Buckets:

hf-doc-build
/

doc-dev

Files

xet

hf-doc-build/doc-dev / course /pr_1021 /zh-CN /chapter6 /6.html

rtrm

about 2 months ago

download

raw

93.8 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"WordPiece tokenization 算法","local":"WordPiece tokenization算法","sections":[{"title":"WordPiece 训练","local":"WordPiece 训练","sections":[],"depth":2},{"title":"tokenization 算法","local":"tokenization 算法","sections":[],"depth":2},{"title":"实现 WordPiece","local":"实现 WordPiece","sections":[],"depth":2}],"depth":1}">
	<link href="/docs/course/pr_1021/zh-CN/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/entry/start.f3a1a511.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/scheduler.37c15a92.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/singletons.9bf55235.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/index.18351ede.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/paths.0ba10750.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/entry/app.c39e37cf.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/index.2bf4358c.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/nodes/0.dad18ce3.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/each.e59479a4.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/nodes/49.02638f23.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/Tip.363c041f.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/Youtube.1e50a667.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/CodeBlock.4e987730.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/CourseFloatingBanner.9ff4c771.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/DocNotebookDropdown.efc1fb7c.js">
	<link rel="modulepreload" href="/docs/course/pr_1021/zh-CN/_app/immutable/chunks/getInferenceSnippets.ebf8be91.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"WordPiece tokenization 算法","local":"WordPiece tokenization算法","sections":[{"title":"WordPiece 训练","local":"WordPiece 训练","sections":[],"depth":2},{"title":"tokenization 算法","local":"tokenization 算法","sections":[],"depth":2},{"title":"实现 WordPiece","local":"实现 WordPiece","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="WordPiece tokenization算法" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#WordPiece tokenization算法"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>WordPiece tokenization 算法</span></h1> <div class="flex space-x-1 absolute z-10 right-0 top-0"><a href="https://discuss.huggingface.co/t/chapter-6-questions" target="_blank"><img alt="Ask a Question" class="!m-0" src="https://img.shields.io/badge/Ask%20a%20question-ffcb4c.svg?logo=data:image/svg+xml;base64,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"></a> <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/master/course/zh-CN/chapter6/section6.ipynb" target="_blank"><img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"></a> <a href="https://studiolab.sagemaker.aws/import/github/huggingface/notebooks/blob/master/course/zh-CN/chapter6/section6.ipynb" target="_blank"><img alt="Open In Studio Lab" class="!m-0" src="https://studiolab.sagemaker.aws/studiolab.svg"></a></div> <p data-svelte-h="svelte-1fqctxb">WordPiece 是 Google 开发的用于 BERT 预训练的分词算法。自此之后，很多基于 BERT 的 Transformer 模型都复用了这种方法，比如 DistilBERT，MobileBERT，Funnel Transformers 和 MPNET。它在训练方面与 BPE 非常类似，但实际的分词方法有所不同。</p> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/qpv6ms_t_1A" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-1jw9ndh">💡 本节详细讲述了 WordPiece，甚至展示了一个完整的实现。如果你只想对这个分词算法有个大概的理解，可以直接跳到最后。</p></div> <h2 class="relative group"><a id="WordPiece 训练" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#WordPiece 训练"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>WordPiece 训练</span></h2> <div class="course-tip course-tip-orange bg-gradient-to-br dark:bg-gradient-to-r before:border-orange-500 dark:before:border-orange-800 from-orange-50 dark:from-gray-900 to-white dark:to-gray-950 border border-orange-50 text-orange-700 dark:text-gray-400"><p data-svelte-h="svelte-1i7swwn">⚠️ Google 从未开源 WordPiece 训练算法的实现，因此以下是我们基于已发表文献的最佳猜测。它可能并非 100％准确的。</p></div> <p data-svelte-h="svelte-11u566e">与BPE 一样，WordPiece 也是从包含模型使用的特殊 tokens 和初始字母表的小词汇表开始的。由于它是通过添加前缀（如 BERT 中的 <code>##</code> ）来识别子词的，每个词最初都会通过在词内部所有字符前添加该前缀进行分割。因此，例如 <code>"word"</code> 将被这样分割：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->w ##o ##r ##d<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-sszbho">因此，初始字母表包含所有出现在单词第一个位置的字符，以及出现在单词内部并带有 WordPiece 前缀的字符。</p> <p data-svelte-h="svelte-1592x9q">然后，同样像 BPE 一样，WordPiece 会学习合并规则。主要的不同之处在于合并对的选择方式。WordPiece 不是选择频率最高的对，而是对每对计算一个得分，使用以下公式：</p> <p><!