course/pr_1021/fr/chapter5/6.html

<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Recherche sémantique avec FAISS&quot;,&quot;local&quot;:&quot;recherche-sémantique-avec-faiss&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Utilisation des enchâssements pour la recherche sémantique&quot;,&quot;local&quot;:&quot;utilisation-des-enchâssements-pour-la-recherche-sémantique&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Chargement et préparation du jeu de données&quot;,&quot;local&quot;:&quot;chargement-et-préparation-du-jeu-de-données&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Création d’enchâssements pour les textes&quot;,&quot;local&quot;:&quot;création-denchâssements-pour-les-textes&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Utilisation de FAISS pour une recherche de similarité efficace&quot;,&quot;local&quot;:&quot;utilisation-de-faiss-pour-une-recherche-de-similarité-efficace&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
		<link href="/docs/course/pr_1021/fr/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/entry/start.e9b4e796.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/scheduler.37c15a92.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/singletons.f2b97a67.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/index.18351ede.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/paths.646b7f0e.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/entry/app.bf6a4ef3.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/index.2bf4358c.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/nodes/0.924470e0.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/each.e59479a4.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/nodes/39.16625819.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/Tip.363c041f.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/Youtube.1e50a667.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/CodeBlock.4e987730.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/CourseFloatingBanner.6add7356.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/FrameworkSwitchCourse.8d4d4ab6.js">
		<link rel="modulepreload" href="/docs/course/pr_1021/fr/_app/immutable/chunks/getInferenceSnippets.ebf8be91.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Recherche sémantique avec FAISS&quot;,&quot;local&quot;:&quot;recherche-sémantique-avec-faiss&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Utilisation des enchâssements pour la recherche sémantique&quot;,&quot;local&quot;:&quot;utilisation-des-enchâssements-pour-la-recherche-sémantique&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Chargement et préparation du jeu de données&quot;,&quot;local&quot;:&quot;chargement-et-préparation-du-jeu-de-données&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Création d’enchâssements pour les textes&quot;,&quot;local&quot;:&quot;création-denchâssements-pour-les-textes&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Utilisation de FAISS pour une recherche de similarité efficace&quot;,&quot;local&quot;:&quot;utilisation-de-faiss-pour-une-recherche-de-similarité-efficace&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="recherche-sémantique-avec-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="#recherche-sémantique-avec-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>Recherche sémantique avec 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> <div class="relative colab-dropdown "> <button class="  " type="button">  <img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"> </button>    </div> <div class="relative colab-dropdown "> <button class="  " type="button">  <img alt="Open In Studio Lab" class="!m-0" src="https://studiolab.sagemaker.aws/studiolab.svg"> </button>    </div></div> <p data-svelte-h="svelte-g453z0">Dans <a href="/course/fr/chapter5/5">section 5</a>, nous avons créé un jeu de données de problèmes et de commentaires GitHub à partir du dépôt 🤗 <em>Datasets</em>. Dans cette section, nous utilisons ces informations pour créer un moteur de recherche qui peut nous aider à trouver des réponses à nos questions les plus urgentes sur la bibliothèque !