Buckets:

hf-doc-build
/

doc-dev

Files

xet

hf-doc-build/doc-dev / course /pr_1069 /it /chapter5 /4.html

rtrm

22 days ago

download

raw

73.2 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Big data? Ci pensa 🤗 Datasets!","local":"big-data-ci-pensa--datasets","sections":[{"title":"Cos’è Pile?","local":"cosè-pile","sections":[],"depth":2},{"title":"La magia del memory mapping","local":"la-magia-del-memory-mapping","sections":[],"depth":2},{"title":"Streaming di dataset","local":"streaming-di-dataset","sections":[],"depth":2}],"depth":1}">
	<link href="/docs/course/pr_1069/it/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/entry/start.693d748d.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/scheduler.37c15a92.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/singletons.60b4c7a2.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/index.18351ede.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/paths.43b6516c.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/entry/app.e9cfd099.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/index.2bf4358c.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/nodes/0.bb8a536c.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/each.e59479a4.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/nodes/36.cf08de02.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/Tip.363c041f.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/Youtube.1e50a667.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/CodeBlock.4e987730.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/CourseFloatingBanner.9ff4c771.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/DocNotebookDropdown.efc1fb7c.js">
	<link rel="modulepreload" href="/docs/course/pr_1069/it/_app/immutable/chunks/getInferenceSnippets.24b50994.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Big data? Ci pensa 🤗 Datasets!","local":"big-data-ci-pensa--datasets","sections":[{"title":"Cos’è Pile?","local":"cosè-pile","sections":[],"depth":2},{"title":"La magia del memory mapping","local":"la-magia-del-memory-mapping","sections":[],"depth":2},{"title":"Streaming di dataset","local":"streaming-di-dataset","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="big-data-ci-pensa--datasets" 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="#big-data-ci-pensa--datasets"><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>Big data? Ci pensa 🤗 Datasets!</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/it/chapter5/section4.ipynb" target="_blank"><img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"></a> <a href="https://studiolab.sagemaker.aws/import/github/huggingface/notebooks/blob/master/course/it/chapter5/section4.ipynb" target="_blank"><img alt="Open In Studio Lab" class="!m-0" src="https://studiolab.sagemaker.aws/studiolab.svg"></a></div> <p data-svelte-h="svelte-ife82b">Al giorno d’oggi non è raro trovarsi a lavorare con dataset grandi diversi gigabyte, soprattutto quando si vuole addestrare un transformer come BERT o GPT-2 da zero. In questi casi, persino <em>caricare</em> i dati può essere un’impresa difficile. Ad esempio, il corpus WebText utilizzato per preaddestrare GPT-2 contiente più di 8 milioni di documenti e 40gb di testo — caricare un dataset del genere sulla RAM del tuo portatile gli farebbe venire un colpo!</p> <p data-svelte-h="svelte-1kyw64i">Per fortuna, 🤗 Datasets è stato sviluppato per superare queste limitazioni, e può risolvere i problemi relativi alla gestione della memoria trattando i dataset come file <em>memory-mapped</em>, e quelli relativi ai limiti del disco rigido attraverso lo <em>stream processing</em> delle voci del corpus.</p> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/JwISwTCPPWo" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-36huww">In questa sezione esploreremo queste funzionalità di 🤗 Datasets con un enorme corpus di 825 GB conosciuto come <a href="https://pile.eleuther.ai" rel="nofollow">Pile</a>. Iniziamo!</p> <h2 class="relative group"><a id="cosè-pile" 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="#cosè-pile"><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>Cos’è Pile?</span></h2> <p data-svelte-h="svelte-e1fzvi">The Pile è un corpus testuale creato da <a href="https://www.eleuther.ai" rel="nofollow">EleutherAI</a> per addestrare modelli di linguaggio su grande scala. Include un grande varietà di dataset, a partire da articoli scientifici, repository di codici da GitHub, e testi dal web filtrati. Il corpus di addestramento è disponibili in <a href="https://the-eye.eu/public/AI/pile/" rel="nofollow">frammenti da 14 GB</a>, ed è possibile scaricare diverse delle <a href="https://the-eye.eu/public/AI/pile_preliminary_components/" rel="nofollow">componenti singole</a>. Iniziamo dando uno sguardo al dataset PubMed Abstracts, un corpus di abstract da 15 milioni di pubblicazioni in ambito biomedico da <a href="https://pubmed.ncbi.nlm.nih.gov/" rel="nofollow">PubMed</a>. Il dataset è in <a href="https://jsonlines.org" rel="nofollow">formato JSON Lines</a> ed è stato compressato usando la libreria <code>zstandard</code>, per cui dobbiamo prima installarla:</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 -->!pip install zstandard<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-ueahrw">Ora, possiamo caricare il dataset utilizzando il meotodo per file remoti che abbiamo visto nella <a href="/course/chapter5/2">sezione 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

