Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Modelli multilingue per l’inferenza","local":"modelli-multilingue-per-linferenza","sections":[{"title":"XLM","local":"xlm","sections":[{"title":"XLM con language embeddings","local":"xlm-con-language-embeddings","sections":[],"depth":3},{"title":"XLM senza language embeddings","local":"xlm-senza-language-embeddings","sections":[],"depth":3}],"depth":2},{"title":"BERT","local":"bert","sections":[],"depth":2},{"title":"XLM-RoBERTa","local":"xlm-roberta","sections":[],"depth":2},{"title":"M2M100","local":"m2m100","sections":[],"depth":2},{"title":"MBart","local":"mbart","sections":[],"depth":2}],"depth":1}"> | |
| <link href="/docs/transformers/pr_33913/it/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/entry/start.cef8f86b.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/scheduler.36a0863c.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/singletons.39cd3b38.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/index.733708bb.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/paths.64312358.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/entry/app.d399d34f.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/index.9c13489a.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/nodes/0.7859fb6f.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/nodes/16.6997de98.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/CodeBlock.05d8ec32.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/DocNotebookDropdown.653c9eec.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/globals.7f7f1b26.js"> | |
| <link rel="modulepreload" href="/docs/transformers/pr_33913/it/_app/immutable/chunks/EditOnGithub.e88f2b7b.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Modelli multilingue per l’inferenza","local":"modelli-multilingue-per-linferenza","sections":[{"title":"XLM","local":"xlm","sections":[{"title":"XLM con language embeddings","local":"xlm-con-language-embeddings","sections":[],"depth":3},{"title":"XLM senza language embeddings","local":"xlm-senza-language-embeddings","sections":[],"depth":3}],"depth":2},{"title":"BERT","local":"bert","sections":[],"depth":2},{"title":"XLM-RoBERTa","local":"xlm-roberta","sections":[],"depth":2},{"title":"M2M100","local":"m2m100","sections":[],"depth":2},{"title":"MBart","local":"mbart","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="modelli-multilingue-per-linferenza" 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="#modelli-multilingue-per-linferenza"><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>Modelli multilingue per l’inferenza</span></h1> <div class="flex space-x-1 absolute z-10 right-0 top-0"> <div class="relative colab-dropdown "> <button class=" " type="button"> <img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"> </button> </div> <div class="relative colab-dropdown "> <button class=" " type="button"> <img alt="Open In Studio Lab" class="!m-0" src="https://studiolab.sagemaker.aws/studiolab.svg"> </button> </div></div> <p data-svelte-h="svelte-1y2uj2f">Ci sono diversi modelli multilingue in 🤗 Transformers, e il loro utilizzo per l’inferenza differisce da quello dei modelli monolingua. Non <em>tutti</em> gli utilizzi dei modelli multilingue sono però diversi. Alcuni modelli, come <a href="https://huggingface.co/google-bert/bert-base-multilingual-uncased" rel="nofollow">google-bert/bert-base-multilingual-uncased</a>, possono essere usati come un modello monolingua. Questa guida ti mostrerà come utilizzare modelli multilingue che utilizzano un modo diverso per fare l’inferenza.</p> <h2 class="relative group"><a id="xlm" 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="#xlm"><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>XLM</span></h2> <p data-svelte-h="svelte-b0zdcx">XLM ha dieci diversi checkpoint, di cui solo uno è monolingua. I nove checkpoint rimanenti possono essere suddivisi in due categorie: i checkpoint che utilizzano i language embeddings e quelli che non li utilizzano.