Hugging Face's logo

Buckets:
hf-doc-build
/
doc-dev

Files
xet
hf-doc-build/doc-dev / course /pr_1007 /ru /chapter5 /6.html
rtrm's picture
rtrm
download
raw
89.8 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Семантический поиск с помощью FAISS&quot;,&quot;local&quot;:&quot;семантический-поиск-с-помощью-faiss&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Использование эмбеддингов для семанического поиска&quot;,&quot;local&quot;:&quot;использование-эмбеддингов-для-семанического-поиска&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Загрузка и подготовка датасета&quot;,&quot;local&quot;:&quot;загрузка-и-подготовка-датасета&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Создание текстовых эмбединнгов&quot;,&quot;local&quot;:&quot;создание-текстовых-эмбединнгов&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Использование FAISS для эффективного семантического поиска&quot;,&quot;local&quot;:&quot;использование-faiss-для-эффективного-семантического-поиска&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
 <link href="/docs/course/pr_1007/ru/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/entry/start.1ce3f942.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/scheduler.37c15a92.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/singletons.14904881.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/index.18351ede.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/paths.c9af421c.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/entry/app.ffdd8c8a.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/index.2bf4358c.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/nodes/0.69f5e653.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/each.e59479a4.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/nodes/38.e04dcf6b.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/Tip.363c041f.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/Youtube.1e50a667.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/CodeBlock.4e987730.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/CourseFloatingBanner.9ff4c771.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/DocNotebookDropdown.efc1fb7c.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/FrameworkSwitchCourse.8d4d4ab6.js">
 <link rel="modulepreload" href="/docs/course/pr_1007/ru/_app/immutable/chunks/getInferenceSnippets.1837c472.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Семантический поиск с помощью FAISS&quot;,&quot;local&quot;:&quot;семантический-поиск-с-помощью-faiss&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Использование эмбеддингов для семанического поиска&quot;,&quot;local&quot;:&quot;использование-эмбеддингов-для-семанического-поиска&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Загрузка и подготовка датасета&quot;,&quot;local&quot;:&quot;загрузка-и-подготовка-датасета&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Создание текстовых эмбединнгов&quot;,&quot;local&quot;:&quot;создание-текстовых-эмбединнгов&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Использование FAISS для эффективного семантического поиска&quot;,&quot;local&quot;:&quot;использование-faiss-для-эффективного-семантического-поиска&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="семантический-поиск-с-помощью-faiss" 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="#семантический-поиск-с-помощью-faiss"><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>Семантический поиск с помощью FAISS</span></h1> <div class="flex space-x-1 absolute z-10 right-0 top-0"><a href="https://discuss.huggingface.co/t/chapter-5-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/chapter5/section6_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/en/chapter5/section6_pt.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-1kqyvl2">В <a href="../chapter5/5">разделе 5</a> мы создали набор данных о issues и комментариях GitHub из репозитория 🤗 Datasets. В этом разделе мы будем использовать эту информацию для создания поисковой системы, которая поможет нам найти ответы на самые насущные вопросы о библиотеке!