chore: import upstream snapshot with attribution
This commit is contained in:
wehub-resource-sync
@@ -0,0 +1,514 @@
|
||||
<!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="チェックポイントとそれらをロードするヘルパーをダウンロードするコード。"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="GPT-ネオックスチェックポイント"/>
|
||||
<meta name="twitter:description" content="チェックポイントとそれらをロードするヘルパーをダウンロードするコード。"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/checkpoint.html"/>
|
||||
<meta property="og:title" content="GPT-ネオックスチェックポイント"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="GPT-ネオックスチェックポイント"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="GPT-ネオックスチェックポイント"/>
|
||||
<meta property="og:description" content="チェックポイントとそれらをロードするヘルパーをダウンロードするコード。"/>
|
||||
|
||||
<title>GPT-ネオックスチェックポイント</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/checkpoint.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">neox</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/neox/checkpoint.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>GPT-ネオックスチェックポイント</h1>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">11</span><span></span><span class="kn">from</span> <span class="nn">pathlib</span> <span class="kn">import</span> <span class="n">Path</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="p">,</span> <span class="n">Union</span><span class="p">,</span> <span class="n">Tuple</span><span class="p">,</span> <span class="n">Optional</span>
|
||||
<span class="lineno">13</span>
|
||||
<span class="lineno">14</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
||||
<span class="lineno">16</span>
|
||||
<span class="lineno">17</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">monit</span><span class="p">,</span> <span class="n">lab</span><span class="p">,</span> <span class="n">logger</span>
|
||||
<span class="lineno">18</span><span class="kn">from</span> <span class="nn">labml.logger</span> <span class="kn">import</span> <span class="n">Text</span><span class="p">,</span> <span class="n">inspect</span>
|
||||
<span class="lineno">19</span><span class="kn">from</span> <span class="nn">labml.utils.download</span> <span class="kn">import</span> <span class="n">download_file</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-1'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-1'>#</a>
|
||||
</div>
|
||||
<p>保護者の URL</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">22</span><span class="n">CHECKPOINTS_URL</span> <span class="o">=</span> <span class="s1">'https://mystic.the-eye.eu/public/AI/models/GPT-NeoX-20B/slim_weights/'</span>
|
||||
<span class="lineno">23</span>
|
||||
<span class="lineno">24</span><span class="n">_CHECKPOINTS_DOWNLOAD_PATH</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Path</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</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>ダウンロードパス</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">28</span><span class="k">def</span> <span class="nf">get_checkpoints_download_path</span><span class="p">():</span>
|
||||
<span class="lineno">29</span> <span class="k">global</span> <span class="n">_CHECKPOINTS_DOWNLOAD_PATH</span>
|
||||
<span class="lineno">30</span>
|
||||
<span class="lineno">31</span> <span class="k">if</span> <span class="n">_CHECKPOINTS_DOWNLOAD_PATH</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">32</span> <span class="k">return</span> <span class="n">_CHECKPOINTS_DOWNLOAD_PATH</span>
|
||||
<span class="lineno">33</span>
|
||||
<span class="lineno">34</span> <span class="n">_CHECKPOINTS_DOWNLOAD_PATH</span> <span class="o">=</span> <span class="n">lab</span><span class="o">.</span><span class="n">get_data_path</span><span class="p">()</span> <span class="o">/</span> <span class="s1">'neox_fast'</span> <span class="o">/</span> <span class="s1">'slim_weights'</span>
|
||||
<span class="lineno">35</span> <span class="k">if</span> <span class="ow">not</span> <span class="n">_CHECKPOINTS_DOWNLOAD_PATH</span><span class="o">.</span><span class="n">exists</span><span class="p">():</span>
|
||||
<span class="lineno">36</span> <span class="n">_CHECKPOINTS_DOWNLOAD_PATH</span> <span class="o">=</span> <span class="n">lab</span><span class="o">.</span><span class="n">get_data_path</span><span class="p">()</span> <span class="o">/</span> <span class="s1">'neox'</span> <span class="o">/</span> <span class="s1">'slim_weights'</span>
|
||||
<span class="lineno">37</span> <span class="n">inspect</span><span class="p">(</span><span class="n">neox_checkpoint_path</span><span class="o">=</span><span class="n">_CHECKPOINTS_DOWNLOAD_PATH</span><span class="p">)</span>
|
||||
<span class="lineno">38</span>
|
||||
<span class="lineno">39</span> <span class="k">return</span> <span class="n">_CHECKPOINTS_DOWNLOAD_PATH</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-3'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-3'>#</a>
|
||||
</div>
|
||||
<h3>ダウンロードするファイルを取得</h3>
|
||||
<ul><p><em>ダウンロードするファイルのリストを返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">42</span><span class="k">def</span> <span class="nf">get_files_to_download</span><span class="p">(</span><span class="n">n_layers</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">44</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">48</span> <span class="n">layers</span> <span class="o">=</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>埋め込みレイヤー</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">50</span> <span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">+</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>トランスフォーマー層</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">52</span> <span class="nb">list</span><span class="p">(</span><span class="nb">range</span><span class="p">(</span><span class="mi">2</span><span class="p">,</span> <span class="mi">2</span> <span class="o">+</span> <span class="n">n_layers</span><span class="p">))</span> <span class="o">+</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>最終正規化層と読み出し層</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">54</span> <span class="p">[</span><span class="mi">47</span><span class="p">,</span> <span class="mi">48</span><span class="p">]</span>
|
||||
<span class="lineno">55</span> <span class="p">)</span>
|
||||
<span class="lineno">56</span>
|
||||
<span class="lineno">57</span> <span class="k">return</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>語彙と構成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">59</span> <span class="p">[</span><span class="s1">'20B_tokenizer.json'</span><span class="p">,</span> <span class="s1">'configs/20B.yml'</span><span class="p">,</span> <span class="s1">'latest'</span><span class="p">]</span> <span class="o">+</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>レイヤーチェックポイント</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">61</span> <span class="p">[</span><span class="sa">f</span><span class="s1">'global_step150000/layer_</span><span class="si">{</span><span class="n">i</span> <span class="si">:</span><span class="s1">02d</span><span class="si">}</span><span class="s1">-model_</span><span class="si">{</span><span class="n">p</span> <span class="si">:</span><span class="s1">02d</span><span class="si">}</span><span class="s1">-model_states.pt'</span> <span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="n">layers</span> <span class="k">for</span> <span class="n">p</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">2</span><span class="p">)]</span> <span class="o">+</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>空の状態 (未使用)</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">63</span> <span class="p">[</span><span class="sa">f</span><span class="s1">'global_step150000/mp_rank_</span><span class="si">{</span><span class="n">i</span> <span class="si">:</span><span class="s1">02d</span><span class="si">}</span><span class="s1">_model_states.pt'</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">8</span><span class="p">)]</span>
|
||||
<span class="lineno">64</span> <span class="p">)</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>
|
||||
<h2>すべてのチェックポイントファイルをダウンロード</h2>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">67</span><span class="k">def</span> <span class="nf">download</span><span class="p">(</span><span class="n">n_layers</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">44</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>ダウンロードするファイルを取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">73</span> <span class="n">files</span> <span class="o">=</span> <span class="n">get_files_to_download</span><span class="p">(</span><span class="n">n_layers</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>繰り返し</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">76</span> <span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">f</span> <span class="ow">in</span> <span class="n">monit</span><span class="o">.</span><span class="n">enum</span><span class="p">(</span><span class="s1">'Download All'</span><span class="p">,</span> <span class="n">files</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>ログ</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">78</span> <span class="n">logger</span><span class="o">.</span><span class="n">log</span><span class="p">([</span><span class="s1">'Downloading '</span><span class="p">,</span> <span class="p">(</span><span class="sa">f</span><span class="s1">'</span><span class="si">{</span><span class="n">i</span> <span class="o">+</span> <span class="mi">1</span> <span class="si">:</span><span class="s1">3d</span><span class="si">}</span><span class="s1">/</span><span class="si">{</span><span class="nb">len</span><span class="p">(</span><span class="n">files</span><span class="p">)</span><span class="si">}</span><span class="s1">'</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">meta</span><span class="p">),</span> <span class="s1">': '</span><span class="p">,</span> <span class="p">(</span><span class="n">f</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">value</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>[ダウンロード]</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">80</span> <span class="n">download_file</span><span class="p">(</span><span class="n">CHECKPOINTS_URL</span> <span class="o">+</span> <span class="n">f</span><span class="p">,</span> <span class="n">get_checkpoints_download_path</span><span class="p">()</span> <span class="o">/</span> <span class="n">f</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
<h3>1 組のチェックポイントファイルをロードする</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">files</span></code>
|
||||
ロードするファイルのペア</li>
|
||||
</ul><p><em>ロードされたパラメータテンソルを返します</em>。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">83</span><span class="k">def</span> <span class="nf">load_checkpoint_files</span><span class="p">(</span><span class="n">files</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">str</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">90</span> <span class="n">checkpoint_path</span> <span class="o">=</span> <span class="n">get_checkpoints_download_path</span><span class="p">()</span> <span class="o">/</span> <span class="s1">'global_step150000'</span>
|
||||
<span class="lineno">91</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">'Load checkpoint'</span><span class="p">):</span>
|
||||
<span class="lineno">92</span> <span class="n">data</span> <span class="o">=</span> <span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">load</span><span class="p">(</span><span class="n">checkpoint_path</span> <span class="o">/</span> <span class="n">f</span><span class="p">)</span> <span class="k">for</span> <span class="n">f</span> <span class="ow">in</span> <span class="n">files</span><span class="p">]</span>
|
||||
<span class="lineno">93</span>
|
||||
<span class="lineno">94</span> <span class="k">return</span> <span class="n">data</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-18'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-18'>#</a>
|
||||
</div>
|
||||
<h3>1 番目の次元に沿ってパーティションをマージしてパラメータをロードします</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">param</span></code>
|
||||
はパラメータです</li>
|
||||
<li><code class="highlight"><span></span><span class="n">key</span></code>
|
||||
はパラメータの名前です</li>
|
||||
<li><code class="highlight"><span></span><span class="n">p1</span></code>
|
||||
第 1 パーティション辞書</li>
|
||||
<li><code class="highlight"><span></span><span class="n">p2</span></code>
|
||||
2 番目のパーティション辞書</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">97</span><span class="k">def</span> <span class="nf">merge_params_dim_0</span><span class="p">(</span><span class="n">param</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">nn</span><span class="o">.</span><span class="n">Parameter</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="nb">str</span><span class="p">,</span> <span class="n">p1</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">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">],</span>
|
||||
<span class="lineno">98</span> <span class="n">p2</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">torch</span><span class="o">.</span><span class="n">Tensor</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">107</span> <span class="n">w1</span><span class="p">,</span> <span class="n">w2</span> <span class="o">=</span> <span class="n">p1</span><span class="p">[</span><span class="n">key</span><span class="p">],</span> <span class="n">p2</span><span class="p">[</span><span class="n">key</span><span class="p">]</span>
|
||||
<span class="lineno">108</span> <span class="n">param</span><span class="o">.</span><span class="n">data</span><span class="p">[:</span><span class="n">w1</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">w1</span>
|
||||
<span class="lineno">109</span> <span class="n">param</span><span class="o">.</span><span class="n">data</span><span class="p">[</span><span class="n">w1</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">w2</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>
|
||||
<h3>2 番目の次元に沿ってパーティションをマージしてパラメータをロードします</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">param</span></code>
|
||||
はパラメータです</li>
|
||||
<li><code class="highlight"><span></span><span class="n">key</span></code>
|
||||
はパラメータの名前です</li>
|
||||
<li><code class="highlight"><span></span><span class="n">p1</span></code>
|
||||
第 1 パーティション辞書</li>
|
||||
<li><code class="highlight"><span></span><span class="n">p2</span></code>
|
||||
2 番目のパーティション辞書</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">112</span><span class="k">def</span> <span class="nf">merge_params_dim_1</span><span class="p">(</span><span class="n">param</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">nn</span><span class="o">.</span><span class="n">Parameter</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="nb">str</span><span class="p">,</span> <span class="n">p1</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">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">],</span>
|
||||
<span class="lineno">113</span> <span class="n">p2</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">torch</span><span class="o">.</span><span class="n">Tensor</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">122</span> <span class="n">w1</span><span class="p">,</span> <span class="n">w2</span> <span class="o">=</span> <span class="n">p1</span><span class="p">[</span><span class="n">key</span><span class="p">],</span> <span class="n">p2</span><span class="p">[</span><span class="n">key</span><span class="p">]</span>
|
||||
<span class="lineno">123</span> <span class="n">param</span><span class="o">.</span><span class="n">data</span><span class="p">[:,</span> <span class="p">:</span><span class="n">w1</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">w1</span>
|
||||
<span class="lineno">124</span> <span class="n">param</span><span class="o">.</span><span class="n">data</span><span class="p">[:,</span> <span class="n">w1</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">w2</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>パーティション化されていないパラメータをロードする</h3>
|
||||
<p>これにより、両方のパーティションが同じであることを確認するためのサニティチェックが行われます。</p>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">param</span></code>
|
||||
はパラメータです</li>
|
||||
<li><code class="highlight"><span></span><span class="n">key</span></code>
|
||||
はパラメータの名前です</li>
|
||||
<li><code class="highlight"><span></span><span class="n">p1</span></code>
|
||||
第 1 パーティション辞書</li>
|
||||
<li><code class="highlight"><span></span><span class="n">p2</span></code>
|
||||
2 番目のパーティション辞書</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">127</span><span class="k">def</span> <span class="nf">merge_params_duplicate</span><span class="p">(</span><span class="n">param</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">nn</span><span class="o">.</span><span class="n">Parameter</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="nb">str</span><span class="p">,</span> <span class="n">p1</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">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">],</span>
|
||||
<span class="lineno">128</span> <span class="n">p2</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">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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">139</span> <span class="n">w1</span><span class="p">,</span> <span class="n">w2</span> <span class="o">=</span> <span class="n">p1</span><span class="p">[</span><span class="n">key</span><span class="p">],</span> <span class="n">p2</span><span class="p">[</span><span class="n">key</span><span class="p">]</span>
|
||||
<span class="lineno">140</span>
|
||||
<span class="lineno">141</span> <span class="n">diff</span> <span class="o">=</span> <span class="nb">sum</span><span class="p">((</span><span class="n">w1</span> <span class="o">-</span> <span class="n">w2</span><span class="p">)</span> <span class="o">**</span> <span class="mi">2</span><span class="p">)</span><span class="o">.</span><span class="n">item</span><span class="p">()</span>
|
||||
<span class="lineno">142</span> <span class="k">assert</span> <span class="n">diff</span> <span class="o"><</span> <span class="mf">1e-4</span><span class="p">,</span> <span class="sa">f</span><span class="s1">'The partitions do not match: </span><span class="si">{</span><span class="n">key</span><span class="si">}</span><span class="s1">'</span>
|
||||
<span class="lineno">143</span>
|
||||
<span class="lineno">144</span> <span class="n">param</span><span class="o">.</span><span class="n">data</span><span class="p">[:]</span> <span class="o">=</span> <span class="p">(</span><span class="n">w1</span> <span class="o">+</span> <span class="n">w2</span><span class="p">)</span> <span class="o">/</span> <span class="mf">2.</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>
|
||||
<h3>分割された負荷バイアスがリデュース時に追加される</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">param</span></code>
|
||||
はパラメータです</li>
|
||||
<li><code class="highlight"><span></span><span class="n">key</span></code>
|
||||
はパラメータの名前です</li>
|
||||
<li><code class="highlight"><span></span><span class="n">p1</span></code>
|
||||
第 1 パーティション辞書</li>
|
||||
<li><code class="highlight"><span></span><span class="n">p2</span></code>
|
||||
2 番目のパーティション辞書</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">147</span><span class="k">def</span> <span class="nf">merge_params_sum</span><span class="p">(</span><span class="n">param</span><span class="p">:</span> <span class="n">Union</span><span class="p">[</span><span class="n">nn</span><span class="o">.</span><span class="n">Parameter</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="nb">str</span><span class="p">,</span> <span class="n">p1</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">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">],</span>
|
||||
<span class="lineno">148</span> <span class="n">p2</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">torch</span><span class="o">.</span><span class="n">Tensor</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">157</span> <span class="n">w1</span><span class="p">,</span> <span class="n">w2</span> <span class="o">=</span> <span class="n">p1</span><span class="p">[</span><span class="n">key</span><span class="p">],</span> <span class="n">p2</span><span class="p">[</span><span class="n">key</span><span class="p">]</span>
|
||||
<span class="lineno">158</span>
|
||||
<span class="lineno">159</span> <span class="n">param</span><span class="o">.</span><span class="n">data</span><span class="p">[:]</span> <span class="o">=</span> <span class="n">w1</span> <span class="o">+</span> <span class="n">w2</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">163</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">164</span> <span class="n">download</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>
|
||||
@@ -0,0 +1,228 @@
|
||||
<!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="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta name="twitter:description" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/evaluation/half_precision.html"/>
|
||||
<meta property="og:title" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta property="og:description" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
|
||||
<title>テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/evaluation/half_precision.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">neox</a>
|
||||
<a class="parent" href="index.html">evaluation</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/neox/evaluation/half_precision.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>テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する</h1>
|
||||
<p>このコードでは、<a href="../index.html">一連のタスクで GPT-Neox</a> を使用して評価します。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">12</span><span></span><span class="kn">import</span> <span class="nn">argparse</span>
|
||||
<span class="lineno">13</span>
|
||||
<span class="lineno">14</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
||||
<span class="lineno">16</span>
|
||||
<span class="lineno">17</span><span class="kn">from</span> <span class="nn">labml_nn.neox.evaluation</span> <span class="kn">import</span> <span class="n">run_eval_harness</span>
|
||||
<span class="lineno">18</span><span class="kn">from</span> <span class="nn">labml_nn.neox.model</span> <span class="kn">import</span> <span class="n">LayerGenerator</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">21</span><span class="k">def</span> <span class="nf">main</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>引数パーサー</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">23</span> <span class="n">parser</span> <span class="o">=</span> <span class="n">argparse</span><span class="o">.</span><span class="n">ArgumentParser</span><span class="p">()</span>
|
||||
<span class="lineno">24</span>
|
||||
<span class="lineno">25</span> <span class="n">parser</span><span class="o">.</span><span class="n">add_argument</span><span class="p">(</span><span class="s2">"--flash"</span><span class="p">,</span> <span class="n">action</span><span class="o">=</span><span class="s1">'store_true'</span><span class="p">,</span> <span class="n">help</span><span class="o">=</span><span class="s2">"whether to use Flash Attention"</span><span class="p">)</span>
|
||||
<span class="lineno">26</span>
|
||||
