Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Tokenizerii rapizi în pipeline-ul de QA","local":"fast-tokenizers-in-the-qa-pipeline","sections":[{"title":"Folosind question-answering pipeline","local":"using-the-question-answering-pipeline","sections":[],"depth":2},{"title":"Folosind un model pentru răspunderea la întrebări","local":"using-a-model-for-question-answering","sections":[],"depth":2},{"title":"Gestionarea contextelor lungi","local":"handling-long-contexts","sections":[],"depth":2}],"depth":1}"> | |
| <link href="/docs/course/pr_1069/rum/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/entry/start.1de7c3d2.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/scheduler.37c15a92.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/singletons.e13b7dfd.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/index.18351ede.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/paths.e130b7b0.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/entry/app.1f82014c.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/index.2bf4358c.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/nodes/0.3c83e1ab.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/nodes/69.99de7488.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/Tip.363c041f.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/Youtube.1e50a667.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/CodeBlock.4e987730.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/CourseFloatingBanner.6add7356.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/FrameworkSwitchCourse.8d4d4ab6.js"> | |
| <link rel="modulepreload" href="/docs/course/pr_1069/rum/_app/immutable/chunks/getInferenceSnippets.24b50994.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Tokenizerii rapizi în pipeline-ul de QA","local":"fast-tokenizers-in-the-qa-pipeline","sections":[{"title":"Folosind question-answering pipeline","local":"using-the-question-answering-pipeline","sections":[],"depth":2},{"title":"Folosind un model pentru răspunderea la întrebări","local":"using-a-model-for-question-answering","sections":[],"depth":2},{"title":"Gestionarea contextelor lungi","local":"handling-long-contexts","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <div class="bg-white leading-none border border-gray-100 rounded-lg flex p-0.5 w-56 text-sm mb-4"><a class="flex justify-center flex-1 py-1.5 px-2.5 focus:outline-none !no-underline rounded-l bg-red-50 dark:bg-transparent text-red-600" href="?fw=pt"><svg class="mr-1.5" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><defs><clipPath id="a"><rect x="3.05" y="0.5" width="25.73" height="31" fill="none"></rect></clipPath></defs><g clip-path="url(#a)"><path d="M24.94,9.51a12.81,12.81,0,0,1,0,18.16,12.68,12.68,0,0,1-18,0,12.81,12.81,0,0,1,0-18.16l9-9V5l-.84.83-6,6a9.58,9.58,0,1,0,13.55,0ZM20.44,9a1.68,1.68,0,1,1,1.67-1.67A1.68,1.68,0,0,1,20.44,9Z" fill="#ee4c2c"></path></g></svg> Pytorch </a><a class="flex justify-center flex-1 py-1.5 px-2.5 focus:outline-none !no-underline rounded-r text-gray-500 filter grayscale" href="?fw=tf"><svg class="mr-1.5" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="0.94em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 274"><path d="M145.726 42.065v42.07l72.861 42.07v-42.07l-72.86-42.07zM0 84.135v42.07l36.43 21.03V105.17L0 84.135zm109.291 21.035l-36.43 21.034v126.2l36.43 21.035v-84.135l36.435 21.035v-42.07l-36.435-21.034V105.17z" fill="#E55B2D"></path><path d="M145.726 42.065L36.43 105.17v42.065l72.861-42.065v42.065l36.435-21.03v-84.14zM255.022 63.1l-36.435 21.035v42.07l36.435-21.035V63.1zm-72.865 84.135l-36.43 21.035v42.07l36.43-21.036v-42.07zm-36.43 63.104l-36.436-21.035v84.135l36.435-21.035V210.34z" fill="#ED8E24"></path><path d="M145.726 0L0 84.135l36.43 21.035l109.296-63.105l72.861 42.07L255.022 63.1L145.726 0zm0 126.204l-36.435 21.03l36.435 21.036l36.43-21.035l-36.43-21.03z" fill="#F8BF3C"></path></svg> TensorFlow </a></div> <h1 class="relative group"><a id="fast-tokenizers-in-the-qa-pipeline" 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="#fast-tokenizers-in-the-qa-pipeline"><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>Tokenizerii rapizi în pipeline-ul de QA</span></h1> <div class="flex space-x-1 absolute z-10 right-0 top-0"><a href="https://discuss.huggingface.co/t/chapter-6-questions" target="_blank"><img alt="Ask a Question" class="!m-0" src="https://img.shields.io/badge/Ask%20a%20question-ffcb4c.svg?logo=data:image/svg+xml;base64,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"></a> <a href="https://colab.research.google.com/github/huggingface/notebooks/blob/master/course/en/chapter6/section3b_pt.ipynb" target="_blank"><img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"></a> <a href="https://studiolab.sagemaker.aws/import/github/huggingface/notebooks/blob/master/course/en/chapter6/section3b_pt.ipynb" target="_blank"><img alt="Open In Studio Lab" class="!m-0" src="https://studiolab.sagemaker.aws/studiolab.svg"></a></div> <p data-svelte-h="svelte-gs77xf">Acum ne vom aprofunda în pipelineul <code>question-answering</code> și să vedem cum putem valorifica offesturile pentru a primi răspunzuri la întrebări la îndemână din context, asemănător cum am făcut cu entitățile grupate în secțiunea precedentă. Pe urmă vom vedea cum vom face față contextelor foarte lungi care ajung truncate. Puteți trece peste această secțiune dacă nu sunteți interesat în întrebarea care răspunde la sarcina aceasta.