	<span class="hljs-comment"># Ci vuole qualche minuto per l'esecuzione, quindi preparati un tè o un caffè nell'attesa :)</span>
	data_files = <span class="hljs-string">"https://the-eye.eu/public/AI/pile_preliminary_components/PUBMED_title_abstracts_2019_baseline.jsonl.zst"</span>
	pubmed_dataset = load_dataset(<span class="hljs-string">"json"</span>, data_files=data_files, split=<span class="hljs-string">"train"</span>)
	pubmed_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">'meta'</span>, <span class="hljs-string">'text'</span>],
	num_rows: <span class="hljs-number">15518009</span>
	})<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-ldt8f5">Possiamo vedere che ci sono 15.518.009 righe e 2 colonne nel nostro dataset — un bel po’!</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-ohmup">✎ Di base, 🤗 Datasets decomprimerà i file necessari a caricare un dataset. Se vuoi risparmiare sullo spazio dell’hard disk, puoi passare <code>DownloadConfig(delete_extracted_True)</code> all’argomento <code>download_config</code> di <code>load_dataset()</code>. Per maggiori dettagli leggi la <a href="https://huggingface.co/docs/datasets/package_reference/builder_classes#datasets.DownloadConfig" rel="nofollow">documentazione</a>.</p></div> <p data-svelte-h="svelte-13o6uv9">Ispezioniamo i contenuti del primo esempio:</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 -->pubmed_dataset[<span class="hljs-number">0</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">'meta'</span>: {<span class="hljs-string">'pmid'</span>: <span class="hljs-number">11409574</span>, <span class="hljs-string">'language'</span>: <span class="hljs-string">'eng'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'Epidemiology of hypoxaemia in children with acute lower respiratory infection.\nTo determine the prevalence of hypoxaemia in children aged under 5 years suffering acute lower respiratory infections (ALRI), the risk factors for hypoxaemia in children under 5 years of age with ALRI, and the association of hypoxaemia with an increased risk of dying in children of the same age ...'</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1kk4pfk">Okay, questo sembra proprio l’abstract di un articolo di medicina. Ora vediamo quanta RAM è stata usata per caricare il dataset!</p> <h2 class="relative group"><a id="la-magia-del-memory-mapping" 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="#la-magia-del-memory-mapping"><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>La magia del memory mapping</span></h2> <p data-svelte-h="svelte-ol6i82">Un modo semplice per calcolare l’uso di memoria su Python è utilizzando la libreria <a href="https://psutil.readthedocs.io/en/latest/" rel="nofollow"><code>psutil</code></a>, che può essere installata con <code>pip</code> come segue:</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 -->!pip install psutil<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1yadugw"><code>psutil</code> offre una classe <code>Process</code> che permette di controllare l’utilizzo della memoria del processo attuale come segue::</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> psutil