<span class="lineno">27</span> <span class="n">opt</span> <span class="o">=</span> <span class="n">parser</span><span class="o">.</span><span class="n">parse_args</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>端末</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">30</span> <span class="n">device</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cuda:0'</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>レイヤーをロード</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">32</span> <span class="n">layers</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">LayerGenerator</span><span class="p">(</span><span class="n">is_clone_layers</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||||
<span class="lineno">33</span> <span class="n">filter_layers</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
|
||||
<span class="lineno">34</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">float16</span><span class="p">,</span>
|
||||
<span class="lineno">35</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span>
|
||||
<span class="lineno">36</span> <span class="n">is_flash_attention</span><span class="o">=</span><span class="n">opt</span><span class="o">.</span><span class="n">flash</span><span class="p">,</span>
|
||||
<span class="lineno">37</span> <span class="p">)</span><span class="o">.</span><span class="n">load</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><code class="highlight"><span></span><span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span></code>
|
||||
モデル作成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">40</span> <span class="n">model</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span><span class="o">*</span><span class="n">layers</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><a href="index.html">評価用ハーネスを実行</a></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">43</span> <span class="nb">print</span><span class="p">(</span><span class="n">run_eval_harness</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="s1">'half_precision'</span><span class="p">,</span> <span class="p">[</span><span class="s1">'lambada'</span><span class="p">],</span> <span class="n">device</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">47</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">48</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>
|
||||
@@ -0,0 +1,950 @@
|
||||
<!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="LM 評価ハーネスを通じて NLP タスクでモデルを評価するコード"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="評価"/>
|
||||
<meta name="twitter:description" content="LM 評価ハーネスを通じて NLP タスクでモデルを評価するコード"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/evaluation/index.html"/>
|
||||
<meta property="og:title" content="評価"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="評価"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="評価"/>
|
||||
<meta property="og:description" content="LM 評価ハーネスを通じて NLP タスクでモデルを評価するコード"/>
|
||||
|
||||
<title>評価</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/evaluation/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">neox</a>
|
||||
<a class="parent" href="index.html">evaluation</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/neox/evaluation/__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>評価</h1>
|
||||
<p><a href="https://github.com/EleutherAI/lm-evaluation-harness">Eleutherai/LM-評価ハーネスでモデルをテストするためのコードです</a>。</p>
|
||||
<ul><li><a href="half_precision.html">単一の GPU での半精度モデルの評価</a></li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">15</span><span></span><span class="kn">import</span> <span class="nn">math</span>
|
||||
<span class="lineno">16</span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">List</span>
|
||||
<span class="lineno">17</span>
|
||||
<span class="lineno">18</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">19</span><span class="kn">import</span> <span class="nn">torch.nn.functional</span> <span class="k">as</span> <span class="nn">F</span>
|
||||
<span class="lineno">20</span><span class="kn">from</span> <span class="nn">lm_eval</span> <span class="kn">import</span> <span class="n">tasks</span><span class="p">,</span> <span class="n">evaluator</span><span class="p">,</span> <span class="n">utils</span>
|
||||
<span class="lineno">21</span><span class="kn">from</span> <span class="nn">lm_eval.base</span> <span class="kn">import</span> <span class="n">BaseLM</span>
|
||||
<span class="lineno">22</span><span class="kn">from</span> <span class="nn">tokenizers</span> <span class="kn">import</span> <span class="n">Tokenizer</span>
|
||||
<span class="lineno">23</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
||||
<span class="lineno">24</span><span class="kn">from</span> <span class="nn">tqdm</span> <span class="kn">import</span> <span class="n">tqdm</span>
|
||||
<span class="lineno">25</span>
|
||||
<span class="lineno">26</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">monit</span>
|
||||
<span class="lineno">27</span><span class="kn">from</span> <span class="nn">labml_nn.neox.tokenizer</span> <span class="kn">import</span> <span class="n">get_tokenizer</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>評価用ハーネスアダプター</h2>
|
||||
<p><a href="https://github.com/EleutherAI/gpt-neox/blob/main/eval_tasks/eval_adapter.py">これはElutherai/GPT-Neoxのアダプターをベースにしています</a></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">30</span><span class="k">class</span> <span class="nc">EvalHarnessAdapter</span><span class="p">(</span><span class="n">BaseLM</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>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">tokenizer</span></code>
|
||||
<a href="huggingface/tokenizers">ハギングフェイストークナイザーです</a></li>
|
||||
<li><code class="highlight"><span></span><span class="n">vocab_size</span></code>
|
||||
はボキャブラリのサイズです (これはトークナイザーのボキャブサイズとは異なります。neoxは埋め込み層モデルを並列化するための追加機能を追加しているからです)。</li>
|
||||
<li><code class="highlight"><span></span><span class="n">batch_size</span></code>
|
||||
はバッチサイズ</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">37</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">tokenizer</span><span class="p">:</span> <span class="n">Tokenizer</span><span class="p">,</span> <span class="n">vocab_size</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">:</span> <span class="nb">int</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">45</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">46</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span> <span class="o">=</span> <span class="n">tokenizer</span>
|
||||
<span class="lineno">47</span> <span class="bp">self</span><span class="o">.</span><span class="n">_eot_token_id</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="o">.</span><span class="n">token_to_id</span><span class="p">(</span><span class="s2">"<|endoftext|>"</span><span class="p">)</span>
|
||||
<span class="lineno">48</span> <span class="bp">self</span><span class="o">.</span><span class="n">_vocab_size</span> <span class="o">=</span> <span class="n">vocab_size</span>
|
||||
<span class="lineno">49</span>
|
||||
<span class="lineno">50</span> <span class="bp">self</span><span class="o">.</span><span class="n">_batch_size</span> <span class="o">=</span> <span class="n">batch_size</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>
|
||||
<p>ボキャブラリーのサイズ</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">52</span> <span class="nd">@property</span>
|
||||
<span class="lineno">53</span> <span class="k">def</span> <span class="nf">device</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||||
<span class="lineno">54</span> <span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">()</span>
|
||||
<span class="lineno">55</span>
|
||||
<span class="lineno">56</span> <span class="nd">@property</span>
|
||||
<span class="lineno">57</span> <span class="k">def</span> <span class="nf">vocab_size</span><span class="p">(</span><span class="bp">self</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">59</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_vocab_size</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>
|
||||
<p>テキスト終了トークン</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">61</span> <span class="nd">@property</span>
|
||||
<span class="lineno">62</span> <span class="k">def</span> <span class="nf">eot_token_id</span><span class="p">(</span><span class="bp">self</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">64</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_eot_token_id</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>
|
||||
<p>最大シーケンス長</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">66</span> <span class="nd">@property</span>
|
||||
<span class="lineno">67</span> <span class="k">def</span> <span class="nf">max_length</span><span class="p">(</span><span class="bp">self</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">69</span> <span class="k">return</span> <span class="mi">2048</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>
|
||||
<p>生成するトークンの最大数</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">71</span> <span class="nd">@property</span>
|
||||
<span class="lineno">72</span> <span class="k">def</span> <span class="nf">max_gen_toks</span><span class="p">(</span><span class="bp">self</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">74</span> <span class="k">return</span> <span class="mi">128</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>
|
||||
<p>バッチサイズ</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">76</span> <span class="nd">@property</span>
|
||||
<span class="lineno">77</span> <span class="k">def</span> <span class="nf">batch_size</span><span class="p">(</span><span class="bp">self</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">81</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_batch_size</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-14'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
<p>与えられたテキストをエンコードする</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">83</span> <span class="k">def</span> <span class="nf">tok_encode</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">string</span><span class="p">:</span> <span class="nb">str</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">87</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="o">.</span><span class="n">encode</span><span class="p">(</span><span class="n">string</span><span class="p">)</span><span class="o">.</span><span class="n">ids</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
<p>トークン ID からテキストをデコード</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">89</span> <span class="k">def</span> <span class="nf">tok_decode</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">tokens</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-17'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-17'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">93</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokenizer</span><span class="o">.</span><span class="n">decode</span><span class="p">(</span><span class="n">tokens</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">95</span> <span class="k">def</span> <span class="nf">_model_call</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">inps</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">96</span> <span class="k">raise</span> <span class="ne">NotImplementedError</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">98</span> <span class="k">def</span> <span class="nf">_model_generate</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">context</span><span class="p">,</span> <span class="n">max_length</span><span class="p">,</span> <span class="n">eos_token_id</span><span class="p">):</span>
|
||||
<span class="lineno">99</span> <span class="k">raise</span> <span class="ne">RuntimeError</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">greedy_until</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">requests</span><span class="p">):</span>
|
||||
<span class="lineno">102</span> <span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">()</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>
|
||||
<h3>次のトークンの対数発生確率を取得</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">requests</span></code>
|
||||
コンテキストと予想される継続を含むリクエストのリスト。</li>
|
||||
</ul><li><code class="highlight"><span></span><span class="n">disable_tqdm</span></code>
|
||||
True の場合、tqdm プログレスバーを無効にします。</li>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">104</span> <span class="nd">@torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">()</span>
|
||||
<span class="lineno">105</span> <span class="k">def</span> <span class="nf">_loglikelihood_tokens</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">requests</span><span class="p">,</span> <span class="n">disable_tqdm</span><span class="o">=</span><span class="kc">False</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>結果について</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">114</span> <span class="n">res</span> <span class="o">=</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>同じ長さのシーケンスが近くなるように、リクエストを長さの降順に並べ替えます</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">118</span> <span class="k">def</span> <span class="nf">_collate</span><span class="p">(</span><span class="n">x</span><span class="p">):</span>
|
||||
<span class="lineno">119</span> <span class="n">toks</span> <span class="o">=</span> <span class="n">x</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">+</span> <span class="n">x</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span>
|
||||
<span class="lineno">120</span> <span class="k">return</span> <span class="o">-</span><span class="nb">len</span><span class="p">(</span><span class="n">toks</span><span class="p">),</span> <span class="nb">tuple</span><span class="p">(</span><span class="n">toks</span><span class="p">)</span>
|
||||
<span class="lineno">121</span>
|
||||
<span class="lineno">122</span> <span class="n">reord</span> <span class="o">=</span> <span class="n">utils</span><span class="o">.</span><span class="n">Reorderer</span><span class="p">(</span><span class="n">requests</span><span class="p">,</span> <span class="n">_collate</span><span class="p">)</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><code class="highlight"><span></span><span class="n">batch_size</span></code>
|
||||
一度に複数のリクエストがあるリクエストをループスルーする</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">125</span> <span class="k">for</span> <span class="n">chunk</span> <span class="ow">in</span> <span class="n">utils</span><span class="o">.</span><span class="n">chunks</span><span class="p">(</span><span class="n">tqdm</span><span class="p">(</span><span class="n">reord</span><span class="o">.</span><span class="n">get_reordered</span><span class="p">(),</span> <span class="n">disable</span><span class="o">=</span><span class="n">disable_tqdm</span><span class="p">),</span> <span class="bp">self</span><span class="o">.</span><span class="n">batch_size</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>
|
||||
<p>バッチの入力を保存するには</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">127</span> <span class="n">inps</span> <span class="o">=</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>バッチの継続</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">129</span> <span class="n">continuations</span> <span class="o">=</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>入力シーケンスの長さ</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">131</span> <span class="n">inplens</span> <span class="o">=</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>バッチ用のパッド入り長さ</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">133</span> <span class="n">padded_length</span> <span class="o">=</span> <span class="kc">None</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>チャンク内の各リクエストをループ処理し、パディング付きの PyTorch テンソルにまとめます。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">135</span> <span class="k">for</span> <span class="n">_</span><span class="p">,</span> <span class="n">context_enc</span><span class="p">,</span> <span class="n">continuation_enc</span> <span class="ow">in</span> <span class="n">chunk</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>コンテキストと続きを連結する</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">137</span> <span class="n">inp</span> <span class="o">=</span> <span class="n">context_enc</span> <span class="o">+</span> <span class="n">continuation_enc</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>サイズが <code class="highlight"><span></span><span class="n">max_length</span></code>
|
||||
</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">139</span> <span class="n">inp</span> <span class="o">=</span> <span class="n">inp</span><span class="p">[</span><span class="o">-</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">max_length</span> <span class="o">+</span> <span class="mi">1</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>最終トークンを削除</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">141</span> <span class="n">inp</span> <span class="o">=</span> <span class="n">inp</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-33'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-33'>#</a>
|
||||
</div>
|
||||
<p>テンソルの作成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">143</span> <span class="n">inp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="n">inp</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></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-34'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-34'>#</a>
|
||||
</div>
|
||||
<p>入力長さ</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">145</span> <span class="n">inplen</span> <span class="o">=</span> <span class="n">inp</span><span class="o">.</span><span class="n">shape</span><span class="p">[</span><span class="mi">0</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>パッドの長さを決めてください。短いシーケンスはパディングされます</p>。
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">149</span> <span class="k">if</span> <span class="n">padded_length</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">150</span> <span class="n">padded_length</span> <span class="o">=</span> <span class="nb">int</span><span class="p">(</span><span class="n">math</span><span class="o">.</span><span class="n">ceil</span><span class="p">(</span><span class="n">inplen</span> <span class="o">/</span> <span class="mi">32</span><span class="p">))</span> <span class="o">*</span> <span class="mi">32</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>padded_length = padded_length が Padded_length でない場合はパディングされた_length、それ以外はプレンなし</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-37'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-37'>#</a>
|
||||
</div>
|
||||
<p>パディング</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">154</span> <span class="n">padding</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">(</span><span class="n">padded_length</span> <span class="o">-</span> <span class="n">inplen</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></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-38'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-38'>#</a>
|
||||
</div>
|
||||
<p>パディングを追加</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">157</span> <span class="n">inp</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cat</span><span class="p">([</span><span class="n">inp</span><span class="p">,</span> <span class="n">padding</span><span class="p">],</span> <span class="n">dim</span><span class="o">=</span><span class="mi">0</span><span class="p">)</span>
|
||||
<span class="lineno">158</span>
|
||||
<span class="lineno">159</span> <span class="n">inps</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">inp</span><span class="p">)</span>
|
||||
<span class="lineno">160</span> <span class="n">continuations</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">continuation_enc</span><span class="p">)</span>
|
||||
<span class="lineno">161</span> <span class="n">inplens</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">inplen</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>モデルロジットを取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">164</span> <span class="n">logits</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">_model_call</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">stack</span><span class="p">(</span><span class="n">inps</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>ログソフトマックスを取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">167</span> <span class="n">multi_logits</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">log_softmax</span><span class="p">(</span><span class="n">logits</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-41'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-41'>#</a>
|
||||
</div>
|
||||
<p>バッチの入力と出力のペアをループ処理します</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">170</span> <span class="k">for</span> <span class="n">logits</span><span class="p">,</span> <span class="n">inplen</span><span class="p">,</span> <span class="n">cont_toks</span> <span class="ow">in</span> <span class="nb">zip</span><span class="p">(</span><span class="n">multi_logits</span><span class="p">,</span> <span class="n">inplens</span><span class="p">,</span> <span class="n">continuations</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-42'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-42'>#</a>
|
||||
</div>
|
||||
<p>予測トークンの数を取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">172</span> <span class="n">contlen</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">cont_toks</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-43'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-43'>#</a>
|
||||
</div>
|
||||
<p>それらのログを取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">174</span> <span class="n">logits</span> <span class="o">=</span> <span class="n">logits</span><span class="p">[</span><span class="n">inplen</span> <span class="o">-</span> <span class="n">contlen</span><span class="p">:</span> <span class="n">inplen</span><span class="p">]</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-44'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-44'>#</a>
|
||||
</div>
|
||||
<p>最も確率の高いトークンを手に入れよう</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">176</span> <span class="n">greedy_tokens</span> <span class="o">=</span> <span class="n">logits</span><span class="o">.</span><span class="n">argmax</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-45'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-45'>#</a>
|
||||
</div>
|
||||
<p>対象トークンを取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">178</span> <span class="n">cont_toks</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="n">cont_toks</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="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">logits</span><span class="o">.</span><span class="n">device</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-46'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-46'>#</a>
|
||||
</div>
|
||||
<p>完全に一致するかどうか</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">180</span> <span class="n">max_equal</span> <span class="o">=</span> <span class="p">(</span><span class="n">greedy_tokens</span> <span class="o">==</span> <span class="n">cont_toks</span><span class="p">)</span><span class="o">.</span><span class="n">all</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-47'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-47'>#</a>
|
||||
</div>
|
||||
<p>対象トークンの対数発生可能性</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">182</span> <span class="n">logits</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">gather</span><span class="p">(</span><span class="n">logits</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="n">cont_toks</span><span class="p">[:,</span> <span class="kc">None</span><span class="p">])</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-48'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-48'>#</a>
|
||||
</div>
|
