Hugging Face's logo

Buckets:
hf-doc-build
/
doc-dev

Files
xet
hf-doc-build/doc-dev / course /pr_1054 /id /chapter6 /4.html
rtrm's picture
rtrm
download
raw
35.7 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Normalisasi dan Pra-tokenisasi&quot;,&quot;local&quot;:&quot;normalization-and-pre-tokenization&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Normalisasi&quot;,&quot;local&quot;:&quot;normalization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Pra-tokenisasi&quot;,&quot;local&quot;:&quot;pre-tokenization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;SentencePiece&quot;,&quot;local&quot;:&quot;sentencepiece&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Tinjauan Algoritma&quot;,&quot;local&quot;:&quot;algorithm-overview&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
 <link href="/docs/course/pr_1054/id/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/entry/start.4f92af03.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/scheduler.36a0863c.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/singletons.7dc7b9a4.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/index.733708bb.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/paths.cf097d06.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/entry/app.19cef1b6.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/index.156fee99.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/nodes/0.1203e4a0.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/each.e59479a4.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/nodes/49.2508bf42.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/Tip.8a648467.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/Youtube.a5d6d567.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/CodeBlock.4cf998e6.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/CourseFloatingBanner.16bb8bff.js">
 <link rel="modulepreload" href="/docs/course/pr_1054/id/_app/immutable/chunks/getInferenceSnippets.472bc46d.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Normalisasi dan Pra-tokenisasi&quot;,&quot;local&quot;:&quot;normalization-and-pre-tokenization&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Normalisasi&quot;,&quot;local&quot;:&quot;normalization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Pra-tokenisasi&quot;,&quot;local&quot;:&quot;pre-tokenization&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;SentencePiece&quot;,&quot;local&quot;:&quot;sentencepiece&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Tinjauan Algoritma&quot;,&quot;local&quot;:&quot;algorithm-overview&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="normalization-and-pre-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="#normalization-and-pre-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>Normalisasi dan Pra-tokenisasi</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/en/chapter6/section4.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/en/chapter6/section4.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-yagy67">Sebelum kita menyelami lebih dalam tiga algoritma tokenisasi subkata yang paling umum digunakan pada model Transformer (Byte-Pair Encoding [BPE], WordPiece, dan Unigram), kita akan terlebih dahulu melihat tahap praproses yang dilakukan oleh setiap tokenizer terhadap teks. Berikut gambaran umum tingkat tinggi tentang langkah-langkah dalam pipeline tokenisasi:</p> <div class="flex justify-center" data-svelte-h="svelte-tutb3n"><img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter6/tokenization_pipeline.svg" alt="Pipeline tokenisasi."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter6/tokenization_pipeline-dark.svg" alt="Pipeline tokenisasi."></div> <p data-svelte-h="svelte-d4nm86">Sebelum membagi teks menjadi subtoken (sesuai dengan modelnya), tokenizer menjalankan dua langkah: <em>normalisasi</em> dan <em>pra-tokenisasi</em>.