	<span class="hljs-comment"># Process.memory_info mostra i dati in byte, quindi convertiamo in megabyte</span>
	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"RAM used: <span class="hljs-subst">{psutil.Process().memory_info().rss / (<span class="hljs-number">1024</span> * <span class="hljs-number">1024</span>):<span class="hljs-number">.2</span>f}</span> MB"</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 -->RAM used: <span class="hljs-number">5678.33</span> MB<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-dgnood">L’attributo <code>rss</code> qui fa riferimento alla <em>grandezza del resident set</em>, che equivale alla frazione di memoria che il processo occupa nella RAM. Questo valore include inoltre la memoria utilizzata dall’interprete Python e dalle librerie caricate, per cui l’ammontare effettivo utilizzato per caricare il dataset è un po’ più piccolo. Per fare un confronto, vediamo quant’è grande il dataset su disco utilizzando l’attributo <code>dataset_size</code>. Come prima, il risultato è espresso in byte, e abbiamo bisogno di convertirlo in gigabyte:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-built_in">print</span>(<span class="hljs-string">f"Number of files in dataset : <span class="hljs-subst">{pubmed_dataset.dataset_size}</span>"</span>)
	size_gb = pubmed_dataset.dataset_size / (<span class="hljs-number">1024</span>**<span class="hljs-number">3</span>)
	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"Dataset size (cache file) : <span class="hljs-subst">{size_gb:<span class="hljs-number">.2</span>f}</span> GB"</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 -->Number of files <span class="hljs-keyword">in</span> dataset : <span class="hljs-number">20979437051</span>
	Dataset size (cache file) : <span class="hljs-number">19.54</span> GB<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1p89ulg">Bene — nonostante sia grande quasi 30 GB, siamo in grado di caricare e accedere al dataset utilizzando molta meno RAM!</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-1p4smth">✏️ <strong>Provaci tu!</strong> Scegli uno dei <a href="https://the-eye.eu/public/AI/pile_preliminary_components/" rel="nofollow">subset</a> di Pile che è più grande della RAM del tuo PC o del tuo portatile, caricalo utilizzando 🤗 Datasets e calcola la quantità di RAM utilizzata. Nota che per avere un valore preciso, dovrai creare un nuovo processo. Puoi trovare le grandezze decompresse di ogni subset nella Tavola 1 dell’<a href="https://arxiv.org/abs/2101.00027" rel="nofollow">articolo su Pile</a></p></div> <p data-svelte-h="svelte-uataba">Se hai dimestichezza con Pandas, questo risultato potrebbe sorprenderti, vista la famosa <a href="https://wesmckinney.com/blog/apache-arrow-pandas-internals/" rel="nofollow">regola di Wes Kinney</a>, ovvero che, in linea di massima, serve una RAM 5-10 volte più grande del dataset che vuoi caricare. Come fa 🤗 Datasets a risolvere questo problema di gestione della memoria? 🤗 Datasets tratta ogni dataset come un <a href="https://it.wikipedia.org/wiki/File_mappato_in_memoria" rel="nofollow">file mappato in memoria</a>, il che permette di avere un mapping tra la RAM e l’archiviazione dei file di sistema, che permette alla librera di accedere e operare su elementi del dataset senza doverli caricare completamente in memoria.</p> <p data-svelte-h="svelte-inyq4t">I file mappati in memoria possono inoltre essre condivisi su più processi, il che permette a metodi come <code>Dataset.map()</code> di poter essere eseguiti in parallelo senza bisogno di spostare o copiare il dataset. Dietro le quinte, tutto ciò è realizzato dal formato di memoria <a href="https://arrow.apache.org" rel="nofollow">Apache Arrow</a> e dalla libreria <a href="https://arrow.apache.org/docs/python/index.html" rel="nofollow"><code>pyarrow</code></a>, che rendono più veloci il caricamento e il processamento dei dati. (per maggiori dettagli su Apache Arrow, e per un confronto con Pandas, dai un’occhiata al <a href="https://towardsdatascience.com/apache-arrow-read-dataframe-with-zero-memory-69634092b1a" rel="nofollow">post di Dejan Simic</a>.) Per vederlo in azione, eseguiamo un piccolo test di velocità con un loop su tutti gli elementi nel dataset PubMed Abstracts:</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> timeit

	code_snippet = <span class="hljs-string">"""batch_size = 1000

	for idx in range(0, len(pubmed_dataset), batch_size):
	_ = pubmed_dataset[idx:idx + batch_size]
	"""</span>