||||
<p>対数推定値の合計と、一致したかどうかを結果に加算します。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">184</span> <span class="n">res</span><span class="o">.</span><span class="n">append</span><span class="p">((</span><span class="nb">float</span><span class="p">(</span><span class="n">logits</span><span class="o">.</span><span class="n">sum</span><span class="p">()),</span> <span class="nb">bool</span><span class="p">(</span><span class="n">max_equal</span><span class="p">)))</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-49'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-49'>#</a>
|
||||
</div>
|
||||
<p>並べ替えて結果を返す</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">187</span> <span class="k">return</span> <span class="n">reord</span><span class="o">.</span><span class="n">get_original</span><span class="p">(</span><span class="n">res</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-50'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-50'>#</a>
|
||||
</div>
|
||||
<h3>与えられた評価を実行</h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">189</span> <span class="nd">@torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">()</span>
|
||||
<span class="lineno">190</span> <span class="k">def</span> <span class="nf">run_eval</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">eval_tasks</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">str</span><span class="p">]):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-51'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-51'>#</a>
|
||||
</div>
|
||||
<p><a href="https://github.com/EleutherAI/lm-evaluation-harness">EleutherAI/LM-評価-ハーネスエバリュエーターを実行</a></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">196</span> <span class="n">results</span> <span class="o">=</span> <span class="n">evaluator</span><span class="o">.</span><span class="n">evaluate</span><span class="p">(</span><span class="n">lm</span><span class="o">=</span><span class="bp">self</span><span class="p">,</span> <span class="n">task_dict</span><span class="o">=</span><span class="n">tasks</span><span class="o">.</span><span class="n">get_task_dict</span><span class="p">(</span><span class="n">eval_tasks</span><span class="p">))</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-52'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-52'>#</a>
|
||||
</div>
|
||||
<p>構成を追加</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">199</span> <span class="n">results</span><span class="p">[</span><span class="s2">"config"</span><span class="p">]</span> <span class="o">=</span> <span class="p">{</span>
|
||||
<span class="lineno">200</span> <span class="s2">"name"</span><span class="p">:</span> <span class="n">name</span><span class="p">,</span>
|
||||
<span class="lineno">201</span> <span class="p">}</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-53'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-53'>#</a>
|
||||
</div>
|
||||
<p></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">204</span> <span class="k">return</span> <span class="n">results</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-54'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-54'>#</a>
|
||||
</div>
|
||||
<h2>評価用ハーネスアダプター</h2>
|
||||
<p><a href="https://github.com/EleutherAI/gpt-neox/blob/main/eval_tasks/eval_adapter.py">これはElutherai/GPT-Neoxのアダプターをベースにしています</a></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">207</span><span class="k">class</span> <span class="nc">NoeXEvalHarnessAdapter</span><span class="p">(</span><span class="n">EvalHarnessAdapter</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-55'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-55'>#</a>
|
||||
</div>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">model</span></code>
|
||||
モデルです</li>
|
||||
<li><code class="highlight"><span></span><span class="n">tokenizer</span></code>
|
||||
<a href="huggingface/tokenizers">ハギングフェイストークナイザーです</a></li>
|
||||
<li><code class="highlight"><span></span><span class="n">vocab_size</span></code>
|
||||
はボキャブラリのサイズです (これはトークナイザーのボキャブサイズとは異なります。neoxは埋め込み層モデルを並列化するための追加機能を追加しているからです)。</li>
|
||||
<li><code class="highlight"><span></span><span class="n">batch_size</span></code>
|
||||
はバッチサイズ</li>
|
||||
<li><code class="highlight"><span></span><span class="n">device</span></code>
|
||||
モデルのデバイスです</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">214</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">model</span><span class="p">:</span> <span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">,</span> <span class="n">tokenizer</span><span class="p">:</span> <span class="n">Tokenizer</span><span class="p">,</span> <span class="n">vocab_size</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</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="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-56'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-56'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">224</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">tokenizer</span><span class="p">,</span> <span class="n">vocab_size</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">)</span>
|
||||
<span class="lineno">225</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span> <span class="o">=</span> <span class="n">model</span>
|
||||
<span class="lineno">226</span> <span class="bp">self</span><span class="o">.</span><span class="n">_device</span> <span class="o">=</span> <span class="n">device</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-57'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-57'>#</a>
|
||||
</div>
|
||||
<p>モデルに電話する</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">228</span> <span class="k">def</span> <span class="nf">_model_call</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">inps</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-58'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-58'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">232</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="p">(</span><span class="n">inps</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-59'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-59'>#</a>
|
||||
</div>
|
||||
<h2>特定のモデルで評価用ハーネスを実行</h2>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">235</span><span class="k">def</span> <span class="nf">run_eval_harness</span><span class="p">(</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="p">,</span> <span class="n">name</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">eval_tasks</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">str</span><span class="p">],</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="p">,</span> <span class="n">batch_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">8</span><span class="p">):</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-60'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-60'>#</a>
|
||||
</div>
|
||||
<p>トークナイザーをロード</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">241</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">'Load tokenizer'</span><span class="p">):</span>
|
||||
<span class="lineno">242</span> <span class="n">tokenizer</span> <span class="o">=</span> <span class="n">get_tokenizer</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-61'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-61'>#</a>
|
||||
</div>
|
||||
<p>何も指定されていない場合はすべてのタスク</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">245</span> <span class="k">if</span> <span class="ow">not</span> <span class="n">eval_tasks</span><span class="p">:</span>
|
||||
<span class="lineno">246</span> <span class="n">eval_tasks</span> <span class="o">=</span> <span class="p">[</span>
|
||||
<span class="lineno">247</span> <span class="s2">"anli_r1"</span><span class="p">,</span>
|
||||
<span class="lineno">248</span> <span class="s2">"anli_r2"</span><span class="p">,</span>
|
||||
<span class="lineno">249</span> <span class="s2">"anli_r3"</span><span class="p">,</span>
|
||||
<span class="lineno">250</span> <span class="s2">"hellaswag"</span><span class="p">,</span>
|
||||
<span class="lineno">251</span> <span class="s2">"lambada"</span><span class="p">,</span>
|
||||
<span class="lineno">252</span> <span class="s2">"piqa"</span><span class="p">,</span>
|
||||
<span class="lineno">253</span> <span class="s2">"winogrande"</span><span class="p">,</span>
|
||||
<span class="lineno">254</span> <span class="s2">"wsc"</span><span class="p">,</span>
|
||||
<span class="lineno">255</span> <span class="s2">"mathqa"</span><span class="p">,</span>
|
||||
<span class="lineno">256</span> <span class="p">]</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-62'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-62'>#</a>
|
||||
</div>
|
||||
<p>アダプターの作成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">259</span> <span class="n">adapter</span> <span class="o">=</span> <span class="n">NoeXEvalHarnessAdapter</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">tokenizer</span><span class="p">,</span> <span class="mi">50_432</span><span class="p">,</span> <span class="n">batch_size</span><span class="p">,</span> <span class="n">device</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-63'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-63'>#</a>
|
||||
</div>
|
||||
<p>実行</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">262</span> <span class="k">return</span> <span class="n">adapter</span><span class="o">.</span><span class="n">run_eval</span><span class="p">(</span><span class="n">name</span><span class="p">,</span> <span class="n">eval_tasks</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>
|
||||
@@ -0,0 +1,239 @@
|
||||
<!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="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta name="twitter:description" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/evaluation/llm_int8.html"/>
|
||||
<meta property="og:title" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
<meta property="og:description" content="テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する"/>
|
||||
|
||||
<title>テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/evaluation/llm_int8.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">neox</a>
|
||||
<a class="parent" href="index.html">evaluation</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/neox/evaluation/llm_int8.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>テストスイートで llm.int8 () 量子化を使用して GPT-Neox を評価する</h1>
|
||||
<p>このコードは、一連のタスクで <a href="../utils/llm_int8.html">llm.int8 () 量子化を使用して</a> <a href="../index.html">GPT-Neox</a> を評価します。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">14</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
||||
<span class="lineno">16</span>
|
||||
<span class="lineno">17</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">monit</span>
|
||||
<span class="lineno">18</span><span class="kn">from</span> <span class="nn">labml_nn.neox.evaluation</span> <span class="kn">import</span> <span class="n">run_eval_harness</span>
|
||||
<span class="lineno">19</span><span class="kn">from</span> <span class="nn">labml_nn.neox.model</span> <span class="kn">import</span> <span class="n">LayerGenerator</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">22</span><span class="k">def</span> <span class="nf">main</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>端末</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">24</span> <span class="n">device</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cuda:0'</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>float16 のレイヤーを CPU にロードします。レイヤーをGPUにロードした後にその場でこれを行うと、CUDAメモリの断片化が発生するため、後でレイヤーをint8に変換します（フラグメンテーションにより約3GBのメモリが失われる可能性があります</p>）。
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">29</span> <span class="n">layer_generator</span> <span class="o">=</span> <span class="n">LayerGenerator</span><span class="p">(</span><span class="n">is_clone_layers</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||||
<span class="lineno">30</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">float16</span><span class="p">,</span>
|
||||
<span class="lineno">31</span> <span class="n">device</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cpu'</span><span class="p">),</span>
|
||||
<span class="lineno">32</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>レイヤーをロード</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">34</span> <span class="n">layers</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">layer_generator</span><span class="o">.</span><span class="n">load</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>これにより、CUDA メモリの断片化が減少します。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">37</span> <span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">monit</span><span class="o">.</span><span class="n">iterate</span><span class="p">(</span><span class="s1">'Convert to int8'</span><span class="p">,</span> <span class="n">layers</span><span class="p">,</span> <span class="n">is_children_silent</span><span class="o">=</span><span class="kc">True</span><span class="p">):</span>
|
||||
<span class="lineno">38</span> <span class="n">layer_generator</span><span class="o">.</span><span class="n">post_load_prepare</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span>
|
||||
<span class="lineno">39</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span>
|
||||
<span class="lineno">40</span> <span class="n">is_llm_int8</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||||
<span class="lineno">41</span> <span class="n">llm_int8_threshold</span><span class="o">=</span><span class="mf">6.0</span><span class="p">,</span>
|
||||
<span class="lineno">42</span> <span class="p">)</span>
|
||||
<span class="lineno">43</span> <span class="n">layer</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</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><code class="highlight"><span></span><span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span></code>
|
||||
モデル作成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">46</span> <span class="n">model</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span><span class="o">*</span><span class="n">layers</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><a href="index.html">評価用ハーネスを実行</a></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">49</span> <span class="nb">print</span><span class="p">(</span><span class="n">run_eval_harness</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="s1">'half_precision'</span><span class="p">,</span> <span class="p">[],</span> <span class="n">device</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">53</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">54</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>
|
||||
@@ -0,0 +1,140 @@
|
||||
<!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="シンプルな GPT ネオックスの実装"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="GPT ネオックス"/>
|
||||
<meta name="twitter:description" content="シンプルな GPT ネオックスの実装"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/index.html"/>
|
||||
<meta property="og:title" content="GPT ネオックス"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="GPT ネオックス"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="GPT ネオックス"/>
|
||||
<meta property="og:description" content="シンプルな GPT ネオックスの実装"/>
|
||||
|
||||
<title>GPT ネオックス</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/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">neox</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/neox/__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>GPT ネオックス</h1>
|
||||
<p>これは、<a href="https://papers.labml.ai/paper/2204.06745">推論と微調整のためのEleuther GPT-Neoxの簡単な実装です</a>。</p>
|
||||
<ul><li><a href="model.html">モデル定義</a></li>
|
||||
<li><a href="tokenizer.html">トークナイザー</a></li>
|
||||
<li><a href="checkpoint.html">チェックポイントのダウンロードと読み込みヘルパー</a></li>
|
||||
<li><a href="utils/index.html">ユーティリティ</a></li>
|
||||
<li><a href="utils/llm_int8.html">llm.int8 () 量子化</a></li></ul>
|
||||
<h3><a href="samples/__init__.py">[サンプル]</a></h3>
|
||||
<ul><li><a href="samples/generate.html">テキストの生成</a></li>
|
||||
<li><a href="samples/finetune.html">パイプラインパラレルによるバイアスの微調整</a></li>
|
||||
<li><a href="samples/llm_int8.html">LLM.int8 () を使用してテキストを生成する</a></li></ul>
|
||||
<h3><a href="evaluation/__init__.py">評価</a></h3>
|
||||
<ul><li><a href="evaluation/half_precision.html">単一の GPU での半精度モデルの評価</a></li>
|
||||
<li><a href="evaluation/llm_int8.html">LLM.int8 () モデルの評価中</a></li></ul>
|
||||
<p><strong><a href="https://www.eleuther.ai">Eleuther GPT-noexの公式ソースコードはeleutherai/gpt-neoxで入手できます</a><a href="https://github.com/eleutherai/gpt-neox">。</a></strong></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre></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>
|
||||
File diff suppressed because one or more lines are too long
@@ -0,0 +1,125 @@
|
||||
<!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&v=4"/>
|
||||
<meta name="twitter:title" content="GPT ネオックス"/>
|
||||
<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/neox/readme.html"/>
|
||||
<meta property="og:title" content="GPT ネオックス"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="GPT ネオックス"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="GPT ネオックス"/>
|
||||
<meta property="og:description" content=""/>
|
||||
|
||||
<title>GPT ネオックス</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/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">neox</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/neox/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>
|
||||
<b>MarkdownException</b> + Small: parse error
|
||||
</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>
|
||||
@@ -0,0 +1,444 @@
|
||||
<!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="フェアスケール・パイプライン・パラレル・モジュールによるGPT-Neoxバイアスの微調整"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="ファインチューン GPT-ネオックス"/>
|
||||
<meta name="twitter:description" content="フェアスケール・パイプライン・パラレル・モジュールによるGPT-Neoxバイアスの微調整"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/samples/finetune.html"/>
|
||||
<meta property="og:title" content="ファインチューン GPT-ネオックス"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="ファインチューン GPT-ネオックス"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="ファインチューン GPT-ネオックス"/>
|
||||
<meta property="og:description" content="フェアスケール・パイプライン・パラレル・モジュールによるGPT-Neoxバイアスの微調整"/>
|
||||
|
||||
<title>ファインチューン GPT-ネオックス</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/samples/finetune.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">neox</a>
|
||||
<a class="parent" href="index.html">samples</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/neox/samples/finetune.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>ファインチューン GPT-ネオックス</h1>
|
||||
<p>これは、パイプライン並列処理でGPT-Neoxを微調整する方法を示しています。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">13</span><span></span><span class="kn">import</span> <span class="nn">fairscale</span>
|
||||
<span class="lineno">14</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">15</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">16</span><span class="kn">import</span> <span class="nn">torch.utils.data</span>
|
||||
<span class="lineno">17</span><span class="kn">import</span> <span class="nn">torch.utils.data</span>
|
||||
<span class="lineno">18</span><span class="kn">import</span> <span class="nn">typing</span>
|
||||
<span class="lineno">19</span><span class="kn">from</span> <span class="nn">torch.utils.data</span> <span class="kn">import</span> <span class="n">DataLoader</span><span class="p">,</span> <span class="n">RandomSampler</span>
|
||||
<span class="lineno">20</span>
|
||||
<span class="lineno">21</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">monit</span><span class="p">,</span> <span class="n">tracker</span><span class="p">,</span> <span class="n">lab</span>
|
||||
<span class="lineno">22</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">23</span><span class="kn">from</span> <span class="nn">labml.logger</span> <span class="kn">import</span> <span class="n">inspect</span>
|
||||
<span class="lineno">24</span><span class="kn">from</span> <span class="nn">labml_nn.neox.utils.text_dataset</span> <span class="kn">import</span> <span class="n">get_training_data</span>
|
||||
<span class="lineno">25</span><span class="kn">from</span> <span class="nn">labml_nn.neox.utils.finetune</span> <span class="kn">import</span> <span class="n">FineTuneBiases</span>
|
||||
<span class="lineno">26</span><span class="kn">from</span> <span class="nn">labml_nn.neox.model</span> <span class="kn">import</span> <span class="n">LayerGenerator</span><span class="p">,</span> <span class="n">NeoXModule</span>
|
||||
<span class="lineno">27</span><span class="kn">from</span> <span class="nn">labml_nn.neox.utils</span> <span class="kn">import</span> <span class="n">balance_layers_simple</span>
|
||||
<span class="lineno">28</span><span class="kn">from</span> <span class="nn">labml_nn.neox.utils.trainer</span> <span class="kn">import</span> <span class="n">PipelineParallelTrainerConf</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>
|
||||
<h3>GPT-Neox レイヤーをロード</h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">31</span><span class="nd">@option</span><span class="p">(</span><span class="n">PipelineParallelTrainerConf</span><span class="o">.</span><span class="n">layers</span><span class="p">,</span> <span class="s1">'PipelineBiases'</span><span class="p">)</span>
|
||||
<span class="lineno">32</span><span class="k">def</span> <span class="nf">neox_layers</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">PipelineParallelTrainerConf</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">36</span> <span class="k">return</span> <span class="nb">list</span><span class="p">(</span><span class="n">LayerGenerator</span><span class="p">(</span><span class="n">is_clone_layers</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">is_clone_layers</span><span class="p">,</span>
|
||||
<span class="lineno">37</span> <span class="n">filter_layers</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">filter_layers</span><span class="p">,</span>
|
||||
<span class="lineno">38</span> <span class="n">dtype</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">dtype</span><span class="p">,</span>
|
||||
<span class="lineno">39</span> <span class="p">)</span><span class="o">.</span><span class="n">load</span><span class="p">())</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-3'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-3'>#</a>
|
||||
</div>
|
||||
<h3>バイアスの微調整器の作成</h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">42</span><span class="nd">@option</span><span class="p">(</span><span class="n">PipelineParallelTrainerConf</span><span class="o">.</span><span class="n">fine_tuner</span><span class="p">,</span> <span class="s1">'PipelineBiases'</span><span class="p">)</span>
|
||||
<span class="lineno">43</span><span class="k">def</span> <span class="nf">fine_tune_biases</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">PipelineParallelTrainerConf</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">48</span> <span class="n">fine_tuner</span> <span class="o">=</span> <span class="n">FineTuneBiases</span><span class="p">(</span><span class="n">typing</span><span class="o">.</span><span class="n">cast</span><span class="p">(</span><span class="n">typing</span><span class="o">.</span><span class="n">List</span><span class="p">[</span><span class="n">NeoXModule</span><span class="p">],</span> <span class="n">c</span><span class="o">.</span><span class="n">layers</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>バイアスをトレーニング可能とマークする</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">50</span> <span class="n">fine_tuner</span><span class="o">.</span><span class="n">set_trainable_params</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">53</span> <span class="k">return</span> <span class="n">fine_tuner</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>パイプライン並列モデルの作成</h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">56</span><span class="nd">@option</span><span class="p">(</span><span class="n">PipelineParallelTrainerConf</span><span class="o">.</span><span class="n">model</span><span class="p">,</span> <span class="s1">'PipelineBiases'</span><span class="p">)</span>