-- HTML_TAG_START --><span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mrow><mi mathvariant="normal">s</mi><mi mathvariant="normal">c</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">e</mi></mrow><mo>=</mo><mo stretchy="false">(</mo><mrow><mi mathvariant="normal">f</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">q</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">f</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">p</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">r</mi></mrow><mo stretchy="false">)</mo><mi mathvariant="normal">/</mi><mo stretchy="false">(</mo><mrow><mi mathvariant="normal">f</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">q</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">f</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">f</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">s</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">l</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">m</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">t</mi></mrow><mo>×</mo><mrow><mi mathvariant="normal">f</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">q</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">f</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">s</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">c</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">d</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">l</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">m</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">t</mi></mrow><mo stretchy="false">)</mo></mrow><annotation encoding="application/x-tex">\mathrm{score} = (\mathrm{freq\_of\_pair}) / (\mathrm{freq\_of\_first\_element} \times\mathrm{freq\_of\_second\_element})</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:0.4306em;"></span><span class="mord"><span class="mord mathrm">score</span></span><span class="mspace" style="margin-right:0.2778em;"></span><span class="mrel">=</span><span class="mspace" style="margin-right:0.2778em;"></span></span><span class="base"><span class="strut" style="height:1.06em;vertical-align:-0.31em;"></span><span class="mopen">(</span><span class="mord"><span class="mord mathrm">freq_of_pair</span></span><span class="mclose">)</span><span class="mord">/</span><span class="mopen">(</span><span class="mord"><span class="mord mathrm">freq_of_first_element</span></span><span class="mspace" style="margin-right:0.2222em;"></span><span class="mbin">×</span><span class="mspace" style="margin-right:0.2222em;"></span></span><span class="base"><span class="strut" style="height:1.06em;vertical-align:-0.31em;"></span><span class="mord"><span class="mord mathrm">freq_of_second_element</span></span><span class="mclose">)</span></span></span></span></span><!-- HTML_TAG_END --></p> <p data-svelte-h="svelte-1cke216">通过将两部分合在一起的频率除以其中各部分的频率的乘积，该算法优先合并那些在词汇表中单独出现出现的对。例如，即使 <code>("un", "##able")</code> 这对在词汇表中出现的频率很高，它也不一定会被合并，因为 <code>"un"</code> 和 <code>"##able"</code> 这两对可能会在很多其他词中出现，频率很高。相比之下，像 <code>("hu", "##gging")</code> 这样的对可能会更快地被合并（假设单词“hugging”在词汇表中出现的频率很高），因为 <code>"hu"</code> 和 <code>"##gging"</code> 可能分别出现的频率较低。</p> <p data-svelte-h="svelte-uw1lny">我们使用与 BPE 示例相同的词汇表：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->(<span class="hljs-string">"hug"</span><span class="hljs-punctuation">,</span> <span class="hljs-number">10</span>)<span class="hljs-punctuation">,</span> (<span class="hljs-string">"pug"</span><span class="hljs-punctuation">,</span> <span class="hljs-number">5</span>)<span class="hljs-punctuation">,</span> (<span class="hljs-string">"pun"</span><span class="hljs-punctuation">,</span> <span class="hljs-number">12</span>)<span class="hljs-punctuation">,</span> (<span class="hljs-string">"bun"</span><span class="hljs-punctuation">,</span> <span class="hljs-number">4</span>)<span class="hljs-punctuation">,</span> (<span class="hljs-string">"hugs"</span><span class="hljs-punctuation">,</span> <span class="hljs-number">5</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1dsawga">经过分割之后将会是：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->(<span class="hljs-string">"h"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-number">10</span>), (<span class="hljs-string">"p"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-number">5</span>), (<span class="hljs-string">"p"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-number">12</span>), (<span class="hljs-string">"b"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-number">4</span>), (<span class="hljs-string">"h"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#s</span>"</span>, <span class="hljs-number">5</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-7tgvkb">所以最初的词汇表将会是 <code>["b", "h", "p", "##g", "##n", "##s", "##u"]</code> （如果我们暂时忽略特殊 tokens ）。出现频率最高的一对是 <code>("##u", "##g")</code> （目前 20 次），但 <code>"##u"</code> 和其他单词一起出现的频率非常高，所以它的分数不是最高的（分数是 1 / 36）。所有带有 <code>"##u"</code> 的对实际上都有相同的分数（1 / 36），所以分数最高的对是 <code>("##g", "##s")</code> —— 唯一没有 <code>"##u"</code> 的对——分数是 1 / 20，所以学习的第一个合并是 <code>("##g", "##s") -> ("##gs")</code> 。</p> <p data-svelte-h="svelte-10vwmky">请注意，当我们合并时，我们会删除两个 tokens 之间的 <code>##</code> ，所以我们将 <code>"##gs"</code> 添加到词汇表中，并将语料库的单词按照改规则进行合并：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->词汇表: [<span class="hljs-string">"b"</span>, <span class="hljs-string">"h"</span>, <span class="hljs-string">"p"</span>, <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#s</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#gs</span>"</span>]
