Hugging Face's logo

Buckets:
hf-doc-build
/
doc-dev

Files
xet
hf-doc-build/doc-dev / course /pr_1021 /zh-CN /chapter2 /4.html
rtrm's picture
rtrm
download
raw
55.8 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Tokenizers&quot;,&quot;local&quot;:&quot;Tokenizers&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;基于单词（Word-based）的 tokenization&quot;,&quot;local&quot;:&quot;基于单词(Word-based)的 tokenization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;基于字符（Character-based）的 tokenization&quot;,&quot;local&quot;:&quot;基于字符(Character-based)的 tokenization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;基于子词（subword）的 tokenization&quot;,&quot;local&quot;:&quot;基于子词（subword）的 tokenization&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;还有更多！&quot;,&quot;local&quot;:&quot;还有更多！&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;加载和保存&quot;,&quot;local&quot;:&quot;加载和保存&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;编码&quot;,&quot;local&quot;:&quot;编码&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;tokenization&quot;,&quot;local&quot;:&quot;tokenization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;从 tokens 到 inputs ID&quot;,&quot;local&quot;:&quot;从 tokens 到 inputs ID&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;解码&quot;,&quot;local&quot;:&quot;解码&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;: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/16.669929b9.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/FrameworkSwitchCourse.8d4d4ab6.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="{&quot;title&quot;:&quot;Tokenizers&quot;,&quot;local&quot;:&quot;Tokenizers&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;基于单词（Word-based）的 tokenization&quot;,&quot;local&quot;:&quot;基于单词(Word-based)的 tokenization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;基于字符（Character-based）的 tokenization&quot;,&quot;local&quot;:&quot;基于字符(Character-based)的 tokenization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;基于子词（subword）的 tokenization&quot;,&quot;local&quot;:&quot;基于子词（subword）的 tokenization&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;还有更多！&quot;,&quot;local&quot;:&quot;还有更多！&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;加载和保存&quot;,&quot;local&quot;:&quot;加载和保存&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;编码&quot;,&quot;local&quot;:&quot;编码&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;tokenization&quot;,&quot;local&quot;:&quot;tokenization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;从 tokens 到 inputs ID&quot;,&quot;local&quot;:&quot;从 tokens 到 inputs ID&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;解码&quot;,&quot;local&quot;:&quot;解码&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <div class="bg-white leading-none border border-gray-100 rounded-lg flex p-0.5 w-56 text-sm mb-4"><a class="flex justify-center flex-1 py-1.5 px-2.5 focus:outline-none !no-underline rounded-l bg-red-50 dark:bg-transparent text-red-600" href="?fw=pt"><svg class="mr-1.5" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><defs><clipPath id="a"><rect x="3.05" y="0.5" width="25.73" height="31" fill="none"></rect></clipPath></defs><g clip-path="url(#a)"><path d="M24.94,9.51a12.81,12.81,0,0,1,0,18.16,12.68,12.68,0,0,1-18,0,12.81,12.81,0,0,1,0-18.16l9-9V5l-.84.83-6,6a9.58,9.58,0,1,0,13.55,0ZM20.44,9a1.68,1.68,0,1,1,1.67-1.67A1.68,1.68,0,0,1,20.44,9Z" fill="#ee4c2c"></path></g></svg> Pytorch </a><a class="flex justify-center flex-1 py-1.5 px-2.5 focus:outline-none !no-underline rounded-r text-gray-500 filter grayscale" href="?fw=tf"><svg class="mr-1.5" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="0.94em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 