Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Processing the data","local":"processing-the-data","sections":[{"title":"Loading a dataset from the Hub","local":"loading-a-dataset-from-the-hub","sections":[],"depth":3},{"title":"Preprocessing a dataset","local":"preprocessing-a-dataset","sections":[{"title":"Dynamic padding","local":"dynamic-padding","sections":[],"depth":5}],"depth":3},{"title":"Section Quiz","local":"section-quiz","sections":[{"title":"1. What is the main advantage of using Dataset.map() with batched=True ?","local":"1-what-is-the-main-advantage-of-using-datasetmap-with-batchedtrue-","sections":[],"depth":3},{"title":"2. Why do we use dynamic padding instead of padding all sequences to the maximum length in the dataset?","local":"2-why-do-we-use-dynamic-padding-instead-of-padding-all-sequences-to-the-maximum-length-in-the-dataset","sections":[],"depth":3},{"title":"3. What does the token_type_ids field represent in BERT tokenization?","local":"3-what-does-the-tokentypeids-field-represent-in-bert-tokenization","sections":[],"depth":3},{"title":"4. When loading a dataset with load_dataset('glue', 'mrpc') , what does the second argument specify?","local":"4-when-loading-a-dataset-with-loaddatasetglue-mrpc--what-does-the-second-argument-specify","sections":[],"depth":3},{"title":"5. What is the purpose of removing columns like ‘sentence1’ and ‘sentence2’ before training?","local":"5-what-is-the-purpose-of-removing-columns-like-sentence1-and-sentence2-before-training","sections":[],"depth":3}],"depth":2}],"depth":1}"> | |
| <link href="/docs/course/pr_1069/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/entry/start.c5306bb2.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/scheduler.37c15a92.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/singletons.bc78d867.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/index.18351ede.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/paths.76894643.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/entry/app.4264f5f8.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/index.7cb9c9b8.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/nodes/0.f5347c47.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/nodes/46.ab9d227e.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/Tip.d10b3fc9.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/Youtube.8666c400.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/CodeBlock.abae2786.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/CourseFloatingBanner.df82c153.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/Question.7e41e492.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/stores.cb4752a8.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/en/_app/immutable/chunks/getInferenceSnippets.f9350a3f.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Processing the data","local":"processing-the-data","sections":[{"title":"Loading a dataset from the Hub","local":"loading-a-dataset-from-the-hub","sections":[],"depth":3},{"title":"Preprocessing a dataset","local":"preprocessing-a-dataset","sections":[{"title":"Dynamic padding","local":"dynamic-padding","sections":[],"depth":5}],"depth":3},{"title":"Section Quiz","local":"section-quiz","sections":[{"title":"1. What is the main advantage of using Dataset.map() with batched=True ?","local":"1-what-is-the-main-advantage-of-using-datasetmap-with-batchedtrue-","sections":[],"depth":3},{"title":"2. Why do we use dynamic padding instead of padding all sequences to the maximum length in the dataset?","local":"2-why-do-we-use-dynamic-padding-instead-of-padding-all-sequences-to-the-maximum-length-in-the-dataset","sections":[],"depth":3},{"title":"3. What does the token_type_ids field represent in BERT tokenization?","local":"3-what-does-the-tokentypeids-field-represent-in-bert-tokenization","sections":[],"depth":3},{"title":"4. When loading a dataset with load_dataset('glue', 'mrpc') , what does the second argument specify?","local":"4-when-loading-a-dataset-with-loaddatasetglue-mrpc--what-does-the-second-argument-specify","sections":[],"depth":3},{"title":"5. What is the purpose of removing columns like ‘sentence1’ and ‘sentence2’ before training?","local":"5-what-is-the-purpose-of-removing-columns-like-sentence1-and-sentence2-before-training","sections":[],"depth":3}],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="processing-the-data" 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="#processing-the-data"><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>Processing the data</span></h1> <div class="flex space-x-1 absolute z-10 right-0 top-0"><a href="https://discuss.huggingface.co/t/chapter-3-questions" target="_blank"><img alt="Ask a Question" class="!m-0" src="https://img.shields.io/badge/Ask%20a%20question-ffcb4c.svg?logo=data:image/svg+xml;base64,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"></a> <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/master/course/en/chapter3/section2.ipynb" target="_blank"><img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"></a> <a href="https://studiolab.sagemaker.aws/import/github/huggingface/notebooks/blob/master/course/en/chapter3/section2.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-1dnua88">Continuing with the example from the <a href="/course/chapter2">previous chapter</a>, here is how we would train a sequence classifier on one batch:</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 | |