	time = timeit.timeit(stmt=code_snippet, number=<span class="hljs-number">1</span>, <span class="hljs-built_in">globals</span>=<span class="hljs-built_in">globals</span>())
	<span class="hljs-built_in">print</span>(
	<span class="hljs-string">f"Iterated over <span class="hljs-subst">{<span class="hljs-built_in">len</span>(pubmed_dataset)}</span> examples (about <span class="hljs-subst">{size_gb:<span class="hljs-number">.1</span>f}</span> GB) in "</span>
	<span class="hljs-string">f"<span class="hljs-subst">{time:<span class="hljs-number">.1</span>f}</span>s, i.e. <span class="hljs-subst">{size_gb/time:<span class="hljs-number">.3</span>f}</span> GB/s"</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">'Iterated over 15518009 examples (about 19.5 GB) in 64.2s, i.e. 0.304 GB/s'</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1fz2u01">Abbiamo usato il modulo di Python <code>timeit</code> per calcolare il tempo di esecuzione impiegato da <code>code_snippet</code>. Tipicamente l’iterazione su un dataset impiega un tempo che va da un decimo di GB al secondo, a diversi GB al secondo. Questo funziona perfettamente per la maggior parte delle applicazioni, ma a volte avrai bisogno di lavorare con un dataset che è troppo grande persino per essere salvato sul tuo portatile. Ad esempio, se cercassimo di scaricare Pile per intero, avremo bisogno di 825 GB di spazio libero su disko! In questi casi, 🤗 Datasets permette di utilizzare processi di streaming che ci permettono di scaricare e accedere al volo ai dati, senza bisogno di scaricare l’intero dataset. Diamo un’occhiata a come funziona.</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-10cqh4y">💡 Nei notebook Jupyter, puoi cronometrare le celle utilizzando la <a href="https://ipython.readthedocs.io/en/stable/interactive/magics.html#magic-timeit" rel="nofollow">funzione magica <code>%%timeit</code></a></p></div> <h2 class="relative group"><a id="streaming-di-dataset" 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="#streaming-di-dataset"><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>Streaming di dataset</span></h2> <p data-svelte-h="svelte-1ejx5xr">Per abilitare lo streaming dei dataset devi semplicemente passare l’argomento <code>streaming=True</code> alla funzione <code>load_dataset()</code>. Ad esempio, carichiamo un’altra volta il dataset PubMed Abstract, ma in modalità streaming:</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 -->pubmed_dataset_streamed = load_dataset(
	<span class="hljs-string">"json"</span>, data_files=data_files, split=<span class="hljs-string">"train"</span>, streaming=<span class="hljs-literal">True</span>
	)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-dqnei4">Invece del solito <code>Dataset</code> che abbiamo incontrato in precedenza in questo capitolo, l’oggetto ritornato con <code>streaming=True' è un </code>IterableDataset<code>. Come suggerito dal nome, per accedere agli elementi di un </code>IterableDataset`, dobbiamo iterare di esso. Possiamo accedere al primo elemento del nostro dataset in streaming come segue:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-built_in">next</span>(<span class="hljs-built_in">iter</span>(pubmed_dataset_streamed))<!-- 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">'meta'</span>: {<span class="hljs-string">'pmid'</span>: <span class="hljs-number">11409574</span>, <span class="hljs-string">'language'</span>: <span class="hljs-string">'eng'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'Epidemiology of hypoxaemia in children with acute lower respiratory infection.\nTo determine the prevalence of hypoxaemia in children aged under 5 years suffering acute lower respiratory infections (ALRI), the risk factors for hypoxaemia in children under 5 years of age with ALRI, and the association of hypoxaemia with an increased risk of dying in children of the same age ...'</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-jp8nni">Gli elementi di un dataset in streaming possono essere processati al volo utilizzando <code>IterableDataset.map()</code>, che è utile durante l’addestramento se hai bisogno di tokenizzare gli input. Il processo è uguale a quello che abbiamo utilizzato per tokenizzare il nostro dataset nel <a href="/course/chapter3">Capitolo 3</a>, con l’unica differenza che ora ritorneremo gli output uno alla volta:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer

	tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"distilbert-base-uncased"</span>)
	tokenized_dataset = pubmed_dataset_streamed.<span class="hljs-built_in">map</span>(<span class="hljs-keyword">lambda</span> x: tokenizer(x[<span class="hljs-string">"text"</span>]))
	<span class="hljs-built_in">next</span>(<span class="hljs-built_in">iter</span>(tokenized_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 -->{<span class="hljs-string">'input_ids'</span>: [<span class="hljs-number">101</span>, <span class="hljs-number">4958</span>, <span class="hljs-number">5178</span>, <span class="hljs-number">4328</span>, <span class="hljs-number">6779</span>, ...], <span class="hljs-string">'attention_mask'</span>: [<span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, ...]}<!-- HTML_TAG_END --></pre></div> <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-shytyc">💡 Per velocizzare la tokenizzazione con lo streaming puoi passare <code>batchet=True</code>, come abbiamo visto nell’ultima sezione. Questo processerà gli esempi per batch. Di default, la grandezza di un batch è 1.000, e può essere specificata attraverso l’argomento <code>batch_size</code>.</p></div> <p data-svelte-h="svelte-18o8mco">È anche possibile mescolare un dataset in streaming utilizzato <code>Iterabledataset.shuffle()</code>, ma a differenza di <code>Dataset.shuffle()</code>, questo metodo mescola solo gli elementi in un <code>buffer_size</code> predefinito:</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 -->shuffled_dataset = pubmed_dataset_streamed.shuffle(buffer_size=<span class="hljs-number">10_000</span>, seed=<span class="hljs-number">42</span>)
	<span class="hljs-built_in">next</span>(<span class="hljs-built_in">iter</span>(shuffled_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 -->{<span class="hljs-string">'meta'</span>: {<span class="hljs-string">'pmid'</span>: <span class="hljs-number">11410799</span>, <span class="hljs-string">'language'</span>: <span class="hljs-string">'eng'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'Randomized study of dose or schedule modification of granulocyte colony-stimulating factor in platinum-based chemotherapy for elderly patients with lung cancer ...'</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-dm7z72">In questo esempio, abbiamo selezionato un esempio casuale dai primi 10.000 esempi nel buffer. Una volta che accediamo a un esempio, il suo posto nel buffer è subito occupato dall’esempio successivo nel corpus (in questo caso l’esempio 10.0001). Puoi inoltre selezionare gli elementi da un dataset in streaming utilizzando le funzioni <code>IterableDataset.take()</code> a <code>IterableDataset.skip()</code>, che funzionano un po’ come <code>Dataset.select()</code>. Ad esempio, per selezionare i primi 5 esempi nel dataset PubMed Abstract dovremmo fare come segue:</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 -->dataset_head = pubmed_dataset_streamed.take(<span class="hljs-number">5</span>)
	<span class="hljs-built_in">list</span>(dataset_head)<!-- 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">'meta'</span>: {<span class="hljs-string">'pmid'</span>: <span class="hljs-number">11409574</span>, <span class="hljs-string">'language'</span>: <span class="hljs-string">'eng'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'Epidemiology of hypoxaemia in children with acute lower respiratory infection ...'</span>},
	{<span class="hljs-string">'meta'</span>: {<span class="hljs-string">'pmid'</span>: <span class="hljs-number">11409575</span>, <span class="hljs-string">'language'</span>: <span class="hljs-string">'eng'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'Clinical signs of hypoxaemia in children with acute lower respiratory infection: indicators of oxygen therapy ...'</span>},