</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="utilisation-des-enchâssements-pour-la-recherche-sémantique" 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="#utilisation-des-enchâssements-pour-la-recherche-sémantique"><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>Utilisation des enchâssements pour la recherche sémantique</span></h2> <p data-svelte-h="svelte-nzpp3o">Comme nous l’avons vu dans le <a href="/course/fr/chapter1">chapitre 1</a>, les modèles de langage basés sur les <em>transformers</em> représentent chaque <em>token</em> dans une étendue de texte sous la forme d’un <em>enchâssement</em>. Il s’avère que l’on peut regrouper les enchâssements individuels pour créer une représentation vectorielle pour des phrases entières, des paragraphes ou (dans certains cas) des documents. Ces enchâssements peuvent ensuite être utilisés pour trouver des documents similaires dans le corpus en calculant la similarité du produit scalaire (ou une autre métrique de similarité) entre chaque enchâssement et en renvoyant les documents avec le plus grand chevauchement.</p> <p data-svelte-h="svelte-dcpi9a">Dans cette section, nous utilisons les enchâssements pour développer un moteur de recherche sémantique. Ces moteurs de recherche offrent plusieurs avantages par rapport aux approches conventionnelles basées sur la correspondance des mots-clés dans une requête avec les documents.</p> <div class="flex justify-center" data-svelte-h="svelte-w2yidl"><img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter5/semantic-search.svg" alt="Recherche sémantique."> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter5/semantic-search-dark.svg" alt="Recherche sémantique."></div>  <h2 class="relative group"><a id="chargement-et-préparation-du-jeu-de-données" 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="#chargement-et-préparation-du-jeu-de-données"><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>Chargement et préparation du jeu de données</span></h2> <p data-svelte-h="svelte-1tqvpff">La première chose que nous devons faire est de télécharger notre jeu de données de problèmes GitHub. Utilisons la bibliothèque 🤗 <em>Hub</em> pour résoudre l’URL où notre fichier est stocké sur le <em>Hub</em> d’Hugging Face :</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-4fgai5">Avec l’URL stocké dans <code>data_files</code>, nous pouvons ensuite charger le jeu de données distant en utilisant la méthode introduite dans <a href="/course/fr/chapter5/2">section 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-19jgbir">Ici, nous avons spécifié l’échantillon <code>train</code> par défaut dans <code>load_dataset()</code>, de sorte que cela renvoie un <code>Dataset</code> au lieu d’un <code>DatasetDict</code>. La première chose à faire est de filtrer les <em>pull requests</em> car celles-ci ont tendance à être rarement utilisées pour répondre aux requêtes des utilisateurs et introduiront du bruit dans notre moteur de recherche. Comme cela devrait être familier maintenant, nous pouvons utiliser la fonction <code>Dataset.filter()</code> pour exclure ces lignes de notre jeu de données. Pendant que nous y sommes, filtrons également les lignes sans commentaires, car celles-ci ne fournissent aucune réponse aux requêtes des utilisateurs :</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-19gal8m">Nous pouvons voir qu’il y a beaucoup de colonnes dans notre jeu de données, dont la plupart n’ont pas besoin de construire notre moteur de recherche. Du point de vue de la recherche, les colonnes les plus informatives sont <code>title</code>, <code>body</code> et <code>comments</code>, tandis que <code>html_url</code> nous fournit un lien vers le problème source. Utilisons la fonction <code>Dataset.remove_columns()</code> pour supprimer le reste :</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-1orepz7">Pour créer nos enchâssements, nous ajoutons à chaque commentaire le titre et le corps du problème, car ces champs contiennent des informations contextuelles utiles. Étant donné que notre colonne <code>comments</code> est actuellement une liste de commentaires pour chaque problème, nous devons « éclater » la colonne afin que chaque ligne se compose d’un <em>tuple</em> <code>(html_url, title, body, comment)</code>. Dans Pandas, nous pouvons le faire avec la fonction <a href="https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.explode.html" rel="nofollow"><code>DataFrame.explode()</code></a>, qui crée une nouvelle ligne pour chaque élément dans une colonne de type liste, tout en répliquant toutes les autres valeurs de colonne. Pour voir cela en action, passons d’abord au format <code>DataFrame</code> de Pandas :</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-1op0ytr">Si nous inspectons la première ligne de ce <code>DataFrame</code>, nous pouvons voir qu’il y a quatre commentaires associés à ce problème :</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-yeckxp">Lorsque nous décomposons <code>df</code>, nous nous attendons à obtenir une ligne pour chacun de ces commentaires. Vérifions si c’est le cas :</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-1hju2sl">Génial, nous pouvons voir que les lignes ont été répliquées, avec la colonne <code>comments</code> contenant les commentaires individuels ! Maintenant que nous en avons fini avec Pandas, nous pouvons rapidement revenir à un <code>Dataset</code> en chargeant le <code>DataFrame</code> en mémoire :</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-wgc24a">D’accord, cela nous a donné quelques milliers de commentaires avec lesquels travailler !