</p> <h2 class="relative group"><a id="normalization" 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="#normalization"><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>Normalisasi</span></h2> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/4IIC2jI9CaU" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-xtkpqm">Langkah normalisasi mencakup pembersihan umum, seperti menghapus spasi yang tidak perlu, mengubah huruf menjadi huruf kecil (lowercasing), dan/atau menghapus aksen. Jika kamu familiar dengan <a href="http://www.unicode.org/reports/tr15/" rel="nofollow">normalisasi Unicode</a> (seperti NFC atau NFKC), tokenizer mungkin juga menerapkannya.</p> <p data-svelte-h="svelte-1skida0">Tokenizer dari 🤗 Transformers memiliki atribut <code>backend_tokenizer</code> yang memberikan akses ke tokenizer dasar dari pustaka 🤗 Tokenizers:</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-uncased&quot;</span>)
<span class="hljs-built_in">print</span>(<span class="hljs-built_in">type</span>(tokenizer.backend_tokenizer))<!-- 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 -->&lt;<span class="hljs-keyword">class</span> <span class="hljs-string">&#x27;tokenizers.Tokenizer&#x27;</span>&gt;<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-16ygsj8">Atribut <code>normalizer</code> dari objek <code>tokenizer</code> memiliki metode <code>normalize_str()</code> yang bisa kita gunakan untuk melihat bagaimana proses normalisasi dilakukan:</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>(tokenizer.backend_tokenizer.normalizer.normalize_str(<span class="hljs-string">&quot;Héllò hôw are ü?&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;hello how are u?&#x27;</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-clu4dc">Dalam contoh ini, karena kita menggunakan checkpoint <code>bert-base-uncased</code>, proses normalisasi mencakup perubahan huruf ke huruf kecil dan penghapusan aksen.</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-pe3j28">✏️ <strong>Coba sendiri!</strong> Muat tokenizer dari checkpoint <code>bert-base-cased</code> dan berikan contoh yang sama. Apa perbedaan utama yang kamu lihat antara versi <code>cased</code> dan <code>uncased</code> dari tokenizer?</p></div> <h2 class="relative group"><a id="pre-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="#pre-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>Pra-tokenisasi</span></h2> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/grlLV8AIXug" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-1guenkd">Seperti yang akan kita lihat pada bagian berikutnya, tokenizer tidak bisa langsung dilatih pada teks mentah. Kita perlu membagi teks terlebih dahulu menjadi entitas kecil seperti kata. Di sinilah langkah pra-tokenisasi berperan. Seperti yang telah dibahas di <a href="/course/chapter2">Bab 2</a>, tokenizer berbasis kata dapat membagi teks menjadi kata berdasarkan spasi dan tanda baca. Kata-kata ini akan menjadi batasan bagi subtoken yang akan dipelajari tokenizer saat pelatihan.</p> <p data-svelte-h="svelte-driqgp">Untuk melihat bagaimana tokenizer cepat (fast tokenizer) melakukan pra-tokenisasi, kita bisa menggunakan metode <code>pre_tokenize_str()</code> dari atribut <code>pre_tokenizer</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 -->tokenizer.backend_tokenizer.pre_tokenizer.pre_tokenize_str(<span class="hljs-string">&quot;Hello, how are you?&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;Hello&#x27;</span>, (<span class="hljs-number">0</span>, <span class="hljs-number">5</span>)), (<span class="hljs-string">&#x27;,&#x27;</span>, (<span class="hljs-number">5</span>, <span class="hljs-number">6</span>)), (<span class="hljs-string">&#x27;how&#x27;</span>, (<span class="hljs-number">7</span>, <span class="hljs-number">10</span>)), (<span class="hljs-string">&#x27;are&#x27;</span>, (<span class="hljs-number">11</span>, <span class="hljs-number">14</span>)), (<span class="hljs-string">&#x27;you&#x27;</span>, (<span class="hljs-number">16</span>, <span class="hljs-number">19</span>)), (<span class="hljs-string">&#x27;?