|
||||
<span class="lineno">57</span><span class="k">def</span> <span class="nf">pipe_model</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">PipelineParallelTrainerConf</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">62</span> <span class="k">if</span> <span class="n">c</span><span class="o">.</span><span class="n">is_checkpointing</span><span class="p">:</span>
|
||||
<span class="lineno">63</span> <span class="k">raise</span> <span class="ne">NotImplementedError</span><span class="p">()</span>
|
||||
<span class="lineno">64</span> <span class="k">else</span><span class="p">:</span>
|
||||
<span class="lineno">65</span> <span class="n">layers</span> <span class="o">=</span> <span class="n">c</span><span class="o">.</span><span class="n">layers</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>ファインチューナーが初期化されていることを確認してください</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">68</span> <span class="n">_</span> <span class="o">=</span> <span class="n">c</span><span class="o">.</span><span class="n">fine_tuner</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>パイプモジュールを作成する</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">71</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">'Pipe'</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>GPU 全体のレイヤー分布を取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">73</span> <span class="n">balance</span> <span class="o">=</span> <span class="n">balance_layers_simple</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">layers</span><span class="p">),</span> <span class="n">c</span><span class="o">.</span><span class="n">n_gpus</span><span class="p">)</span>
|
||||
<span class="lineno">74</span> <span class="n">inspect</span><span class="p">(</span><span class="n">balance</span><span class="o">=</span><span class="n">balance</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>各 GPU のデバイス</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">76</span> <span class="n">devices</span> <span class="o">=</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="sa">f</span><span class="s1">'cuda:</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">'</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="n">c</span><span class="o">.</span><span class="n">n_gpus</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>フェアスケールパイプモジュールを作成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">78</span> <span class="n">pipe_model</span> <span class="o">=</span> <span class="n">fairscale</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">Pipe</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span><span class="o">*</span><span class="n">layers</span><span class="p">),</span>
|
||||
<span class="lineno">79</span> <span class="n">balance</span><span class="o">=</span><span class="n">balance</span><span class="p">,</span>
|
||||
<span class="lineno">80</span> <span class="n">devices</span><span class="o">=</span><span class="n">devices</span><span class="p">,</span>
|
||||
<span class="lineno">81</span> <span class="n">chunks</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">chunks</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">84</span> <span class="k">return</span> <span class="n">pipe_model</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-15'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-15'>#</a>
|
||||
</div>
|
||||
<h4>小さなシェイクスピアデータセット</h4>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">87</span><span class="nd">@option</span><span class="p">(</span><span class="n">PipelineParallelTrainerConf</span><span class="o">.</span><span class="n">train_loader</span><span class="p">)</span>
|
||||
<span class="lineno">88</span><span class="k">def</span> <span class="nf">tiny_shakespeare</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">PipelineParallelTrainerConf</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">92</span> <span class="n">dataset</span> <span class="o">=</span> <span class="n">get_training_data</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">max_seq_len</span><span class="p">)</span>
|
||||
<span class="lineno">93</span>
|
||||
<span class="lineno">94</span> <span class="k">return</span> <span class="n">DataLoader</span><span class="p">(</span><span class="n">dataset</span><span class="p">,</span>
|
||||
<span class="lineno">95</span> <span class="n">batch_size</span><span class="o">=</span><span class="n">c</span><span class="o">.</span><span class="n">batch_size</span><span class="p">,</span>
|
||||
<span class="lineno">96</span> <span class="n">sampler</span><span class="o">=</span><span class="n">RandomSampler</span><span class="p">(</span><span class="n">dataset</span><span class="p">,</span> <span class="n">replacement</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">99</span><span class="k">def</span> <span class="nf">main</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>実験を作成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">101</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">'pipe_neox_biases'</span><span class="p">,</span>
|
||||
<span class="lineno">102</span> <span class="n">writers</span><span class="o">=</span><span class="p">{</span><span class="s1">'screen'</span><span class="p">,</span> <span class="s1">'web_api'</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>コンフィグを初期化</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">105</span> <span class="n">conf</span> <span class="o">=</span> <span class="n">PipelineParallelTrainerConf</span><span class="p">()</span>
|
||||
<span class="lineno">106</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="lineno">107</span> <span class="s1">'learning_rate'</span><span class="p">:</span> <span class="mf">3e-4</span><span class="p">,</span>
|
||||
<span class="lineno">108</span> <span class="s1">'is_checkpointing'</span><span class="p">:</span> <span class="kc">False</span><span class="p">,</span>
|
||||
<span class="lineno">109</span> <span class="s1">'max_seq_len'</span><span class="p">:</span> <span class="mi">128</span><span class="p">,</span>
|
||||
<span class="lineno">110</span> <span class="s1">'batch_size'</span><span class="p">:</span> <span class="mi">64</span><span class="p">,</span>
|
||||
<span class="lineno">111</span> <span class="s1">'chunks'</span><span class="p">:</span> <span class="mi">8</span><span class="p">,</span>
|
||||
<span class="lineno">112</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>実験を始める</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">115</span> <span class="k">with</span> <span class="n">experiment</span><span class="o">.</span><span class="n">start</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>モデルを初期化します。これをループの前に行って、ログをクリーンアップしてください。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">117</span> <span class="n">_</span> <span class="o">=</span> <span class="n">conf</span><span class="o">.</span><span class="n">model</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>列車</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">120</span> <span class="k">for</span> <span class="n">epoch</span> <span class="ow">in</span> <span class="n">monit</span><span class="o">.</span><span class="n">loop</span><span class="p">(</span><span class="n">conf</span><span class="o">.</span><span class="n">epochs</span><span class="p">):</span>
|
||||
<span class="lineno">121</span> <span class="n">conf</span><span class="o">.</span><span class="n">train_epoch</span><span class="p">()</span>
|
||||
<span class="lineno">122</span> <span class="n">tracker</span><span class="o">.</span><span class="n">new_line</span><span class="p">()</span>
|
||||
<span class="lineno">123</span> <span class="n">torch</span><span class="o">.</span><span class="n">save</span><span class="p">(</span><span class="n">conf</span><span class="o">.</span><span class="n">fine_tuner</span><span class="o">.</span><span class="n">state_dict</span><span class="p">(),</span> <span class="nb">str</span><span class="p">(</span><span class="n">lab</span><span class="o">.</span><span class="n">get_data_path</span><span class="p">()</span> <span class="o">/</span> <span class="s1">'fine_tune.pt'</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">127</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">128</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>
|
||||
@@ -0,0 +1,406 @@
|
||||
<!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-ネオックスでテキストを生成"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="GPT-ネオックスでテキストを生成"/>
|
||||
<meta name="twitter:description" content="GPT-ネオックスでテキストを生成"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/samples/generate.html"/>
|
||||
<meta property="og:title" content="GPT-ネオックスでテキストを生成"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="GPT-ネオックスでテキストを生成"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="GPT-ネオックスでテキストを生成"/>
|
||||
<meta property="og:description" content="GPT-ネオックスでテキストを生成"/>
|
||||
|
||||
<title>GPT-ネオックスでテキストを生成</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/samples/generate.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">neox</a>
|
||||
<a class="parent" href="index.html">samples</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/neox/samples/generate.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>GPT-ネオックスでテキストを生成</h1>
|
||||
<p>これは、単一のGPUでGPT-Neoxからテキストを生成する方法を示しています。</p>
|
||||
<p>これには、45 GB 以上のメモリを搭載した GPU が必要です。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">15</span><span></span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-1'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-1'>#</a>
|
||||
</div>
|
||||
<p>輸入</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">16</span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">List</span>
|
||||
<span class="lineno">17</span>
|
||||
<span class="lineno">18</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">19</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
||||
<span class="lineno">20</span>
|
||||
<span class="lineno">21</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">monit</span>
|
||||
<span class="lineno">22</span><span class="kn">from</span> <span class="nn">labml_nn.neox.model</span> <span class="kn">import</span> <span class="n">LayerGenerator</span>
|
||||
<span class="lineno">23</span><span class="kn">from</span> <span class="nn">labml_nn.neox.utils</span> <span class="kn">import</span> <span class="n">get_tokens</span><span class="p">,</span> <span class="n">print_tokens</span>
|
||||
<span class="lineno">24</span><span class="kn">from</span> <span class="nn">labml_nn.neox.utils.cache</span> <span class="kn">import</span> <span class="n">get_cache</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>ロードするレイヤーのリスト。これはテストに使用されます。<code class="highlight"><span></span><span class="p">{</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">}</span></code>
|
||||
のようにレイヤーのサブセットを割り当てて、最初のレイヤーのみをトランスフォーマーレイヤーに読み込むことができます</p>。
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">29</span><span class="n">LAYERS</span> <span class="o">=</span> <span class="kc">None</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>完了を促すプロンプト</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">32</span><span class="n">PROMPT</span> <span class="o">=</span> <span class="s1">'Einstein was born in the German Empire, but moved to Switzerland in 1895, forsaking his German'</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>次のトークンを予測</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">model</span></code>
|
||||
モデルです</li>
|
||||
<li><code class="highlight"><span></span><span class="n">ids</span></code>
|
||||
は入力トークン ID</li>
|
||||
<li><code class="highlight"><span></span><span class="n">device</span></code>
|
||||
モデルのデバイスです</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">35</span><span class="k">def</span> <span class="nf">infer</span><span class="p">(</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="p">,</span> <span class="n">ids</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">device</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</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">44</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-6'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-6'>#</a>
|
||||
</div>
|
||||
<p>トークンを入手</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">46</span> <span class="n">x</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="n">ids</span><span class="p">)[</span><span class="kc">None</span><span class="p">,</span> <span class="p">:]</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</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>評価モデル</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">48</span> <span class="n">x</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="n">x</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>予測トークンを返す</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">51</span> <span class="k">return</span> <span class="n">x</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">max</span><span class="p">(</span><span class="n">dim</span><span class="o">=-</span><span class="mi">1</span><span class="p">)[</span><span class="mi">1</span><span class="p">]</span><span class="o">.</span><span class="n">tolist</span><span class="p">()</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-9'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-9'>#</a>
|
||||
</div>
|
||||
<h2>テキストを生成</h2>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">54</span><span class="k">def</span> <span class="nf">generate</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><a href="../utils/cache.html">生成を高速化するために中間キーと値のペアをキャッシュするようにキャッシュを設定</a></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">60</span> <span class="n">cache</span> <span class="o">=</span> <span class="n">get_cache</span><span class="p">()</span>
|
||||
<span class="lineno">61</span> <span class="n">cache</span><span class="o">.</span><span class="n">set</span><span class="p">(</span><span class="s1">'use_cache'</span><span class="p">,</span> <span class="kc">True</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>端末</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">64</span> <span class="n">device</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cuda:0'</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>レイヤーをロード</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">67</span> <span class="n">layers</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">LayerGenerator</span><span class="p">(</span><span class="n">is_clone_layers</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||||
<span class="lineno">68</span> <span class="n">filter_layers</span><span class="o">=</span><span class="n">LAYERS</span><span class="p">,</span>
|
||||
<span class="lineno">69</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">float16</span><span class="p">,</span>
|
||||
<span class="lineno">70</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span>
|
||||
<span class="lineno">71</span> <span class="p">)</span><span class="o">.</span><span class="n">load</span><span class="p">())</span>
|
||||
<span class="lineno">72</span>
|
||||
<span class="lineno">73</span> <span class="n">model</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span><span class="o">*</span><span class="n">layers</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>トークン ID を取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">76</span> <span class="n">ids</span> <span class="o">=</span> <span class="n">get_tokens</span><span class="p">(</span><span class="n">PROMPT</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>モデルを実行</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">79</span> <span class="n">cache</span><span class="o">.</span><span class="n">set</span><span class="p">(</span><span class="s1">'state_ids'</span><span class="p">,</span> <span class="p">(</span><span class="kc">None</span><span class="p">,</span> <span class="mi">1</span><span class="p">))</span>
|
||||
<span class="lineno">80</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">'Infer'</span><span class="p">):</span>
|
||||
<span class="lineno">81</span> <span class="n">next_token</span> <span class="o">=</span> <span class="n">infer</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">ids</span><span class="p">,</span> <span class="n">device</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-15'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-15'>#</a>
|
||||
</div>
|
||||
<p>予測トークンを追加</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">84</span> <span class="n">ids</span> <span class="o">+=</span> <span class="p">[</span><span class="n">next_token</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>トークンを100個予測する</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">87</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">100</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>キャッシュされたアクティベーションを使用するように状態を設定します</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">89</span> <span class="n">cache</span><span class="o">.</span><span class="n">set</span><span class="p">(</span><span class="s1">'state_ids'</span><span class="p">,</span> <span class="p">(</span><span class="n">i</span><span class="p">,</span> <span class="n">i</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>次のトークンを入手してください。以前のトークンのキーと値のペアをキャッシュするので、最後のトークンのみをモデルにフィードすることに注意してください</p>。
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">92</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">'Infer'</span><span class="p">):</span>
|
||||
<span class="lineno">93</span> <span class="n">next_token</span> <span class="o">=</span> <span class="n">infer</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="p">[</span><span class="n">next_token</span><span class="p">],</span> <span class="n">device</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-19'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-19'>#</a>
|
||||
</div>
|
||||
<p>予測トークンを追加</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">95</span> <span class="n">ids</span> <span class="o">+=</span> <span class="p">[</span><span class="n">next_token</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>プリント</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">97</span> <span class="n">print_tokens</span><span class="p">(</span><span class="n">ids</span><span class="p">,</span> <span class="p">[</span><span class="n">ids</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">101</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">102</span> <span class="n">generate</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>
|
||||
@@ -0,0 +1,129 @@
|
||||
<!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="推論と微調整用のサンプル"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="[サンプル]"/>
|
||||
<meta name="twitter:description" content="推論と微調整用のサンプル"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/samples/index.html"/>
|
||||
<meta property="og:title" content="[サンプル]"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="[サンプル]"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="[サンプル]"/>
|
||||
<meta property="og:description" content="推論と微調整用のサンプル"/>
|
||||
|
||||
<title>[サンプル]</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/samples/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">neox</a>
|
||||
<a class="parent" href="index.html">samples</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/neox/samples/__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>[サンプル]</h1>
|
||||
<ul><li><a href="generate.html">テキストの生成</a></li>
|
||||
<li><a href="finetune.html">パイプライン並列トレーニングによるバイアスの微調整</a></li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre></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>
|
||||
@@ -0,0 +1,348 @@
|
||||
<!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="LLM.int8 () 量子化を使用して GPT-Neox でテキストを生成"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="LLM.int8 () 量子化を使用して GPT-Neox でテキストを生成"/>
|
||||
<meta name="twitter:description" content="LLM.int8 () 量子化を使用して GPT-Neox でテキストを生成"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/samples/llm_int8.html"/>
|
||||
<meta property="og:title" content="LLM.int8 () 量子化を使用して GPT-Neox でテキストを生成"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="LLM.int8 () 量子化を使用して GPT-Neox でテキストを生成"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="LLM.int8 () 量子化を使用して GPT-Neox でテキストを生成"/>
|
||||
<meta property="og:description" content="LLM.int8 () 量子化を使用して GPT-Neox でテキストを生成"/>
|
||||
|
||||
<title>LLM.int8 () 量子化を使用して GPT-Neox でテキストを生成</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/samples/llm_int8.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">neox</a>
|
||||
<a class="parent" href="index.html">samples</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/neox/samples/llm_int8.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>LLM.int8 () 量子化を使用して GPT-Neox でテキストを生成</h1>
|
||||
<p>これは、<a href="../utils/llm_int8.html">LLM.int8</a> () 量子化を使用して GPT-Neox からテキストを生成する方法を示しています。</p>
|
||||
<p>これには 24 GB のメモリを搭載した GPU が必要です。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">15</span><span></span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">16</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
||||
<span class="lineno">17</span>
|
||||
<span class="lineno">18</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">monit</span>
|
||||
<span class="lineno">19</span><span class="kn">from</span> <span class="nn">labml_nn.neox.model</span> <span class="kn">import</span> <span class="n">LayerGenerator</span>
|
||||
<span class="lineno">20</span><span class="kn">from</span> <span class="nn">labml_nn.neox.samples.generate</span> <span class="kn">import</span> <span class="n">PROMPT</span><span class="p">,</span> <span class="n">infer</span>
|
||||
<span class="lineno">21</span><span class="kn">from</span> <span class="nn">labml_nn.neox.utils</span> <span class="kn">import</span> <span class="n">get_tokens</span><span class="p">,</span> <span class="n">print_tokens</span>
|
||||
<span class="lineno">22</span><span class="kn">from</span> <span class="nn">labml_nn.neox.utils.cache</span> <span class="kn">import</span> <span class="n">get_cache</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>テキストを生成</h2>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">25</span><span class="k">def</span> <span class="nf">generate</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><a href="../utils/cache.html">生成を高速化するために中間キーと値のペアをキャッシュするようにキャッシュを設定</a></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">31</span> <span class="n">cache</span> <span class="o">=</span> <span class="n">get_cache</span><span class="p">()</span>
|
||||
<span class="lineno">32</span> <span class="n">cache</span><span class="o">.</span><span class="n">set</span><span class="p">(</span><span class="s1">'use_cache'</span><span class="p">,</span> <span class="kc">True</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>端末</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">35</span> <span class="n">device</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cuda:0'</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>float16 のレイヤーを CPU にロードします。レイヤーをGPUにロードした後にその場でこれを行うと、CUDAメモリの断片化が発生するため、後でレイヤーをint8に変換します（フラグメンテーションにより約3GBのメモリが失われる可能性があります</p>）。