| <span class="hljs-keyword">from</span> torch.optim <span class="hljs-keyword">import</span> AdamW | |
| <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, AutoModelForSequenceClassification | |
| <span class="hljs-comment"># Same as before</span> | |
| checkpoint = <span class="hljs-string">"bert-base-uncased"</span> | |
| tokenizer = AutoTokenizer.from_pretrained(checkpoint) | |
| model = AutoModelForSequenceClassification.from_pretrained(checkpoint) | |
| sequences = [ | |
| <span class="hljs-string">"I've been waiting for a HuggingFace course my whole life."</span>, | |
| <span class="hljs-string">"This course is amazing!"</span>, | |
| ] | |
| batch = tokenizer(sequences, padding=<span class="hljs-literal">True</span>, truncation=<span class="hljs-literal">True</span>, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-comment"># This is new</span> | |
| batch[<span class="hljs-string">"labels"</span>] = torch.tensor([<span class="hljs-number">1</span>, <span class="hljs-number">1</span>]) | |
| optimizer = AdamW(model.parameters()) | |
| loss = model(**batch).loss | |
| loss.backward() | |
| optimizer.step()<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-k42f7q">Of course, just training the model on two sentences is not going to yield very good results. To get better results, you will need to prepare a bigger dataset.</p> <p data-svelte-h="svelte-zjffq0">In this section we will use as an example the MRPC (Microsoft Research Paraphrase Corpus) dataset, introduced in a <a href="https://www.aclweb.org/anthology/I05-5002.pdf" rel="nofollow">paper</a> by William B. Dolan and Chris Brockett. The dataset consists of 5,801 pairs of sentences, with a label indicating if they are paraphrases or not (i.e., if both sentences mean the same thing). We’ve selected it for this chapter because it’s a small dataset, so it’s easy to experiment with training on it.</p> <h3 class="relative group"><a id="loading-a-dataset-from-the-hub" 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="#loading-a-dataset-from-the-hub"><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>Loading a dataset from the Hub</span></h3> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/_BZearw7f0w" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-1l7s9ct">The Hub doesn’t just contain models; it also has multiple datasets in lots of different languages. You can browse the datasets <a href="https://huggingface.co/datasets" rel="nofollow">here</a>, and we recommend you try to load and process a new dataset once you have gone through this section (see the general documentation <a href="https://huggingface.co/docs/datasets/loading" rel="nofollow">here</a>). But for now, let’s focus on the MRPC dataset! This is one of the 10 datasets composing the <a href="https://gluebenchmark.com/" rel="nofollow">GLUE benchmark</a>, which is an academic benchmark that is used to measure the performance of ML models across 10 different text classification tasks.</p> <p data-svelte-h="svelte-a0uedu">The 🤗 Datasets library provides a very simple command to download and cache a dataset on the Hub. We can download the MRPC dataset like this:</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-wn0wjo">💡 <strong>Additional Resources</strong>: For more dataset loading techniques and examples, check out the <a href="https://huggingface.co/docs/datasets/" rel="nofollow">🤗 Datasets documentation</a>.</p></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-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset | |
| raw_datasets = load_dataset(<span class="hljs-string">"glue"</span>, <span class="hljs-string">"mrpc"</span>) | |
| raw_datasets<!-- 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 -->DatasetDict({ | |
| train: Dataset({ | |
| features: [<span class="hljs-string">'sentence1'</span>, <span class="hljs-string">'sentence2'</span>, <span class="hljs-string">'label'</span>, <span class="hljs-string">'idx'</span>], | |
| num_rows: <span class="hljs-number">3668</span> | |
| }) | |
| validation: Dataset({ | |
| features: [<span class="hljs-string">'sentence1'</span>, <span class="hljs-string">'sentence2'</span>, <span class="hljs-string">'label'</span>, <span class="hljs-string">'idx'</span>], | |
| num_rows: <span class="hljs-number">408</span> | |
| }) | |
| test: Dataset({ | |
| features: [<span class="hljs-string">'sentence1'</span>, <span class="hljs-string">'sentence2'</span>, <span class="hljs-string">'label'</span>, <span class="hljs-string">'idx'</span>], | |
| num_rows: <span class="hljs-number">1725</span> | |
| }) | |