274"><path d="M145.726 42.065v42.07l72.861 42.07v-42.07l-72.86-42.07zM0 84.135v42.07l36.43 21.03V105.17L0 84.135zm109.291 21.035l-36.43 21.034v126.2l36.43 21.035v-84.135l36.435 21.035v-42.07l-36.435-21.034V105.17z" fill="#E55B2D"></path><path d="M145.726 42.065L36.43 105.17v42.065l72.861-42.065v42.065l36.435-21.03v-84.14zM255.022 63.1l-36.435 21.035v42.07l36.435-21.035V63.1zm-72.865 84.135l-36.43 21.035v42.07l36.43-21.036v-42.07zm-36.43 63.104l-36.436-21.035v84.135l36.435-21.035V210.34z" fill="#ED8E24"></path><path d="M145.726 0L0 84.135l36.43 21.035l109.296-63.105l72.861 42.07L255.022 63.1L145.726 0zm0 126.204l-36.435 21.03l36.435 21.036l36.43-21.035l-36.43-21.03z" fill="#F8BF3C"></path></svg> TensorFlow </a></div> <h1 class="relative group"><a id="Tokenizers" 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="#Tokenizers"><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>Tokenizers</span></h1> <div class="flex space-x-1 absolute z-10 right-0 top-0"><a href="https://discuss.huggingface.co/t/chapter-2-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/chapter2/section4_pt.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/chapter2/section4_pt.ipynb" target="_blank"><img alt="Open In Studio Lab" class="!m-0" src="https://studiolab.sagemaker.aws/studiolab.svg"></a></div> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/VFp38yj8h3A" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-1abez2d">tokenizer 是 NLP 管道的核心组件之一。它们有一个非常明确的目的：将文本转换为模型可以处理的数据。模型只能处理数字，因此 tokenizer 需要将我们的文本输入转换为数字。在本节中，我们将确切地探讨 tokenization 管道中发生的事情。</p> <p data-svelte-h="svelte-n4gsny">在 NLP 任务中，通常处理的原始数据是文本。这里是一个例子：</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 -->Jim Henson was <span class="hljs-selector-tag">a</span> puppeteer<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-vxq3bx">但是，模型只能处理数字，因此我们需要找到一种将原始文本转换为数字的方法。这就是 tokenizer 所做的，并且有很多方法可以解决这个问题。目标是找到最有意义的表达方式 —— 即对模型来说最有意义的方式 —— 如果可能，还要找到最简洁的表达方式。</p> <p data-svelte-h="svelte-133k26">让我们看一下 tokenization 算法的一些示例，并尝试回答一些你可能对 tokenization 有的疑问。</p> <h2 class="relative group"><a id="基于单词(Word-based)的 tokenization" 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="#基于单词(Word-based)的 tokenization"><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>基于单词（Word-based）的 tokenization</span></h2> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/nhJxYji1aho" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-4g4obf">想到的第一种 tokenizer 是基于词（word-based）的 tokenization。它通常很容易配置和使用，只需几条规则，并且通常会产生不错的结果。例如，在下图中，目标是将原始文本拆分为单词并为每个单词找到一个数字表示：</p> <div class="flex justify-center" data-svelte-h="svelte-1qoje79"><img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter2/word_based_tokenization.svg" alt="An example of word-based tokenization."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter2/word_based_tokenization-dark.svg" alt="An example of word-based tokenization."></div> <p data-svelte-h="svelte-i2sgro">有多种方法可以拆分文本。例如，我们可以通过使用 Python 的 <code>split()</code> 函数，使用空格将文本分割为单词：</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 -->tokenized_text = <span class="hljs-string">&quot;Jim Henson was a puppeteer&quot;</span>.split()