</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-19dmo8j">✏️ <strong>Essayez !</strong> Voyez si vous pouvez utiliser <code>Dataset.map()</code> pour exploser la colonne <code>comments</code> de <code>issues_dataset</code> <em>sans</em> recourir à l’utilisation de Pandas. C’est un peu délicat. La section <a href="https://huggingface.co/docs/datasets/about_map_batch#batch-mapping" rel="nofollow">« Batch mapping »</a> de la documentation 🤗 <em>Datasets</em> peut être utile pour cette tâche.</p></div> <p data-svelte-h="svelte-1yihtam">Maintenant que nous avons un commentaire par ligne, créons une nouvelle colonne <code>comments_length</code> contenant le nombre de mots par commentaire :</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-1o9cnrz">Nous pouvons utiliser cette nouvelle colonne pour filtrer les commentaires courts incluant généralement des éléments tels que « cc @lewtun » ou « Merci ! » qui ne sont pas pertinents pour notre moteur de recherche. Il n’y a pas de nombre précis à sélectionner pour le filtre mais 15 mots semblent être un bon début :</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-813lg6">Après avoir un peu nettoyé notre jeu de données, concaténons le titre, la description et les commentaires du problème dans une nouvelle colonne <code>text</code>. Comme d’habitude, nous allons écrire une fonction simple que nous pouvons passer à <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-1ib5u0u">Nous sommes enfin prêts à créer des enchâssements ! Jetons un coup d’œil.</p>  <h2 class="relative group"><a id="création-denchâssements-pour-les-textes" 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="#création-denchâssements-pour-les-textes"><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>Création d’enchâssements pour les textes</span></h2> <p data-svelte-h="svelte-t003t9">Nous avons vu dans <a href="/course/fr/chapter2">chapitre 2</a> que nous pouvons obtenir des enchâssements de <em>tokens</em> en utilisant la classe <code>AutoModel</code>. Tout ce que nous avons à faire est de choisir un <em>checkpoint</em> approprié à partir duquel charger le modèle. Heureusement, il existe une bibliothèque appelée <code>sentence-transformers</code> dédiée à la création d’enchâssements. Comme décrit dans la <a href="https://www.sbert.net/examples/applications/semantic-search/README.html#symmetric-vs-asymmetric-semantic-search" rel="nofollow">documentation de la bibliothèque</a>, notre cas d’utilisation est un exemple de <em>recherche sémantique asymétrique</em>. En effet, nous avons une requête courte dont nous aimerions trouver la réponse dans un document plus long, par exemple un commentaire à un problème. Le <a href="https://www.sbert.net/docs/pretrained_models.html#model-overview" rel="nofollow">tableau de présentation des modèles</a> de la documentation indique que le <em>checkpoint</em> <code>multi-qa-mpnet-base-dot-v1</code> a les meilleures performances pour la recherche sémantique. Utilisons donc le pour notre application. Nous allons également charger le <em>tokenizer</em> en utilisant le même <em>checkpoint</em> :</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-1m0h1ny">Pour accélérer le processus, il est utile de placer le modèle et les entrées sur un périphérique GPU, alors faisons-le maintenant :</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-x2hn4x">Comme nous l’avons mentionné précédemment, nous aimerions représenter chaque entrée dans notre corpus de problèmes GitHub comme un vecteur unique. Nous devons donc regrouper ou faire la moyenne de nos enchâssements de <em>tokens</em> d’une manière ou d’une autre. Une approche populaire consiste à effectuer un <em>regroupement CLS</em> sur les sorties de notre modèle, où nous collectons simplement le dernier état caché pour le <em>token</em> spécial <code>[CLS]</code>. La fonction suivante fait ça pour nous :</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-15c9368">Ensuite, nous