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">40</span> <span class="n">layer_generator</span> <span class="o">=</span> <span class="n">LayerGenerator</span><span class="p">(</span><span class="n">is_clone_layers</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||||
<span class="lineno">41</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">float16</span><span class="p">,</span>
|
||||
<span class="lineno">42</span> <span class="n">device</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cpu'</span><span class="p">),</span>
|
||||
<span class="lineno">43</span> <span class="n">is_llm_int8</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
||||
<span class="lineno">44</span> <span class="p">)</span>
|
||||
<span class="lineno">45</span> <span class="n">layers</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">layer_generator</span><span class="o">.</span><span class="n">load</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>これにより、CUDA メモリの断片化が減少します。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">48</span> <span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">monit</span><span class="o">.</span><span class="n">iterate</span><span class="p">(</span><span class="s1">'Convert to int8'</span><span class="p">,</span> <span class="n">layers</span><span class="p">,</span> <span class="n">is_children_silent</span><span class="o">=</span><span class="kc">True</span><span class="p">):</span>
|
||||
<span class="lineno">49</span> <span class="n">layer_generator</span><span class="o">.</span><span class="n">post_load_prepare</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span>
|
||||
<span class="lineno">50</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">,</span>
|
||||
<span class="lineno">51</span> <span class="n">is_llm_int8</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
||||
<span class="lineno">52</span> <span class="n">llm_int8_threshold</span><span class="o">=</span><span class="mf">6.0</span><span class="p">,</span>
|
||||
<span class="lineno">53</span> <span class="p">)</span>
|
||||
<span class="lineno">54</span> <span class="n">layer</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</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><code class="highlight"><span></span><span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span></code>
|
||||
モデル作成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">57</span> <span class="n">model</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span><span class="o">*</span><span class="n">layers</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>デバッグ用にキャッシュをクリアしてメモリの概要を印刷</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">60</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">empty_cache</span><span class="p">()</span>
|
||||
<span class="lineno">61</span> <span class="nb">print</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">memory_summary</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>トークン ID を取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">64</span> <span class="n">ids</span> <span class="o">=</span> <span class="n">get_tokens</span><span class="p">(</span><span class="n">PROMPT</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>モデルを実行します。<a href="generate.html"><code class="highlight"><span></span><span class="n">infer</span></code>
|
||||
</a>で定義されている関数を使用します <a href="generate.html"><code class="highlight"><span></span><span class="n">generate</span><span class="o">.</span><span class="n">py</span></code>
|
||||
</a></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">68</span> <span class="n">cache</span><span class="o">.</span><span class="n">set</span><span class="p">(</span><span class="s1">'state_ids'</span><span class="p">,</span> <span class="p">(</span><span class="kc">None</span><span class="p">,</span> <span class="mi">1</span><span class="p">))</span>
|
||||
<span class="lineno">69</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">'Infer'</span><span class="p">):</span>
|
||||
<span class="lineno">70</span> <span class="n">next_token</span> <span class="o">=</span> <span class="n">infer</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">ids</span><span class="p">,</span> <span class="n">device</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-10'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-10'>#</a>
|
||||
</div>
|
||||
<p>予測トークンを追加</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">73</span> <span class="n">ids</span> <span class="o">+=</span> <span class="p">[</span><span class="n">next_token</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>トークンを100個予測する</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">76</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">100</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>キャッシュされたアクティベーションを使用するように状態を設定します</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">78</span> <span class="n">cache</span><span class="o">.</span><span class="n">set</span><span class="p">(</span><span class="s1">'state_ids'</span><span class="p">,</span> <span class="p">(</span><span class="n">i</span><span class="p">,</span> <span class="n">i</span> <span class="o">+</span> <span class="mi">1</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>次のトークンを入手してください。以前のトークンのキーと値のペアをキャッシュするので、最後のトークンのみをモデルにフィードすることに注意してください</p>。
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">81</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">'Infer'</span><span class="p">):</span>
|
||||
<span class="lineno">82</span> <span class="n">next_token</span> <span class="o">=</span> <span class="n">infer</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="p">[</span><span class="n">next_token</span><span class="p">],</span> <span class="n">device</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-14'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
<p>予測トークンを追加</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">84</span> <span class="n">ids</span> <span class="o">+=</span> <span class="p">[</span><span class="n">next_token</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>プリント</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">86</span> <span class="n">print_tokens</span><span class="p">(</span><span class="n">ids</span><span class="p">,</span> <span class="p">[</span><span class="n">ids</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">90</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">91</span> <span class="n">generate</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>
|
||||
@@ -0,0 +1,157 @@
|
||||
<!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-Neox トークナイザーをロードします"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="GPT-ネオックストークナイザー"/>
|
||||
<meta name="twitter:description" content="GPT-Neox トークナイザーをロードします"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/tokenizer.html"/>
|
||||
<meta property="og:title" content="GPT-ネオックストークナイザー"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="GPT-ネオックストークナイザー"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="GPT-ネオックストークナイザー"/>
|
||||
<meta property="og:description" content="GPT-Neox トークナイザーをロードします"/>
|
||||
|
||||
<title>GPT-ネオックストークナイザー</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/tokenizer.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">neox</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/neox/tokenizer.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>GPT-ネオックストークナイザー</h1>
|
||||
<p>これにより、ダウンロードしたボキャブラリーから Hugging Face トークナイザーが初期化されます。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">13</span><span></span><span class="kn">from</span> <span class="nn">tokenizers</span> <span class="kn">import</span> <span class="n">Tokenizer</span>
|
||||
<span class="lineno">14</span>
|
||||
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">lab</span><span class="p">,</span> <span class="n">monit</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>
|
||||
<h3>NeoX トークナイザーをロード</h3>
|
||||
<ul><p><em>トークナイザーを返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">18</span><span class="nd">@monit</span><span class="o">.</span><span class="n">func</span><span class="p">(</span><span class="s1">'Load NeoX Tokenizer'</span><span class="p">)</span>
|
||||
<span class="lineno">19</span><span class="k">def</span> <span class="nf">get_tokenizer</span><span class="p">()</span> <span class="o">-></span> <span class="n">Tokenizer</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">25</span> <span class="n">vocab_file</span> <span class="o">=</span> <span class="n">lab</span><span class="o">.</span><span class="n">get_data_path</span><span class="p">()</span> <span class="o">/</span> <span class="s1">'neox'</span> <span class="o">/</span> <span class="s1">'slim_weights'</span> <span class="o">/</span> <span class="s1">'20B_tokenizer.json'</span>
|
||||
<span class="lineno">26</span> <span class="n">tokenizer</span> <span class="o">=</span> <span class="n">Tokenizer</span><span class="o">.</span><span class="n">from_file</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">vocab_file</span><span class="p">))</span>
|
||||
<span class="lineno">27</span>
|
||||
<span class="lineno">28</span> <span class="k">return</span> <span class="n">tokenizer</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>
|
||||
@@ -0,0 +1,396 @@
|
||||
<!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="推論を高速化するための中間アクティベーション用のキャッシュ。"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="中間アクティベーション用キャッシュ"/>
|
||||
<meta name="twitter:description" content="推論を高速化するための中間アクティベーション用のキャッシュ。"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/utils/cache.html"/>
|
||||
<meta property="og:title" content="中間アクティベーション用キャッシュ"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="中間アクティベーション用キャッシュ"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="中間アクティベーション用キャッシュ"/>
|
||||
<meta property="og:description" content="推論を高速化するための中間アクティベーション用のキャッシュ。"/>
|
||||
|
||||
<title>中間アクティベーション用キャッシュ</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/utils/cache.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">neox</a>
|
||||
<a class="parent" href="index.html">utils</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/neox/utils/cache.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>中間アクティベーション用キャッシュ</h1>
|
||||
<p>推論中、モデルはトークンごとに出力します。このシンプルなキャッシュを使ってキーと値のアテンションレイヤーを格納するので、以前のトークンでそれらを再計算する必要がありません</p>。
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">15</span><span></span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Any</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>キャッシュ</h2>
|
||||
<p>これにより、キーと値のキャッシュが維持され、値のプッシュとポップが同じ順序でキューに入れられます。キューは複数のアテンションレイヤーがあるので便利です</p>。
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">18</span><span class="k">class</span> <span class="nc">Cache</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">26</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">27</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cache</span> <span class="o">=</span> <span class="p">{}</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-3'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-3'>#</a>
|
||||
</div>
|
||||
<h3>キャッシュをクリア</h3>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">29</span> <span class="k">def</span> <span class="nf">clear_all</span><span class="p">(</span><span class="bp">self</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>
|
||||
|
||||
</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">_cache</span> <span class="o">=</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>値をキューにプッシュ</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">name</span></code>
|
||||
はキューの名前です</li>
|
||||
<li><code class="highlight"><span></span><span class="n">value</span></code>
|
||||
プッシュする値です</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">35</span> <span class="k">def</span> <span class="nf">push</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">value</span><span class="p">:</span> <span class="n">Any</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>存在しない場合は空のキューを作成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">44</span> <span class="k">if</span> <span class="n">name</span> <span class="ow">not</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="p">:</span>
|
||||
<span class="lineno">45</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="p">[</span><span class="n">name</span><span class="p">]</span> <span class="o">=</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>キューにプッシュ</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">48</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="p">[</span><span class="n">name</span><span class="p">]</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">value</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>キューのサイズを返す</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">name</span></code>
|
||||
はキューの名前です</li>
|
||||
<p><em>キューが存在する場合はキューのサイズを返します</em>。それ以外の場合は None</p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">50</span> <span class="k">def</span> <span class="nf">q_size</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</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">58</span> <span class="k">if</span> <span class="n">name</span> <span class="ow">not</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="p">:</span>
|
||||
<span class="lineno">59</span> <span class="k">return</span> <span class="kc">None</span>
|
||||
<span class="lineno">60</span>
|
||||
<span class="lineno">61</span> <span class="k">if</span> <span class="nb">type</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="p">[</span><span class="n">name</span><span class="p">])</span> <span class="o">!=</span> <span class="nb">list</span><span class="p">:</span>
|
||||
<span class="lineno">62</span> <span class="k">return</span> <span class="kc">None</span>
|
||||
<span class="lineno">63</span>
|
||||
<span class="lineno">64</span> <span class="k">return</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="p">[</span><span class="n">name</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>キューからポップする</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">name</span></code>
|
||||
はキューの名前です</li>
|
||||
<p><em>値を返す</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">66</span> <span class="k">def</span> <span class="nf">pop</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">name</span><span class="p">:</span> <span class="nb">str</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">73</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="p">[</span><span class="n">name</span><span class="p">]</span><span class="o">.</span><span class="n">pop</span><span class="p">(</span><span class="mi">0</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>
|
||||
<h3>値をキャッシュする</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">key</span></code>
|
||||
キャッシュされる値の名前</li>
|
||||
<li><code class="highlight"><span></span><span class="n">value</span></code>
|
||||
は値です</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">75</span> <span class="k">def</span> <span class="nf">set</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">key</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">value</span><span class="p">:</span> <span class="n">Any</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">82</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="p">[</span><span class="n">key</span><span class="p">]</span> <span class="o">=</span> <span class="n">value</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-14'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
<h3>キャッシュから値を取得</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">key</span></code>
|
||||
キャッシュ時に使用される名前です</li>
|
||||
<li><code class="highlight"><span></span><span class="n">default</span></code>
|
||||
キャッシュが空の場合のデフォルト値です</li>
|
||||
<p><em>キャッシュされた値を返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">84</span> <span class="k">def</span> <span class="nf">get</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">key</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">default</span><span class="p">:</span> <span class="n">Any</span> <span class="o">=</span> <span class="kc">None</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">92</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">key</span><span class="p">,</span> <span class="n">default</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
<h3>キャッシュ値をクリアする</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">key</span></code>
|
||||
キャッシュ時に使用される名前です</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">94</span> <span class="k">def</span> <span class="nf">clear</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">key</span><span class="p">:</span> <span class="nb">str</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">100</span> <span class="k">del</span> <span class="bp">self</span><span class="o">.</span><span class="n">_cache</span><span class="p">[</span><span class="n">key</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>キャッシュ用シングルトン</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">104</span><span class="n">_INSTANCE</span> <span class="o">=</span> <span class="kc">None</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>
|
||||
<h3>キャッシュインスタンスを取得</h3>
|
||||
<ul><p><em>キャッシュインスタンスを返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">107</span><span class="k">def</span> <span class="nf">get_cache</span><span class="p">()</span> <span class="o">-></span> <span class="n">Cache</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">113</span> <span class="k">global</span> <span class="n">_INSTANCE</span>
|
||||
<span class="lineno">114</span>
|
||||
<span class="lineno">115</span> <span class="k">if</span> <span class="n">_INSTANCE</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">116</span> <span class="n">_INSTANCE</span> <span class="o">=</span> <span class="n">Cache</span><span class="p">()</span>
|
||||
<span class="lineno">117</span>
|
||||
<span class="lineno">118</span> <span class="k">return</span> <span class="n">_INSTANCE</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>
|
||||
@@ -0,0 +1,293 @@
|
||||
<!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&v=4"/>
|
||||
<meta name="twitter:title" content="finetune.py"/>
|
||||
<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/neox/utils/finetune.html"/>
|
||||
<meta property="og:title" content="finetune.py"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="finetune.py"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="finetune.py"/>
|
||||
<meta property="og:description" content=""/>
|
||||
|
||||
<title>finetune.py</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/utils/finetune.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">neox</a>
|
||||
<a class="parent" href="index.html">utils</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/neox/utils/finetune.py" 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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">1</span><span></span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">List</span><span class="p">,</span> <span class="n">Dict</span>
|
||||
<span class="lineno">2</span>
|
||||
<span class="lineno">3</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">4</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</span>
|
||||
<span class="lineno">5</span>
|
||||
<span class="lineno">6</span><span class="kn">from</span> <span class="nn">labml_nn.neox.model</span> <span class="kn">import</span> <span class="n">TransformerLayer</span><span class="p">,</span> <span class="n">NeoXModule</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">9</span><span class="k">class</span> <span class="nc">FineTuner</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">10</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">layers</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="n">NeoXModule</span><span class="p">]):</span>
|
||||
<span class="lineno">11</span> <span class="bp">self</span><span class="o">.</span><span class="n">layers</span> <span class="o">=</span> <span class="n">layers</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">13</span> <span class="k">def</span> <span class="nf">get_trainable_params</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">]:</span>
|
||||
<span class="lineno">14</span> <span class="n">params</span> <span class="o">=</span> <span class="p">{}</span>
|
||||
<span class="lineno">15</span> <span class="k">for</span> <span class="n">i</span><span class="p">,</span> <span class="n">layer</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">layers</span><span class="p">):</span>
|
||||
<span class="lineno">16</span> <span class="n">params</span><span class="o">.</span><span class="n">update</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">get_layer_trainable_params</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="n">prefix</span><span class="o">=</span><span class="sa">f</span><span class="s1">'layer_</span><span class="si">{</span><span class="n">i</span> <span class="si">:</span><span class="s1">02d</span><span class="si">}</span><span class="s1">'</span><span class="p">))</span>
|
||||
<span class="lineno">17</span>
|
||||
<span class="lineno">18</span> <span class="k">return</span> <span class="n">params</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-4'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-4'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">20</span> <span class="k">def</span> <span class="nf">get_layer_trainable_params</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">layer</span><span class="p">:</span> <span class="n">NeoXModule</span><span class="p">,</span> <span class="n">prefix</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">]:</span>
|
||||
<span class="lineno">21</span> <span class="k">raise</span> <span class="ne">NotImplementedError</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">23</span> <span class="k">def</span> <span class="nf">set_trainable_params</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||||
<span class="lineno">24</span> <span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">layers</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><code class="highlight"><span></span><span class="n">requires_grad</span></code>
|
||||
<code class="highlight"><span></span><span class="kc">False</span></code>
|
||||
レイヤー全体でに設定します。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">26</span> <span class="n">layer</span><span class="o">.</span><span class="n">requires_grad_</span><span class="p">(</span><span class="kc">False</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">28</span> <span class="k">for</span> <span class="n">p</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">get_trainable_params</span><span class="p">()</span><span class="o">.</span><span class="n">values</span><span class="p">():</span>
|
||||
<span class="lineno">29</span> <span class="n">p</span><span class="o">.</span><span class="n">requires_grad_</span><span class="p">(</span><span class="kc">True</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">31</span> <span class="k">def</span> <span class="nf">state_dict</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||||