&#x27;</span>, (<span class="hljs-number">19</span>, <span class="hljs-number">20</span>))]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1tkswau">Perhatikan bahwa tokenizer sudah mencatat offset, yang memungkinkan kita mendapatkan pemetaan offset seperti yang digunakan pada bagian sebelumnya. Di sini tokenizer mengabaikan dua spasi dan menggantinya dengan satu, tetapi offset tetap mencerminkan jarak sebenarnya.</p> <p data-svelte-h="svelte-74iyxy">Karena kita menggunakan tokenizer BERT, proses pra-tokenisasi mencakup pemisahan berdasarkan spasi dan tanda baca. Tokenizer lain dapat memiliki aturan berbeda. Misalnya, jika kita menggunakan tokenizer GPT-2:</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 = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;gpt2&quot;</span>)
tokenizer.backend_tokenizer.pre_tokenizer.pre_tokenize_str(<span class="hljs-string">&quot;Hello, how are you?&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-12h5k79">Tokenizer ini juga membagi berdasarkan spasi dan tanda baca, tetapi mempertahankan spasi dan menggantinya dengan simbol <code>Ġ</code>, memungkinkan pemulihan spasi asli saat mendekode token:</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;Hello&#x27;</span>, (<span class="hljs-number">0</span>, <span class="hljs-number">5</span>)), (<span class="hljs-string">&#x27;,&#x27;</span>, (<span class="hljs-number">5</span>, <span class="hljs-number">6</span>)), (<span class="hljs-string">&#x27;Ġhow&#x27;</span>, (<span class="hljs-number">6</span>, <span class="hljs-number">10</span>)), (<span class="hljs-string">&#x27;Ġare&#x27;</span>, (<span class="hljs-number">10</span>, <span class="hljs-number">14</span>)), (<span class="hljs-string">&#x27;Ġ&#x27;</span>, (<span class="hljs-number">14</span>, <span class="hljs-number">15</span>)), (<span class="hljs-string">&#x27;Ġyou&#x27;</span>, (<span class="hljs-number">15</span>, <span class="hljs-number">19</span>)),
 (<span class="hljs-string">&#x27;?&#x27;</span>, (<span class="hljs-number">19</span>, <span class="hljs-number">20</span>))]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-bj7o7n">Juga perhatikan bahwa tidak seperti tokenizer BERT, tokenizer ini tidak mengabaikan spasi ganda.</p> <p data-svelte-h="svelte-1wbfhmm">Sebagai contoh terakhir, mari kita lihat tokenizer T5 yang berbasis pada algoritma SentencePiece:</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 = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;t5-small&quot;</span>)
tokenizer.backend_tokenizer.pre_tokenizer.pre_tokenize_str(<span class="hljs-string">&quot;Hello, how are you?&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;▁Hello,&#x27;</span>, (<span class="hljs-number">0</span>, <span class="hljs-number">6</span>)), (<span class="hljs-string">&#x27;▁how&#x27;</span>, (<span class="hljs-number">7</span>, <span class="hljs-number">10</span>)), (<span class="hljs-string">&#x27;▁are&#x27;</span>, (<span class="hljs-number">11</span>, <span class="hljs-number">14</span>)), (<span class="hljs-string">&#x27;▁you?&#x27;</span>, (<span class="hljs-number">16</span>, <span class="hljs-number">20</span>))]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-zldbr5">Seperti tokenizer GPT-2, tokenizer ini mempertahankan spasi dan menggantinya dengan token khusus (<code>▁</code>), tetapi tokenizer T5 hanya membagi berdasarkan spasi, bukan tanda baca. Ia juga menambahkan spasi di awal kalimat (sebelum <code>Hello</code>) dan mengabaikan spasi ganda.</p> <p data-svelte-h="svelte-1l7wzof">Setelah melihat bagaimana beberapa tokenizer memproses teks, kita bisa mulai menjelajahi algoritma dasarnya. Kita akan mulai dengan meninjau sekilas SentencePiece; kemudian, dalam tiga bagian berikutnya, kita akan mempelajari bagaimana tiga algoritma utama tokenisasi subkata bekerja.