</p> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/OATCgQtNX2o" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <h2 class="relative group"><a id="использование-эмбеддингов-для-семанического-поиска" 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="#использование-эмбеддингов-для-семанического-поиска"><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>Использование эмбеддингов для семанического поиска</span></h2> <p data-svelte-h="svelte-u7ovj5">Как мы видели в <a href="../chapter1/1">Главе 1</a>, языковые модели на основе Transformer представляют каждую лексему в текстовом фрагменте как <em>эмбеддинг-вектор</em>. Оказывается, можно «объединить» отдельные вложения, чтобы создать векторное представление для целых предложений, абзацев или (в некоторых случаях) документов. Затем эти вложения можно использовать для поиска похожих документов в корпусе путем вычисления скалярного произведения (или какой-либо другой метрики сходства) между каждым вложением и возврата документов с наибольшим перекрытием.</p> <p data-svelte-h="svelte-dm8pdq">В этом разделе мы будем использовать вложения для разработки семантической поисковой системы. Эти поисковые системы предлагают несколько преимуществ по сравнению с традиционными подходами, основанными на сопоставлении ключевых слов в запросе с документами.</p> <div class="flex justify-center" data-svelte-h="svelte-yxatr"><img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter5/semantic-search.svg" alt="Semantic search."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter5/semantic-search-dark.svg" alt="Semantic search."></div> <h2 class="relative group"><a id="загрузка-и-подготовка-датасета" 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="#загрузка-и-подготовка-датасета"><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>Загрузка и подготовка датасета</span></h2> <p data-svelte-h="svelte-8ar22o">Первое, что нам нужно сделать, это загрузить наш набор данных об issues GitHub, поэтому давайте воспользуемся библиотекой 🤗 Hub для получения URL-адреса, по которому наш файл хранится в Hugging Face Hub:</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> huggingface_hub <span class="hljs-keyword">import</span> hf_hub_url
data_files = hf_hub_url(
 repo_id=<span class="hljs-string">&quot;lewtun/github-issues&quot;</span>,
 filename=<span class="hljs-string">&quot;datasets-issues-with-comments.jsonl&quot;</span>,
 repo_type=<span class="hljs-string">&quot;dataset&quot;</span>,
)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1ucdqs">С URL-адресом, сохраненным в <code>data_files</code>, мы можем загрузить удаленный набор данных, используя метод, представленный в <a href="../chapter5/2">раздел 2</a>:</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> datasets <span class="hljs-keyword">import</span> load_dataset
issues_dataset = load_dataset(<span class="hljs-string">&quot;json&quot;</span>, data_files=data_files, split=<span class="hljs-string">&quot;train&quot;</span>)
issues_dataset<!-- 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 -->Dataset({
 features: [<span class="hljs-string">&#x27;url&#x27;</span>, <span class="hljs-string">&#x27;repository_url&#x27;</span>, <span class="hljs-string">&#x27;labels_url&#x27;</span>, <span class="hljs-string">&#x27;comments_url&#x27;</span>, <span class="hljs-string">&#x27;events_url&#x27;</span>, <span class="hljs-string">&#x27;html_url&#x27;</span>, <span class="hljs-string">&#x27;id&#x27;</span>, <span class="hljs-string">&#x27;node_id&#x27;</span>, <span class="hljs-string">&#x27;number&#x27;</span>, <span class="hljs-string">&#x27;title&#x27;</span>, <span class="hljs-string">&#x27;user&#x27;</span>, <span class="hljs-string">&#x27;labels&#x27;</span>, <span class="hljs-string">&#x27;state&#x27;</span>, <span class="hljs-string">&#x27;locked&#x27;</span>, <span class="hljs-string">&#x27;assignee&#x27;</span>, <span class="hljs-string">&#x27;assignees&#x27;</span>, <span class="hljs-string">&#x27;milestone&#x27;</span>, <span class="hljs-string">&#x27;comments&#x27;</span>, <span class="hljs-string">&#x27;created_at&#x27;</span>, <span class="hljs-string">&#x27;updated_at&#x27;</span>, <span class="hljs-string">&#x27;closed_at&#x27;</span>, <span class="hljs-string">&#x27;author_association&#x27;</span>, <span class="hljs-string">&#x27;active_lock_reason&#x27;</span>, <span class="hljs-string">&#x27;pull_request&#x27;</span>, <span class="hljs-string">&#x27;body&#x27;</span>, <span class="hljs-string">&#x27;performed_via_github_app&#x27;</span>, <span class="hljs-string">&#x27;is_pull_request&#x27;</span>],
 num_rows: <span class="hljs-number">2855</span>