<span class="hljs-built_in">print</span>(tokenized_text)<!-- 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">&#x27;Jim&#x27;</span>, <span class="hljs-string">&#x27;Henson&#x27;</span>, <span class="hljs-string">&#x27;was&#x27;</span>, <span class="hljs-string">&#x27;a&#x27;</span>, <span class="hljs-string">&#x27;puppeteer&#x27;</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1yavz1o">此外，还有一些基于单词的 tokenizer 的变体，对标点符号有额外的规则。使用这类 tokenizer，我们最终可以得到一些非常大的“词汇表（vocabulary）”，其中词汇表的大小由我们在语料库中拥有的独立 tokens 的总数确定。</p> <p data-svelte-h="svelte-16z9r95">每个单词都分配了一个 ID，从 0 开始一直到词汇表的大小。模型使用这些 ID 来识别每个词。</p> <p data-svelte-h="svelte-1q8zp9w">如果我们想用基于单词的 tokenizer 完全覆盖一种语言，我们需要为语言中的每个单词设置一个标识符，这将生成大量的 tokens。例如，英语中有超过 500,000 个单词，因此要构建从每个单词到 ID 的映射，我们需要跟踪这么多 ID。此外，像“dog”这样的词与“dogs”这样的词的表示方式不同，模型最初无法知道“dog”和“dogs”是相似的：它会将这两个词识别为不相关。这同样适用于其他相似的词，例如“run”和“running”，模型最初也不会看到它们的相似性。</p> <p data-svelte-h="svelte-d40e1g">最后，我们需要一个自定义 token 来表示不在我们词汇表中的单词。这被称为“unknown” token，通常表示为“[UNK]”或“&lt;unk&gt;”。如果你看到 tokenizer 产生了很多这样的 token 这通常是一个不好的迹象，因为它无法检索到一个词的合理表示，并且你会在转化过程中丢失信息。制作词汇表时的其中一个目标是 tokenizer 将尽可能少的单词标记为未知 tokens。</p> <p data-svelte-h="svelte-cdxdpa">减少未知 tokens 数量的一种方法是使用更深一层的 tokenizer 即基于字符（character-based）的 tokenizer</p> <h2 class="relative group"><a id="基于字符(Character-based)的 tokenization" 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="#基于字符(Character-based)的 tokenization"><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>基于字符（Character-based）的 tokenization</span></h2> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/ssLq_EK2jLE" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-17nuoh1">基于字符的 tokenizer 将文本拆分为字符，而不是单词。这有两个主要好处：</p> <ul data-svelte-h="svelte-a5azzl"><li>词汇量要小得多。</li> <li>unknown tokens （out-of-vocabulary）要少得多，因为每个单词都可以由字符构建。</li></ul> <p data-svelte-h="svelte-84gmqk">但在此过程中也有一些问题，关于空格和标点符号：</p> <div class="flex justify-center" data-svelte-h="svelte-cwc099"><img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter2/character_based_tokenization.svg" alt="An example of character-based tokenization."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter2/character_based_tokenization-dark.svg" alt="An example of character-based tokenization."></div> <p data-svelte-h="svelte-2svf75">这种方法也不是完美的。由于现在表示是基于字符而不是单词，因此人们可能会争辩说，从直觉上讲，它的意义不大：每个字符本身并没有多大意义，但是单词则不然。然而，这又因语言而异；例如，在中文中，每个字符比拉丁语言中的字符包含更多的信息。</p> <p data-svelte-h="svelte-aul5t5">另一件要考虑的因素是，这样做会导致我们的模型需要处理大量的 tokens：虽然一个单词在基于单词的 tokenizer 中只是一个 token，但当它被转换为字符时，很可能就变成了 10 个或更多的 tokens</p> <p data-svelte-h="svelte-5nvlhz">为了两全其美，我们可以使用结合这两种方法的第三种技术：基于子词（subword）的 tokenization。</p> <h2 class="relative group"><a id="基于子词（subword）的 tokenization" 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="#基于子词（subword）的 tokenization"><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>基于子词（subword）的 tokenization</span></h2> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/zHvTiHr506c" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-9utx71">基于子词（subword）的 tokenization 算法依赖于这样一个原则：常用词不应被分解为更小的子词，但罕见词应被分解为有意义的子词。</p> <p data-svelte-h="svelte-25ax2q">例如，“annoyingly”可能被视为一个罕见的词，可以分解为“annoying”和“ly”。这两者都可能作为独立的子词并且出现得更频繁，同时“annoyingly”的含义通过“annoying”和“ly”的复合含义得以保留。</p> <p data-svelte-h="svelte-9s7qet">这里有一个例子，展示了基于子词的 tokenization 算法如何将序列“Let’s do tokenization!”分词：</p> <div class="flex justify-center" data-svelte-h="svelte-l1uqeh"><img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter2/bpe_subword.svg" alt="A subword tokenization algorithm."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter2/bpe_subword-dark.svg" alt="A subword tokenization algorithm."></div> <p data-svelte-h="svelte-7sxyh7">这些子词最终提供了大量的语义信息：例如，在上面的例子中，“tokenization”被分割成“token”和“ization”，这两个 tokens 在保持空间效率的同时具有语义意义（只需要两个 tokens 就能表示一个长词）。这让我们能够在词汇量小的情况下获得相对良好的覆盖率，并且几乎没有未知的 token。</p> <p data-svelte-h="svelte-1f0xucg">这种方法在土耳其语等粘着型语言（agglutinative languages）中特别有用，你可以通过将子词串在一起来形成（几乎）任意长的复杂词。</p> <h3 class="relative group"><a id="还有更多！" 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="#还有更多！"><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>还有更多！</span></h3> <p data-svelte-h="svelte-2alht4">不出所料，还有更多的技术。