</p> <iframe class="w-full xl:w-4/6 h-80" src="https://www.youtube-nocookie.com/embed/_wxyB3j3mk4" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe> <h2 class="relative group"><a id="using-the-question-answering-pipeline" 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="#using-the-question-answering-pipeline"><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>Folosind question-answering pipeline</span></h2> <p data-svelte-h="svelte-1drz3d6">Cum am văzut în <a href="/course/chapter1">Capitolul 1</a>, noi putem folosi pipelineul <code>question-answering</code> ca acesta pentru a răspunde la o întrebare:</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> pipeline | |
| question_answerer = pipeline(<span class="hljs-string">"question-answering"</span>) | |
| context = <span class="hljs-string">""" | |
| 🤗 Transformers is backed by the three most popular deep learning libraries — Jax, PyTorch, and TensorFlow — with a seamless integration | |
| between them. It's straightforward to train your models with one before loading them for inference with the other. | |
| """</span> | |
| question = <span class="hljs-string">"Which deep learning libraries back 🤗 Transformers?"</span> | |
| question_answerer(question=question, context=context)<!-- 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">'score'</span>: <span class="hljs-number">0.97773</span>, | |
| <span class="hljs-string">'start'</span>: <span class="hljs-number">78</span>, | |
| <span class="hljs-string">'end'</span>: <span class="hljs-number">105</span>, | |
| <span class="hljs-string">'answer'</span>: <span class="hljs-string">'Jax, PyTorch and TensorFlow'</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1jof88z">Spre deosebire de alte pipelineuri, care nu put trunca și face split la text care este mai lung decât lungimea maxim acceptată de model(și, prin urmare, pot pierde informații la sfârșitul unui document), accest pipeline poate face față contextelor foarte lungi și va returna răspunsul la întrebare chiar dacă aceasta se află la sfârșit:</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 -->long_context = <span class="hljs-string">""" | |
| 🤗 Transformers: State of the Art NLP | |
| 🤗 Transformers provides thousands of pretrained models to perform tasks on texts such as classification, information extraction, | |
| question answering, summarization, translation, text generation and more in over 100 languages. | |
| Its aim is to make cutting-edge NLP easier to use for everyone. | |
| 🤗 Transformers provides APIs to quickly download and use those pretrained models on a given text, fine-tune them on your own datasets and | |
| then share them with the community on our model hub. At the same time, each python module defining an architecture is fully standalone and | |
| can be modified to enable quick research experiments. | |
| Why should I use transformers? | |
| 1. Easy-to-use state-of-the-art models: | |
| - High performance on NLU and NLG tasks. | |
| - Low barrier to entry for educators and practitioners. | |
| - Few user-facing abstractions with just three classes to learn. | |
| - A unified API for using all our pretrained models. | |
| - Lower compute costs, smaller carbon footprint: | |
| 2. Researchers can share trained models instead of always retraining. | |
| - Practitioners can reduce compute time and production costs. | |
| - Dozens of architectures with over 10,000 pretrained models, some in more than 100 languages. | |
| 3. Choose the right framework for every part of a model's lifetime: | |
| - Train state-of-the-art models in 3 lines of code. | |
| - Move a single model between TF2.0/PyTorch frameworks at will. | |
| - Seamlessly pick the right framework for training, evaluation and production. | |
| 4. Easily customize a model or an example to your needs: | |
| - We provide examples for each architecture to reproduce the results published by its original authors. | |
| - Model internals are exposed as consistently as possible. | |
| - Model files can be used independently of the library for quick experiments. | |
| 🤗 Transformers is backed by the three most popular deep learning libraries — Jax, PyTorch and TensorFlow — with a seamless integration | |
| between them. It's straightforward to train your models with one before loading them for inference with the other. | |
| """</span> | |
| question_answerer(question=question, context=long_context)<!-- 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">'score'</span>: <span class="hljs-number">0.97149</span>, | |
| <span class="hljs-string">'start'</span>: <span class="hljs-number">1892</span>, | |
| <span class="hljs-string">'end'</span>: <span class="hljs-number">1919</span>, | |