allons créer une fonction utile qui va tokeniser une liste de documents, placer les tenseurs dans le GPU, les donner au modèle et enfin appliquer le regroupement CLS aux sorties :</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-1wcjyg0">Nous pouvons tester le fonctionnement de la fonction en lui donnant la première entrée textuelle de notre corpus et en inspectant la forme de sortie :</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-fug72j">Super ! Nous avons converti la première entrée de notre corpus en un vecteur à 768 dimensions. Nous pouvons utiliser <code>Dataset.map()</code> pour appliquer notre fonction <code>get_embeddings()</code> à chaque ligne de notre corpus. Créons donc une nouvelle colonne <code>embeddings</code> comme suit :</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-1lvl2ge">Notez que nous avons converti les enchâssements en tableaux NumPy. C’est parce que 🤗 <em>Datasets</em> nécessite ce format lorsque nous essayons de les indexer avec FAISS, ce que nous ferons ensuite.</p>  <h2 class="relative group"><a id="utilisation-de-faiss-pour-une-recherche-de-similarité-efficace" 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="#utilisation-de-faiss-pour-une-recherche-de-similarité-efficace"><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>Utilisation de FAISS pour une recherche de similarité efficace</span></h2> <p data-svelte-h="svelte-z2qr3l">Maintenant que nous avons un jeu de données d’incorporations, nous avons besoin d’un moyen de les rechercher. Pour ce faire, nous utiliserons une structure de données spéciale dans 🤗 <em>Datasets</em> appelée <em>FAISS index</em>. <a href="https://faiss.ai/" rel="nofollow">FAISS</a> (abréviation de <em>Facebook AI Similarity Search</em>) est une bibliothèque qui fournit des algorithmes efficaces pour rechercher et regrouper rapidement des vecteurs d’intégration.</p> <p data-svelte-h="svelte-nuxq3m">L’idée de base derrière FAISS est de créer une structure de données spéciale appelée un <em>index</em> qui permet de trouver quels plongements sont similaires à un plongement d’entrée. Créer un index FAISS dans 🤗 <em>Datasets</em> est simple — nous utilisons la fonction <code>Dataset.add_faiss_index()</code> et spécifions quelle colonne de notre jeu de données nous aimerions indexer :</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-155710n">Nous pouvons maintenant effectuer des requêtes sur cet index en effectuant une recherche des voisins les plus proches avec la fonction <code>Dataset.get_nearest_examples()</code>. Testons cela en enchâssant d’abord une question comme suit :</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-ydla52">Tout comme avec les documents, nous avons maintenant un vecteur de 768 dimensions représentant la requête. Nous pouvons le comparer à l’ensemble du corpus pour trouver les enchâssements les plus similaires :</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-n43esl">La fonction <code>Dataset.get_nearest_examples()</code> renvoie un <em>tuple</em> de scores qui classent le chevauchement entre la requête et le document, et un jeu correspondant d’échantillons (ici, les 5 meilleures correspondances). Collectons-les dans un <code>pandas.DataFrame</code> afin de pouvoir les trier facilement :</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-yc6jv5">Nous pouvons maintenant parcourir les premières lignes pour voir dans quelle mesure notre requête correspond aux commentaires disponibles :</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-1o63v1">Pas mal ! Notre deuxième résultat semble correspondre à la requête.</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-1ad6xd7">✏️ <strong>Essayez !</strong> Créez votre propre requête et voyez si vous pouvez trouver une réponse dans les documents récupérés. Vous devrez peut-être augmenter le paramètre <code>k</code> dans <code>Dataset.get_nearest_examples()</code> pour élargir la recherche.</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/fr/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_snjr9y = {
						assets: "/docs/course/pr_1021/fr",
						base: "/docs/course/pr_1021/fr",
						env: {}
					};

					const element = document.currentScript.parentElement;

					const data = [null,null];

					Promise.all([
						import("/docs/course/pr_1021/fr/_app/immutable/entry/start.e9b4e796.js"),
						import("/docs/course/pr_1021/fr/_app/immutable/entry/app.bf6a4ef3.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 39],
							data,
							form: null,
							error: null
						});
					});
				}
			</script>