})<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1ab3s1i">Здесь мы указали подвыборку <code>train</code> по умолчанию в <code>load_dataset()</code>, поэтому он возвращает <code>Dataset</code> вместо <code>DatasetDict</code>. Первым делом нужно отфильтровать запросы на pull-requests, поскольку они, как правило, редко используются для ответов на вопросы пользователей и создают шум в нашей поисковой системе. Как должно быть уже известно, мы можем использовать функцию <code>Dataset.filter()</code>, чтобы исключить эти строки из нашего набора данных. Давайте также отфильтруем строки без комментариев, поскольку они не дают ответов на запросы пользователей:</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 -->issues_dataset = issues_dataset.<span class="hljs-built_in">filter</span>(
 <span class="hljs-keyword">lambda</span> x: (x[<span class="hljs-string">&quot;is_pull_request&quot;</span>] == <span class="hljs-literal">False</span> <span class="hljs-keyword">and</span> <span class="hljs-built_in">len</span>(x[<span class="hljs-string">&quot;comments&quot;</span>]) &gt; <span class="hljs-number">0</span>)
)
issues_dataset<!-- 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 -->Dataset({
 features: [<span class="hljs-string">&#x27;url&#x27;</span>, <span class="hljs-string">&#x27;repository_url&#x27;</span>, <span class="hljs-string">&#x27;labels_url&#x27;</span>, <span class="hljs-string">&#x27;comments_url&#x27;</span>, <span class="hljs-string">&#x27;events_url&#x27;</span>, <span class="hljs-string">&#x27;html_url&#x27;</span>, <span class="hljs-string">&#x27;id&#x27;</span>, <span class="hljs-string">&#x27;node_id&#x27;</span>, <span class="hljs-string">&#x27;number&#x27;</span>, <span class="hljs-string">&#x27;title&#x27;</span>, <span class="hljs-string">&#x27;user&#x27;</span>, <span class="hljs-string">&#x27;labels&#x27;</span>, <span class="hljs-string">&#x27;state&#x27;</span>, <span class="hljs-string">&#x27;locked&#x27;</span>, <span class="hljs-string">&#x27;assignee&#x27;</span>, <span class="hljs-string">&#x27;assignees&#x27;</span>, <span class="hljs-string">&#x27;milestone&#x27;</span>, <span class="hljs-string">&#x27;comments&#x27;</span>, <span class="hljs-string">&#x27;created_at&#x27;</span>, <span class="hljs-string">&#x27;updated_at&#x27;</span>, <span class="hljs-string">&#x27;closed_at&#x27;</span>, <span class="hljs-string">&#x27;author_association&#x27;</span>, <span class="hljs-string">&#x27;active_lock_reason&#x27;</span>, <span class="hljs-string">&#x27;pull_request&#x27;</span>, <span class="hljs-string">&#x27;body&#x27;</span>, <span class="hljs-string">&#x27;performed_via_github_app&#x27;</span>, <span class="hljs-string">&#x27;is_pull_request&#x27;</span>],
 num_rows: <span class="hljs-number">771</span>
})<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-g25nn8">Мы видим, что в нашем наборе данных много столбцов, большинство из которых нам не нужно для создания нашей поисковой системы. С точки зрения поиска наиболее информативными столбцами являются <code>title</code>, <code>body</code> и <code>comments</code>, а <code>html_url</code> содержит нам ссылку на исходную проблему. Давайте воспользуемся функцией <code>Dataset.remove_columns()</code>, чтобы удалить остальные столбцы:</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 -->columns = issues_dataset.column_names
columns_to_keep = [<span class="hljs-string">&quot;title&quot;</span>, <span class="hljs-string">&quot;body&quot;</span>, <span class="hljs-string">&quot;html_url&quot;</span>, <span class="hljs-string">&quot;comments&quot;</span>]
columns_to_remove = <span class="hljs-built_in">set</span>(columns_to_keep).symmetric_difference(columns)
issues_dataset = issues_dataset.remove_columns(columns_to_remove)
issues_dataset<!-- 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 -->Dataset({
 features: [<span class="hljs-string">&#x27;html_url&#x27;</span>, <span class="hljs-string">&#x27;title&#x27;</span>, <span class="hljs-string">&#x27;comments&#x27;</span>, <span class="hljs-string">&#x27;body&#x27;</span>],
 num_rows: <span class="hljs-number">771</span>
})<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-kf244j">Чтобы создать наши эмбеддинги, мы дополним каждый комментарий заголовком и телом проблемы, поскольку эти поля часто содержат полезную контекстную информацию. Поскольку наш столбец <code>comments</code> в настоящее время представляет собой список комментариев для каждой проблемы, нам нужно «развернуть» столбец, чтобы каждая строка состояла из кортежа <code>(html_url, title, body, comment)</code>. В Pandas мы можем сделать это с помощью функции <a href="https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.explode.html" rel="nofollow"><code>DataFrame.explode()</code></a>, которая создает новую строку для каждого элемента в столбце, похожем на список, при копировании всех других значений столбца. Чтобы увидеть это в действии, давайте сначала переключимся на формат Pandas <code>DataFrame</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 -->issues_dataset.set_format(<span class="hljs-string">&quot;pandas&quot;</span>)