| <span class="hljs-string">'answer'</span>: <span class="hljs-string">'Jax, PyTorch and TensorFlow'</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-41pc3s">Hai să vedem cum el face toate astea!</p> <h2 class="relative group"><a id="using-a-model-for-question-answering" 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="#using-a-model-for-question-answering"><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>Folosind un model pentru răspunderea la întrebări</span></h2> <p data-svelte-h="svelte-hx0mps">Ca în cazul oricărui altui pipeline, începem prin tokenizarea datelor de intrare și apoi le trimitem prin model. Checkpointul utilizat în mod implicit pentru pipelineul <code>question-answering</code> este <a href="https://huggingface.co/distilbert-base-cased-distilled-squad" rel="nofollow"><code>distilbert-base-cased-distilled-squad</code></a> (“squad” din nume provine de la datasetul pe care modelul a fost ajustat; vom vorbi mai multe despre datasetul SQuAD în <a href="/course/chapter7/7">Capitolul 7</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> transformers <span class="hljs-keyword">import</span> AutoTokenizer, AutoModelForQuestionAnswering | |
| model_checkpoint = <span class="hljs-string">"distilbert-base-cased-distilled-squad"</span> | |
| tokenizer = AutoTokenizer.from_pretrained(model_checkpoint) | |
| model = AutoModelForQuestionAnswering.from_pretrained(model_checkpoint) | |
| inputs = tokenizer(question, context, return_tensors=<span class="hljs-string">"pt"</span>) | |
| outputs = model(**inputs)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1o0rxfe">Observați că noi tokenizăm întrebrea și contextul ca o perecehe, cu întrebarea prima.</p> <div class="flex justify-center" data-svelte-h="svelte-47wedv"><img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter6/question_tokens.svg" alt="An example of tokenization of question and context"> <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface-course/documentation-images/resolve/main/en/chapter6/question_tokens-dark.svg" alt="An example of tokenization of question and context"></div> <p data-svelte-h="svelte-1k6k3aj">Modelele create pentru răspunderea la întrebări funcționează puțin diferit de modelele pe care le-am văzut până acum. Folosind imaginea de mai sus ca exemplu, modelul a fost antrenat pentru a prezice indicele tokenului cu care începe răspunsului (aici 21) și indicele simbolului la care se termină răspunsul (aici 24). Acesta este motivul pentru care modelele respective nu returnează un singur tensor de logits, ci două: unul pentru logits-ul corespunzători tokenului cu care începe răspunsului și unul pentru logits-ul corespunzător tokenului de sfârșit al răspunsului. Deoarece în acest caz avem un singur input care conține 66 de token-uri, obținem:</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 -->start_logits = outputs.start_logits | |
| end_logits = outputs.end_logits | |
| <span class="hljs-built_in">print</span>(start_logits.shape, end_logits.shape)<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->torch.Size([<span class="hljs-number">1</span>, <span class="hljs-number">66</span>]) torch.Size([<span class="hljs-number">1</span>, <span class="hljs-number">66</span>])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1eehlns">Pentru a converti acești logits în probabilități, vom aplica o funcție softmax - dar înainte de aceasta, trebuie să ne asigurăm că mascăm indicii care nu fac parte din context. Inputul nostru este <code>[CLS] întrebare [SEP] context [SEP]</code>, deci trebuie să mascăm token-urile întrebării, precum și tokenul <code>[SEP]</code>. Cu toate acestea, vom păstra simbolul <code>[CLS]</code>, deoarece unele modele îl folosesc pentru a indica faptul că răspunsul nu se află în context.</p> <p data-svelte-h="svelte-doo8ql">Deoarece vom aplica ulterior un softmax, trebuie doar să înlocuim logiturile pe care dorim să le mascăm cu un număr negativ mare. Aici, folosim <code>-10000</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">import</span> torch | |
| sequence_ids = inputs.sequence_ids() | |
| <span class="hljs-comment"># Mask everything apart from the tokens of the context</span> | |
| mask = [i != <span class="hljs-number">1</span> <span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> sequence_ids] | |
| <span class="hljs-comment"># Unmask the [CLS] token</span> | |
| mask[<span class="hljs-number">0</span>] = <span class="hljs-literal">False</span> | |
| mask = torch.tensor(mask)[<span class="hljs-literal">None</span>] | |
| start_logits[mask] = -<span class="hljs-number">10000</span> | |
| end_logits[mask] = -<span class="hljs-number">10000</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1p1eafu">Acum că am mascat în mod corespunzător logiturile corespunzătoare pozițiilor pe care nu dorim să le prezicem, putem aplica softmax:</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 -->start_probabilities = torch.nn.functional.softmax(start_logits, dim=-<span class="hljs-number">1</span>)[<span class="hljs-number">0</span>] | |
| end_probabilities = torch.nn.functional.softmax(end_logits, dim=-<span class="hljs-number">1</span>)[<span class="hljs-number">0</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-pz25ke">La acest stadiu, am putea lua argmax al probabilităților de început și de sfârșit - dar am putea ajunge la un indice de început care este mai mare decât indicele de sfârșit, deci trebuie să luăm câteva precauții suplimentare. Vom calcula probabilitățile fiecărui <code>start_index</code> și <code>end_index</code> posibil în cazul în care <code>start_index <= end_index</code>, apoi vom lua un tuple <code>(start_index, end_index)</code> cu cea mai mare probabilitate.</p> <p>Presupunând că evenimentele “The answer starts at <code data-svelte-h="svelte-8eqb3b">start_index</code>” și “The answer ends at <code data-svelte-h="svelte-9cistc">end_index</code>” sunt independente, probabilitatea ca răspunsul să înceapă la <code data-svelte-h="svelte-8eqb3b">start_index</code> și să se termine la <code data-svelte-h="svelte-9cistc">end_index</code> este: | |