| })<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1c0bj8y">As you can see, we get a <code>DatasetDict</code> object which contains the training set, the validation set, and the test set. Each of those contains several columns (<code>sentence1</code>, <code>sentence2</code>, <code>label</code>, and <code>idx</code>) and a variable number of rows, which are the number of elements in each set (so, there are 3,668 pairs of sentences in the training set, 408 in the validation set, and 1,725 in the test set).</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-124lsw3">This command downloads and caches the dataset, by default in <em>~/.cache/huggingface/datasets</em>. Recall from Chapter 2 that you can customize your cache folder by setting the <code>HF_HOME</code> environment variable.</p></div> <p data-svelte-h="svelte-65xqw4">We can access each pair of sentences in our <code>raw_datasets</code> object by indexing, like with a dictionary:</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 -->raw_train_dataset = raw_datasets[<span class="hljs-string">"train"</span>] | |
| raw_train_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">'idx'</span>: <span class="hljs-number">0</span>, | |
| <span class="hljs-string">'label'</span>: <span class="hljs-number">1</span>, | |
| <span class="hljs-string">'sentence1'</span>: <span class="hljs-string">'Amrozi accused his brother , whom he called " the witness " , of deliberately distorting his evidence .'</span>, | |
| <span class="hljs-string">'sentence2'</span>: <span class="hljs-string">'Referring to him as only " the witness " , Amrozi accused his brother of deliberately distorting his evidence .'</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1o36uf9">We can see the labels are already integers, so we won’t have to do any preprocessing there. To know which integer corresponds to which label, we can inspect the <code>features</code> of our <code>raw_train_dataset</code>. This will tell us the type of each column:</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 -->raw_train_dataset.features<!-- 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">'sentence1'</span>: Value(dtype=<span class="hljs-string">'string'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), | |
| <span class="hljs-string">'sentence2'</span>: Value(dtype=<span class="hljs-string">'string'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), | |
| <span class="hljs-string">'label'</span>: ClassLabel(num_classes=<span class="hljs-number">2</span>, names=[<span class="hljs-string">'not_equivalent'</span>, <span class="hljs-string">'equivalent'</span>], names_file=<span class="hljs-literal">None</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), | |
| <span class="hljs-string">'idx'</span>: Value(dtype=<span class="hljs-string">'int32'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>)}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1c3slq1">Behind the scenes, <code>label</code> is of type <code>ClassLabel</code>, and the mapping of integers to label name is stored in the <em>names</em> folder. <code>0</code> corresponds to <code>not_equivalent</code>, and <code>1</code> corresponds to <code>equivalent</code>.</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-vjl0ht">✏️ <strong>Try it out!</strong> Look at element 15 of the training set and element 87 of the validation set. What are their labels?</p></div> <h3 class="relative group"><a id="preprocessing-a-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="#preprocessing-a-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>Preprocessing a dataset</span></h3> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/0u3ioSwev3s" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-cvke6e">To preprocess the dataset, we need to convert the text to numbers the model can make sense of. As you saw in the <a href="/course/chapter2">previous chapter</a>, this is done with a tokenizer. We can feed the tokenizer one sentence or a list of sentences, so we can directly tokenize all the first sentences and all the second sentences of each pair like this:</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 | |
| checkpoint = <span class="hljs-string">"bert-base-uncased"</span> | |
| tokenizer = AutoTokenizer.from_pretrained(checkpoint) | |
| tokenized_sentences_1 = tokenizer(raw_datasets[<span class="hljs-string">"train"</span>][<span class="hljs-string">"sentence1"</span>]) | |
| tokenized_sentences_2 = tokenizer(raw_datasets[<span class="hljs-string">"train"</span>][<span class="hljs-string">"sentence2"</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-19b7iu4">💡 <strong>Deep Dive</strong>: For more advanced tokenization techniques and understanding how different tokenizers work, explore the <a href="https://huggingface.co/docs/transformers/main/en/tokenizer_summary" rel="nofollow">🤗 Tokenizers documentation</a> and the <a href="https://huggingface.co/learn/cookbook/en/advanced_rag#tokenization-strategies" rel="nofollow">tokenization guide in the cookbook</a>.</p></div> <p data-svelte-h="svelte-17gxq0h">However, we can’t just pass two sequences to the model and get a prediction of whether the two sentences are paraphrases or not. We need to handle the two sequences as a pair, and apply the appropriate preprocessing. Fortunately, the tokenizer can also take a pair of sequences and prepare it the way our BERT model expects:</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 -->inputs = tokenizer(<span class="hljs-string">"This is the first sentence."</span>, <span class="hljs-string">"This is the second one."</span>) | |