仅举几例：</p> <ul data-svelte-h="svelte-189jw7e"><li>Byte-level BPE，用于 GPT-2</li> <li>WordPiece，用于 BERT</li> <li>SentencePiece or Unigram，用于多个多语言模型</li></ul> <p data-svelte-h="svelte-b0yqzn">你现在应该对 tokenizer 的工作原理有足够的了解，可以开始使用 API 了。</p> <h2 class="relative group"><a id="加载和保存" 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="#加载和保存"><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>加载和保存</span></h2> <p data-svelte-h="svelte-7jx849">加载和保存 tokenizer 就像使用模型一样简单。实际上，它基于相同的两种方法： <code>from_pretrained()</code> 和 <code>save_pretrained()</code> 。这些方法会加载或保存分词器使用的算法（有点像模型的架构（architecture））以及其词汇表（有点像模型的权重（weights））。</p> <p data-svelte-h="svelte-jadsr1">加载使用与 BERT 相同的 checkpoint 训练的 BERT tokenizer 与加载模型的方式相同，只是换成了 <code>Bert tokenizer</code> 类：</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> BertTokenizer
tokenizer = BertTokenizer.from_pretrained(<span class="hljs-string">&quot;bert-base-cased&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-ixa086">如同 <code>AutoModel</code> ， <code>AutoTokenizer</code> 类将根据 checkpoint 名称在库中获取正确的 tokenizer 类，并且可以直接与任何 checkpoint 一起使用：</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">&quot;bert-base-cased&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-48exol">现在我们可以像在上一节中显示的那样使用 tokenizer：</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 -->tokenizer(<span class="hljs-string">&quot;Using a Transformer network is simple&quot;</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">&#x27;input_ids&#x27;</span>: [<span class="hljs-number">101</span>, <span class="hljs-number">7993</span>, <span class="hljs-number">170</span>, <span class="hljs-number">11303</span>, <span class="hljs-number">1200</span>, <span class="hljs-number">2443</span>, <span class="hljs-number">1110</span>, <span class="hljs-number">3014</span>, <span class="hljs-number">102</span>],
 <span class="hljs-string">&#x27;token_type_ids&#x27;</span>: [<span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>],
 <span class="hljs-string">&#x27;attention_mask&#x27;</span>: [<span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>]}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1gq3icf">保存 tokenizer 与保存模型完全相同：</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 -->tokenizer.save_pretrained(<span class="hljs-string">&quot;directory_on_my_computer&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-6sn0le">我们将在 <a href="/Couse/chapter3">第三章</a> 中将更多地谈论 <code>token_type_ids</code> ，稍后我们将解释 <code>attention_mask</code> 。首先，让我们看看如何生成 <code>input_ids</code> 。为此，我们需要查看 tokenizer 的内部是如何实现的。</p> <h2 class="relative group"><a id="编码" 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="#编码"><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>编码</span></h2> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/Yffk5aydLzg" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-cfz385">将文本翻译成数字被称为编码（encoding）。编码分两步完成：分词，然后转换为 inputs ID。</p> <p data-svelte-h="svelte-1lc1ns9">正如我们所见，第一步是将文本拆分为单词（或部分单词、标点符号等），通常称为 tokens 不同的分词器使用的算法也不一样，这就是为什么我们需要使用模型名称来实例化 tokenizer，以确保我们使用模型预训练时使用的相同的算法。</p> <p data-svelte-h="svelte-1w05jja">第二步是将这些 tokens 转换为数字，这样我们就可以用它们构建一个张量并将它们提供给模型。为此，tokenizer 有一个词汇表（vocabulary），这是我们在使用 <code>from_pretrained()</code> 方法实例化它时下载的部分。同样，我们需要使用与预训练模型时相同的词汇表。</p> <p data-svelte-h="svelte-1fh6myb">为了更好地理解这两个步骤，我们将分别探讨它们。请注意，我们将单独执行部分 tokenization 管道的方法来向你展示这些步骤的中间结果，但在实践中，你应该直接在你的输入上调用 tokenizer（如第 2 小节所示）。</p> <h3 class="relative group"><a id="tokenization" 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"><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</span></h3> <p data-svelte-h="svelte-5k5zyu">tokenization 过程由 tokenizer 的 <code>tokenize()</code> 方法实现：</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">&quot;bert-base-cased&quot;</span>)