| <!-- HTML_TAG_START --><span class="katex-display"><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML" display="block"><semantics><mrow><mrow><mi mathvariant="normal">s</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">p</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">b</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">b</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">l</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">s</mi></mrow><mo stretchy="false">[</mo><mrow><mi mathvariant="normal">s</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">d</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">x</mi></mrow><mo stretchy="false">]</mo><mo>×</mo><mrow><mi mathvariant="normal">e</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">d</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">p</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">b</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">b</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">l</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">s</mi></mrow><mo stretchy="false">[</mo><mrow><mi mathvariant="normal">e</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">d</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">d</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">x</mi></mrow><mo stretchy="false">]</mo></mrow><annotation encoding="application/x-tex">\mathrm{start\_probabilities}[\mathrm{start\_index}] \times \mathrm{end\_probabilities}[\mathrm{end\_index}]</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:1.06em;vertical-align:-0.31em;"></span><span class="mord"><span class="mord mathrm">start_probabilities</span></span><span class="mopen">[</span><span class="mord"><span class="mord mathrm">start_index</span></span><span class="mclose">]</span><span class="mspace" style="margin-right:0.2222em;"></span><span class="mbin">×</span><span class="mspace" style="margin-right:0.2222em;"></span></span><span class="base"><span class="strut" style="height:1.06em;vertical-align:-0.31em;"></span><span class="mord"><span class="mord mathrm">end_probabilities</span></span><span class="mopen">[</span><span class="mord"><span class="mord mathrm">end_index</span></span><span class="mclose">]</span></span></span></span></span><!-- HTML_TAG_END --></p> <p>Deci, pentru a calcula toate scorurile, trebuie doar să calculăm toate produsele<!-- HTML_TAG_START --><span class="katex"><span class="katex-mathml"><math xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mrow><mi mathvariant="normal">s</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">p</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">b</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">b</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">l</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">s</mi></mrow><mo stretchy="false">[</mo><mrow><mi mathvariant="normal">s</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">d</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">x</mi></mrow><mo stretchy="false">]</mo><mo>×</mo><mrow><mi mathvariant="normal">e</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">d</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">p</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">b</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">b</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">l</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">s</mi></mrow><mo stretchy="false">[</mo><mrow><mi mathvariant="normal">e</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">d</mi><mi mathvariant="normal">_</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">n</mi><mi mathvariant="normal">d</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">x</mi></mrow><mo stretchy="false">]</mo></mrow><annotation encoding="application/x-tex">\mathrm{start\_probabilities}[\mathrm{start\_index}] \times \mathrm{end\_probabilities}[\mathrm{end\_index}]</annotation></semantics></math></span><span class="katex-html" aria-hidden="true"><span class="base"><span class="strut" style="height:1.06em;vertical-align:-0.31em;"></span><span class="mord"><span class="mord mathrm">start_probabilities</span></span><span class="mopen">[</span><span class="mord"><span class="mord mathrm">start_index</span></span><span class="mclose">]</span><span class="mspace" style="margin-right:0.2222em;"></span><span class="mbin">×</span><span class="mspace" style="margin-right:0.2222em;"></span></span><span class="base"><span class="strut" style="height:1.06em;vertical-align:-0.31em;"></span><span class="mord"><span class="mord mathrm">end_probabilities</span></span><span class="mopen">[</span><span class="mord"><span class="mord mathrm">end_index</span></span><span class="mclose">]</span></span></span></span><!-- HTML_TAG_END --> unde <code data-svelte-h="svelte-1kp3szf">start_index <= end_index</code>.