df = issues_dataset[:]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-garzmz">Если мы проверим первую строку в этом <code>DataFrame</code>, мы увидим, что есть четыре комментария, связанных с этой проблемой:</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 -->df[<span class="hljs-string">&quot;comments&quot;</span>][<span class="hljs-number">0</span>].tolist()<!-- 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;the bug code locate in ：\r\n if data_args.task_name is not None:\r\n # Downloading and loading a dataset from the hub.\r\n datasets = load_dataset(&quot;glue&quot;, data_args.task_name, cache_dir=model_args.cache_dir)&#x27;</span>,
 <span class="hljs-string">&#x27;Hi @jinec,\r\n\r\nFrom time to time we get this kind of `ConnectionError` coming from the github.com website: https://raw.githubusercontent.com\r\n\r\nNormally, it should work if you wait a little and then retry.\r\n\r\nCould you please confirm if the problem persists?&#x27;</span>,
 <span class="hljs-string">&#x27;cannot connect，even by Web browser，please check that there is some problems。&#x27;</span>,
 <span class="hljs-string">&#x27;I can access https://raw.githubusercontent.com/huggingface/datasets/1.7.0/datasets/glue/glue.py without problem...&#x27;</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-81f1dh">Когда мы «развернем» <code>df</code>, мы ожидаем получить по одной строке для каждого из этих комментариев. Проверим, так ли это:</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 -->comments_df = df.explode(<span class="hljs-string">&quot;comments&quot;</span>, ignore_index=<span class="hljs-literal">True</span>)
comments_df.head(<span class="hljs-number">4</span>)<!-- HTML_TAG_END --></pre></div> <table border="1" class="dataframe" style="table-layout: fixed; word-wrap:break-word; width: 100%;" data-svelte-h="svelte-1g5whzd"><thead><tr style="text-align: right;"><th></th> <th>html_url</th> <th>title</th> <th>comments</th> <th>body</th></tr></thead> <tbody><tr><th>0</th> <td>https://github.com/huggingface/datasets/issues/2787</td> <td>ConnectionError: Couldn&#39;t reach https://raw.githubusercontent.com</td> <td>the bug code locate in ：\r\n if data_args.task_name is not None...</td> <td>Hello,\r\nI am trying to run run_glue.py and it gives me this error...</td></tr> <tr><th>1</th> <td>https://github.com/huggingface/datasets/issues/2787</td> <td>ConnectionError: Couldn&#39;t reach https://raw.githubusercontent.com</td> <td>Hi @jinec,\r\n\r\nFrom time to time we get this kind of `ConnectionError` coming from the github.com website: https://raw.githubusercontent.com...</td> <td>Hello,\r\nI am trying to run run_glue.py and it gives me this error...</td></tr> <tr><th>2</th> <td>https://github.com/huggingface/datasets/issues/2787</td> <td>ConnectionError: Couldn&#39;t reach https://raw.githubusercontent.com</td> <td>cannot connect，even by Web browser，please check that there is some problems。</td> <td>Hello,\r\nI am trying to run run_glue.py and it gives me this error...</td></tr> <tr><th>3</th> <td>https://github.com/huggingface/datasets/issues/2787</td> <td>ConnectionError: Couldn&#39;t reach https://raw.githubusercontent.com</td> <td>I can access https://raw.githubusercontent.com/huggingface/datasets/1.7.0/datasets/glue/glue.py without problem...</td> <td>Hello,\r\nI am trying to run run_glue.py and it gives me this error...</td></tr></tbody></table> <p data-svelte-h="svelte-s611ju">Отлично, мы видим, что строки были скопированы, а столбец <code>comments</code> содержит отдельные комментарии! Теперь, когда мы закончили с Pandas, мы можем быстро вернуться к <code>Dataset</code>, загрузив <code>DataFrame</code> в память:</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> datasets <span class="hljs-keyword">import</span> Dataset
comments_dataset = Dataset.from_pandas(comments_df)
comments_dataset<!-- 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 -->Dataset({