sequence = <span class="hljs-string">&quot;Using a Transformer network is simple&quot;</span>
tokens = tokenizer.tokenize(sequence)
<span class="hljs-built_in">print</span>(tokens)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-v6pc2g">这个方法的输出是一个字符串列表，或者说 tokens</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">&#x27;Using&#x27;</span>, <span class="hljs-string">&#x27;a&#x27;</span>, <span class="hljs-string">&#x27;transform&#x27;</span>, <span class="hljs-string">&#x27;##er&#x27;</span>, <span class="hljs-string">&#x27;network&#x27;</span>, <span class="hljs-string">&#x27;is&#x27;</span>, <span class="hljs-string">&#x27;simple&#x27;</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-d05xm3">这个 tokenizer 是一个基于子词的 tokenizer：它对词进行拆分，直到获得可以用其词汇表表示的 tokens。以 <code>transformer</code> 为例，它分为两个 tokens <code>transform</code> 和 <code>##er</code> 。</p> <h3 class="relative group"><a id="从 tokens 到 inputs ID" 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="#从 tokens 到 inputs ID"><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>从 tokens 到 inputs ID</span></h3> <p data-svelte-h="svelte-3upetb">inputs ID 的转换由 tokenizer 的 <code>convert_tokens_to_ids()</code> 方法实现：</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 -->ids = tokenizer.convert_tokens_to_ids(tokens)
<span class="hljs-built_in">print</span>(ids)<!-- 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-number">7993</span>, <span class="hljs-number">170</span>, <span class="hljs-number">11303</span>, <span class="hljs-number">1200</span>, <span class="hljs-number">2443</span>, <span class="hljs-number">1110</span>, <span class="hljs-number">3014</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-6fung9">这些输出，一旦转换为适当的框架张量，就可以用作模型的输入，如本章前面所示。</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-fqcvbv">✏️ <strong>试试看！</strong> 请将我们在第 2 节中使用的输入句子（“I’ve been waiting for a HuggingFace course my whole life.”和“I hate this so much!”）执行最后两个步骤（分词和转换为 inputs ID）。检查你获得的 inputs ID 是否与我们在第二节中获得的一致！</p></div> <h2 class="relative group"><a id="解码" 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="#解码"><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>解码</span></h2> <p data-svelte-h="svelte-170m9z3">解码（Decoding） 正好相反：从 inputs ID 到一个字符串。这以通过 <code>decode()</code> 方法实现：</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 -->decoded_string = tokenizer.decode([<span class="hljs-number">7993</span>, <span class="hljs-number">170</span>, <span class="hljs-number">11303</span>, <span class="hljs-number">1200</span>, <span class="hljs-number">2443</span>, <span class="hljs-number">1110</span>, <span class="hljs-number">3014</span>])
<span class="hljs-built_in">print</span>(decoded_string)<!-- 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">&#x27;Using a Transformer network is simple&#x27;</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1lilk66">请注意， <code>decode</code> 方法不仅将索引转换回 tokens，还将属于相同单词的 tokens 组合在一起以生成可读的句子。当我们使用预测新文本的模型（根据提示生成的文本，或序列到序列问题（如翻译或摘要））时，这样的功能将非常有用。</p> <p data-svelte-h="svelte-1uwy6la">到现在为止，你应该了解 tokenizer 可以处理的原子操作：分词、转换为 ID 以及将 ID 转换回字符串。然而，我们只是瞥到了冰山一角。在下一节中，我们将继续探讨它能力的极限，并看看如何克服它们。</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/chapter2/4.mdx" target="_blank"><span data-svelte-h="svelte-1kd6by1">&lt;</span> <span data-svelte-h="svelte-x0xyl0">&gt;</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, 16],
 data,
 form: null,
 error: null
 });
 });
 }
 </script>

Xet Storage Details

Size:
55.8 kB
·
Xet hash:
33692dd410b5baa0abb217bb4da89e914121eedf1906c654237afcb834c3a4ad

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