<span class="lineno">32</span> <span class="k">return</span> <span class="p">{</span><span class="n">n</span><span class="p">:</span> <span class="n">p</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">cpu</span><span class="p">()</span> <span class="k">for</span> <span class="n">n</span><span class="p">,</span> <span class="n">p</span> <span class="ow">in</span> <span class="bp">self</span><span class="o">.</span><span class="n">get_trainable_params</span><span class="p">()</span><span class="o">.</span><span class="n">items</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">34</span> <span class="k">def</span> <span class="nf">load_state_dict</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">state_dict</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">torch</span><span class="o">.</span><span class="n">Tensor</span><span class="p">]):</span>
|
||||
<span class="lineno">35</span> <span class="n">params</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">get_trainable_params</span><span class="p">()</span>
|
||||
<span class="lineno">36</span> <span class="k">for</span> <span class="n">n</span><span class="p">,</span> <span class="n">p</span> <span class="ow">in</span> <span class="n">params</span><span class="o">.</span><span class="n">items</span><span class="p">():</span>
|
||||
<span class="lineno">37</span> <span class="n">p</span><span class="o">.</span><span class="n">data</span><span class="p">[:]</span> <span class="o">=</span> <span class="n">state_dict</span><span class="p">[</span><span class="n">n</span><span class="p">]</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">p</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
|
||||
<span class="lineno">38</span>
|
||||
<span class="lineno">39</span> <span class="k">for</span> <span class="n">n</span> <span class="ow">in</span> <span class="n">state_dict</span><span class="o">.</span><span class="n">keys</span><span class="p">():</span>
|
||||
<span class="lineno">40</span> <span class="k">assert</span> <span class="n">n</span> <span class="ow">in</span> <span class="n">params</span><span class="p">,</span> <span class="n">n</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-10'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-10'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">43</span><span class="k">class</span> <span class="nc">FineTuneBiases</span><span class="p">(</span><span class="n">FineTuner</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">44</span> <span class="k">def</span> <span class="nf">get_layer_trainable_params</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">layer</span><span class="p">:</span> <span class="n">NeoXModule</span><span class="p">,</span> <span class="n">prefix</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="n">Dict</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">Parameter</span><span class="p">]:</span>
|
||||
<span class="lineno">45</span> <span class="n">params</span> <span class="o">=</span> <span class="p">{}</span>
|
||||
<span class="lineno">46</span>
|
||||
<span class="lineno">47</span> <span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="n">TransformerLayer</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>注意出力で加算するので、mlpバイアスをトレーニングする必要はありません。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">49</span> <span class="n">params</span><span class="p">[</span><span class="sa">f</span><span class="s1">'</span><span class="si">{</span><span class="n">prefix</span><span class="si">}</span><span class="s1">.attention.output.bias'</span><span class="p">]</span> <span class="o">=</span> <span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">output</span><span class="o">.</span><span class="n">bias</span>
|
||||
<span class="lineno">50</span> <span class="n">params</span><span class="p">[</span><span class="sa">f</span><span class="s1">'</span><span class="si">{</span><span class="n">prefix</span><span class="si">}</span><span class="s1">.attention.qkv_lin.bias'</span><span class="p">]</span> <span class="o">=</span> <span class="n">layer</span><span class="o">.</span><span class="n">attention</span><span class="o">.</span><span class="n">qkv_lin</span><span class="o">.</span><span class="n">bias</span>
|
||||
<span class="lineno">51</span> <span class="n">params</span><span class="p">[</span><span class="sa">f</span><span class="s1">'</span><span class="si">{</span><span class="n">prefix</span><span class="si">}</span><span class="s1">.ffn.dense_h_h4.bias'</span><span class="p">]</span> <span class="o">=</span> <span class="n">layer</span><span class="o">.</span><span class="n">ffn</span><span class="o">.</span><span class="n">dense_h_h4</span><span class="o">.</span><span class="n">bias</span>
|
||||
<span class="lineno">52</span> <span class="k">else</span><span class="p">:</span>
|
||||
<span class="lineno">53</span> <span class="k">pass</span>
|
||||
<span class="lineno">54</span>
|
||||
<span class="lineno">55</span> <span class="k">return</span> <span class="n">params</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>
|
||||
@@ -0,0 +1,387 @@
|
||||
<!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="ユーティリティとヘルパー関数"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="ユーティリティとヘルパー"/>
|
||||
<meta name="twitter:description" content="ユーティリティとヘルパー関数"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/utils/index.html"/>
|
||||
<meta property="og:title" content="ユーティリティとヘルパー"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="ユーティリティとヘルパー"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="ユーティリティとヘルパー"/>
|
||||
<meta property="og:description" content="ユーティリティとヘルパー関数"/>
|
||||
|
||||
<title>ユーティリティとヘルパー</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/utils/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">neox</a>
|
||||
<a class="parent" href="index.html">utils</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/neox/utils/__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>ユーティリティとヘルパー</h1>
|
||||
<ul><li><a href="cache.html">中間アクティベーション用のキャッシュ (推論を高速化するため)</a></li>
|
||||
<li><a href="finetune.html">微調整用ツール</a></li>
|
||||
<li><a href="trainer.html">トレーナー</a></li>
|
||||
<li><a href="text_dataset.html">テキストデータセット</a></li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">15</span><span></span><span class="kn">import</span> <span class="nn">typing</span>
|
||||
<span class="lineno">16</span><span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">List</span><span class="p">,</span> <span class="n">Optional</span>
|
||||
<span class="lineno">17</span>
|
||||
<span class="lineno">18</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">19</span>
|
||||
<span class="lineno">20</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">logger</span>
|
||||
<span class="lineno">21</span><span class="kn">from</span> <span class="nn">labml.logger</span> <span class="kn">import</span> <span class="n">Text</span>
|
||||
<span class="lineno">22</span><span class="kn">from</span> <span class="nn">labml_nn.neox.tokenizer</span> <span class="kn">import</span> <span class="n">get_tokenizer</span>
|
||||
<span class="lineno">23</span>
|
||||
<span class="lineno">24</span><span class="k">if</span> <span class="n">typing</span><span class="o">.</span><span class="n">TYPE_CHECKING</span><span class="p">:</span>
|
||||
<span class="lineno">25</span> <span class="kn">from</span> <span class="nn">tokenizers</span> <span class="kn">import</span> <span class="n">Tokenizer</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-1'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-1'>#</a>
|
||||
</div>
|
||||
<p>トークナイザーシングルトン</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">28</span><span class="n">_TOKENIZER</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="s1">'Tokenizer'</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</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>トークン ID を取得</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">text</span></code>
|
||||
トークン化するテキストです</li>
|
||||
<p>トークン ID <em>を返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">31</span><span class="k">def</span> <span class="nf">get_tokens</span><span class="p">(</span><span class="n">text</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></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-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">38</span> <span class="k">global</span> <span class="n">_TOKENIZER</span>
|
||||
<span class="lineno">39</span> <span class="k">if</span> <span class="n">_TOKENIZER</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">40</span> <span class="n">_TOKENIZER</span> <span class="o">=</span> <span class="n">get_tokenizer</span><span class="p">()</span>
|
||||
<span class="lineno">41</span> <span class="k">return</span> <span class="n">_TOKENIZER</span><span class="o">.</span><span class="n">encode_batch</span><span class="p">([</span><span class="n">text</span><span class="p">])[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">ids</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>モデル出力からトークンを印刷</h3>
|
||||
<p>Pretty は、モデルからの出力と一緒にターゲットトークンを出力します。</p>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">ids</span></code>
|
||||
ターゲットトークン ID</li>
|
||||
<li><code class="highlight"><span></span><span class="n">xs</span></code>
|
||||
はモデルの出力です</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">44</span><span class="k">def</span> <span class="nf">print_token_outputs</span><span class="p">(</span><span class="n">ids</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="o">*</span><span class="n">xs</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-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">53</span> <span class="n">ids</span> <span class="o">=</span> <span class="n">ids</span> <span class="o">+</span> <span class="p">[</span><span class="o">-</span><span class="mi">1</span><span class="p">]</span>
|
||||
<span class="lineno">54</span> <span class="n">xs</span> <span class="o">=</span> <span class="p">[[</span><span class="o">-</span><span class="mi">1</span><span class="p">]</span> <span class="o">+</span> <span class="n">x</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">max</span><span class="p">(</span><span class="n">dim</span><span class="o">=-</span><span class="mi">1</span><span class="p">)[</span><span class="mi">1</span><span class="p">]</span><span class="o">.</span><span class="n">tolist</span><span class="p">()</span> <span class="k">for</span> <span class="n">x</span> <span class="ow">in</span> <span class="n">xs</span><span class="p">]</span>
|
||||
<span class="lineno">55</span>
|
||||
<span class="lineno">56</span> <span class="n">print_tokens</span><span class="p">(</span><span class="n">ids</span><span class="p">,</span> <span class="n">xs</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>トークンの印刷</h3>
|
||||
<p>Pretty は比較用のトークンを印刷します</p>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">target</span></code>
|
||||
ターゲットトークン ID</li>
|
||||
<li><code class="highlight"><span></span><span class="n">others</span></code>
|
||||
はモデルからサンプリングされた出力です</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">59</span><span class="k">def</span> <span class="nf">print_tokens</span><span class="p">(</span><span class="n">target</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">others</span><span class="p">:</span> <span class="n">List</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-7'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-7'>#</a>
|
||||
</div>
|
||||
<p>トークナイザーをロード</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">70</span> <span class="k">global</span> <span class="n">_TOKENIZER</span>
|
||||
<span class="lineno">71</span> <span class="k">if</span> <span class="n">_TOKENIZER</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">72</span> <span class="n">_TOKENIZER</span> <span class="o">=</span> <span class="n">get_tokenizer</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>トークンを文字列のリストに変換</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">75</span> <span class="n">text</span> <span class="o">=</span> <span class="p">[]</span>
|
||||
<span class="lineno">76</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="nb">len</span><span class="p">(</span><span class="n">target</span><span class="p">)):</span>
|
||||
<span class="lineno">77</span> <span class="n">tokens</span> <span class="o">=</span> <span class="p">[</span><span class="n">_TOKENIZER</span><span class="o">.</span><span class="n">decode</span><span class="p">([</span><span class="n">target</span><span class="p">[</span><span class="n">i</span><span class="p">]])</span> <span class="k">if</span> <span class="n">target</span><span class="p">[</span><span class="n">i</span><span class="p">]</span> <span class="o">!=</span> <span class="o">-</span><span class="mi">1</span> <span class="k">else</span> <span class="s1">'---'</span><span class="p">]</span>
|
||||
<span class="lineno">78</span> <span class="k">for</span> <span class="n">j</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">others</span><span class="p">)):</span>
|
||||
<span class="lineno">79</span> <span class="n">tokens</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">_TOKENIZER</span><span class="o">.</span><span class="n">decode</span><span class="p">([</span><span class="n">others</span><span class="p">[</span><span class="n">j</span><span class="p">][</span><span class="n">i</span><span class="p">]])</span> <span class="k">if</span> <span class="n">others</span><span class="p">[</span><span class="n">j</span><span class="p">][</span><span class="n">i</span><span class="p">]</span> <span class="o">!=</span> <span class="o">-</span><span class="mi">1</span> <span class="k">else</span> <span class="s1">'---'</span><span class="p">)</span>
|
||||
<span class="lineno">80</span>
|
||||
<span class="lineno">81</span> <span class="n">text</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">tokens</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>統計情報</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">84</span> <span class="n">correct</span> <span class="o">=</span> <span class="p">[</span><span class="mi">0</span> <span class="k">for</span> <span class="n">_</span> <span class="ow">in</span> <span class="n">others</span><span class="p">]</span>
|
||||
<span class="lineno">85</span> <span class="n">total</span> <span class="o">=</span> <span class="mi">0</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>トークンを繰り返し処理</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">88</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="nb">len</span><span class="p">(</span><span class="n">target</span><span class="p">)):</span>
|
||||
<span class="lineno">89</span> <span class="n">parts</span> <span class="o">=</span> <span class="p">[(</span><span class="sa">f</span><span class="s1">'</span><span class="si">{</span><span class="n">i</span><span class="si">}</span><span class="s1">: '</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">meta</span><span class="p">)]</span>
|
||||
<span class="lineno">90</span> <span class="n">parts</span> <span class="o">+=</span> <span class="p">[(</span><span class="s1">'"'</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">),</span> <span class="p">(</span><span class="n">text</span><span class="p">[</span><span class="n">i</span><span class="p">][</span><span class="mi">0</span><span class="p">],</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">),</span> <span class="p">(</span><span class="s1">'"'</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">),</span> <span class="s1">'</span><span class="se">\t</span><span class="s1">'</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>空のターゲット</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">93</span> <span class="k">if</span> <span class="n">target</span><span class="p">[</span><span class="n">i</span><span class="p">]</span> <span class="o">==</span> <span class="o">-</span><span class="mi">1</span><span class="p">:</span>
|
||||
<span class="lineno">94</span> <span class="k">for</span> <span class="n">j</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">others</span><span class="p">)):</span>
|
||||
<span class="lineno">95</span> <span class="n">parts</span> <span class="o">+=</span> <span class="p">[(</span><span class="s1">'"'</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">),</span> <span class="p">(</span><span class="n">text</span><span class="p">[</span><span class="n">i</span><span class="p">][</span><span class="n">j</span> <span class="o">+</span> <span class="mi">1</span><span class="p">],</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">),</span> <span class="p">(</span><span class="s1">'"'</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">),</span> <span class="s1">'</span><span class="se">\t</span><span class="s1">'</span><span class="p">]</span>
|
||||
<span class="lineno">96</span>
|
||||
<span class="lineno">97</span> <span class="n">logger</span><span class="o">.</span><span class="n">log</span><span class="p">(</span><span class="n">parts</span><span class="p">)</span>
|
||||
<span class="lineno">98</span> <span class="k">continue</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>トークンの数</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">101</span> <span class="n">total</span> <span class="o">+=</span> <span class="mi">1</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>その他の出力</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">104</span> <span class="k">for</span> <span class="n">j</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">others</span><span class="p">)):</span>
|
||||
<span class="lineno">105</span> <span class="n">correct</span><span class="p">[</span><span class="n">j</span><span class="p">]</span> <span class="o">+=</span> <span class="mi">1</span> <span class="k">if</span> <span class="n">others</span><span class="p">[</span><span class="n">j</span><span class="p">][</span><span class="n">i</span><span class="p">]</span> <span class="o">==</span> <span class="n">target</span><span class="p">[</span><span class="n">i</span><span class="p">]</span> <span class="k">else</span> <span class="mi">0</span>
|
||||
<span class="lineno">106</span>
|
||||
<span class="lineno">107</span> <span class="n">parts</span> <span class="o">+=</span> <span class="p">[(</span><span class="s1">'"'</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">),</span>
|
||||
<span class="lineno">108</span> <span class="p">(</span><span class="n">text</span><span class="p">[</span><span class="n">i</span><span class="p">][</span><span class="n">j</span> <span class="o">+</span> <span class="mi">1</span><span class="p">],</span> <span class="n">Text</span><span class="o">.</span><span class="n">success</span> <span class="k">if</span> <span class="n">others</span><span class="p">[</span><span class="n">j</span><span class="p">][</span><span class="n">i</span><span class="p">]</span> <span class="o">==</span> <span class="n">target</span><span class="p">[</span><span class="n">i</span><span class="p">]</span> <span class="k">else</span> <span class="n">Text</span><span class="o">.</span><span class="n">danger</span><span class="p">),</span>
|
||||
<span class="lineno">109</span> <span class="p">(</span><span class="s1">'"'</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">subtle</span><span class="p">),</span> <span class="s1">'</span><span class="se">\t</span><span class="s1">'</span><span class="p">]</span>
|
||||
<span class="lineno">110</span>
|
||||
<span class="lineno">111</span> <span class="n">logger</span><span class="o">.</span><span class="n">log</span><span class="p">(</span><span class="n">parts</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>統計情報</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">114</span> <span class="n">parts</span> <span class="o">=</span> <span class="p">[(</span><span class="sa">f</span><span class="s1">'</span><span class="si">{</span><span class="n">total</span><span class="si">}</span><span class="s1">'</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">highlight</span><span class="p">),</span> <span class="s1">'</span><span class="se">\t</span><span class="s1">'</span><span class="p">]</span>
|
||||
<span class="lineno">115</span> <span class="k">for</span> <span class="n">j</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">others</span><span class="p">)):</span>
|
||||
<span class="lineno">116</span> <span class="n">parts</span> <span class="o">+=</span> <span class="p">[(</span><span class="sa">f</span><span class="s1">'</span><span class="si">{</span><span class="n">correct</span><span class="p">[</span><span class="n">j</span><span class="p">]</span><span class="si">}</span><span class="s1">'</span><span class="p">,</span> <span class="n">Text</span><span class="o">.</span><span class="n">value</span><span class="p">),</span> <span class="s1">'</span><span class="se">\t</span><span class="s1">'</span><span class="p">]</span>
|
||||
<span class="lineno">117</span> <span class="n">logger</span><span class="o">.</span><span class="n">log</span><span class="p">(</span><span class="n">parts</span><span class="p">)</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-15'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-15'>#</a>
|
||||
</div>
|
||||
<h3>バランスレイヤー</h3>
|
||||
<p>に分割<code class="highlight"><span></span><span class="n">n_chunks</span></code>
|
||||
. <code class="highlight"><span></span><span class="n">n_layers</span></code>
|
||||
これはパイプラインの並列トレーニングに使用されます。</p>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">n_layers</span></code>
|
||||
はレイヤーの数</li>
|
||||
<li><code class="highlight"><span></span><span class="n">n_chunks</span></code>
|
||||
はチャンクの数</li>
|
||||
<p>Returns は、<em>各チャンクのレイヤー数のリストを返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">120</span><span class="k">def</span> <span class="nf">balance_layers_simple</span><span class="p">(</span><span class="n">n_layers</span><span class="p">:</span> <span class="nb">int</span><span class="p">,</span> <span class="n">n_chunks</span><span class="p">:</span> <span class="nb">int</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">130</span> <span class="n">balance</span> <span class="o">=</span> <span class="p">[]</span>
|
||||
<span class="lineno">131</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="n">n_chunks</span><span class="p">):</span>
|
||||
<span class="lineno">132</span> <span class="n">balance</span><span class="o">.</span><span class="n">append</span><span class="p">((</span><span class="n">n_layers</span> <span class="o">-</span> <span class="nb">sum</span><span class="p">(</span><span class="n">balance</span><span class="p">))</span> <span class="o">//</span> <span class="p">(</span><span class="n">n_chunks</span> <span class="o">-</span> <span class="n">i</span><span class="p">))</span>
|
||||
<span class="lineno">133</span>
|
||||
<span class="lineno">134</span> <span class="k">return</span> <span class="nb">list</span><span class="p">(</span><span class="nb">reversed</span><span class="p">(</span><span class="n">balance</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>
|
||||
@@ -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="nn.linear 層を 8 ビットの整数層に変換します。"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="GPT ネオックスの llm.int8 ()"/>
|
||||
<meta name="twitter:description" content="nn.linear 層を 8 ビットの整数層に変換します。"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/utils/llm_int8.html"/>
|
||||
<meta property="og:title" content="GPT ネオックスの llm.int8 ()"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="GPT ネオックスの llm.int8 ()"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="GPT ネオックスの llm.int8 ()"/>
|
||||
<meta property="og:description" content="nn.linear 層を 8 ビットの整数層に変換します。"/>
|
||||
|
||||
<title>GPT ネオックスの llm.int8 ()</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/utils/llm_int8.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">neox</a>