</p> <p data-svelte-h="svelte-i54l9f">Mai întâi hai să calculăm toate produsele posibile:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->scores = start_probabilities[:, <span class="hljs-literal">None</span>] * end_probabilities[<span class="hljs-literal">None</span>, :]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1wrvb02">Apoi vom masca valorile în care <code>start_index > end_index</code> prin stabilirea lor la <code>0</code> (celelalte probabilități sunt toate numere pozitive). Funcția <code>torch.triu()</code> returnează partea triunghiulară superioară a tensorului 2D trecut ca argument, deci va face această mascare pentru noi:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->scores = torch.triu(score)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1lao0yt">Acum trebuie doar să obținem indicele maximului. Deoarece PyTorch va returna indicele în tensorul aplatizat, trebuie să folosim operațiile floor division <code>//</code> și modulusul <code>%</code> pentru a obține <code>start_index</code> și <code>end_index</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 -->max_index = scores.argmax().item() | |
| start_index = max_index // scores.shape[<span class="hljs-number">1</span>] | |
| end_index = max_index % scores.shape[<span class="hljs-number">1</span>] | |
| <span class="hljs-built_in">print</span>(scores[start_index, end_index])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-128j8ii">Nu am terminat încă, dar cel puțin avem deja scorul corect pentru răspuns (puteți verifica acest lucru comparându-l cu primul rezultat din secțiunea anterioară):</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-number">0.97773</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-1cmcxlu">✏️ <strong>Încercați!</strong> Calculați indicii de început și de sfârșit pentru cele mai probabile cinci răspunsuri.</p></div> <p data-svelte-h="svelte-2ksf6i">Avem <code>start_index</code> și <code>end_index</code> ale răspunsului în termeni de tokens, deci acum trebuie doar să convertim în character indices în context. Acesta este momentul în care offseturile vor fi foarte utile. Putem să le luăm și să le folosim așa cum am făcut în sarcina de clasificare a tokenurilor:</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_with_offsets = tokenizer(question, context, return_offsets_mapping=<span class="hljs-literal">True</span>) | |
| offsets = inputs_with_offsets[<span class="hljs-string">"offset_mapping"</span>] | |
| start_char, _ = offsets[start_index] | |
| _, end_char = offsets[end_index] | |
| answer = context[start_char:end_char]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1jq8cgv">Acum trebuie doar să formatăm totul pentru a obține rezultatul nostru:</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 -->result = { | |
| <span class="hljs-string">"answer"</span>: answer, | |
| <span class="hljs-string">"start"</span>: start_char, | |
| <span class="hljs-string">"end"</span>: end_char, | |
| <span class="hljs-string">"score"</span>: scores[start_index, end_index], | |
| } | |
| <span class="hljs-built_in">print</span>(result)<!-- 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">'answer'</span>: <span class="hljs-string">'Jax, PyTorch and TensorFlow'</span>, | |
| <span class="hljs-string">'start'</span>: <span class="hljs-number">78</span>, | |
| <span class="hljs-string">'end'</span>: <span class="hljs-number">105</span>, | |
| <span class="hljs-string">'score'</span>: <span class="hljs-number">0.97773</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-sb5jce">Grozav! Este la fel ca în primul nostru exemplu!</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-1tm427z">✏️ <strong>Încercați!</strong> Utilizați cele mai bune scoruri pe care le-ați calculat anterior pentru a afișa cele mai probabile cinci răspunsuri. Pentru a vă verifica rezultatele, întoarceți-vă la primul pipeline și introduceți <code>top_k=5</code> atunci când îl apelați.</p></div> <h2 class="relative group"><a id="handling-long-contexts" 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="#handling-long-contexts"><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>Gestionarea contextelor lungi</span></h2> <p data-svelte-h="svelte-1b2qe2s">Dacă încercăm să tokenizăm întrebarea și contextul lung pe care le-am folosit ca un exemplu anterior, vom obține un număr de tokenuri mai mare decât lungimea maximă utilizată în pipelineul <code>question-answering</code> (care este 384):</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(question, long_context) | |
| <span class="hljs-built_in">print</span>(<span class="hljs-built_in">len</span>(inputs[<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">461</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1a66x7q">Prin urmare, va trebui să trunchiem inputurile la lungimea maximă. Există mai multe modalități prin care putem face acest lucru, dar nu dorim să trunchiem întrebarea, ci doar contextul. Deoarece contextul este a doua propoziție, vom utiliza strategia de trunchiere <code>"only_second"</code>. Problema care apare atunci este că răspunsul la întrebare poate să nu fie în contextul trunchiat. Aici, de exemplu, am ales o întrebare la care răspunsul se află spre sfârșitul contextului, iar atunci când îl trunchiem, răspunsul nu este prezent:</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(question, long_context, max_length=<span class="hljs-number">384</span>, truncation=<span class="hljs-string">"only_second"</span>) | |