 features: [<span class="hljs-string">&#x27;html_url&#x27;</span>, <span class="hljs-string">&#x27;title&#x27;</span>, <span class="hljs-string">&#x27;comments&#x27;</span>, <span class="hljs-string">&#x27;body&#x27;</span>],
 num_rows: <span class="hljs-number">2842</span>
})<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-iyjal4">Хорошо, это дало нам несколько тысяч комментариев для работы!</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-3jhlkr">✏️ <strong>Попробуйте!</strong> Посмотрите, сможете ли вы использовать <code>Dataset.map()</code>, чтобы развернуть столбец <code>comments</code> столбца <code>issues_dataset</code> <em>без</em> использования Pandas. Это немного сложно; вы можете найти раздел <a href="https://huggingface.co/docs/datasets/about_map_batch#batch-mapping" rel="nofollow">“Batch mapping”</a> документации 🤗 Datasets, полезным для этой задачи.</p></div> <p data-svelte-h="svelte-opu2kq">Теперь, когда у нас есть один комментарий в строке, давайте создадим новый столбец <code>comments_length</code>, содержащий количество слов в комментарии:</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 -->comments_dataset = comments_dataset.<span class="hljs-built_in">map</span>(
 <span class="hljs-keyword">lambda</span> x: {<span class="hljs-string">&quot;comment_length&quot;</span>: <span class="hljs-built_in">len</span>(x[<span class="hljs-string">&quot;comments&quot;</span>].split())}
)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-7k4b3c">Мы можем использовать этот новый столбец для фильтрации коротких комментариев, которые обычно содержат такие слова, как «cc @lewtun» или Thanks!», которые не имеют отношения к нашей поисковой системе. Нет точного числа для выбора порога числа слов, но около 15 слов кажется хорошим началом:</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 -->comments_dataset = comments_dataset.<span class="hljs-built_in">filter</span>(<span class="hljs-keyword">lambda</span> x: x[<span class="hljs-string">&quot;comment_length&quot;</span>] &gt; <span class="hljs-number">15</span>)
comments_dataset<!-- 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 -->Dataset({
 features: [<span class="hljs-string">&#x27;html_url&#x27;</span>, <span class="hljs-string">&#x27;title&#x27;</span>, <span class="hljs-string">&#x27;comments&#x27;</span>, <span class="hljs-string">&#x27;body&#x27;</span>, <span class="hljs-string">&#x27;comment_length&#x27;</span>],
 num_rows: <span class="hljs-number">2098</span>
})<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-qjfuzk">Немного очистив наш набор данных, давайте соединим название issue, описание и комментарии вместе в новом столбце <code>text</code>. Как обычно, мы напишем простую функцию, которую мы можем передать в <code>Dataset.map()</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 --><span class="hljs-keyword">def</span> <span class="hljs-title function_">concatenate_text</span>(<span class="hljs-params">examples</span>):
 <span class="hljs-keyword">return</span> {
 <span class="hljs-string">&quot;text&quot;</span>: examples[<span class="hljs-string">&quot;title&quot;</span>]
 + <span class="hljs-string">&quot; \n &quot;</span>
 + examples[<span class="hljs-string">&quot;body&quot;</span>]
 + <span class="hljs-string">&quot; \n &quot;</span>
 + examples[<span class="hljs-string">&quot;comments&quot;</span>]
 }
comments_dataset = comments_dataset.<span class="hljs-built_in">map</span>(concatenate_text)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-12xaa86">Наконец-то мы готовы создать несколько эмбеддингов’! Давайте взглянем.</p> <h2 class="relative group"><a id="создание-текстовых-эмбединнгов" 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="#создание-текстовых-эмбединнгов"><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>Создание текстовых эмбединнгов</span></h2> <p data-svelte-h="svelte-m3fi2j">В <a href="../chapter2/1">Главе 2</a> мы видели, что можно получить эмбеддингов токенов с помощью класса AutoModel. Все, что нам нужно сделать, это выбрать подходящую контрольную точку для загрузки модели. К счастью, есть библиотека под названием <code>sentence-transformers</code>, предназначенная для создания эмбеддингов. Как описано в <a href="https://www.sbert.net/examples/applications/semantic-search/README.html#%D1%81%D0%B8%D0%BC%D0%BC%D0%B5%D1%82%D1%80%D0%B8%D1%87%D0%BD%D1%8B%D0%B9-vs-%D0%B0%D1%81%D0%B8%D0%BC%D0%BC%D0%B5%D1%82%D1%80%D0%B8%D1%87%D0%BD%D1%8B%D0%B9-semantic-search" rel="nofollow">документации</a> библиотеки, наш вариант использования является примером <em>асимметричного семантического поиска</em> потому что у нас есть короткий запрос, ответ на который мы хотели бы найти в более длинном документе, например, в комментарии к проблеме. В удобной <a href="https://www.sbert.net/docs/pretrained_models.html#model-overview" rel="nofollow">таблице обзора модели</a> в документации указано, что контрольная точка <code>multi-qa-mpnet-base-dot-v1</code> имеет лучшую производительность для семантического поиска, поэтому мы будем использовать её для нашего приложения. Мы также загрузим токенизатор, используя ту же контрольную точку:</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, AutoModel
