class="o">.</span><span class="n">linear</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="n">d_model</span><span class="p">,</span> <span class="n">heads</span> <span class="o">*</span> <span class="n">d_k</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="n">bias</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>ヘッド数</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">49</span> <span class="bp">self</span><span class="o">.</span><span class="n">heads</span> <span class="o">=</span> <span class="n">heads</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>各ヘッドのベクトルの次元数</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">51</span> <span class="bp">self</span><span class="o">.</span><span class="n">d_k</span> <span class="o">=</span> <span class="n">d_k</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">53</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="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-7'>
<div class='docs'>
<div class='section-link'>
<a href='#section-7'>#</a>
</div>
<p><code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">d_model</span><span class="p">]</span></code>
<code class="highlight"><span></span><span class="p">[</span><span class="n">batch_size</span><span class="p">,</span> <span class="n">d_model</span><span class="p">]</span></code>
入力の形状はまたはです。線形変換を最後の次元に適用し、それを頭に分割します。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">57</span> <span class="n">head_shape</span> <span class="o">=</span> <span class="n">x</span><span class="o">.</span><span class="n">shape</span><span class="p">[:</span><span class="o">-</span><span class="mi">1</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>線形変換</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">60</span> <span class="n">x</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">linear</span><span class="p">(</span><span class="n">x</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>最後のディメンションをヘッドに分割</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">63</span> <span class="n">x</span> <span class="o">=</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="n">head_shape</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">d_k</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><code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">heads</span><span class="p">,</span> <span class="n">d_k</span><span class="p">]</span></code>
出力の形状があるか <code class="highlight"><span></span><span class="p">[</span><span class="n">batch_size</span><span class="p">,</span> <span class="n">heads</span><span class="p">,</span> <span class="n">d_model</span><span class="p">]</span></code>
</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">66</span> <span class="k">return</span> <span class="n">x</span></pre></div>
</div>
</div>
<div class='section' id='section-11'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-11'>#</a>
</div>
<p><a id="MHA"></a></p>
<h2>マルチヘッドアテンションモジュール</h2>
<p><code class="highlight"><span></span><span class="n">query</span></code>
与えられたベクトルやベクトルに対して、スケーリングされたマルチヘッド・アテンションを計算します。<code class="highlight"><span></span><span class="n">key</span></code>
<code class="highlight"><span></span><span class="n">value</span></code>
</p>
<p><span ><span class="katex-display"><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="mop"><span class="mord mathnormal">A</span><span class="mord mathnormal">tt</span><span class="mord mathnormal">e</span><span class="mord mathnormal">n</span><span class="mord mathnormal">t</span><span class="mord mathnormal">i</span><span class="mord mathnormal">o</span><span class="mord mathnormal">n</span></span><span class="mopen">(</span><span class="mord mathnormal">Q</span><span class="mpunct">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord mathnormal" style="margin-right:0.07153em;">K</span><span class="mpunct">,</span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord mathnormal" style="margin-right:0.22222em;">V</span><span class="mclose">)</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:3.0000299999999998em;vertical-align:-1.25003em;"></span><span class="mord"><span class="mop op-limits"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.6944399999999998em;"><span style="top:-2.20556em;margin-left:0em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathnormal mtight">se</span><span class="mord mathnormal mtight" style="margin-right:0.03588em;">q</span></span></span></span><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span><span class="mop"><span class="mop"><span class="mord coloredeq eqh" style=""><span class="mord mathnormal" style="">so</span><span class="mord mathnormal" style="margin-right:0.10764em">f</span><span class="mord mathnormal" style="">t</span><span class="mord mathnormal" style="">ma</span><span class="mord mathnormal" style="">x</span></span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:1.030548em;"><span></span></span></span></span></span></span><span class="mord"><span class="delimsizing size4">(</span></span><span class="mord coloredeq eqe" style=""><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:1.5261079999999998em;"><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 coloredeq eqi" 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 mathnormal mtight" style="margin-right:0.03148em">k</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 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=""><span class="mord mathnormal coloredeq eqg" style="">Q</span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="margin-right:0.07153em">K</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.849108em;"><span style="top:-3.063em;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></span></span></span></span></span></span></span></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><span class="mord"><span class="delimsizing size4">)</span></span><span class="mord mathnormal" style="margin-right:0.22222em;">V</span></span></span></span></span></span></p>