	语料库: (<span class="hljs-string">"h"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-number">10</span>), (<span class="hljs-string">"p"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-number">5</span>), (<span class="hljs-string">"p"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-number">12</span>), (<span class="hljs-string">"b"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-number">4</span>), (<span class="hljs-string">"h"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#gs</span>"</span>, <span class="hljs-number">5</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1tt5wup">此时， <code>"##u"</code> 出现在所有可能的对中，因此它们最终都具有相同的分数。在这种情况下，第一个对会被合并，于是我们得到了 <code>("h", "##u") -> "hu"</code> 规则：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->词汇表: [<span class="hljs-string">"b"</span>, <span class="hljs-string">"h"</span>, <span class="hljs-string">"p"</span>, <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#s</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#gs</span>"</span>, <span class="hljs-string">"hu"</span>]
	语料库: (<span class="hljs-string">"hu"</span> <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-number">10</span>), (<span class="hljs-string">"p"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-number">5</span>), (<span class="hljs-string">"p"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-number">12</span>), (<span class="hljs-string">"b"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-number">4</span>), (<span class="hljs-string">"hu"</span> <span class="hljs-string">"#<span class="hljs-subst">#gs</span>"</span>, <span class="hljs-number">5</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1nj9dhv">然后，下一个最佳得分的对是 <code>("hu", "##g")</code> 和 <code>("hu", "##gs")</code> （得分为 1/15，而所有其他配对的得分为 1/21），因此得分最高的第一对合并：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->词汇表: [<span class="hljs-string">"b"</span>, <span class="hljs-string">"h"</span>, <span class="hljs-string">"p"</span>, <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#s</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span>, <span class="hljs-string">"#<span class="hljs-subst">#gs</span>"</span>, <span class="hljs-string">"hu"</span>, <span class="hljs-string">"hug"</span>]
	语料库: (<span class="hljs-string">"hug"</span>, <span class="hljs-number">10</span>), (<span class="hljs-string">"p"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#g</span>"</span>, <span class="hljs-number">5</span>), (<span class="hljs-string">"p"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-number">12</span>), (<span class="hljs-string">"b"</span> <span class="hljs-string">"#<span class="hljs-subst">#u</span>"</span> <span class="hljs-string">"#<span class="hljs-subst">#n</span>"</span>, <span class="hljs-number">4</span>), (<span class="hljs-string">"hu"</span> <span class="hljs-string">"#<span class="hljs-subst">#gs</span>"</span>, <span class="hljs-number">5</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1ib8ifw">然后我们就按此方式继续，直到我们达到所需的词汇表大小。</p> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-1q0b1w">✏️ <strong>现在轮到你了！</strong> 下一个合并规则是什么？</p></div> <h2 class="relative group"><a id="tokenization 算法" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#tokenization 算法"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>tokenization 算法</span></h2> <p data-svelte-h="svelte-1g7xvnn">WordPiece 和 BPE 的分词方式有所不同，WordPiece 只保存最终词汇表，而不保存学习到的合并规则。WordPiece 从待分词的词开始，找到词汇表中最长的子词，然后在其处分割。例如，如果我们使用上述示例中学习到的词汇表，对于词 <code>"hugs"</code> ，从开始处的最长子词在词汇表中是 <code>"hug"</code> ，所以我们在那里分割，得到 <code>["hug", "##s"]</code> 。然后我们继续处理 <code>"##s"</code> ，它在词汇表中，所以 <code>"hugs"</code> 的分词结果是 <code>["hug", "##s"]</code> 。</p> <p data-svelte-h="svelte-1om85t3">如果使用 BPE，我们会按照学习到的合并规则进行合并，并将其分词为 <code>["hu", "##gs"]</code> ，不同的字词分词算法所以最终得到的编码是不同的。</p> <p data-svelte-h="svelte-trjwfq">再举一个例子，让我们看看 <code>"bugs"</code> 将如何分词的。 <code>"b"</code> 是从词汇表中单词开头开始的最长子词，所以我们在那里分割并得到 <code>["b", "##ugs"]</code> 。然后 <code>"##u"</code> 是词汇表中从 <code>"##ugs"</code> 开始的最长的子词，所以我们在那里拆分并得到 <code>["b", "##u, "##gs"]</code> 。最后， <code>"##gs"</code> 在词汇表中，因此 <code>"bugs"</code> 的分词结果是： <code>["b", "##u, "##gs"]</code> 。</p> <p data-svelte-h="svelte-160l13j">当分词过程中无法在词汇库中找到该子词时，整个词会被标记为 unknown（未知）—— 例如， <code>"mug"</code> 将被标记为 <code>["[UNK]"]</code> ， <code>"bum"</code> 也是如此（即使我们的词汇表中包含 <code>"b"</code> 和 <code>"##u"</code> 开始，但是 <code>"##m"</code> 不在词汇表中，因此最终的分词结果只会是 <code>["[UNK]"]</code> ，而不是 <code>["b", "##u", "[UNK]"]</code> ）。这是与 BPE 的另一个区别，BPE 只会将不在词汇库中的单个字符标记为 unknown。</p> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-1yqamlc">✏️ <strong>现在轮到你了！</strong> <code>"pugs"</code> 将被如何分词？</p></div> <h2 class="relative group"><a id="实现 WordPiece" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#实现 WordPiece"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>实现 WordPiece</span></h2> <p data-svelte-h="svelte-1hl7db2">现在让我们看一下 WordPiece 算法的实现。与 BPE 一样，这只是教学示例，你不能在大型语料库上使用。</p> <p data-svelte-h="svelte-17nhne6">我们将使用与 BPE 示例中相同的语料库：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->corpus = [
	<span class="hljs-string">"This is the Hugging Face Course."</span>,
	<span class="hljs-string">"This chapter is about tokenization."</span>,
	<span class="hljs-string">"This section shows several tokenizer algorithms."</span>,
	<span class="hljs-string">"Hopefully, you will be able to understand how they are trained and generate tokens."</span>,
	]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1qxzm28">首先，我们需要将语料库预分词为单词。由于我们正在复刻 WordPiece tokenizer （如 BERT），因此我们将使用 <code>bert-base-cased</code> tokenizer 进行预分词：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer

	tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"bert-base-cased"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1a86zo">然后我们在进行预分词的同时，计算语料库中每个单词的频率：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> collections <span class="hljs-keyword">import</span> defaultdict

	word_freqs = defaultdict(<span class="hljs-built_in">int</span>)
	<span class="hljs-keyword">for</span> text <span class="hljs-keyword">in</span> corpus:
	words_with_offsets = tokenizer.backend_tokenizer.pre_tokenizer.pre_tokenize_str(text)
	new_words = [word <span class="hljs-keyword">for</span> word, offset <span class="hljs-keyword">in</span> words_with_offsets]
	<span class="hljs-keyword">for</span> word <span class="hljs-keyword">in</span> new_words:
	word_freqs[word] += <span class="hljs-number">1</span>

	word_freqs<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->defaultdict(
	<span class="hljs-built_in">int</span>, {<span class="hljs-string">'This'</span>: <span class="hljs-number">3</span>, <span class="hljs-string">'is'</span>: <span class="hljs-number">2</span>, <span class="hljs-string">'the'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'Hugging'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'Face'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'Course'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'.'</span>: <span class="hljs-number">4</span>, <span class="hljs-string">'chapter'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'about'</span>: <span class="hljs-number">1</span>,
	<span class="hljs-string">'tokenization'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'section'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'shows'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'several'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'tokenizer'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'algorithms'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'Hopefully'</span>: <span class="hljs-number">1</span>,
	<span class="hljs-string">','</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'you'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'will'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'be'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'able'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'to'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'understand'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'how'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'they'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'are'</span>: <span class="hljs-number">1</span>,
	<span class="hljs-string">'trained'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'and'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'generate'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'tokens'</span>: <span class="hljs-number">1</span>})<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1mul76c">如我们之前看到的，字母表是一个独特的集合，由所有单词的第一个字母以及所有以 <code>##</code> 为前缀和在单词中的其他字母组成：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->alphabet = []
	<span class="hljs-keyword">for</span> word <span class="hljs-keyword">in</span> word_freqs.keys():
	<span class="hljs-keyword">if</span> word[<span class="hljs-number">0</span>] <span class="hljs-keyword">not</span> <span class="hljs-keyword">in</span> alphabet:
	alphabet.append(word[<span class="hljs-number">0</span>])
	<span class="hljs-keyword">for</span> letter <span class="hljs-keyword">in</span> word[<span class="hljs-number">1</span>:]:
	<span class="hljs-keyword">if</span> <span class="hljs-string">f"##<span class="hljs-subst">{letter}</span>"</span> <span class="hljs-keyword">not</span> <span class="hljs-keyword">in</span> alphabet:
	alphabet.append(<span class="hljs-string">f"##<span class="hljs-subst">{letter}</span>"</span>)