| inputs<!-- 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">2023</span>, <span class="hljs-number">2003</span>, <span class="hljs-number">1996</span>, <span class="hljs-number">2034</span>, <span class="hljs-number">6251</span>, <span class="hljs-number">1012</span>, <span class="hljs-number">102</span>, <span class="hljs-number">2023</span>, <span class="hljs-number">2003</span>, <span class="hljs-number">1996</span>, <span class="hljs-number">2117</span>, <span class="hljs-number">2028</span>, <span class="hljs-number">1012</span>, <span class="hljs-number">102</span>], | |
| <span class="hljs-string">'token_type_ids'</span>: [<span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>], | |
| <span class="hljs-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>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>] | |
| }<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-s9c1sv">We discussed the <code>input_ids</code> and <code>attention_mask</code> keys in <a href="/course/chapter2">Chapter 2</a>, but we put off talking about <code>token_type_ids</code>. In this example, this is what tells the model which part of the input is the first sentence and which is the second sentence.</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-1j17f76">✏️ <strong>Try it out!</strong> Take element 15 of the training set and tokenize the two sentences separately and as a pair. What’s the difference between the two results?</p></div> <p data-svelte-h="svelte-17ajvt1">If we decode the IDs inside <code>input_ids</code> back to words:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->tokenizer.convert_ids_to_tokens(inputs[<span class="hljs-string">"input_ids"</span>])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1oivwa">we will get:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->[<span class="hljs-string">'[CLS]'</span>, <span class="hljs-string">'this'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'first'</span>, <span class="hljs-string">'sentence'</span>, <span class="hljs-string">'.'</span>, <span class="hljs-string">'[SEP]'</span>, <span class="hljs-string">'this'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'second'</span>, <span class="hljs-string">'one'</span>, <span class="hljs-string">'.'</span>, <span class="hljs-string">'[SEP]'</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1tvepx7">So we see the model expects the inputs to be of the form <code>[CLS] sentence1 [SEP] sentence2 [SEP]</code> when there are two sentences. Aligning this with the <code>token_type_ids</code> gives us:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->[<span class="hljs-string">'[CLS]'</span>, <span class="hljs-string">'this'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'first'</span>, <span class="hljs-string">'sentence'</span>, <span class="hljs-string">'.'</span>, <span class="hljs-string">'[SEP]'</span>, <span class="hljs-string">'this'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'second'</span>, <span class="hljs-string">'one'</span>, <span class="hljs-string">'.'</span>, <span class="hljs-string">'[SEP]'</span>] | |
| [ <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>, <span class="hljs-number">1</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-10hqnfn">As you can see, the parts of the input corresponding to <code>[CLS] sentence1 [SEP]</code> all have a token type ID of <code>0</code>, while the other parts, corresponding to <code>sentence2 [SEP]</code>, all have a token type ID of <code>1</code>.</p> <p data-svelte-h="svelte-hspdt8">Note that if you select a different checkpoint, you won’t necessarily have the <code>token_type_ids</code> in your tokenized inputs (for instance, they’re not returned if you use a DistilBERT model). They are only returned when the model will know what to do with them, because it has seen them during its pretraining.</p> <p data-svelte-h="svelte-11u1x7y">Here, BERT is pretrained with token type IDs, and on top of the masked language modeling objective we talked about in <a href="/course/chapter1">Chapter 1</a>, it has an additional objective called <em>next sentence prediction</em>. The goal with this task is to model the relationship between pairs of sentences.</p> <p data-svelte-h="svelte-1e7se1j">With next sentence prediction, the model is provided pairs of sentences (with randomly masked tokens) and asked to predict whether the second sentence follows the first. To make the task non-trivial, half of the time the sentences follow each other in the original document they were extracted from, and the other half of the time the two sentences come from two different documents.