	{<span class="hljs-string">'meta'</span>: {<span class="hljs-string">'pmid'</span>: <span class="hljs-number">11409576</span>, <span class="hljs-string">'language'</span>: <span class="hljs-string">'eng'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">"Hypoxaemia in children with severe pneumonia in Papua New Guinea ..."</span>},
	{<span class="hljs-string">'meta'</span>: {<span class="hljs-string">'pmid'</span>: <span class="hljs-number">11409577</span>, <span class="hljs-string">'language'</span>: <span class="hljs-string">'eng'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'Oxygen concentrators and cylinders ...'</span>},
	{<span class="hljs-string">'meta'</span>: {<span class="hljs-string">'pmid'</span>: <span class="hljs-number">11409578</span>, <span class="hljs-string">'language'</span>: <span class="hljs-string">'eng'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'Oxygen supply in rural africa: a personal experience ...'</span>}]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1lfebd3">Allo stesso modo, è possibile utilizzare la funzione <code>IterableDataset.skip()</code> per creare sezioni di addestramento e di validazione da un dataset mescolato, come segue:</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-comment"># Salta i primi 1.000 esempi, il resto viene incluso nell'insieme di addestramento</span>
	train_dataset = shuffled_dataset.skip(<span class="hljs-number">1000</span>)
	<span class="hljs-comment"># Includi i primi 1.000 esempi nell'insieme di validazione</span>
	validation_dataset = shuffled_dataset.take(<span class="hljs-number">1000</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-vvs954">Concludiamo la nostra ricognizione dello streaming di dataset con un’applicazione comune: la combinazione di più dataset per creare un unico corpus. 🤗 Datasets fornisce una funzione <code>interleave_datasets()</code>, che converte una lista di oggetti <code>IterableDataset</code> in un unico <code>IterableDataset</code>, dove gli elementi del nuovo dataset sono ottenuti alternando tra gli esempi forniti. Questa funzione è particolarmente utile quando cerchiamo di combinare dataset di grandi dimensioni, come esempio possiamo utilizzare in streaming la sezione FreeLaw del Pile, un dataset di 51 GB di pareri legali dai tribunali statunitensi:</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 -->law_dataset_streamed = load_dataset(
	<span class="hljs-string">"json"</span>,
	data_files=<span class="hljs-string">"https://the-eye.eu/public/AI/pile_preliminary_components/FreeLaw_Opinions.jsonl.zst"</span>,
	split=<span class="hljs-string">"train"</span>,
	streaming=<span class="hljs-literal">True</span>,
	)
	<span class="hljs-built_in">next</span>(<span class="hljs-built_in">iter</span>(law_dataset_streamed))<!-- 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">'meta'</span>: {<span class="hljs-string">'case_ID'</span>: <span class="hljs-string">'110921.json'</span>,
	<span class="hljs-string">'case_jurisdiction'</span>: <span class="hljs-string">'scotus.tar.gz'</span>,
	<span class="hljs-string">'date_created'</span>: <span class="hljs-string">'2010-04-28T17:12:49Z'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'\n461 U.S. 238 (1983)\nOLIM ET AL.\nv.\nWAKINEKONA\nNo. 81-1581.\nSupreme Court of United States.\nArgued January 19, 1983.\nDecided April 26, 1983.\nCERTIORARI TO THE UNITED STATES COURT OF APPEALS FOR THE NINTH CIRCUIT\n*239 Michael A. Lilly, First Deputy Attorney General of Hawaii, argued the cause for petitioners. With him on the brief was James H. Dannenberg, Deputy Attorney General...'</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-30gh6j">Questo dataset è abbastanza grande da mettere sotto sforzo la RAM di molto portatili, ma siamo riusciti a caricarlo e accedervi senza alcun problema! Ora cominiamo gli esempi di FreeLaw e di PubMed Abstracts con la funzione <code>interleave_datasets()</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">from</span> itertools <span class="hljs-keyword">import</span> islice
	<span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> interleave_datasets