model_ckpt = <span class="hljs-string">&quot;sentence-transformers/multi-qa-mpnet-base-dot-v1&quot;</span>
tokenizer = AutoTokenizer.from_pretrained(model_ckpt)
model = AutoModel.from_pretrained(model_ckpt)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-inum91">Чтобы ускорить процесс построения эмбеддингов, рекомендуется переместить модель и входные данные на устройстве с графическим процессором, поэтому давайте сделаем это сейчас:</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">import</span> torch
device = torch.device(<span class="hljs-string">&quot;cuda&quot;</span>)
model.to(device)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1hrvssd">Как мы упоминали ранее, мы хотели бы представить каждую запись в нашем корпусе issues GitHub как единый вектор, поэтому нам нужно каким-то образом «объединить» или усреднить наши вложения токенов. Одним из популярных подходов является выполнение <em>CLS pooling</em> выходных данных нашей модели, когда мы просто собираем последнее скрытое состояние для специального токена <code>[CLS]</code>. Следующая функция поможет нам:</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_">cls_pooling</span>(<span class="hljs-params">model_output</span>):
 <span class="hljs-keyword">return</span> model_output.last_hidden_state[:, <span class="hljs-number">0</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-x8q51b">Далее мы создадим вспомогательную функцию, которая разметит список документов, поместит тензоры в GPU, передаст их в модель и, наконец, применит CLS pooling к выходным данным:</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_">get_embeddings</span>(<span class="hljs-params">text_list</span>):
 encoded_input = tokenizer(
 text_list, padding=<span class="hljs-literal">True</span>, truncation=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>
 )
 encoded_input = {k: v.to(device) <span class="hljs-keyword">for</span> k, v <span class="hljs-keyword">in</span> encoded_input.items()}
 model_output = model(**encoded_input)
 <span class="hljs-keyword">return</span> cls_pooling(model_output)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1arf6dc">Мы можем проверить работу функции, передав ей первую текстовую запись в нашем корпусе и проверив размерности данных на выходе:</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 -->embedding = get_embeddings(comments_dataset[<span class="hljs-string">&quot;text&quot;</span>][<span class="hljs-number">0</span>])
embedding.shape<!-- 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 -->torch.Size([<span class="hljs-number">1</span>, <span class="hljs-number">768</span>])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1lqreox">Отлично, мы преобразовали первую запись в нашем корпусе в 768-мерный вектор! Мы можем использовать <code>Dataset.map()</code>, чтобы применить нашу функцию <code>get_embeddings()</code> к каждой строке в нашем корпусе, поэтому давайте создадим новый столбец <code>embeddings</code> следующим образом:</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 -->embeddings_dataset = comments_dataset.<span class="hljs-built_in">map</span>(
 <span class="hljs-keyword">lambda</span> x: {<span class="hljs-string">&quot;embeddings&quot;</span>: get_embeddings(x[<span class="hljs-string">&quot;text&quot;</span>]).detach().cpu().numpy()[<span class="hljs-number">0</span>]}