	alphabet.sort()
	alphabet

	<span class="hljs-built_in">print</span>(alphabet)<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->[<span class="hljs-string">'##a'</span>, <span class="hljs-string">'##b'</span>, <span class="hljs-string">'##c'</span>, <span class="hljs-string">'##d'</span>, <span class="hljs-string">'##e'</span>, <span class="hljs-string">'##f'</span>, <span class="hljs-string">'##g'</span>, <span class="hljs-string">'##h'</span>, <span class="hljs-string">'##i'</span>, <span class="hljs-string">'##k'</span>, <span class="hljs-string">'##l'</span>, <span class="hljs-string">'##m'</span>, <span class="hljs-string">'##n'</span>, <span class="hljs-string">'##o'</span>, <span class="hljs-string">'##p'</span>, <span class="hljs-string">'##r'</span>, <span class="hljs-string">'##s'</span>,
	<span class="hljs-string">'##t'</span>, <span class="hljs-string">'##u'</span>, <span class="hljs-string">'##v'</span>, <span class="hljs-string">'##w'</span>, <span class="hljs-string">'##y'</span>, <span class="hljs-string">'##z'</span>, <span class="hljs-string">','</span>, <span class="hljs-string">'.'</span>, <span class="hljs-string">'C'</span>, <span class="hljs-string">'F'</span>, <span class="hljs-string">'H'</span>, <span class="hljs-string">'T'</span>, <span class="hljs-string">'a'</span>, <span class="hljs-string">'b'</span>, <span class="hljs-string">'c'</span>, <span class="hljs-string">'g'</span>, <span class="hljs-string">'h'</span>, <span class="hljs-string">'i'</span>, <span class="hljs-string">'s'</span>, <span class="hljs-string">'t'</span>, <span class="hljs-string">'u'</span>,
	<span class="hljs-string">'w'</span>, <span class="hljs-string">'y'</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-6gvlr8">我们还在该词汇表的开头添加了模型使用的特殊 tokens，在使用 BERT 的情况下，特殊 tokens 是 <code>["[PAD]", "[UNK]", "[CLS]", "[SEP]", "[MASK]"]</code> ：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->vocab = [<span class="hljs-string">"[PAD]"</span>, <span class="hljs-string">"[UNK]"</span>, <span class="hljs-string">"[CLS]"</span>, <span class="hljs-string">"[SEP]"</span>, <span class="hljs-string">"[MASK]"</span>] + alphabet.copy()<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-15qyvc2">接下来我们需要将每个单词进行分割，除了第一个字母外，其他字母都需要以 <code>##</code> 为前缀：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->splits = {
	word: [c <span class="hljs-keyword">if</span> i == <span class="hljs-number">0</span> <span class="hljs-keyword">else</span> <span class="hljs-string">f"##<span class="hljs-subst">{c}</span>"</span> <span class="hljs-keyword">for</span> i, c <span class="hljs-keyword">in</span> <span class="hljs-built_in">enumerate</span>(word)]
	<span class="hljs-keyword">for</span> word <span class="hljs-keyword">in</span> word_freqs.keys()
	}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-3upalw">现在我们已经准备好训练了，让我们编写一个函数来计算每对的分数。我们需要在训练的每个步骤中使用它：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">def</span> <span class="hljs-title function_">compute_pair_scores</span>(<span class="hljs-params">splits</span>):
	letter_freqs = defaultdict(<span class="hljs-built_in">int</span>)
	pair_freqs = defaultdict(<span class="hljs-built_in">int</span>)
	<span class="hljs-keyword">for</span> word, freq <span class="hljs-keyword">in</span> word_freqs.items():
	split = splits[word]
	<span class="hljs-keyword">if</span> <span class="hljs-built_in">len</span>(split) == <span class="hljs-number">1</span>:
	letter_freqs[split[<span class="hljs-number">0</span>]] += freq
	<span class="hljs-keyword">continue</span>
	<span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> <span class="hljs-built_in">range</span>(<span class="hljs-built_in">len</span>(split) - <span class="hljs-number">1</span>):
	pair = (split[i], split[i + <span class="hljs-number">1</span>])
	letter_freqs[split[i]] += freq
	pair_freqs[pair] += freq
	letter_freqs[split[-<span class="hljs-number">1</span>]] += freq