<p>簡単に言うと、クエリに一致するキーを見つけ、それらのキーの値を取得します。</p>
<p>クエリとキーのドット積がどの程度一致しているかを示す指標として使用します。<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 coloredeq eqh" style=""><span class="mord mathnormal" style="">so</span><span class="mord mathnormal" style="margin-right:0.10764em">f</span><span class="mord mathnormal" style="">t</span><span class="mord mathnormal" style="">ma</span><span class="mord mathnormal" style="">x</span></span></span></span></span></span>撮影前にドットプロダクトをスケーリングします。<span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.383108em;vertical-align:-0.538em;"></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:0.845108em;"><span style="top:-2.5864385em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord sqrt mtight"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.8622307142857143em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord mtight" style="padding-left:0.833em;"><span class="mord mtight coloredeq eqi" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">d</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.3448em;"><span style="top:-2.3487714285714287em;margin-left:0em;margin-right:0.07142857142857144em;"><span class="pstrut" style="height:2.5em;"></span><span class="sizing reset-size3 size1 mtight" style=""><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15122857142857138em;"><span></span></span></span></span></span></span></span></span></span><span style="top:-2.8222307142857144em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail mtight" 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:0.17776928571428574em;"><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:-3.394em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mtight coloredeq eqj" style=""><span class="mord mtight" style="">1</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.538em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span></span></span></span>これは、ドット積値が大きい場合に softmax のグラデーションが非常に小さくなる原因とならないようにするためです</p><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.84444em;vertical-align:-0.15em;"></span><span class="mord coloredeq eqi" style=""><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 mathnormal mtight" style="margin-right:0.03148em">k</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>Softmax は、シーケンス (または時間) の軸に沿って計算されます。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">69</span><span class="k">class</span> <span class="nc">MultiHeadAttention</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">):</span></pre></div>
</div>
</div>
<div class='section' id='section-12'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-12'>#</a>
</div>
<ul><li><code class="highlight"><span></span><span class="n">heads</span></code>
は頭の数です。</li>
<li><code class="highlight"><span></span><span class="n">d_model</span></code>
<code class="highlight"><span></span><span class="n">query</span></code>
<code class="highlight"><span></span><span class="n">key</span></code>
<code class="highlight"><span></span><span class="n">value</span></code>
およびベクトル内の特徴の数です。</li></ul>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">90</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">heads</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">d_model</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">dropout_prob</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">0.1</span><span class="p">,</span> <span class="n">bias</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</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>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">96</span> <span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</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>ヘッドあたりの機能数</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">99</span> <span class="bp">self</span><span class="o">.</span><span class="n">d_k</span> <span class="o">=</span> <span class="n">d_model</span> <span class="o">//</span> <span class="n">heads</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>ヘッド数</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">101</span> <span class="bp">self</span><span class="o">.</span><span class="n">heads</span> <span class="o">=</span> <span class="n">heads</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>これらは<code class="highlight"><span></span><span class="n">query</span></code>
、、<code class="highlight"><span></span><span class="n">key</span></code>
<code class="highlight"><span></span><span class="n">value</span></code>