|
||||
<a class="parent" href="index.html">utils</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/neox/utils/llm_int8.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>GPT ネオックスの llm.int ()</h1>
|
||||
<p>これは、レイヤーを llm.int8 () <code class="highlight"><span></span><span class="n">nn</span><span class="o">.</span><span class="n">Linear</span></code>
|
||||
線形レイヤーに変換するユーティリティ関数を実装します。</p>
|
||||
<p><a href="https://papers.labml.ai/paper/eb2bcaee1d0011edaa66a71c10a887e7">LLM.int8 () の論文では</a>、大規模な言語モデルでパフォーマンスを低下させることなく、外れ値を処理する際に int8 量子化を使用してメモリ使用量を削減できることが示されています。重みと入力をスケーリングされた8ビット整数に変換し、行列乗算を行ってint32の結果を生成し、それをfloat16に変換して再スケーリングします。その結果、大規模な言語モデルでは、一部の特徴によって極端な値 (外れ値) がモデルの出力の大部分を占める可能性があることが示されています。これらの機能は8ビットの整数空間に制限されるため、モデルのパフォーマンスが低下します。解決策として、これらの外れ値（指定されたしきい値を超える）を選択し、その乗算をfloat16空間で個別に計算します。外れ値の割合は約 0.01% なので、メモリ使用量は増加せず、モデルのパフォーマンス低下を防ぎます</p>。
|
||||
<p><a href="../model.html#post_load_prepare">GPT-NoEx レイヤーを変換するコードは model.py で定義されています。</a></p>
|
||||
<p>int8 量子化による GPT-Neox の使用例を以下に示します。</p>
|
||||
<ul><li><a href="../samples/llm_int8.html">テキストを生成</a></li>
|
||||
<li><a href="../evaluation/llm_int8.html">評価テストを実行</a></li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">33</span><span></span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-1'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-1'>#</a>
|
||||
</div>
|
||||
<p><a href="https://github.com/timdettmers/bitsandbytes"><code class="highlight"><span></span><span class="n">bitsandbytes</span></code>
|
||||
</a>パッケージをインポート</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">34</span><span class="k">try</span><span class="p">:</span>
|
||||
<span class="lineno">35</span> <span class="kn">from</span> <span class="nn">bitsandbytes.nn</span> <span class="kn">import</span> <span class="n">Linear8bitLt</span><span class="p">,</span> <span class="n">Int8Params</span>
|
||||
<span class="lineno">36</span><span class="k">except</span> <span class="ne">ImportError</span><span class="p">:</span>
|
||||
<span class="lineno">37</span> <span class="k">raise</span> <span class="ne">ImportError</span><span class="p">(</span><span class="s1">'''Please install `bitsandbytes` with `pip install bitsandbytes -U`'''</span><span class="p">)</span>
|
||||
<span class="lineno">38</span>
|
||||
<span class="lineno">39</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">40</span><span class="kn">from</span> <span class="nn">torch</span> <span class="kn">import</span> <span class="n">nn</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>
|
||||
<h2><code class="highlight"><span></span><span class="n">nn</span><span class="o">.</span><span class="n">Linear</span></code>
|
||||
レイヤーを llm.int8 () リニアレイヤーに変換します</h2>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">linear_module</span></code>
|
||||
<code class="highlight"><span></span><span class="n">nn</span><span class="o">.</span><span class="n">Linear</span></code>
|
||||
変換するレイヤーです</li>
|
||||
<li><code class="highlight"><span></span><span class="n">device</span></code>
|
||||
モデルのデバイスです</li>
|
||||
<li><code class="highlight"><span></span><span class="n">threshold</span></code>
|
||||
<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>外れ値の検出に使用する閾値です</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">43</span><span class="k">def</span> <span class="nf">make_llm_int8_linear</span><span class="p">(</span><span class="n">linear_module</span><span class="p">:</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</span><span class="p">,</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="p">,</span> <span class="n">threshold</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">6.0</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">53</span> <span class="k">assert</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">linear_module</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">Linear</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>空の Linear8BitLT モジュールを作成します。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">56</span> <span class="n">int8_lin</span> <span class="o">=</span> <span class="n">Linear8bitLt</span><span class="p">(</span>
|
||||
<span class="lineno">57</span> <span class="n">linear_module</span><span class="o">.</span><span class="n">in_features</span><span class="p">,</span>
|
||||
<span class="lineno">58</span> <span class="n">linear_module</span><span class="o">.</span><span class="n">out_features</span><span class="p">,</span>
|
||||
<span class="lineno">59</span> <span class="n">linear_module</span><span class="o">.</span><span class="n">bias</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">,</span>
|
||||
<span class="lineno">60</span> <span class="n">has_fp16_weights</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
||||
<span class="lineno">61</span> <span class="n">threshold</span><span class="o">=</span><span class="n">threshold</span><span class="p">,</span>
|
||||
<span class="lineno">62</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>ウェイトをクオンタイズ</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">65</span> <span class="n">int8_lin</span><span class="o">.</span><span class="n">_parameters</span><span class="p">[</span><span class="s1">'weight'</span><span class="p">]</span> <span class="o">=</span> <span class="n">Int8Params</span><span class="p">(</span><span class="n">linear_module</span><span class="o">.</span><span class="n">weight</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">cpu</span><span class="p">(),</span>
|
||||
<span class="lineno">66</span> <span class="n">requires_grad</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
|
||||
<span class="lineno">67</span> <span class="n">has_fp16_weights</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</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>float16 空間にバイアスを設定します。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">70</span> <span class="k">if</span> <span class="n">linear_module</span><span class="o">.</span><span class="n">bias</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">71</span> <span class="n">int8_lin</span><span class="o">.</span><span class="n">_parameters</span><span class="p">[</span><span class="s1">'bias'</span><span class="p">]</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">linear_module</span><span class="o">.</span><span class="n">bias</span><span class="o">.</span><span class="n">data</span><span class="p">,</span>
|
||||
<span class="lineno">72</span> <span class="n">requires_grad</span><span class="o">=</span><span class="kc">False</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">75</span> <span class="k">return</span> <span class="n">int8_lin</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>
|
||||
@@ -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="テキストデータセットをロードしてGPT-Neoxを微調整します"/>
|
||||
|
||||
<meta name="twitter:card" content="summary"/>
|
||||
<meta name="twitter:image:src" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta name="twitter:title" content="GPT-Neox 用テキストデータセット"/>
|
||||
<meta name="twitter:description" content="テキストデータセットをロードしてGPT-Neoxを微調整します"/>
|
||||
<meta name="twitter:site" content="@labmlai"/>
|
||||
<meta name="twitter:creator" content="@labmlai"/>
|
||||
|
||||
<meta property="og:url" content="https://nn.labml.ai/neox/utils/text_dataset.html"/>
|
||||
<meta property="og:title" content="GPT-Neox 用テキストデータセット"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="GPT-Neox 用テキストデータセット"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="GPT-Neox 用テキストデータセット"/>
|
||||
<meta property="og:description" content="テキストデータセットをロードしてGPT-Neoxを微調整します"/>
|
||||
|
||||
<title>GPT-Neox 用テキストデータセット</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/utils/text_dataset.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">neox</a>
|
||||
<a class="parent" href="index.html">utils</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/neox/utils/text_dataset.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>GPT-Neox 用テキストデータセット</h1>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">10</span><span></span><span class="kn">from</span> <span class="nn">pathlib</span> <span class="kn">import</span> <span class="n">PurePath</span><span class="p">,</span> <span class="n">Path</span>
|
||||
<span class="lineno">11</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">12</span>
|
||||
<span class="lineno">13</span><span class="kn">import</span> <span class="nn">torch</span>
|
||||
<span class="lineno">14</span><span class="kn">import</span> <span class="nn">torch.utils.data</span>
|
||||
<span class="lineno">15</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">lab</span>
|
||||
<span class="lineno">16</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">monit</span>
|
||||
<span class="lineno">17</span><span class="kn">from</span> <span class="nn">labml.logger</span> <span class="kn">import</span> <span class="n">inspect</span>
|
||||
<span class="lineno">18</span><span class="kn">from</span> <span class="nn">labml.utils.download</span> <span class="kn">import</span> <span class="n">download_file</span>
|
||||
<span class="lineno">19</span>
|
||||
<span class="lineno">20</span><span class="kn">from</span> <span class="nn">labml_nn.neox.tokenizer</span> <span class="kn">import</span> <span class="n">get_tokenizer</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>
|
||||
<h3>テキストファイルの読み込み</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">path</span></code>
|
||||
テキストファイルの場所です</li>
|
||||
<li><code class="highlight"><span></span><span class="n">url</span></code>
|
||||
ファイルをダウンロードする URL</li>
|
||||
<li><code class="highlight"><span></span><span class="n">filter_subset</span></code>
|
||||
フィルタリングする文字の数です。大規模なデータセットを試すときのテスト時にこれを使用してください</li>
|
||||
<p><em>テキストコンテンツを返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">23</span><span class="k">def</span> <span class="nf">load_text</span><span class="p">(</span><span class="n">path</span><span class="p">:</span> <span class="n">PurePath</span><span class="p">,</span> <span class="n">url</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span> <span class="o">*</span><span class="p">,</span> <span class="n">filter_subset</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">int</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-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">34</span> <span class="n">path</span> <span class="o">=</span> <span class="n">Path</span><span class="p">(</span><span class="n">path</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>存在しない場合はダウンロード</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">37</span> <span class="k">if</span> <span class="ow">not</span> <span class="n">path</span><span class="o">.</span><span class="n">exists</span><span class="p">():</span>
|
||||
<span class="lineno">38</span> <span class="k">if</span> <span class="ow">not</span> <span class="n">url</span><span class="p">:</span>
|
||||
<span class="lineno">39</span> <span class="k">raise</span> <span class="ne">FileNotFoundError</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">path</span><span class="p">))</span>
|
||||
<span class="lineno">40</span> <span class="k">else</span><span class="p">:</span>
|
||||
<span class="lineno">41</span> <span class="n">download_file</span><span class="p">(</span><span class="n">url</span><span class="p">,</span> <span class="n">path</span><span class="p">)</span>
|
||||
<span class="lineno">42</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="s2">"Load data"</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>データを読み込む</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">45</span> <span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="nb">str</span><span class="p">(</span><span class="n">path</span><span class="p">),</span> <span class="s1">'r'</span><span class="p">)</span> <span class="k">as</span> <span class="n">f</span><span class="p">:</span>
|
||||
<span class="lineno">46</span> <span class="n">text</span> <span class="o">=</span> <span class="n">f</span><span class="o">.</span><span class="n">read</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>フィルタ</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">48</span> <span class="k">if</span> <span class="n">filter_subset</span><span class="p">:</span>
|
||||
<span class="lineno">49</span> <span class="n">text</span> <span class="o">=</span> <span class="n">text</span><span class="p">[:</span><span class="n">filter_subset</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">52</span> <span class="k">return</span> <span class="n">text</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>
|
||||
<h2>GPT-Neox を微調整するためのデータセット</h2>
|
||||
<p>これは非常に大きなデータセットには最適化されていません。</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">55</span><span class="k">class</span> <span class="nc">NeoXDataset</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">Dataset</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>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">tokens</span></code>
|
||||
トークン ID のリストです</li>
|
||||
<li><code class="highlight"><span></span><span class="n">seq_len</span></code>
|
||||
は 1 つのトレーニングサンプルのシーケンス長です</li></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">62</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">tokens</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">seq_len</span><span class="p">:</span> <span class="nb">int</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">68</span> <span class="bp">self</span><span class="o">.</span><span class="n">seq_len</span> <span class="o">=</span> <span class="n">seq_len</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>サンプル数</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">70</span> <span class="n">n_samples</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">tokens</span><span class="p">)</span> <span class="o">//</span> <span class="n">seq_len</span>
|
||||
<span class="lineno">71</span> <span class="bp">self</span><span class="o">.</span><span class="n">n_samples</span> <span class="o">=</span> <span class="n">n_samples</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>切り捨て</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">73</span> <span class="n">tokens</span> <span class="o">=</span> <span class="n">tokens</span><span class="p">[:</span><span class="n">n_samples</span> <span class="o">*</span> <span class="n">seq_len</span> <span class="o">+</span> <span class="mi">1</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>PyTorch テンソルの作成</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">75</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokens</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="n">tokens</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">77</span> <span class="k">def</span> <span class="fm">__len__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||||
<span class="lineno">78</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">n_samples</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-14'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
<h3>サンプルを入手</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">idx</span></code>
|
||||
サンプルのインデックスです</li>
|
||||
<p><em>入力とターゲットを返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">80</span> <span class="k">def</span> <span class="fm">__getitem__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">idx</span><span class="p">:</span> <span class="nb">int</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">87</span> <span class="n">offset</span> <span class="o">=</span> <span class="n">idx</span> <span class="o">*</span> <span class="bp">self</span><span class="o">.</span><span class="n">seq_len</span>
|
||||
<span class="lineno">88</span> <span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokens</span><span class="p">[</span><span class="n">offset</span><span class="p">:</span><span class="n">offset</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">seq_len</span><span class="p">],</span> <span class="bp">self</span><span class="o">.</span><span class="n">tokens</span><span class="p">[</span><span class="n">offset</span> <span class="o">+</span> <span class="mi">1</span><span class="p">:</span><span class="n">offset</span> <span class="o">+</span> <span class="mi">1</span> <span class="o">+</span> <span class="bp">self</span><span class="o">.</span><span class="n">seq_len</span><span class="p">]</span>
|
||||
<span class="lineno">89</span>
|
||||
<span class="lineno">90</span>
|
||||
<span class="lineno">91</span><span class="n">DATASETS</span> <span class="o">=</span> <span class="p">{</span>
|
||||
<span class="lineno">92</span> <span class="s1">'tiny_shakespeare'</span><span class="p">:</span> <span class="p">{</span>
|
||||
<span class="lineno">93</span> <span class="s1">'file'</span><span class="p">:</span> <span class="s1">'tiny_shakespeare.txt'</span><span class="p">,</span>
|
||||
<span class="lineno">94</span> <span class="s1">'url'</span><span class="p">:</span> <span class="s1">'https://raw.githubusercontent.com/karpathy/char-rnn/master/data/tinyshakespeare/input.txt'</span>
|
||||
<span class="lineno">95</span> <span class="p">}</span>
|
||||
<span class="lineno">96</span><span class="p">}</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs doc-strings'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
<h3>データセットの読み込み</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">seq_len</span></code>
|
||||
は 1 つのトレーニングサンプルのシーケンス長です</li>
|
||||
<li><code class="highlight"><span></span><span class="n">dataset_name</span></code>
|
||||
はデータセットの名前</li>
|
||||
<p><em>データセットを返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">99</span><span class="k">def</span> <span class="nf">get_training_data</span><span class="p">(</span><span class="n">seq_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">32</span><span class="p">,</span> <span class="n">dataset_name</span><span class="p">:</span> <span class="nb">str</span> <span class="o">=</span> <span class="s1">'tiny_shakespeare'</span><span class="p">,</span> <span class="n">truncate</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="o">-</span><span class="mi">1</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">108</span> <span class="n">ds</span> <span class="o">=</span> <span class="n">DATASETS</span><span class="p">[</span><span class="n">dataset_name</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>コンテンツを読み込む</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">110</span> <span class="n">text</span> <span class="o">=</span> <span class="n">load_text</span><span class="p">(</span><span class="n">lab</span><span class="o">.</span><span class="n">get_data_path</span><span class="p">()</span> <span class="o">/</span> <span class="n">ds</span><span class="p">[</span><span class="s1">'file'</span><span class="p">],</span> <span class="n">ds</span><span class="p">[</span><span class="s1">'url'</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>トークン化</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">112</span> <span class="n">tokenizer</span> <span class="o">=</span> <span class="n">get_tokenizer</span><span class="p">()</span>
|
||||
<span class="lineno">113</span> <span class="n">tokens</span> <span class="o">=</span> <span class="n">tokenizer</span><span class="o">.</span><span class="n">encode_batch</span><span class="p">([</span><span class="n">text</span><span class="p">])[</span><span class="mi">0</span><span class="p">]</span>
|
||||
<span class="lineno">114</span>
|
||||
<span class="lineno">115</span> <span class="k">if</span> <span class="n">truncate</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
||||
<span class="lineno">116</span> <span class="n">token_ids</span> <span class="o">=</span> <span class="n">tokens</span><span class="o">.</span><span class="n">ids</span><span class="p">[:</span><span class="n">truncate</span> <span class="o">*</span> <span class="n">seq_len</span><span class="p">]</span>
|
||||
<span class="lineno">117</span> <span class="k">else</span><span class="p">:</span>
|
||||
<span class="lineno">118</span> <span class="n">token_ids</span> <span class="o">=</span> <span class="n">tokens</span><span class="o">.</span><span class="n">ids</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">121</span> <span class="k">return</span> <span class="n">NeoXDataset</span><span class="p">(</span><span class="n">token_ids</span><span class="p">,</span> <span class="n">seq_len</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">124</span><span class="k">def</span> <span class="nf">_test</span><span class="p">():</span>
|
||||
<span class="lineno">125</span> <span class="n">dataset</span> <span class="o">=</span> <span class="n">get_training_data</span><span class="p">()</span>
|
||||
<span class="lineno">126</span>
|
||||
<span class="lineno">127</span> <span class="n">inspect</span><span class="p">(</span><span class="n">tokens</span><span class="o">=</span><span class="nb">len</span><span class="p">(</span><span class="n">dataset</span><span class="o">.</span><span class="n">tokens</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">131</span><span class="k">if</span> <span class="vm">__name__</span> <span class="o">==</span> <span class="s1">'__main__'</span><span class="p">:</span>
|
||||
<span class="lineno">132</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>
|
||||
@@ -0,0 +1,537 @@
|
||||
<!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&v=4"/>
|
||||
<meta name="twitter:title" content="trainer.py"/>
|
||||
<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/neox/utils/trainer.html"/>
|
||||
<meta property="og:title" content="trainer.py"/>
|
||||
<meta property="og:image" content="https://avatars1.githubusercontent.com/u/64068543?s=400&v=4"/>
|
||||
<meta property="og:site_name" content="trainer.py"/>
|
||||
<meta property="og:type" content="object"/>
|
||||
<meta property="og:title" content="trainer.py"/>
|
||||
<meta property="og:description" content=""/>
|
||||
|
||||
<title>trainer.py</title>
|
||||
<link rel="shortcut icon" href="/icon.png"/>
|
||||
<link rel="stylesheet" href="../../pylit.css?v=1">
|
||||
<link rel="canonical" href="https://nn.labml.ai/neox/utils/trainer.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">neox</a>
|
||||
<a class="parent" href="index.html">utils</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/neox/utils/trainer.py" 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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">1</span><span></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">Set</span><span class="p">,</span> <span class="n">List</span>
|
||||
<span class="lineno">2</span>
|
||||
<span class="lineno">3</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">4</span><span class="kn">import</span> <span class="nn">torch.optim</span>
|
||||
<span class="lineno">5</span><span class="kn">import</span> <span class="nn">torch.utils.data</span>
|
||||
<span class="lineno">6</span><span class="kn">from</span> <span class="nn">torch.cuda</span> <span class="kn">import</span> <span class="n">amp</span>
|
||||
<span class="lineno">7</span><span class="kn">from</span> <span class="nn">torch.cuda.amp</span> <span class="kn">import</span> <span class="n">GradScaler</span>
|
||||
<span class="lineno">8</span>
|
||||
<span class="lineno">9</span><span class="kn">from</span> <span class="nn">labml</span> <span class="kn">import</span> <span class="n">monit</span><span class="p">,</span> <span class="n">tracker</span>
|
||||
<span class="lineno">10</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="lineno">11</span><span class="kn">from</span> <span class="nn">labml_nn.neox.utils.finetune</span> <span class="kn">import</span> <span class="n">FineTuner</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>