	scores = {
	pair: freq / (letter_freqs[pair[<span class="hljs-number">0</span>]] * letter_freqs[pair[<span class="hljs-number">1</span>]])
	<span class="hljs-keyword">for</span> pair, freq <span class="hljs-keyword">in</span> pair_freqs.items()
	}
	<span class="hljs-keyword">return</span> scores<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-cihmgg">让我们来看看在初始分割后的部分字典：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->pair_scores = compute_pair_scores(splits)
	<span class="hljs-keyword">for</span> i, key <span class="hljs-keyword">in</span> <span class="hljs-built_in">enumerate</span>(pair_scores.keys()):
	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"<span class="hljs-subst">{key}</span>: <span class="hljs-subst">{pair_scores[key]}</span>"</span>)
	<span class="hljs-keyword">if</span> i >= <span class="hljs-number">5</span>:
	<span class="hljs-keyword">break</span><!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->(<span class="hljs-string">'T'</span>, <span class="hljs-string">'##h'</span>): <span class="hljs-number">0.125</span>
	(<span class="hljs-string">'##h'</span>, <span class="hljs-string">'##i'</span>): <span class="hljs-number">0.03409090909090909</span>
	(<span class="hljs-string">'##i'</span>, <span class="hljs-string">'##s'</span>): <span class="hljs-number">0.02727272727272727</span>
	(<span class="hljs-string">'i'</span>, <span class="hljs-string">'##s'</span>): <span class="hljs-number">0.1</span>
	(<span class="hljs-string">'t'</span>, <span class="hljs-string">'##h'</span>): <span class="hljs-number">0.03571428571428571</span>
	(<span class="hljs-string">'##h'</span>, <span class="hljs-string">'##e'</span>): <span class="hljs-number">0.011904761904761904</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1tie2v3">现在，只需要一个快速循环就可以找到得分最高的对：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->best_pair = <span class="hljs-string">""</span>
	max_score = <span class="hljs-literal">None</span>
	<span class="hljs-keyword">for</span> pair, score <span class="hljs-keyword">in</span> pair_scores.items():
	<span class="hljs-keyword">if</span> max_score <span class="hljs-keyword">is</span> <span class="hljs-literal">None</span> <span class="hljs-keyword">or</span> max_score < score:
	best_pair = pair
	max_score = score