)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1o2epj6">Обратите внимание, что мы преобразовали вложения в массивы NumPy — это потому, что 🤗 Datasets требуют этого формата, когда мы пытаемся проиндексировать их с помощью FAISS, что мы сделаем дальше.</p> <h2 class="relative group"><a id="использование-faiss-для-эффективного-семантического-поиска" 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="#использование-faiss-для-эффективного-семантического-поиска"><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>Использование FAISS для эффективного семантического поиска</span></h2> <p data-svelte-h="svelte-l8ge7o">Теперь, когда у нас есть датасет с эмбеддингами, нам нужен способ поиска по ним. Для этого мы будем использовать специальную структуру данных из 🤗 Datasets, называемую <em>FAISS index</em>. <a href="https://faiss.ai/" rel="nofollow">FAISS</a> (сокращение от Facebook AI Similarity Search) — это библиотека, предоставляющая эффективные алгоритмы для быстрого поиска и кластеризации эмбеддингов.</p> <p data-svelte-h="svelte-u31vl5">Основная идея FAISS состоит в том, чтобы создать специальную структуру данных, называемую <em>index</em>, которая позволяет найти, какие эмбеддинги подобны входным эмбеддингам. Создать индекс FAISS в 🤗 Datasets очень просто — мы используем функцию <code>Dataset.add_faiss_index()</code> и указываем, какой столбец нашего набора данных мы хотим проиндексировать:</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 -->embeddings_dataset.add_faiss_index(column=<span class="hljs-string">&quot;embeddings&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1uma1g4">Теперь мы можем выполнять запросы к этому индексу, выполняя поиск ближайшего соседа с помощью функции <code>Dataset.get_nearest_examples()</code>. Давайте проверим это, сначала внедрив вопрос следующим образом:</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 -->question = <span class="hljs-string">&quot;How can I load a dataset offline?&quot;</span>
question_embedding = get_embeddings([question]).cpu().detach().numpy()
question_embedding.shape<!-- 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 -->torch.Size([<span class="hljs-number">1</span>, <span class="hljs-number">768</span>])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-eam6de">Как и в случае с документами, теперь у нас есть 768-мерный вектор, представляющий запрос, который мы можем сравнить со всем корпусом, чтобы найти наиболее похожие объекты:</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 -->scores, samples = embeddings_dataset.get_nearest_examples(
 <span class="hljs-string">&quot;embeddings&quot;</span>, question_embedding, k=<span class="hljs-number">5</span>
)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-8i9jhj">Функция <code>Dataset.get_nearest_examples()</code> возвращает набор оценок, которые ранжируют совпадение между запросом и документом, и соответствующий набор образцов (здесь 5 лучших совпадений). Давайте соберем их в <code>pandas.DataFrame</code>, чтобы мы могли легко их отсортировать:</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">import</span> pandas <span class="hljs-keyword">as</span> pd
samples_df = pd.DataFrame.from_dict(samples)
samples_df[<span class="hljs-string">&quot;scores&quot;</span>] = scores
samples_df.sort_values(<span class="hljs-string">&quot;scores&quot;</span>, ascending=<span class="hljs-literal">False</span>, inplace=<span class="hljs-literal">True</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-j9fzh3">Теперь мы можем пройтись по первым нескольким строкам, чтобы увидеть, насколько наш запрос соответствует имеющимся комментариям:</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">for</span> _, row <span class="hljs-keyword">in</span> samples_df.iterrows():
 <span class="hljs-built_in">print</span>(<span class="hljs-string">f&quot;COMMENT: <span class="hljs-subst">{row.comments}</span>&quot;</span>)
 <span class="hljs-built_in">print</span>(<span class="hljs-string">f&quot;SCORE: <span class="hljs-subst">{row.scores}</span>&quot;</span>)
 <span class="hljs-built_in">print</span>(<span class="hljs-string">f&quot;TITLE: <span class="hljs-subst">{row.title}</span>&quot;</span>)
 <span class="hljs-built_in">print</span>(<span class="hljs-string">f&quot;URL: <span class="hljs-subst">{row.html_url}</span>&quot;</span>)
 <span class="hljs-built_in">print</span>(<span class="hljs-string">&quot;=&quot;</span> * <span class="hljs-number">50</span>)
 <span class="hljs-built_in">print</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">&quot;&quot;&quot;
COMMENT: Requiring online connection is a deal breaker in some cases unfortunately so it&#x27;d be great if offline mode is added similar to how `transformers` loads models offline fine.
@mandubian&#x27;s second bullet point suggests that there&#x27;s a workaround allowing you to use your offline (custom?) dataset with `datasets`. Could you please elaborate on how that should look like?