</p> <h2 class="relative group"><a id="sentencepiece" 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="#sentencepiece"><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>SentencePiece</span></h2> <p data-svelte-h="svelte-10lepug"><a href="https://github.com/google/sentencepiece" rel="nofollow">SentencePiece</a> adalah algoritma tokenisasi untuk praproses teks yang dapat digunakan dengan model apa pun yang akan kita bahas. Algoritma ini memperlakukan teks sebagai rangkaian karakter Unicode, dan mengganti spasi dengan karakter khusus, <code>▁</code>. Digunakan bersama algoritma Unigram (lihat <a href="/course/chapter6/7">bagian 7</a>), algoritma ini bahkan tidak memerlukan langkah pra-tokenisasi, yang sangat berguna untuk bahasa yang tidak menggunakan spasi seperti Bahasa Tionghoa atau Jepang.</p> <p data-svelte-h="svelte-1izmor8">Fitur utama lain dari SentencePiece adalah <em>tokenisasi yang dapat dibalik</em> (reversible tokenization): karena tidak ada perlakuan khusus untuk spasi, proses decoding token cukup dilakukan dengan menggabungkan kembali token dan mengganti <code>▁</code> dengan spasi — ini akan menghasilkan teks yang telah dinormalisasi. Seperti yang sudah kita lihat, tokenizer BERT menghapus spasi ganda, sehingga tokenisasi miliknya tidak dapat dibalik.</p> <h2 class="relative group"><a id="algorithm-overview" 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="#algorithm-overview"><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>Tinjauan Algoritma</span></h2> <p data-svelte-h="svelte-xkb0o4">Pada bagian berikutnya, kita akan membahas tiga algoritma utama tokenisasi subkata: BPE (digunakan oleh GPT-2 dan lainnya), WordPiece (misalnya digunakan oleh BERT), dan Unigram (digunakan oleh T5 dan lainnya). Sebelum mulai, berikut adalah ringkasan cara kerja masing-masing. Jangan ragu untuk kembali ke tabel ini setelah membaca setiap bagian jika masih belum sepenuhnya memahami.</p> <table data-svelte-h="svelte-1o2jhlv"><thead><tr><th>Model</th> <th>BPE</th> <th>WordPiece</th> <th>Unigram</th></tr></thead> <tbody><tr><td>Pelatihan</td> <td>Mulai dari kosakata kecil dan mempelajari aturan penggabungan token</td> <td>Mulai dari kosakata kecil dan mempelajari aturan penggabungan token</td> <td>Mulai dari kosakata besar dan mempelajari aturan penghapusan token</td></tr> <tr><td>Langkah pelatihan</td> <td>Menggabungkan pasangan token paling sering muncul</td> <td>Menggabungkan pasangan dengan skor terbaik berdasarkan frekuensi, mengutamakan pasangan dengan token individual yang lebih jarang</td> <td>Menghapus semua token dalam kosakata untuk meminimalkan loss pada seluruh korpus</td></tr> <tr><td>Yang dipelajari</td> <td>Aturan penggabungan dan kosakata</td> <td>Hanya kosakata</td> <td>Kosakata dengan skor untuk tiap token</td></tr> <tr><td>Proses encoding</td> <td>Membagi kata menjadi karakter, lalu menggabungkan sesuai aturan</td> <td>Mencari subkata terpanjang dari awal yang ada di kosakata, lalu lanjutkan</td> <td>Mencari pembagian token paling mungkin berdasarkan skor pelatihan</td></tr></tbody></table> <p data-svelte-h="svelte-qnjx4h">Sekarang, mari kita mulai dengan BPE!</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/id/chapter6/4.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_ojy514 = {
 assets: "/docs/course/pr_1054/id",
 base: "/docs/course/pr_1054/id",
 env: {}
 };
 const element = document.currentScript.parentElement;
 const data = [null,null];
 Promise.all([
 import("/docs/course/pr_1054/id/_app/immutable/entry/start.4f92af03.js"),
 import("/docs/course/pr_1054/id/_app/immutable/entry/app.19cef1b6.js")
 ]).then(([kit, app]) => {
 kit.start(app, element, {
 node_ids: [0, 49],
 data,
 form: null,
 error: null
 });
 });
 }
 </script>

Xet Storage Details

Size:
35.7 kB
·
Xet hash:
ea65a3775e5245b899c5c993afdb8d02ec8e9a368935e6c80586a035754f9c0e

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