</p> <p data-svelte-h="svelte-1ujm1f5">In general, you don’t need to worry about whether or not there are <code>token_type_ids</code> in your tokenized inputs: as long as you use the same checkpoint for the tokenizer and the model, everything will be fine as the tokenizer knows what to provide to its model.</p> <p data-svelte-h="svelte-57wndc">Now that we have seen how our tokenizer can deal with one pair of sentences, we can use it to tokenize our whole dataset: like in the <a href="/course/chapter2">previous chapter</a>, we can feed the tokenizer a list of pairs of sentences by giving it the list of first sentences, then the list of second sentences. This is also compatible with the padding and truncation options we saw in <a href="/course/chapter2">Chapter 2</a>. So, one way to preprocess the training dataset is:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->tokenized_dataset = tokenizer( | |
| raw_datasets[<span class="hljs-string">"train"</span>][<span class="hljs-string">"sentence1"</span>], | |
| raw_datasets[<span class="hljs-string">"train"</span>][<span class="hljs-string">"sentence2"</span>], | |
| padding=<span class="hljs-literal">True</span>, | |
| truncation=<span class="hljs-literal">True</span>, | |
| )<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-12m6ruz">This works well, but it has the disadvantage of returning a dictionary (with our keys, <code>input_ids</code>, <code>attention_mask</code>, and <code>token_type_ids</code>, and values that are lists of lists). It will also only work if you have enough RAM to store your whole dataset during the tokenization (whereas the datasets from the 🤗 Datasets library are <a href="https://arrow.apache.org/" rel="nofollow">Apache Arrow</a> files stored on the disk, so you only keep the samples you ask for loaded in memory).</p> <p data-svelte-h="svelte-ykgx4y">To keep the data as a dataset, we will use the <a href="https://huggingface.co/docs/datasets/package_reference/main_classes#datasets.Dataset.map" rel="nofollow"><code>Dataset.map()</code></a> method. This also allows us some extra flexibility, if we need more preprocessing done than just tokenization. The <code>map()</code> method works by applying a function on each element of the dataset, so let’s define a function that tokenizes our inputs:</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_">tokenize_function</span>(<span class="hljs-params">example</span>): | |
| <span class="hljs-keyword">return</span> tokenizer(example[<span class="hljs-string">"sentence1"</span>], example[<span class="hljs-string">"sentence2"</span>], truncation=<span class="hljs-literal">True</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1jfj8zf">This function takes a dictionary (like the items of our dataset) and returns a new dictionary with the keys <code>input_ids</code>, <code>attention_mask</code>, and <code>token_type_ids</code>. Note that it also works if the <code>example</code> dictionary contains several samples (each key as a list of sentences) since the <code>tokenizer</code> works on lists of pairs of sentences, as seen before. This will allow us to use the option <code>batched=True</code> in our call to <code>map()</code>, which will greatly speed up the tokenization. The <code>tokenizer</code> is backed by a tokenizer written in Rust from the <a href="https://github.com/huggingface/tokenizers" rel="nofollow">🤗 Tokenizers</a> library. This tokenizer can be very fast, but only if we give it lots of inputs at once.</p> <p data-svelte-h="svelte-115iydu">Note that we’ve left the <code>padding</code> argument out in our tokenization function for now. This is because padding all the samples to the maximum length is not efficient: it’s better to pad the samples when we’re building a batch, as then we only need to pad to the maximum length in that batch, and not the maximum length in the entire dataset. This can save a lot of time and processing power when the inputs have very variable lengths!</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-zltst">📚 <strong>Performance Tips</strong>: Learn more about efficient data processing techniques in the <a href="https://huggingface.co/docs/datasets/about_arrow" rel="nofollow">🤗 Datasets performance guide</a>.</p></div> <p data-svelte-h="svelte-oo95cw">Here is how we apply the tokenization function on all our datasets at once. We’re using <code>batched=True</code> in our call to <code>map</code> so the function is applied to multiple elements of our dataset at once, and not on each element separately. This allows for faster preprocessing.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->tokenized_datasets = raw_datasets.<span class="hljs-built_in">map</span>(tokenize_function, batched=<span class="hljs-literal">True</span>) | |
| tokenized_datasets<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-12x3mv1">The way the 🤗 Datasets library applies this processing is by adding new fields to the datasets, one for each key in the dictionary returned by the preprocessing function:</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 -->DatasetDict({ | |
| train: Dataset({ | |