|
||||
<h3>トレーニング可能なパラメータを取得</h3>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">model</span></code>
|
||||
トレーニングするモデルです</li>
|
||||
<p><em>トレーニング用のパラメータのリストを返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">14</span><span class="k">def</span> <span class="nf">get_trainable_params</span><span class="p">(</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="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>すべてのパラメータを取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">23</span> <span class="n">params</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">model</span><span class="o">.</span><span class="n">parameters</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>グラデーションを必要とするフィルターパラメーター</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">25</span> <span class="n">trainable_params</span> <span class="o">=</span> <span class="p">[</span><span class="n">p</span> <span class="k">for</span> <span class="n">p</span> <span class="ow">in</span> <span class="n">params</span> <span class="k">if</span> <span class="n">p</span><span class="o">.</span><span class="n">requires_grad</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></p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">28</span> <span class="k">return</span> <span class="n">trainable_params</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">31</span><span class="k">class</span> <span class="nc">TrainerConf</span><span class="p">(</span><span class="n">BaseConfigs</span><span class="p">):</span>
|
||||
<span class="lineno">32</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">33</span> <span class="n">layers</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</span><span class="p">]</span>
|
||||
<span class="lineno">34</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">Optimizer</span> <span class="o">=</span> <span class="s1">'Adam'</span>
|
||||
<span class="lineno">35</span> <span class="n">train_loader</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">DataLoader</span>
|
||||
<span class="lineno">36</span> <span class="n">valid_loader</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">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">DataLoader</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">,</span>
|
||||
<span class="lineno">37</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">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cuda:0'</span><span class="p">)</span>
|
||||
<span class="lineno">38</span> <span class="n">scaler</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">GradScaler</span><span class="p">]</span> <span class="o">=</span> <span class="s1">'Default'</span>
|
||||
<span class="lineno">39</span> <span class="n">is_amp</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span>
|
||||
<span class="lineno">40</span> <span class="n">dtype</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">dtype</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">float16</span>
|
||||
<span class="lineno">41</span>
|
||||
<span class="lineno">42</span> <span class="n">is_clone_layers</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span>
|
||||
<span class="lineno">43</span>
|
||||
<span class="lineno">44</span> <span class="n">loss_func</span><span class="p">:</span> <span class="n">nn</span><span class="o">.</span><span class="n">Module</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">45</span> <span class="n">checkpoints_per_epoch</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">0</span>
|
||||
<span class="lineno">46</span> <span class="n">samples_per_epoch</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">0</span>
|
||||
<span class="lineno">47</span>
|
||||
<span class="lineno">48</span> <span class="n">grad_norm</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">float</span><span class="p">]</span> <span class="o">=</span> <span class="mf">1.0</span>
|
||||
<span class="lineno">49</span> <span class="n">learning_rate</span><span class="p">:</span> <span class="nb">float</span> <span class="o">=</span> <span class="mf">3e-4</span>
|
||||
<span class="lineno">50</span> <span class="n">max_seq_len</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">1024</span>
|
||||
<span class="lineno">51</span> <span class="n">batch_size</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">64</span>
|
||||
<span class="lineno">52</span> <span class="n">epochs</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">16</span>
|
||||
<span class="lineno">53</span>
|
||||
<span class="lineno">54</span> <span class="n">n_gpus</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">device_count</span><span class="p">()</span>
|
||||
<span class="lineno">55</span>
|
||||
<span class="lineno">56</span> <span class="n">filter_layers</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">Set</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</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>
|
||||
<ul><li><code class="highlight"><span></span><span class="n">dataset_split</span></code>
|
||||
電車/有効</li>
|
||||
<li><code class="highlight"><span></span><span class="n">sample</span></code>
|
||||
サンプルです</li>
|
||||
<p>損失、出力、<em>ターゲットを返します</em></p></ul>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">58</span> <span class="k">def</span> <span class="nf">get_loss</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">sample</span><span class="p">,</span> <span class="n">dataset_split</span><span class="p">:</span> <span class="nb">str</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">64</span> <span class="n">data</span><span class="p">,</span> <span class="n">target</span> <span class="o">=</span> <span class="n">sample</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>フォワードパス</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">67</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">'Forward pass'</span><span class="p">):</span>
|
||||
<span class="lineno">68</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="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>ターゲットを出力と同じデバイスに移動</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">70</span> <span class="n">target</span> <span class="o">=</span> <span class="n">target</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">output</span><span class="o">.</span><span class="n">device</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>損失の計算</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">72</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="o">.</span><span class="n">view</span><span class="p">(</span><span class="n">target</span><span class="o">.</span><span class="n">numel</span><span class="p">(),</span> <span class="o">-</span><span class="mi">1</span><span class="p">),</span> <span class="n">target</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="lineno">73</span>
|
||||
<span class="lineno">74</span> <span class="k">return</span> <span class="n">loss</span><span class="p">,</span> <span class="n">output</span><span class="p">,</span> <span class="n">target</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-11'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-11'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">76</span> <span class="k">def</span> <span class="nf">train</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
|
||||
<span class="lineno">77</span> <span class="k">for</span> <span class="n">epoch</span> <span class="ow">in</span> <span class="n">monit</span><span class="o">.</span><span class="n">loop</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">epochs</span><span class="p">):</span>
|
||||
<span class="lineno">78</span> <span class="bp">self</span><span class="o">.</span><span class="n">train_epoch</span><span class="p">()</span>
|
||||
<span class="lineno">79</span> <span class="n">tracker</span><span class="o">.</span><span class="n">new_line</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>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">81</span> <span class="k">def</span> <span class="nf">sample</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">idx</span><span class="p">):</span>
|
||||
<span class="lineno">82</span> <span class="k">pass</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">84</span> <span class="k">def</span> <span class="nf">save_checkpoint</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">idx</span><span class="p">):</span>
|
||||
<span class="lineno">85</span> <span class="k">pass</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-14'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-14'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">87</span> <span class="k">def</span> <span class="nf">get_iterators</span><span class="p">(</span><span class="bp">self</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>バッチを繰り返し処理する</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">89</span> <span class="n">iterators</span> <span class="o">=</span> <span class="p">[(</span><span class="s1">'train'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">train_loader</span><span class="p">)]</span>
|
||||
<span class="lineno">90</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">valid_loader</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">91</span> <span class="n">iterators</span><span class="o">.</span><span class="n">append</span><span class="p">((</span><span class="s1">'valid'</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">valid_loader</span><span class="p">))</span>
|
||||
<span class="lineno">92</span>
|
||||
<span class="lineno">93</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">samples_per_epoch</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
||||
<span class="lineno">94</span> <span class="n">iterators</span><span class="o">.</span><span class="n">append</span><span class="p">((</span><span class="bp">self</span><span class="o">.</span><span class="n">sample</span><span class="p">,</span> <span class="p">[</span><span class="n">i</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="bp">self</span><span class="o">.</span><span class="n">samples_per_epoch</span><span class="p">)]))</span>
|
||||
<span class="lineno">95</span>
|
||||
<span class="lineno">96</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">checkpoints_per_epoch</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
|
||||
<span class="lineno">97</span> <span class="n">iterators</span><span class="o">.</span><span class="n">append</span><span class="p">((</span><span class="bp">self</span><span class="o">.</span><span class="n">save_checkpoint</span><span class="p">,</span> <span class="p">[</span><span class="n">i</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="bp">self</span><span class="o">.</span><span class="n">checkpoints_per_epoch</span><span class="p">)]))</span>
|
||||
<span class="lineno">98</span>
|
||||
<span class="lineno">99</span> <span class="k">return</span> <span class="n">iterators</span></pre></div>
|
||||
</div>
|
||||
</div>
|
||||
<div class='section' id='section-16'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-16'>#</a>
|
||||
</div>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">101</span> <span class="k">def</span> <span class="nf">train_epoch</span><span class="p">(</span><span class="bp">self</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>鉄道模型を設定</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">103</span> <span class="bp">self</span><span class="o">.</span><span class="n">model</span><span class="o">.</span><span class="n">train</span><span class="p">()</span>
|
||||
<span class="lineno">104</span>
|
||||
<span class="lineno">105</span> <span class="n">iterators</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">get_iterators</span><span class="p">()</span>
|
||||
<span class="lineno">106</span> <span class="k">for</span> <span class="n">split_name</span><span class="p">,</span> <span class="n">sample</span> <span class="ow">in</span> <span class="n">monit</span><span class="o">.</span><span class="n">mix</span><span class="p">(</span><span class="mi">1024</span><span class="p">,</span> <span class="o">*</span><span class="n">iterators</span><span class="p">):</span>
|
||||
<span class="lineno">107</span> <span class="k">if</span> <span class="n">split_name</span> <span class="o">==</span> <span class="s1">'train'</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>グラデーションをゼロに設定</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">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
|
||||
<span class="lineno">110</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add_global_step</span><span class="p">()</span>
|
||||
<span class="lineno">111</span>
|
||||
<span class="lineno">112</span> <span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">set_grad_enabled</span><span class="p">(</span><span class="n">split_name</span> <span class="o">==</span> <span class="s1">'train'</span><span class="p">):</span>
|
||||
<span class="lineno">113</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">is_amp</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>フォワードパス</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">115</span> <span class="k">with</span> <span class="n">amp</span><span class="o">.</span><span class="n">autocast</span><span class="p">():</span>
|
||||
<span class="lineno">116</span> <span class="n">loss</span><span class="p">,</span> <span class="n">output</span><span class="p">,</span> <span class="n">target</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">get_loss</span><span class="p">(</span><span class="n">sample</span><span class="p">,</span> <span class="n">split_name</span><span class="p">)</span>
|
||||
<span class="lineno">117</span> <span class="k">else</span><span class="p">:</span>
|
||||
<span class="lineno">118</span> <span class="n">loss</span><span class="p">,</span> <span class="n">output</span><span class="p">,</span> <span class="n">target</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">get_loss</span><span class="p">(</span><span class="n">sample</span><span class="p">,</span> <span class="n">split_name</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>予測を取得</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">121</span> <span class="n">pred</span> <span class="o">=</span> <span class="n">output</span><span class="o">.</span><span class="n">argmax</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-21'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-21'>#</a>
|
||||
</div>
|
||||
<p>精度を計算</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">123</span> <span class="n">accuracy</span> <span class="o">=</span> <span class="n">pred</span><span class="o">.</span><span class="n">eq</span><span class="p">(</span><span class="n">target</span><span class="p">)</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span><span class="o">.</span><span class="n">item</span><span class="p">()</span> <span class="o">/</span> <span class="p">(</span><span class="n">target</span> <span class="o">!=</span> <span class="o">-</span><span class="mi">100</span><span class="p">)</span><span class="o">.</span><span class="n">sum</span><span class="p">()</span>
|
||||
<span class="lineno">124</span>
|
||||
<span class="lineno">125</span> <span class="n">tracker</span><span class="o">.</span><span class="n">add</span><span class="p">({</span><span class="sa">f</span><span class="s1">'loss.</span><span class="si">{</span><span class="n">split_name</span><span class="si">}</span><span class="s1">'</span><span class="p">:</span> <span class="n">loss</span><span class="p">,</span> <span class="sa">f</span><span class="s1">'acc.</span><span class="si">{</span><span class="n">split_name</span><span class="si">}</span><span class="s1">'</span><span class="p">:</span> <span class="n">accuracy</span> <span class="o">*</span> <span class="mi">100</span><span class="p">})</span>
|
||||
<span class="lineno">126</span>
|
||||
<span class="lineno">127</span> <span class="k">if</span> <span class="n">split_name</span> <span class="o">==</span> <span class="s1">'train'</span><span class="p">:</span>
|
||||
<span class="lineno">128</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</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>バックワードパス</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">130</span> <span class="n">loss</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span><span class="o">.</span><span class="n">scale</span><span class="p">(</span><span class="n">loss</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>トラッカー追加 ({'損失.scaled': 損失})</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">133</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">'Backward pass'</span><span class="p">):</span>
|
||||
<span class="lineno">134</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-24'>
|
||||
<div class='docs'>
|
||||
<div class='section-link'>
|
||||
<a href='#section-24'>#</a>
|
||||
</div>
|
||||
<p>最適化</p>
|
||||
|
||||
</div>
|
||||
<div class='code'>
|
||||
<div class="highlight"><pre><span class="lineno">137</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">'Optimize'</span><span class="p">):</span>
|
||||
<span class="lineno">138</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">139</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>
|
||||
<span class="lineno">140</span> <span class="k">else</span><span class="p">:</span>
|
||||
<span class="lineno">141</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span><span class="o">.</span><span class="n">unscale_</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="lineno">142</span> <span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">grad_norm</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
|
||||
<span class="lineno">143</span> <span class="n">torch</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">clip_grad_norm_</span><span class="p">(</span><span class="n">get_trainable_params</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="bp">self</span><span class="o">.</span><span class="n">grad_norm</span><span class="p">)</span>
|
||||
<span class="lineno">144</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span><span class="o">.</span><span class="n">step</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="lineno">145</span> <span class="bp">self</span><span class="o">.</span><span class="n">scaler</span><span class="o">.</span><span class="n">update</span><span class="p">()</span>
|
||||
<span class="lineno">146</span>
|
||||
<span class="lineno">147</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-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">150</span><span class="nd">@option</span><span class="p">(</span><span class="n">TrainerConf</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="s1">'Adam'</span><span class="p">)</span>
|
||||
<span class="lineno">151</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">TrainerConf</span><span class="p">):</span>
|
||||
<span class="lineno">152</span> <span class="k">if</span> <span class="n">c</span><span class="o">.</span><span class="n">dtype</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">float32</span><span class="p">:</span>
|
||||
<span class="lineno">153</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">Adam</span><span class="p">(</span><span class="n">get_trainable_params</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">model</span><span class="p">),</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="lineno">154</span> <span class="k">elif</span> <span class="n">c</span><span class="o">.</span><span class="n">dtype</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">float16</span><span class="p">:</span>
|
||||
<span class="lineno">155</span> <span class="kn">from</span> <span class="nn">labml_nn.optimizers.adam_fp16</span> <span class="kn">import</span> <span class="n">AdamFP16</span>
|
||||
<span class="lineno">156</span> <span class="k">return</span> <span class="n">AdamFP16</span><span class="p">(</span><span class="n">get_trainable_params</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">model</span><span class="p">),</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="lineno">157</span> <span class="k">else</span><span class="p">:</span>
|
||||
<span class="lineno">158</span> <span class="k">raise</span> <span class="ne">NotImplementedError</span><span class="p">()</span>
|
||||
<span class="lineno">159</span>
|
||||
<span class="lineno">160</span>
|
||||
<span class="lineno">161</span><span class="nd">@option</span><span class="p">(</span><span class="n">TrainerConf</span><span class="o">.</span><span class="n">optimizer</span><span class="p">,</span> <span class="s1">'SGD'</span><span class="p">)</span>
|
||||
<span class="lineno">162</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">TrainerConf</span><span class="p">):</span>
|
||||
<span class="lineno">163</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">get_trainable_params</span><span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">model</span><span class="p">),</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="lineno">164</span>
|
||||
<span class="lineno">165</span>
|
||||
<span class="lineno">166</span><span class="nd">@option</span><span class="p">(</span><span class="n">TrainerConf</span><span class="o">.</span><span class="n">scaler</span><span class="p">,</span> <span class="s1">'Default'</span><span class="p">)</span>
|
||||
<span class="lineno">167</span><span class="k">def</span> <span class="nf">grad_scaler</span><span class="p">(</span><span class="n">c</span><span class="p">:</span> <span class="n">TrainerConf</span><span class="p">):</span>
|
||||
<span class="lineno">168</span> <span class="k">if</span> <span class="ow">not</span> <span class="n">c</span><span class="o">.</span><span class="n">is_amp</span><span class="p">:</span>
|
||||
<span class="lineno">169</span> <span class="k">return</span> <span class="kc">None</span>
|
||||
<span class="lineno">170</span>
|
||||
<span class="lineno">171</span> <span class="k">if</span> <span class="n">c</span><span class="o">.</span><span class="n">dtype</span> <span class="o">==</span> <span class="n">torch</span><span class="o">.</span><span class="n">float16</span><span class="p">:</span>
|
||||
<span class="lineno">172</span> <span class="kn">from</span> <span class="nn">labml_nn.optimizers.adam_fp16</span> <span class="kn">import</span> <span class="n">GradScalerFP16</span>
|
||||
<span class="lineno">173</span> <span class="k">return</span> <span class="n">GradScalerFP16</span><span class="p">()</span>
|
||||
<span class="lineno">174</span> <span class="k">else</span><span class="p">:</span>
|
||||
<span class="lineno">175</span> <span class="k">return</span> <span class="n">GradScaler</span><span class="p">()</span>
|
||||
<span class="lineno">176</span>
|
||||
<span class="lineno">177</span>
|
||||
<span class="lineno">178</span><span class="k">class</span> <span class="nc">PipelineParallelTrainerConf</span><span class="p">(</span><span class="n">TrainerConf</span><span class="p">):</span>
|
||||
<span class="lineno">179</span> <span class="n">is_checkpointing</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span>
|
||||
<span class="lineno">180</span> <span class="n">chunks</span><span class="p">:</span> <span class="nb">int</span>
|
||||
<span class="lineno">181</span>
|
||||
<span class="lineno">182</span> <span class="n">fine_tuner</span><span class="p">:</span> <span class="n">FineTuner</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>
|
||||
Reference in New Issue
Block a user