のベクトルを変えて、多面的な注意を促します。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">104</span> <span class="bp">self</span><span class="o">.</span><span class="n">query</span> <span class="o">=</span> <span class="n">PrepareForMultiHeadAttention</span><span class="p">(</span><span class="n">d_model</span><span class="p">,</span> <span class="n">heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">d_k</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="n">bias</span><span class="p">)</span>
<span class="lineno">105</span> <span class="bp">self</span><span class="o">.</span><span class="n">key</span> <span class="o">=</span> <span class="n">PrepareForMultiHeadAttention</span><span class="p">(</span><span class="n">d_model</span><span class="p">,</span> <span class="n">heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">d_k</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="n">bias</span><span class="p">)</span>
<span class="lineno">106</span> <span class="bp">self</span><span class="o">.</span><span class="n">value</span> <span class="o">=</span> <span class="n">PrepareForMultiHeadAttention</span><span class="p">(</span><span class="n">d_model</span><span class="p">,</span> <span class="n">heads</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">d_k</span><span class="p">,</span> <span class="n">bias</span><span class="o">=</span><span class="kc">True</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>時間軸に沿った注目のソフトマックス <code class="highlight"><span></span><span class="n">key</span></code>
</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">109</span> <span class="bp">self</span><span class="o">.</span><span class="n">softmax</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Softmax</span><span class="p">(</span><span class="n">dim</span><span class="o">=</span><span class="mi">1</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>出力レイヤー</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">112</span> <span class="bp">self</span><span class="o">.</span><span class="n">output</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="n">d_model</span><span class="p">,</span> <span class="n">d_model</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>ドロップアウト</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">114</span> <span class="bp">self</span><span class="o">.</span><span class="n">dropout</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Dropout</span><span class="p">(</span><span class="n">dropout_prob</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>ソフトマックス前のスケーリングファクター</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">116</span> <span class="bp">self</span><span class="o">.</span><span class="n">scale</span> <span class="o">=</span> <span class="mi">1</span> <span class="o">/</span> <span class="n">math</span><span class="o">.</span><span class="n">sqrt</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">d_k</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>必要に応じてロギングやその他の計算に使用できるように、アテンションを保存します</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">119</span> <span class="bp">self</span><span class="o">.</span><span class="n">attn</span> <span class="o">=</span> <span class="kc">None</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>クエリとキー間のスコアの計算</h3>
<p>この方法は、相対的注意力などの他のバリエーションではオーバーライドできます。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">121</span> <span class="k">def</span> <span class="nf">get_scores</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">query</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">key</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-23'>
<div class='docs'>
<div class='section-link'>
<a href='#section-23'>#</a>
</div>
<p><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.043548em;vertical-align:-0.19444em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="">Q</span><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.07153em">K</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.849108em;"><span style="top:-3.063em;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></span></span></span></span></span></span></span></span></span></span></span></span>計算または <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:0.969438em;vertical-align:-0.286108em;"></span><span class="mord"><span class="mord mathnormal" style="margin-right:0.05764em;">S</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.3361079999999999em;"><span style="top:-2.5500000000000003em;margin-left:-0.05764em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathnormal mtight">ijbh</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.286108em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.2777777777777778em;"></span><span class="mrel">=</span><span class="mspace" style="margin-right:0.2777777777777778em;"></span></span><span class="base"><span class="strut" style="height:1.0497100000000001em;vertical-align:-0.29971000000000003em;"></span><span class="mop"><span class="mop op-symbol small-op" style="position:relative;top:-0.0000050000000000050004em;"></span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.1863979999999999em;"><span style="top:-2.40029em;margin-left:0em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mathnormal mtight">d</span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.29971000000000003em;"><span></span></span></span></span></span></span><span