| features: [<span class="hljs-string">'attention_mask'</span>, <span class="hljs-string">'idx'</span>, <span class="hljs-string">'input_ids'</span>, <span class="hljs-string">'label'</span>, <span class="hljs-string">'sentence1'</span>, <span class="hljs-string">'sentence2'</span>, <span class="hljs-string">'token_type_ids'</span>], | |
| num_rows: <span class="hljs-number">3668</span> | |
| }) | |
| validation: Dataset({ | |
| features: [<span class="hljs-string">'attention_mask'</span>, <span class="hljs-string">'idx'</span>, <span class="hljs-string">'input_ids'</span>, <span class="hljs-string">'label'</span>, <span class="hljs-string">'sentence1'</span>, <span class="hljs-string">'sentence2'</span>, <span class="hljs-string">'token_type_ids'</span>], | |
| num_rows: <span class="hljs-number">408</span> | |
| }) | |
| test: Dataset({ | |
| features: [<span class="hljs-string">'attention_mask'</span>, <span class="hljs-string">'idx'</span>, <span class="hljs-string">'input_ids'</span>, <span class="hljs-string">'label'</span>, <span class="hljs-string">'sentence1'</span>, <span class="hljs-string">'sentence2'</span>, <span class="hljs-string">'token_type_ids'</span>], | |
| num_rows: <span class="hljs-number">1725</span> | |
| }) | |
| })<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-b2zdkp">You can even use multiprocessing when applying your preprocessing function with <code>map()</code> by passing along a <code>num_proc</code> argument. We didn’t do this here because the 🤗 Tokenizers library already uses multiple threads to tokenize our samples faster, but if you are not using a fast tokenizer backed by this library, this could speed up your preprocessing.</p> <p data-svelte-h="svelte-k426ib">Our <code>tokenize_function</code> returns a dictionary with the keys <code>input_ids</code>, <code>attention_mask</code>, and <code>token_type_ids</code>, so those three fields are added to all splits of our dataset. Note that we could also have changed existing fields if our preprocessing function returned a new value for an existing key in the dataset to which we applied <code>map()</code>.</p> <p data-svelte-h="svelte-1onxd79">The last thing we will need to do is pad all the examples to the length of the longest element when we batch elements together — a technique we refer to as <em>dynamic padding</em>.</p> <h5 class="relative group"><a id="dynamic-padding" 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="#dynamic-padding"><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>Dynamic padding</span></h5> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/7q5NyFT8REg" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <p data-svelte-h="svelte-1gicira">The function that is responsible for putting together samples inside a batch is called a <em>collate function</em>. It’s an argument you can pass when you build a <code>DataLoader</code>, the default being a function that will just convert your samples to PyTorch tensors and concatenate them (recursively if your elements are lists, tuples, or dictionaries). This won’t be possible in our case since the inputs we have won’t all be of the same size. We have deliberately postponed the padding, to only apply it as necessary on each batch and avoid having over-long inputs with a lot of padding. This will speed up training by quite a bit, but note that if you’re training on a TPU it can cause problems — TPUs prefer fixed shapes, even when that requires extra padding.</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-1szjgfl">🚀 <strong>Optimization Guide</strong>: For more details on optimizing training performance, including padding strategies and TPU considerations, see the <a href="https://huggingface.co/docs/transformers/main/en/performance" rel="nofollow">🤗 Transformers performance documentation</a>.</p></div> <p data-svelte-h="svelte-1b915t7">To do this in practice, we have to define a collate function that will apply the correct amount of padding to the items of the dataset we want to batch together. Fortunately, the 🤗 Transformers library provides us with such a function via <code>DataCollatorWithPadding</code>. It takes a tokenizer when you instantiate it (to know which padding token to use, and whether the model expects padding to be on the left or on the right of the inputs) and will do everything you need:</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> DataCollatorWithPadding | |
| data_collator = DataCollatorWithPadding(tokenizer=tokenizer)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1dw0kzt">To test this new toy, let’s grab a few samples from our training set that we would like to batch together. Here, we remove the columns <code>idx</code>, <code>sentence1</code>, and <code>sentence2</code> as they won’t be needed and contain strings (and we can’t create tensors with strings) and have a look at the lengths of each entry in the batch:</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 -->samples = tokenized_datasets[<span class="hljs-string">"train"</span>][:<span class="hljs-number">8</span>] | |