| <span class="hljs-built_in">print</span>(tokenizer.decode(inputs[<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-string">""" | |
| [CLS] Which deep learning libraries back [UNK] Transformers? [SEP] [UNK] Transformers : State of the Art NLP | |
| [UNK] Transformers provides thousands of pretrained models to perform tasks on texts such as classification, information extraction, | |
| question answering, summarization, translation, text generation and more in over 100 languages. | |
| Its aim is to make cutting-edge NLP easier to use for everyone. | |
| [UNK] Transformers provides APIs to quickly download and use those pretrained models on a given text, fine-tune them on your own datasets and | |
| then share them with the community on our model hub. At the same time, each python module defining an architecture is fully standalone and | |
| can be modified to enable quick research experiments. | |
| Why should I use transformers? | |
| 1. Easy-to-use state-of-the-art models: | |
| - High performance on NLU and NLG tasks. | |
| - Low barrier to entry for educators and practitioners. | |
| - Few user-facing abstractions with just three classes to learn. | |
| - A unified API for using all our pretrained models. | |
| - Lower compute costs, smaller carbon footprint: | |
| 2. Researchers can share trained models instead of always retraining. | |
| - Practitioners can reduce compute time and production costs. | |
| - Dozens of architectures with over 10,000 pretrained models, some in more than 100 languages. | |
| 3. Choose the right framework for every part of a model's lifetime: | |
| - Train state-of-the-art models in 3 lines of code. | |
| - Move a single model between TF2.0/PyTorch frameworks at will. | |
| - Seamlessly pick the right framework for training, evaluation and production. | |
| 4. Easily customize a model or an example to your needs: | |
| - We provide examples for each architecture to reproduce the results published by its original authors. | |
| - Model internal [SEP] | |
| """</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-wbvi95">Aceasta înseamnă că modelul va avea dificultăți în a alege răspunsul corect. Pentru a rezolva acest lucru, pipelineul <code>question-answering</code> ne permite să împărțim contextul în bucăți mai mici, specificând lungimea maximă. Pentru a ne asigura că nu împărțim contextul exact în locul nepotrivit pentru a face posibilă găsirea răspunsului, aceasta include și o anumită suprapunere între bucăți.</p> <p data-svelte-h="svelte-15swwgh">Putem cere tokenizerului (rapid sau lent) să facă acest lucru pentru noi adăugând <code>return_overflowing_tokens=True</code>, și putem specifica suprapunerea dorită cu argumentul <code>stride</code>. Iată un exemplu, folosind o propoziție mai mică:</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 -->sentence = <span class="hljs-string">"This sentence is not too long but we are going to split it anyway."</span> | |
| inputs = tokenizer( | |
| sentence, truncation=<span class="hljs-literal">True</span>, return_overflowing_tokens=<span class="hljs-literal">True</span>, max_length=<span class="hljs-number">6</span>, stride=<span class="hljs-number">2</span> | |
| ) | |
| <span class="hljs-keyword">for</span> ids <span class="hljs-keyword">in</span> inputs[<span class="hljs-string">"input_ids"</span>]: | |
| <span class="hljs-built_in">print</span>(tokenizer.decode(ids))<!-- 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">'[CLS] This sentence is not [SEP]'</span> | |
| <span class="hljs-string">'[CLS] is not too long [SEP]'</span> | |
| <span class="hljs-string">'[CLS] too long but we [SEP]'</span> | |
| <span class="hljs-string">'[CLS] but we are going [SEP]'</span> | |
| <span class="hljs-string">'[CLS] are going to split [SEP]'</span> | |
| <span class="hljs-string">'[CLS] to split it anyway [SEP]'</span> | |
| <span class="hljs-string">'[CLS] it anyway. [SEP]'</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-10ynzq4">După cum putem vedea, propoziția a fost împărțită în bucăți astfel încât fiecare intrare din <code>inputs["input_ids"]</code> să aibă cel mult 6 token-uri (aici ar trebui să adăugăm padding pentru ca ultima intrare să aibă aceeași dimensiune ca celelalte) și există o suprapunere de 2 tokenuri între fiecare intrare.</p> <p data-svelte-h="svelte-14khn33">Să aruncăm o privire mai atentă la rezultatul tokenizării:</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>(inputs.keys())<!-- 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 -->dict_keys([<span class="hljs-string">'input_ids'</span>, <span class="hljs-string">'attention_mask'</span>, <span class="hljs-string">'overflow_to_sample_mapping'</span>])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-mjj8y4">Așa cum era de așteptat, obținem ID-uri de intrare și un attention mask. Ultima cheie, <code>overflow_to_sample_mapping</code>, este o hartă care ne spune cărei propoziții îi corespunde fiecare dintre rezultate - aici avem 7 rezultate care provin toate din (singura) propoziție pe care am transmis-o tokenizerului:</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>(inputs[<span class="hljs-string">"overflow_to_sample_mapping"</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">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>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1l99crl">Acest lucru este mai util atunci când tokenizăm mai multe propoziții împreună. De exemplu, aceasta:</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 -->sentences = [ | |