SCORE: 25.505046844482422
TITLE: Discussion using datasets in offline mode
URL: https://github.com/huggingface/datasets/issues/824
==================================================
COMMENT: The local dataset builders (csv, text , json and pandas) are now part of the `datasets` package since #1726 :)
You can now use them offline
\`\`\`python
datasets = load_dataset(&quot;text&quot;, data_files=data_files)
\`\`\`
We&#x27;ll do a new release soon
SCORE: 24.555509567260742
TITLE: Discussion using datasets in offline mode
URL: https://github.com/huggingface/datasets/issues/824
==================================================
COMMENT: I opened a PR that allows to reload modules that have already been loaded once even if there&#x27;s no internet.
Let me know if you know other ways that can make the offline mode experience better. I&#x27;d be happy to add them :)
I already note the &quot;freeze&quot; modules option, to prevent local modules updates. It would be a cool feature.
----------
&gt; @mandubian&#x27;s second bullet point suggests that there&#x27;s a workaround allowing you to use your offline (custom?) dataset with `datasets`. Could you please elaborate on how that should look like?
Indeed `load_dataset` allows to load remote dataset script (squad, glue, etc.) but also you own local ones.
For example if you have a dataset script at `./my_dataset/my_dataset.py` then you can do
\`\`\`python
load_dataset(&quot;./my_dataset&quot;)
\`\`\`
and the dataset script will generate your dataset once and for all.
----------
About I&#x27;m looking into having `csv`, `json`, `text`, `pandas` dataset builders already included in the `datasets` package, so that they are available offline by default, as opposed to the other datasets that require the script to be downloaded.
cf #1724
SCORE: 24.14896583557129
TITLE: Discussion using datasets in offline mode
URL: https://github.com/huggingface/datasets/issues/824
==================================================
COMMENT: &gt; here is my way to load a dataset offline, but it **requires** an online machine
&gt;
&gt; 1. (online machine)
&gt;
&gt; ```
&gt;
&gt; import datasets
&gt;
&gt; data = datasets.load_dataset(...)
&gt;
&gt; data.save_to_disk(/YOUR/DATASET/DIR)
&gt;
&gt; ```
&gt;
&gt; 2. copy the dir from online to the offline machine
&gt;
&gt; 3. (offline machine)
&gt;
&gt; ```
&gt;
&gt; import datasets
&gt;
&gt; data = datasets.load_from_disk(/SAVED/DATA/DIR)
&gt;
&gt; ```
&gt;
&gt;
&gt;
&gt; HTH.
SCORE: 22.893993377685547
TITLE: Discussion using datasets in offline mode
URL: https://github.com/huggingface/datasets/issues/824
==================================================
COMMENT: here is my way to load a dataset offline, but it **requires** an online machine
1. (online machine)
\`\`\`
import datasets
data = datasets.load_dataset(...)
data.save_to_disk(/YOUR/DATASET/DIR)
\`\`\`
2. copy the dir from online to the offline machine
3. (offline machine)
\`\`\`
import datasets
data = datasets.load_from_disk(/SAVED/DATA/DIR)
\`\`\`
HTH.
SCORE: 22.406635284423828
TITLE: Discussion using datasets in offline mode
URL: https://github.com/huggingface/datasets/issues/824
==================================================
&quot;&quot;&quot;</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-bi04jq">Неплохо! Наше второе обращение, кажется, соответствует запросу.</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-slam50">✏️ <strong>Попробуйте!</strong> Создайте свой собственный запрос и посмотрите, сможете ли вы найти ответ в найденных документах. Возможно, вам придется увеличить параметр <code>k</code> в <code>Dataset.get_nearest_examples()</code>, чтобы расширить поиск.</p></div> <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/ru/chapter5/6.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_v9b91u = {
 assets: "/docs/course/pr_1007/ru",
 base: "/docs/course/pr_1007/ru",
 env: {}
 };
 const element = document.currentScript.parentElement;
 const data = [null,null];
 Promise.all([
 import("/docs/course/pr_1007/ru/_app/immutable/entry/start.1ce3f942.js"),
 import("/docs/course/pr_1007/ru/_app/immutable/entry/app.ffdd8c8a.js")
 ]).then(([kit, app]) => {
 kit.start(app, element, {
 node_ids: [0, 38],
 data,
 form: null,
 error: null
 });
 });
 }
 </script>

Xet Storage Details

Size:
89.8 kB
·
Xet hash:
4acf850ba596dad2c0631fbb33044ad40ad16ca104ef632c04ba6adf2b2d8fe4

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