| samples = {k: v <span class="hljs-keyword">for</span> k, v <span class="hljs-keyword">in</span> samples.items() <span class="hljs-keyword">if</span> k <span class="hljs-keyword">not</span> <span class="hljs-keyword">in</span> [<span class="hljs-string">"idx"</span>, <span class="hljs-string">"sentence1"</span>, <span class="hljs-string">"sentence2"</span>]} | |
| [<span class="hljs-built_in">len</span>(x) <span class="hljs-keyword">for</span> x <span class="hljs-keyword">in</span> samples[<span class="hljs-string">"input_ids"</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-number">50</span>, <span class="hljs-number">59</span>, <span class="hljs-number">47</span>, <span class="hljs-number">67</span>, <span class="hljs-number">59</span>, <span class="hljs-number">50</span>, <span class="hljs-number">62</span>, <span class="hljs-number">32</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1g0r85d">No surprise, we get samples of varying length, from 32 to 67. Dynamic padding means the samples in this batch should all be padded to a length of 67, the maximum length inside the batch. Without dynamic padding, all of the samples would have to be padded to the maximum length in the whole dataset, or the maximum length the model can accept. Let’s double-check that our <code>data_collator</code> is dynamically padding the batch properly:</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 -->batch = data_collator(samples) | |
| {k: v.shape <span class="hljs-keyword">for</span> k, v <span class="hljs-keyword">in</span> batch.items()}<!-- 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">'attention_mask'</span>: torch.Size([<span class="hljs-number">8</span>, <span class="hljs-number">67</span>]), | |
| <span class="hljs-string">'input_ids'</span>: torch.Size([<span class="hljs-number">8</span>, <span class="hljs-number">67</span>]), | |
| <span class="hljs-string">'token_type_ids'</span>: torch.Size([<span class="hljs-number">8</span>, <span class="hljs-number">67</span>]), | |
| <span class="hljs-string">'labels'</span>: torch.Size([<span class="hljs-number">8</span>])}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-3l70up">Looking good! Now that we’ve gone from raw text to batches our model can deal with, we’re ready to fine-tune it!</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-1ydnchh">✏️ <strong>Try it out!</strong> Replicate the preprocessing on the GLUE SST-2 dataset. It’s a little bit different since it’s composed of single sentences instead of pairs, but the rest of what we did should look the same. For a harder challenge, try to write a preprocessing function that works on any of the GLUE tasks.</p> <p data-svelte-h="svelte-1dy141p">📖 <strong>Additional Practice</strong>: Check out these hands-on examples from the <a href="https://huggingface.co/docs/transformers/main/en/notebooks" rel="nofollow">🤗 Transformers examples</a>.</p></div> <p data-svelte-h="svelte-87542j">Perfect! Now that we have preprocessed our data with the latest best practices from the 🤗 Datasets library, we’re ready to move on to training our model using the modern Trainer API. The next section will show you how to fine-tune your model effectively using the latest features and optimizations available in the Hugging Face ecosystem.</p> <h2 class="relative group"><a id="section-quiz" 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="#section-quiz"><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>Section Quiz</span></h2> <p data-svelte-h="svelte-itfd98">Test your understanding of data processing concepts:</p> <h3 class="relative group"><a id="1-what-is-the-main-advantage-of-using-datasetmap-with-batchedtrue-" 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="#1-what-is-the-main-advantage-of-using-datasetmap-with-batchedtrue-"><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>1. What is the main advantage of using Dataset.map() with batched=True ?</span></h3> <div><form><label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="0"> <!-- HTML_TAG_START -->It uses less memory.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="1"> <!-- HTML_TAG_START -->It processes multiple examples at once, making tokenization much faster.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="2"> <!-- HTML_TAG_START -->It automatically handles padding for you.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="3"> <!-- HTML_TAG_START -->It converts the data to PyTorch tensors.<!-- HTML_TAG_END --></label> <div class="flex flex-row items-center mt-3"><button class="btn px-4 mr-4" type="submit" disabled>Submit</button> </div></form></div> <h3 class="relative group"><a id="2-why-do-we-use-dynamic-padding-instead-of-padding-all-sequences-to-the-maximum-length-in-the-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="#2-why-do-we-use-dynamic-padding-instead-of-padding-all-sequences-to-the-maximum-length-in-the-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>2. Why do we use dynamic padding instead of padding all sequences to the maximum length in the dataset?</span></h3> <div><form><label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="0"> <!-- HTML_TAG_START -->Dynamic padding is required by the model architecture.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="1"> <!