	<span class="hljs-built_in">print</span>(best_pair, max_score)<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->(<span class="hljs-string">'a'</span>, <span class="hljs-string">'##b'</span>) <span class="hljs-number">0.2</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-97mtnh">所以第一个要学习的合并是 <code>('a', '##b') -> 'ab'</code> ，并且我们添加 <code>'ab'</code> 到词汇表中：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->vocab.append(<span class="hljs-string">"ab"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-os95fs">接下来，我们需要对 <code>splits</code> 字典进行这种合并。让我们为此写另一个函数：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">def</span> <span class="hljs-title function_">merge_pair</span>(<span class="hljs-params">a, b, splits</span>):
	<span class="hljs-keyword">for</span> word <span class="hljs-keyword">in</span> word_freqs:
	split = splits[word]
	<span class="hljs-keyword">if</span> <span class="hljs-built_in">len</span>(split) == <span class="hljs-number">1</span>:
	<span class="hljs-keyword">continue</span>
	i = <span class="hljs-number">0</span>
	<span class="hljs-keyword">while</span> i < <span class="hljs-built_in">len</span>(split) - <span class="hljs-number">1</span>:
	<span class="hljs-keyword">if</span> split[i] == a <span class="hljs-keyword">and</span> split[i + <span class="hljs-number">1</span>] == b:
	merge = a + b[<span class="hljs-number">2</span>:] <span class="hljs-keyword">if</span> b.startswith(<span class="hljs-string">"##"</span>) <span class="hljs-keyword">else</span> a + b
	split = split[:i] + [merge] + split[i + <span class="hljs-number">2</span> :]
	<span class="hljs-keyword">else</span>:
	i += <span class="hljs-number">1</span>
	splits[word] = split
	<span class="hljs-keyword">return</span> splits<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1y048j">我们可以看看第一次合并的结果：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->splits = merge_pair(<span class="hljs-string">"a"</span>, <span class="hljs-string">"##b"</span>, splits)
	splits[<span class="hljs-string">"about"</span>]<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->[<span class="hljs-string">'ab'</span>, <span class="hljs-string">'##o'</span>, <span class="hljs-string">'##u'</span>, <span class="hljs-string">'##t'</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-7tzu4v">现在我们有了合并循环的所有代码。让我们设定词汇表的大小为 70：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->vocab_size = <span class="hljs-number">70</span>
	<span class="hljs-keyword">while</span> <span class="hljs-built_in">len</span>(vocab) < vocab_size:
	scores = compute_pair_scores(splits)
	best_pair, max_score = <span class="hljs-string">""</span>, <span class="hljs-literal">None</span>
	<span class="hljs-keyword">for</span> pair, score <span class="hljs-keyword">in</span> scores.items():
	<span class="hljs-keyword">if</span> max_score <span class="hljs-keyword">is</span> <span class="hljs-literal">None</span> <span class="hljs-keyword">or</span> max_score < score:
	best_pair = pair
	max_score = score
	splits = merge_pair(*best_pair, splits)
	new_token = (
	best_pair[<span class="hljs-number">0</span>] + best_pair[<span class="hljs-number">1</span>][<span class="hljs-number">2</span>:]
	<span class="hljs-keyword">if</span> best_pair[<span class="hljs-number">1</span>].startswith(<span class="hljs-string">"##"</span>)
	<span class="hljs-keyword">else</span> best_pair[<span class="hljs-number">0</span>] + best_pair[<span class="hljs-number">1</span>]
	)
	vocab.append(new_token)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-q8i5ms">然后我们可以查看生成的词汇表：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-built_in">print</span>(vocab)<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->[<span class="hljs-string">'[PAD]'</span>, <span class="hljs-string">'[UNK]'</span>, <span class="hljs-string">'[CLS]'</span>, <span class="hljs-string">'[SEP]'</span>, <span class="hljs-string">'[MASK]'</span>, <span class="hljs-string">'##a'</span>, <span class="hljs-string">'##b'</span>, <span class="hljs-string">'##c'</span>, <span class="hljs-string">'##d'</span>, <span class="hljs-string">'##e'</span>, <span class="hljs-string">'##f'</span>, <span class="hljs-string">'##g'</span>, <span class="hljs-string">'##h'</span>, <span class="hljs-string">'##i'</span>, <span class="hljs-string">'##k'</span>,
	<span class="hljs-string">'##l'</span>, <span class="hljs-string">'##m'</span>, <span class="hljs-string">'##n'</span>, <span class="hljs-string">'##o'</span>, <span class="hljs-string">'##p'</span>, <span class="hljs-string">'##r'</span>, <span class="hljs-string">'##s'</span>, <span class="hljs-string">'##t'</span>, <span class="hljs-string">'##u'</span>, <span class="hljs-string">'##v'</span>, <span class="hljs-string">'##w'</span>, <span class="hljs-string">'##y'</span>, <span class="hljs-string">'##z'</span>, <span class="hljs-string">','</span>, <span class="hljs-string">'.'</span>, <span class="hljs-string">'C'</span>, <span class="hljs-string">'F'</span>, <span class="hljs-string">'H'</span>,
	<span class="hljs-string">'T'</span>, <span class="hljs-string">'a'</span>, <span class="hljs-string">'b'</span>, <span class="hljs-string">'c'</span>, <span class="hljs-string">'g'</span>, <span class="hljs-string">'h'</span>, <span class="hljs-string">'i'</span>, <span class="hljs-string">'s'</span>, <span class="hljs-string">'t'</span>, <span class="hljs-string">'u'</span>, <span class="hljs-string">'w'</span>, <span class="hljs-string">'y'</span>, <span class="hljs-string">'ab'</span>,<span class="hljs-string">'##fu'</span>, <span class="hljs-string">'Fa'</span>, <span class="hljs-string">'Fac'</span>, <span class="hljs-string">'##ct'</span>, <span class="hljs-string">'##ful'</span>, <span class="hljs-string">'##full'</span>, <span class="hljs-string">'##fully'</span>,