| <span class="hljs-string">"This sentence is not too long but we are going to split it anyway."</span>, | |
| <span class="hljs-string">"This sentence is shorter but will still get split."</span>, | |
| ] | |
| inputs = tokenizer( | |
| sentences, truncation=<span class="hljs-literal">True</span>, return_overflowing_tokens=<span class="hljs-literal">True</span>, max_length=<span class="hljs-number">6</span>, stride=<span class="hljs-number">2</span> | |
| ) | |
| <span class="hljs-built_in">print</span>(inputs[<span class="hljs-string">"overflow_to_sample_mapping"</span>])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-b6k3qh">gets 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-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>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1l5swnq">ceea ce înseamnă că prima propoziție este împărțită în 7 fragmente ca înainte, iar următoarele 4 fragmente provin din a doua propoziție.</p> <p data-svelte-h="svelte-xddee7">Acum să ne întoarcem la contextul nostru lung. În mod implicit, pipelineul <code>question-answering</code> utilizează o lungime maximă de 384, așa cum am menționat mai devreme, și un stride de 128, care corespund modului în care modelul a fost fine-tuned (puteți ajusta acești parametri prin trecerea argumentelor <code>max_seq_len</code> și <code>stride</code> atunci când apelați pipelineul). Astfel, vom utiliza acești parametri la tokenizare. Vom adăuga, de asemenea, padding (pentru a avea sampleuri de aceeași lungime, astfel încât să putem construi tensori), precum și pentru a solicita offsets:</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( | |
| question, | |
| long_context, | |
| stride=<span class="hljs-number">128</span>, | |
| max_length=<span class="hljs-number">384</span>, | |
| padding=<span class="hljs-string">"longest"</span>, | |
| truncation=<span class="hljs-string">"only_second"</span>, | |
| return_overflowing_tokens=<span class="hljs-literal">True</span>, | |
| return_offsets_mapping=<span class="hljs-literal">True</span>, | |
| )<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-ehoc4e">Aceste “inputuri” vor conține ID-urile de input și attention maskurile așteptate de model, precum și offseturile și “overflow_to_sample_mapping” despre care tocmai am vorbit. Deoarece cei doi nu sunt parametri utilizați de model, îi vom scoate din <code>inputs</code> (și nu vom stoca harta, deoarece nu este utilă aici) înainte de a-l converti într-un tensor:</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.pop(<span class="hljs-string">"overflow_to_sample_mapping"</span>) | |
| offsets = inputs.pop(<span class="hljs-string">"offset_mapping"</span>) | |
| inputs = inputs.convert_to_tensors(<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-built_in">print</span>(inputs[<span class="hljs-string">"input_ids"</span>].shape)<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->torch.Size([<span class="hljs-number">2</span>, <span class="hljs-number">384</span>])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-zlhaau">Contextul nostru lung a fost împărțit în două, ceea ce înseamnă că, după ce trece prin modelul nostru, vom avea două seturi de logits de început și de sfârșit:</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 -->outputs = model(**inputs) | |
| start_logits = outputs.start_logits | |
| end_logits = outputs.end_logits | |
| <span class="hljs-built_in">print</span>(start_logits.shape, end_logits.shape)<!-- HTML_TAG_END --></pre></div> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START -->torch.Size([<span class="hljs-number">2</span>, <span class="hljs-number">384</span>]) torch.Size([<span class="hljs-number">2</span>, <span class="hljs-number">384</span>])<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-w1s80s">Ca și înainte, mai întâi mascăm tokenii care nu fac parte din context înainte de a lua softmax. De asemenea, mascăm toți padding tokens(marcate de attention mask):</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 -->sequence_ids = inputs.sequence_ids() | |
| <span class="hljs-comment"># Mask everything apart from the tokens of the context</span> | |
| mask = [i != <span class="hljs-number">1</span> <span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> sequence_ids] | |
| <span class="hljs-comment"># Unmask the [CLS] token</span> | |
| mask[<span class="hljs-number">0</span>] = <span class="hljs-literal">False</span> | |
| <span class="hljs-comment"># Mask all the [PAD] tokens</span> | |
| mask = torch.logical_or(torch.tensor(mask)[<span class="hljs-literal">None</span>], (inputs[<span class="hljs-string">"attention_mask"</span>] == <span class="hljs-number">0</span>)) | |
| start_logits[mask] = -<span class="hljs-number">10000</span> | |
| end_logits[mask] = -<span class="hljs-number">10000</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-xc3z5h">Then we can use the softmax to convert our logits to probabilities:</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 -->start_probabilities = torch.nn.functional.softmax(start_logits, dim=-<span class="hljs-number">1</span>) | |