-- HTML_TAG_START -->It reduces computational overhead by only padding to the maximum length in each batch.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="2"> <!-- HTML_TAG_START -->It improves model accuracy.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="3"> <!-- HTML_TAG_START -->It's required when using the DataCollatorWithPadding.<!-- HTML_TAG_END --></label> <div class="flex flex-row items-center mt-3"><button class="btn px-4 mr-4" type="submit" disabled>Submit</button> </div></form></div> <h3 class="relative group"><a id="3-what-does-the-tokentypeids-field-represent-in-bert-tokenization" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#3-what-does-the-tokentypeids-field-represent-in-bert-tokenization"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>3. What does the token_type_ids field represent in BERT tokenization?</span></h3> <div><form><label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="0"> <!-- HTML_TAG_START -->The position of each token in the sequence.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="1"> <!-- HTML_TAG_START -->Which sentence each token belongs to when processing sentence pairs.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="2"> <!-- HTML_TAG_START -->The attention mask for each token.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="3"> <!-- HTML_TAG_START -->The vocabulary ID of each token.<!-- HTML_TAG_END --></label> <div class="flex flex-row items-center mt-3"><button class="btn px-4 mr-4" type="submit" disabled>Submit</button> </div></form></div> <h3 class="relative group"><a id="4-when-loading-a-dataset-with-loaddatasetglue-mrpc--what-does-the-second-argument-specify" 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="#4-when-loading-a-dataset-with-loaddatasetglue-mrpc--what-does-the-second-argument-specify"><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>4. When loading a dataset with load_dataset('glue', 'mrpc') , what does the second argument specify?</span></h3> <div><form><label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="0"> <!-- HTML_TAG_START -->The version of the dataset to load.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="1"> <!-- HTML_TAG_START -->The specific task or subset within the GLUE benchmark.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="2"> <!-- HTML_TAG_START -->The split of the dataset (train/validation/test).<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="3"> <!-- HTML_TAG_START -->The format to return the data in.<!-- HTML_TAG_END --></label> <div class="flex flex-row items-center mt-3"><button class="btn px-4 mr-4" type="submit" disabled>Submit</button> </div></form></div> <h3 class="relative group"><a id="5-what-is-the-purpose-of-removing-columns-like-sentence1-and-sentence2-before-training" 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="#5-what-is-the-purpose-of-removing-columns-like-sentence1-and-sentence2-before-training"><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>5. What is the purpose of removing columns like ‘sentence1’ and ‘sentence2’ before training?</span></h3> <div><form><label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="0"> <!-- HTML_TAG_START -->To save memory during training.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="1"> <!-- HTML_TAG_START -->The model doesn't expect these raw text columns and would throw an error.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="2"> <!-- HTML_TAG_START -->These columns are not needed for evaluation.<!-- HTML_TAG_END --></label> <label class="block"><input autocomplete="off" class="form-input -mt-1.5 mr-2" name="choice" type="checkbox" value="3"> <!-- HTML_TAG_START -->It improves training speed significantly.<!-- HTML_TAG_END --></label> <div class="flex flex-row items-center mt-3"><button class="btn px-4 mr-4" type="submit" disabled>Submit</button> </div></form></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-l314m0">💡 <strong>Key Takeaways:</strong></p> <ul data-svelte-h="svelte-bdhi0x"><li>Use <code>batched=True</code> with <code>Dataset.map()</code> for significantly faster preprocessing</li> <li>Dynamic padding with <code>DataCollatorWithPadding</code> is more efficient than fixed-length padding</li> <li>Always preprocess your data to match what your model expects (numerical tensors, correct column names)</li> <li>The 🤗 Datasets library provides powerful tools for efficient data processing at scale</li></ul></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/en/chapter3/2.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_1y0degu = { | |
| assets: "/docs/course/pr_1069/en", | |
| base: "/docs/course/pr_1069/en", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/course/pr_1069/en/_app/immutable/entry/start.c5306bb2.js"), | |
| import("/docs/course/pr_1069/en/_app/immutable/entry/app.4264f5f8.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 46], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 86.4 kB
- Xet hash:
- e26e3b2c26bb20f0968f8b96b6703515ed328e0b207d8976f13e797f98d0761c
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.