class="mspace" style="margin-right:0.16666666666666666em;"></span><span class="mord"><span class="mord mathnormal">Q</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"><span class="mord mtight"><span class="mord mathnormal mtight">ibh</span><span class="mord mathnormal mtight">d</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 class="mord"><span class="mord mathnormal" style="margin-right:0.07153em;">K</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.3361079999999999em;"><span style="top:-2.5500000000000003em;margin-left:-0.07153em;margin-right:0.05em;"><span class="pstrut" style="height:2.7em;"></span><span class="sizing reset-size6 size3 mtight"><span class="mord mtight"><span class="mord mathnormal mtight">jbh</span><span class="mord mathnormal mtight">d</span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.286108em;"><span></span></span></span></span></span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">129</span> <span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">einsum</span><span class="p">(</span><span class="s1">&#39;ibhd,jbhd-&gt;ijbh&#39;</span><span class="p">,</span> <span class="n">query</span><span class="p">,</span> <span class="n">key</span><span class="p">)</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><code class="highlight"><span></span><span class="n">mask</span></code>
には形状があり<code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len_q</span><span class="p">,</span> <span class="n">seq_len_k</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">]</span></code>
、最初の次元はクエリ次元です。<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 eqj" style=""><span class="mord" style="">1</span></span></span></span></span></span>クエリディメンションがそれと等しい場合はブロードキャストされます</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">131</span> <span class="k">def</span> <span class="nf">prepare_mask</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">mask</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">query_shape</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span> <span class="n">key_shape</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</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">137</span> <span class="k">assert</span> <span class="n">mask</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">==</span> <span class="mi">1</span> <span class="ow">or</span> <span class="n">mask</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">==</span> <span class="n">query_shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="lineno">138</span> <span class="k">assert</span> <span class="n">mask</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">==</span> <span class="n">key_shape</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="lineno">139</span> <span class="k">assert</span> <span class="n">mask</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span> <span class="o">==</span> <span class="mi">1</span> <span class="ow">or</span> <span class="n">mask</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span> <span class="o">==</span> <span class="n">query_shape</span><span class="p">[</span><span class="mi">1</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>すべての頭に同じマスクをかけました。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">142</span> <span class="n">mask</span> <span class="o">=</span> <span class="n">mask</span><span class="o">.</span><span class="n">unsqueeze</span><span class="p">(</span><span class="o">-</span><span class="mi">1</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>生成されるマスクには形状があります <code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len_q</span><span class="p">,</span> <span class="n">seq_len_k</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">heads</span><span class="p">]</span></code>
</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">145</span> <span class="k">return</span> <span class="n">mask</span></pre></div>
</div>
</div>
<div class='section' id='section-28'>
<div class='docs doc-strings'>
<div class='section-link'>
<a href='#section-28'>#</a>
</div>
<p><code class="highlight"><span></span><span class="n">query</span></code>
<code class="highlight"><span></span><span class="n">key</span></code>
<code class="highlight"><span></span><span class="n">value</span></code>
およびは、<em>クエリ</em><em>キー</em><em>および値のベクトルのコレクションを格納するテンソルです</em>。形があります<code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">d_model</span><span class="p">]</span></code>
</p>
<p><code class="highlight"><span></span><span class="n">mask</span></code>
<code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">]</span></code>
形状があり、バッチの場合<code class="highlight"><span></span><span class="n">b</span></code>
<code class="highlight"><span></span><span class="n">mask</span><span class="p">[</span><span class="n">i</span><span class="p">,</span> <span class="n">j</span><span class="p">,</span> <span class="n">b</span><span class="p">]</span></code>
<code class="highlight"><span></span><span class="n">i</span></code>
その位置のクエリがその位置のキー値にアクセスできるかどうかを示します。<code class="highlight"><span></span><span class="n">j</span></code>
</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">147</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="o">*</span><span class="p">,</span>