	<span class="hljs-string">'Th'</span>, <span class="hljs-string">'ch'</span>, <span class="hljs-string">'##hm'</span>, <span class="hljs-string">'cha'</span>, <span class="hljs-string">'chap'</span>, <span class="hljs-string">'chapt'</span>, <span class="hljs-string">'##thm'</span>, <span class="hljs-string">'Hu'</span>, <span class="hljs-string">'Hug'</span>, <span class="hljs-string">'Hugg'</span>, <span class="hljs-string">'sh'</span>, <span class="hljs-string">'th'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'##thms'</span>, <span class="hljs-string">'##za'</span>, <span class="hljs-string">'##zat'</span>,
	<span class="hljs-string">'##ut'</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-e50qfc">如我们所见，相较于 BPE（字节对编码），此分词器在学习单词部分作为 tokens 时稍快一些。</p> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-syspc2">💡 在同一语料库上使用 <code>train_new_from_iterator()</code> 不会产生完全相同的词汇表。这是因为 🤗 Tokenizers 库没有为训练实现 WordPiece（因为我们不完全确定它的真实实现方式），而是使用了 BPE。</p></div> <p data-svelte-h="svelte-ddxha3">要对新文本进行分词，我们先预分词，再进行分割，然后在每个词上使用分词算法。也就是说，我们寻找从第一个词开始的最大子词并将其分割，然后我们对第二部分重复此过程，以此类推，对该词以及文本中的后续词进行分割：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">def</span> <span class="hljs-title function_">encode_word</span>(<span class="hljs-params">word</span>):
	tokens = []
	<span class="hljs-keyword">while</span> <span class="hljs-built_in">len</span>(word) > <span class="hljs-number">0</span>:
	i = <span class="hljs-built_in">len</span>(word)
	<span class="hljs-keyword">while</span> i > <span class="hljs-number">0</span> <span class="hljs-keyword">and</span> word[:i] <span class="hljs-keyword">not</span> <span class="hljs-keyword">in</span> vocab:
	i -= <span class="hljs-number">1</span>
	<span class="hljs-keyword">if</span> i == <span class="hljs-number">0</span>:
	<span class="hljs-keyword">return</span> [<span class="hljs-string">"[UNK]"</span>]
	tokens.append(word[:i])
	word = word[i:]
	<span class="hljs-keyword">if</span> <span class="hljs-built_in">len</span>(word) > <span class="hljs-number">0</span>:
	word = <span class="hljs-string">f"##<span class="hljs-subst">{word}</span>"</span>
	<span class="hljs-keyword">return</span> tokens<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-4xc89b">让我们使用词汇表中的一个词和一个不在词汇表中的词上测试一下：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-built_in">print</span>(encode_word(<span class="hljs-string">"Hugging"</span>))
	<span class="hljs-built_in">print</span>(encode_word(<span class="hljs-string">"HOgging"</span>))<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->[<span class="hljs-string">'Hugg'</span>, <span class="hljs-string">'##i'</span>, <span class="hljs-string">'##n'</span>, <span class="hljs-string">'##g'</span>]
	[<span class="hljs-string">'[UNK]'</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1qypl47">现在，让我们编写一个对文本分词的函数：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">def</span> <span class="hljs-title function_">tokenize</span>(<span class="hljs-params">text</span>):
	pre_tokenize_result = tokenizer._tokenizer.pre_tokenizer.pre_tokenize_str(text)
	pre_tokenized_text = [word <span class="hljs-keyword">for</span> word, offset <span class="hljs-keyword">in</span> pre_tokenize_result]
	encoded_words = [encode_word(word) <span class="hljs-keyword">for</span> word <span class="hljs-keyword">in</span> pre_tokenized_text]
	<span class="hljs-keyword">return</span> <span class="hljs-built_in">sum</span>(encoded_words, [])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-35pr86">我们可以在任何文本上尝试：</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->tokenize(<span class="hljs-string">"This is the Hugging Face course!"</span>)<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->[<span class="hljs-string">'Th'</span>, <span class="hljs-string">'##i'</span>, <span class="hljs-string">'##s'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'th'</span>, <span class="hljs-string">'##e'</span>, <span class="hljs-string">'Hugg'</span>, <span class="hljs-string">'##i'</span>, <span class="hljs-string">'##n'</span>, <span class="hljs-string">'##g'</span>, <span class="hljs-string">'Fac'</span>, <span class="hljs-string">'##e'</span>, <span class="hljs-string">'c'</span>, <span class="hljs-string">'##o'</span>, <span class="hljs-string">'##u'</span>, <span class="hljs-string">'##r'</span>, <span class="hljs-string">'##s'</span>,
	<span class="hljs-string">'##e'</span>, <span class="hljs-string">'[UNK]'</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-tt696e">这就是 WordPiece 算法的全部内容！现在让我们来看看 Unigram。</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/course/blob/main/chapters/zh-CN/chapter6/6.mdx" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span data-svelte-h="svelte-x0xyl0">></span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p>

	<script>
	{
	__sveltekit_9aawfw = {
	assets: "/docs/course/pr_1021/zh-CN",
	base: "/docs/course/pr_1021/zh-CN",
	env: {}
	};

	const element = document.currentScript.parentElement;

	const data = [null,null];

	Promise.all([
	import("/docs/course/pr_1021/zh-CN/_app/immutable/entry/start.f3a1a511.js"),
	import("/docs/course/pr_1021/zh-CN/_app/immutable/entry/app.c39e37cf.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 49],
	data,
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 93.8 kB
Xet hash:: 8dca3933078dfa2b348a7ceb6efd34ca627e076f6e53a21e8ef6f4928c8300f5

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.