	combined_dataset = interleave_datasets([pubmed_dataset_streamed, law_dataset_streamed])
	<span class="hljs-built_in">list</span>(islice(combined_dataset, <span class="hljs-number">2</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">'meta'</span>: {<span class="hljs-string">'pmid'</span>: <span class="hljs-number">11409574</span>, <span class="hljs-string">'language'</span>: <span class="hljs-string">'eng'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'Epidemiology of hypoxaemia in children with acute lower respiratory infection ...'</span>},
	{<span class="hljs-string">'meta'</span>: {<span class="hljs-string">'case_ID'</span>: <span class="hljs-string">'110921.json'</span>,
	<span class="hljs-string">'case_jurisdiction'</span>: <span class="hljs-string">'scotus.tar.gz'</span>,
	<span class="hljs-string">'date_created'</span>: <span class="hljs-string">'2010-04-28T17:12:49Z'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'\n461 U.S. 238 (1983)\nOLIM ET AL.\nv.\nWAKINEKONA\nNo. 81-1581.\nSupreme Court of United States.\nArgued January 19, 1983.\nDecided April 26, 1983.\nCERTIORARI TO THE UNITED STATES COURT OF APPEALS FOR THE NINTH CIRCUIT\n*239 Michael A. Lilly, First Deputy Attorney General of Hawaii, argued the cause for petitioners. With him on the brief was James H. Dannenberg, Deputy Attorney General...'</span>}]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-qtvrif">Abbiamo utilizzato la funzione <code>islice()</code> del modulo Python <code>itertools</code> per selezionare i primi due esempi dai dataset combinati, e abbiamo visto che corrispondono ai primi esempi di ognuno dei due dataset originali.</p> <p data-svelte-h="svelte-13odihg">Infine, se vuoi processare il Pile in streaming, in tutti i suoi 825 GB, puoi recuperare tutti i file preparati, come segue:</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 -->base_url = <span class="hljs-string">"https://the-eye.eu/public/AI/pile/"</span>
	data_files = {
	<span class="hljs-string">"train"</span>: [base_url + <span class="hljs-string">"train/"</span> + <span class="hljs-string">f"<span class="hljs-subst">{idx:02d}</span>.jsonl.zst"</span> <span class="hljs-keyword">for</span> idx <span class="hljs-keyword">in</span> <span class="hljs-built_in">range</span>(<span class="hljs-number">30</span>)],
	<span class="hljs-string">"validation"</span>: base_url + <span class="hljs-string">"val.jsonl.zst"</span>,
	<span class="hljs-string">"test"</span>: base_url + <span class="hljs-string">"test.jsonl.zst"</span>,
	}
	pile_dataset = load_dataset(<span class="hljs-string">"json"</span>, data_files=data_files, streaming=<span class="hljs-literal">True</span>)
	<span class="hljs-built_in">next</span>(<span class="hljs-built_in">iter</span>(pile_dataset[<span class="hljs-string">"train"</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">'meta'</span>: {<span class="hljs-string">'pile_set_name'</span>: <span class="hljs-string">'Pile-CC'</span>},
	<span class="hljs-string">'text'</span>: <span class="hljs-string">'It is done, and submitted. You can play “Survival of the Tastiest” on Android, and on the web...'</span>}<!-- HTML_TAG_END --></pre></div> <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-1ir16xe">✏️ <strong>Prova tu!</strong> Usa uno dei corpora Common Crawl come <a href="https://huggingface.co/datasets/mc4" rel="nofollow"><code>mc4</code></a> oppure <a href="https://huggingface.co/datasets/oscar" rel="nofollow"><code>oscar</code></a> per crare un dataset multilingue in streaming, che rappresenta le proporzioni delle lingue parlate in un paese a tua scelta. Ad esempio, le quattro lingue ufficiali in Svizzera sono il tedesco, il francesce, l’italiano e il romancio, per cui potresti creare un corpus della Svizzera raccogliendo i campioni da Oscar, secondo la percentuale di parlanti di ognuna.</p></div> <p data-svelte-h="svelte-3idnlg">Ora hai a tua disposizione tutti gli strumenti per caricare e processare dataset di ogni tipo — ma a meno che tu non sia estremamente fortunato, arriverà un momento nel tuo cammino in cui dovrai effettivamente creare un dataset per risolvere i tuoi problemi. Questo sarà argomento della prossima sezione!</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/course/blob/main/chapters/it/chapter5/4.mdx" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span data-svelte-h="svelte-x0xyl0">></span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p>

	<script>
	{
	__sveltekit_u8ez91 = {
	assets: "/docs/course/pr_1069/it",
	base: "/docs/course/pr_1069/it",
	env: {}
	};

	const element = document.currentScript.parentElement;

	const data = [null,null];

	Promise.all([
	import("/docs/course/pr_1069/it/_app/immutable/entry/start.693d748d.js"),
	import("/docs/course/pr_1069/it/_app/immutable/entry/app.e9cfd099.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 36],
	data,
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 73.2 kB
Xet hash:: 7220027cbd2721b205bf63bd580df004182b691500417ceaa539323c2df37b68

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