<span class="lineno">148</span> <span class="n">query</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">149</span> <span class="n">key</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">150</span> <span class="n">value</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">151</span> <span class="n">mask</span><span class="p">:</span> <span class="n">Optional</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="o">=</span> <span class="kc">None</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><code class="highlight"><span></span><span class="n">query</span></code>
<code class="highlight"><span></span><span class="n">key</span></code>
<code class="highlight"><span></span><span class="n">value</span></code>
そして形がある <code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">d_model</span><span class="p">]</span></code>
</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">163</span> <span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">_</span> <span class="o">=</span> <span class="n">query</span><span class="o">.</span><span class="n">shape</span>
<span class="lineno">164</span>
<span class="lineno">165</span> <span class="k">if</span> <span class="n">mask</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="lineno">166</span> <span class="n">mask</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">prepare_mask</span><span class="p">(</span><span class="n">mask</span><span class="p">,</span> <span class="n">query</span><span class="o">.</span><span class="n">shape</span><span class="p">,</span> <span class="n">key</span><span class="o">.</span><span class="n">shape</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><code class="highlight"><span></span><span class="n">query</span></code>
<code class="highlight"><span></span><span class="n">key</span></code>
<code class="highlight"><span></span><span class="n">value</span></code>
注意力計算の準備をして<code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">heads</span><span class="p">,</span> <span class="n">d_k</span><span class="p">]</span></code>
これで形ができあがります。</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">170</span> <span class="n">query</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">query</span><span class="p">(</span><span class="n">query</span><span class="p">)</span>
<span class="lineno">171</span> <span class="n">key</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">key</span><span class="p">(</span><span class="n">key</span><span class="p">)</span>
<span class="lineno">172</span> <span class="n">value</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">value</span><span class="p">(</span><span class="n">value</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><span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.043548em;vertical-align:-0.19444em;"></span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="">Q</span><span class="mord" style=""><span class="mord mathnormal" style="margin-right:0.07153em">K</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.849108em;"><span style="top:-3.063em;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></span></span></span></span></span></span></span></span></span></span></span></span>アテンションスコアを計算します。<code class="highlight"><span></span><span class="p">[</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">heads</span><span class="p">]</span></code>
これにより形状のテンソルが得られます</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">176</span> <span class="n">scores</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">get_scores</span><span class="p">(</span><span class="n">query</span><span class="p">,</span> <span class="n">key</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-32'>
<div class='docs'>
<div class='section-link'>
<a href='#section-32'>#</a>
</div>
<p>スケールスコア <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:1.633028em;vertical-align:-0.538em;"></span><span class="mord coloredeq eqe" style=""><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:1.095028em;"><span style="top:-2.5864385em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord sqrt mtight" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.8622307142857143em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord mtight" style="padding-left:0.833em"><span class="mord mtight" style=""><span class="mord mtight coloredeq eqi" style=""><span class="mord mathnormal mtight" style="">d</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.3448em;"><span style="top:-2.3487714285714287em;margin-left:0em;margin-right:0.07142857142857144em;"><span class="pstrut" style="height:2.5em;"></span><span class="sizing reset-size3 size1 mtight" style=""><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15122857142857138em;"><span></span></span></span></span></span></span></span></span></span><span style="top:-2.8222307142857144em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail mtight" 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:0.17776928571428574em;"><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:-3.446108em;"><span class="pstrut" style="height:3em;"></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 eqg" style="">Q</span><span class="mord mtight coloredeq eqg" style=""><span class="mord mathnormal mtight" style="margin-right:0.07153em">K</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.9270285714285713em;"><span style="top:-2.931em;margin-right:0.07142857142857144em;"><span class="pstrut" style="height:2.5em;"></span><span class="sizing reset-size3 size1 mtight" style=""><span class="mord mtight" style=""></span></span></span></span></span></span></span></span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.538em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">179</span> <span class="n">scores</span> <span class="o">*=</span> <span class="bp">self</span><span class="o">.</span><span class="n">scale</span></pre></div>