| end_probabilities = torch.nn.functional.softmax(end_logits, dim=-<span class="hljs-number">1</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-6l5dky">Următorul pas este similar cu ceea ce am făcut pentru contextul mic, dar îl repetăm pentru fiecare dintre cele două chunkuri. Atribuim un scor tuturor intervalelor posibile de răspuns, apoi luăm intervalul cu cel mai bun scor:</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 -->candidates = [] | |
| <span class="hljs-keyword">for</span> start_probs, end_probs <span class="hljs-keyword">in</span> <span class="hljs-built_in">zip</span>(start_probabilities, end_probabilities): | |
| scores = start_probs[:, <span class="hljs-literal">None</span>] * end_probs[<span class="hljs-literal">None</span>, :] | |
| idx = torch.triu(scores).argmax().item() | |
| start_idx = idx // scores.shape[<span class="hljs-number">1</span>] | |
| end_idx = idx % scores.shape[<span class="hljs-number">1</span>] | |
| score = scores[start_idx, end_idx].item() | |
| candidates.append((start_idx, end_idx, score)) | |
| <span class="hljs-built_in">print</span>(candidates)<!-- 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">0</span>, <span class="hljs-number">18</span>, <span class="hljs-number">0.33867</span>), (<span class="hljs-number">173</span>, <span class="hljs-number">184</span>, <span class="hljs-number">0.97149</span>)]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-wybiq2">Cei doi candidați corespund celor mai bune răspunsuri pe care modelul le-a putut găsi în fiecare parte. Modelul este mult mai încrezător că răspunsul corect se află în a doua parte (ceea ce este un semn bun!). Acum trebuie doar să facem map celor două intervale de tokenuri cu intervalele de caractere din context (trebuie să o punem în corespondență doar pe a doua pentru a avea răspunsul nostru, dar este interesant să vedem ce a ales modelul în prima parte).</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-s7zbv9">✏️ <strong>Încercați!</strong> Adaptați codul de mai sus pentru a returna scorurile și spanurile intervalele pentru cele mai probabile cinci răspunsuri (în total, nu pe chunk).</p></div> <p data-svelte-h="svelte-1e1qzy4"><code>offsets</code>-urile pe care le-am luat mai devreme este de fapt o listă de offsets, cu o listă pentru fiecare chunk de text:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-keyword">for</span> candidate, offset <span class="hljs-keyword">in</span> <span class="hljs-built_in">zip</span>(candidates, offsets): | |
| start_token, end_token, score = candidate | |
| start_char, _ = offset[start_token] | |
| _, end_char = offset[end_token] | |
| answer = long_context[start_char:end_char] | |
| result = {<span class="hljs-string">"answer"</span>: answer, <span class="hljs-string">"start"</span>: start_char, <span class="hljs-string">"end"</span>: end_char, <span class="hljs-string">"score"</span>: score} | |
| <span class="hljs-built_in">print</span>(result)<!-- 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">'answer'</span>: <span class="hljs-string">'\n🤗 Transformers: State of the Art NLP'</span>, <span class="hljs-string">'start'</span>: <span class="hljs-number">0</span>, <span class="hljs-string">'end'</span>: <span class="hljs-number">37</span>, <span class="hljs-string">'score'</span>: <span class="hljs-number">0.33867</span>} | |
| {<span class="hljs-string">'answer'</span>: <span class="hljs-string">'Jax, PyTorch and TensorFlow'</span>, <span class="hljs-string">'start'</span>: <span class="hljs-number">1892</span>, <span class="hljs-string">'end'</span>: <span class="hljs-number">1919</span>, <span class="hljs-string">'score'</span>: <span class="hljs-number">0.97149</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1d49c5t">Dacă ignorăm primul rezultat, obținem același rezultat ca și pipelineul noastru pentru acest context lung - yay!</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-gpytty">✏️ <strong>Încercați!</strong> Utilizați cele mai bune scoruri pe care le-ați calculat înainte pentru a afișa cele mai probabile cinci răspunsuri (pentru întregul context, nu pentru fiecare chunk). Pentru a vă verifica rezultatele, întoarceți-vă la primul pipeline și introduceți <code>top_k=5</code> atunci când îl apelați.</p></div> <p data-svelte-h="svelte-1r5vmkg">Aici se încheie scufundarea noastră în capacitățile tokenizerului. Vom pune toate acestea din nou în practică în capitolul următor, când vă vom arăta cum să ajustați un model pentru o serie de sarcini NLP comune.</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/rum/chapter6/3b.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_1ftlxhy = { | |
| assets: "/docs/course/pr_1069/rum", | |
| base: "/docs/course/pr_1069/rum", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/course/pr_1069/rum/_app/immutable/entry/start.1de7c3d2.js"), | |
| import("/docs/course/pr_1069/rum/_app/immutable/entry/app.1f82014c.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 69], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 99 kB
- Xet hash:
- 1f3b6a069b226df9226daa5720def09ee672202f12bd827cb62d6eb734dfebb0
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.