</div>
</div>
<div class='section' id='section-33'>
<div class='docs'>
<div class='section-link'>
<a href='#section-33'>#</a>
</div>
<p>マスクを適用</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">182</span> <span class="k">if</span> <span class="n">mask</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="lineno">183</span> <span class="n">scores</span> <span class="o">=</span> <span class="n">scores</span><span class="o">.</span><span class="n">masked_fill</span><span class="p">(</span><span class="n">mask</span> <span class="o">==</span> <span class="mi">0</span><span class="p">,</span> <span class="nb">float</span><span class="p">(</span><span class="s1">&#39;-inf&#39;</span><span class="p">))</span></pre></div>
</div>
</div>
<div class='section' id='section-34'>
<div class='docs'>
<div class='section-link'>
<a href='#section-34'>#</a>
</div>
<p><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 coloredeq eqh" style=""><span class="mord mathnormal" style="">so</span><span class="mord mathnormal" style="margin-right:0.10764em">f</span><span class="mord mathnormal" style="">t</span><span class="mord mathnormal" style="">ma</span><span class="mord mathnormal" style="">x</span></span></span></span></span></span>キーシーケンス次元に沿って注目 <span ><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:3.0000299999999998em;vertical-align:-1.25003em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mop op-limits" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.6944399999999998em;"><span style="top:-2.20556em;margin-left:0em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">se</span><span class="mord mathnormal mtight" style="margin-right:0.03588em">q</span></span></span></span><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span><span class="mop" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqh" style="">so</span><span class="mord mathnormal coloredeq eqh" style="margin-right:0.10764em">f</span><span class="mord mathnormal coloredeq eqh" style="">t</span><span class="mord mathnormal coloredeq eqh" style="">ma</span><span class="mord mathnormal coloredeq eqh" style="">x</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:1.030548em;"><span></span></span></span></span></span></span><span class="mord" style=""><span class="delimsizing size4" style=""><span style="">(</span></span></span><span class="mord" style=""><span class="mord coloredeq eqe" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.095028em;"><span style="top:-2.5864385em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord sqrt mtight" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.8622307142857143em;"><span class="svg-align" style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span class="mord mtight" style="padding-left:0.833em"><span class="mord mtight" style=""><span class="mord mtight coloredeq eqi" style=""><span class="mord mathnormal mtight" style="">d</span><span class="msupsub"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.3448em;"><span style="top:-2.3487714285714287em;margin-left:0em;margin-right:0.07142857142857144em;"><span class="pstrut" style="height:2.5em;"></span><span class="sizing reset-size3 size1 mtight" style=""><span class="mord mathnormal mtight" style="margin-right:0.03148em">k</span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.15122857142857138em;"><span></span></span></span></span></span></span></span></span></span><span style="top:-2.8222307142857144em;"><span class="pstrut" style="height:3em;"></span><span class="hide-tail mtight" 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:0.17776928571428574em;"><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:-3.446108em;"><span class="pstrut" style="height:3em;"></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 eqg" style="">Q</span><span class="mord mtight coloredeq eqg" style=""><span class="mord mathnormal mtight" style="margin-right:0.07153em">K</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.9270285714285713em;"><span style="top:-2.931em;margin-right:0.07142857142857144em;"><span class="pstrut" style="height:2.5em;"></span><span class="sizing reset-size3 size1 mtight" style=""><span class="mord mtight" style=""></span></span></span></span></span></span></span></span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:0.538em;"><span></span></span></span></span></span><span class="mclose nulldelimiter"></span></span></span><span class="mord" style=""><span class="delimsizing size4" style=""><span style="">)</span></span></span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">187</span> <span class="n">attn</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">softmax</span><span class="p">(</span><span class="n">scores</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-35'>
<div class='docs'>
<div class='section-link'>
<a href='#section-35'>#</a>
</div>
<p>デバッグ時の注意事項を保存</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">190</span> <span class="n">tracker</span><span class="o">.</span><span class="n">debug</span><span class="p">(</span><span class="s1">&#39;attn&#39;</span><span class="p">,</span> <span class="n">attn</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-36'>
<div class='docs'>
<div class='section-link'>
<a href='#section-36'>#</a>
</div>
<p>ドロップアウトを適用</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">193</span> <span class="n">attn</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">dropout</span><span class="p">(</span><span class="n">attn</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-37'>
<div class='docs'>
<div class='section-link'>
<a href='#section-37'>#</a>
</div>
<p>値による乗算 <span ><span class="katex-display"><span class="katex"><span aria-hidden="true" class="katex-html"><span class="base"><span class="strut" style="height:3.0000299999999998em;vertical-align:-1.25003em;"></span><span class="mord coloredeq eqc" style=""><span class="mord" style=""><span class="mop op-limits" style=""><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:0.6944399999999998em;"><span style="top:-2.20556em;margin-left:0em;"><span class="pstrut" style="height:3em;"></span><span class="sizing reset-size6 size3 mtight" style=""><span class="mord mtight" style=""><span class="mord mathnormal mtight" style="">se</span><span class="mord mathnormal mtight" style="margin-right:0.03588em">q</span></span></span></span><span style="top:-3em;"><span class="pstrut" style="height:3em;"></span><span><span class="mop" style=""><span class="mord" style=""><span class="mord mathnormal coloredeq eqh" style="">so</span><span class="mord mathnormal coloredeq eqh" style="margin-right:0.10764em">f</span><span class="mord mathnormal coloredeq eqh" style="">t</span><span class="mord mathnormal coloredeq eqh" style="">ma</span><span class="mord mathnormal coloredeq eqh" style="">x</span></span></span></span></span></span><span class="vlist-s"></span></span><span class="vlist-r"><span class="vlist" style="height:1.030548em;"><span></span></span></span></span></span></span><span class="mord" style=""><span class="delimsizing size4" style=""><span style="">(</span></span></span><span class="mord" style=""><span class="mord coloredeq eqe" style=""><span class="mopen nulldelimiter"></span><span class="mfrac"><span class="vlist-t vlist-t2"><span class="vlist-r"><span class="vlist" style="height:1.5261079999999998em;"><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 coloredeq eqi" 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 mathnormal mtight" style="margin-right:0.03148em">k</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 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=""><span class="mord mathnormal coloredeq eqg" style="">Q</span><span class="mord coloredeq eqg" style=""><span class="mord mathnormal" style="margin-right:0.07153em">K</span><span class="msupsub"><span class="vlist-t"><span class="vlist-r"><span class="vlist" style="height:0.849108em;"><span style="top:-3.063em;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></span></span></span></span></span></span></span></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><span class="mord" style=""><span class="delimsizing size4" style=""><span style="">)</span></span></span></span><span class="mord mathnormal" style="margin-right:0.22222em;">V</span></span></span></span></span></span></p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">197</span> <span class="n">x</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">einsum</span><span class="p">(</span><span class="s2">&quot;ijbh,jbhd-&gt;ibhd&quot;</span><span class="p">,</span> <span class="n">attn</span><span class="p">,</span> <span class="n">value</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-38'>
<div class='docs'>
<div class='section-link'>
<a href='#section-38'>#</a>
</div>
<p>他の計算に注意を向けておく</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">200</span> <span class="bp">self</span><span class="o">.</span><span class="n">attn</span> <span class="o">=</span> <span class="n">attn</span><span class="o">.</span><span class="n">detach</span><span class="p">()</span></pre></div>
</div>
</div>
<div class='section' id='section-39'>
<div class='docs'>
<div class='section-link'>
<a href='#section-39'>#</a>
</div>
<p>複数のヘッドを連結</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">203</span> <span class="n">x</span> <span class="o">=</span> <span class="n">x</span><span class="o">.</span><span class="n">reshape</span><span class="p">(</span><span class="n">seq_len</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">)</span></pre></div>
</div>
</div>
<div class='section' id='section-40'>
<div class='docs'>
<div class='section-link'>
<a href='#section-40'>#</a>
</div>
<p>出力レイヤー</p>
</div>
<div class='code'>
<div class="highlight"><pre><span class="lineno">206</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">output</span><span class="p">(</span><span class="n">x</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>