</p> <h3 class="relative group"><a id="xlm-con-language-embeddings" 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="#xlm-con-language-embeddings"><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>XLM con language embeddings</span></h3> <p data-svelte-h="svelte-6flhpc">I seguenti modelli XLM utilizzano gli embeddings linguistici per specificare la lingua utilizzata per l’inferenza:</p> <ul data-svelte-h="svelte-hwtd9h"><li><code>FacebookAI/xlm-mlm-ende-1024</code> (Modellazione mascherata del linguaggio (Masked language modeling, in inglese), Inglese-Tedesco)</li> <li><code>FacebookAI/xlm-mlm-enfr-1024</code> (Modellazione mascherata del linguaggio, Inglese-Francese)</li> <li><code>FacebookAI/xlm-mlm-enro-1024</code> (Modellazione mascherata del linguaggio, Inglese-Rumeno)</li> <li><code>FacebookAI/xlm-mlm-xnli15-1024</code> (Modellazione mascherata del linguaggio, lingue XNLI)</li> <li><code>FacebookAI/xlm-mlm-tlm-xnli15-1024</code> (Modellazione mascherata del linguaggio + traduzione, lingue XNLI)</li> <li><code>FacebookAI/xlm-clm-enfr-1024</code> (Modellazione causale del linguaggio, Inglese-Francese)</li> <li><code>FacebookAI/xlm-clm-ende-1024</code> (Modellazione causale del linguaggio, Inglese-Tedesco)</li></ul> <p data-svelte-h="svelte-1o9s8ae">Gli embeddings linguistici sono rappresentati come un tensore delle stesse dimensioni dell’ <code>input_ids</code> passato al modello. I valori in questi tensori dipendono dal linguaggio usato e sono identificati dagli attributi <code>lang2id</code> e <code>id2lang</code> del tokenizer.</p> <p data-svelte-h="svelte-tnbb9o">In questo esempio, carica il checkpoint <code>FacebookAI/xlm-clm-enfr-1024</code> (Modellazione causale del linguaggio, Inglese-Francese):</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-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> XLMTokenizer, XLMWithLMHeadModel | |
| <span class="hljs-meta">>>> </span>tokenizer = XLMTokenizer.from_pretrained(<span class="hljs-string">"FacebookAI/xlm-clm-enfr-1024"</span>) | |
| <span class="hljs-meta">>>> </span>model = XLMWithLMHeadModel.from_pretrained(<span class="hljs-string">"FacebookAI/xlm-clm-enfr-1024"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1lvsgdy">L’attributo <code>lang2id</code> del tokenizer mostra il linguaggio del modello e il suo ids:</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-meta">>>> </span><span class="hljs-built_in">print</span>(tokenizer.lang2id) | |
| {<span class="hljs-string">'en'</span>: <span class="hljs-number">0</span>, <span class="hljs-string">'fr'</span>: <span class="hljs-number">1</span>}<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1gs48v1">Poi, crea un esempio di input:</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-meta">>>> </span>input_ids = torch.tensor([tokenizer.encode(<span class="hljs-string">"Wikipedia was used to"</span>)]) <span class="hljs-comment"># batch size of 1</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1lqmqx6">Imposta l’id del linguaggio a <code>"en"</code> e usalo per definire il language embedding. Il language embedding è un tensore riempito con <code>0</code> perché questo è il language id per l’inglese. Questo tensore dovrebbe avere la stessa dimensione di <code>input_ids</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-meta">>>> </span>language_id = tokenizer.lang2id[<span class="hljs-string">"en"</span>] <span class="hljs-comment"># 0</span> | |
| <span class="hljs-meta">>>> </span>langs = torch.tensor([language_id] * input_ids.shape[<span class="hljs-number">1</span>]) <span class="hljs-comment"># torch.tensor([0, 0, 0, ..., 0])</span> | |
| <span class="hljs-meta">>>> </span><span class="hljs-comment"># We reshape it to be of size (batch_size, sequence_length)</span> | |
| <span class="hljs-meta">>>> </span>langs = langs.view(<span class="hljs-number">1</span>, -<span class="hljs-number">1</span>) <span class="hljs-comment"># is now of shape [1, sequence_length] (we have a batch size of 1)</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1257czm">Adesso puoi inserire <code>input_ids</code> e language embedding nel modello:</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-meta">>>> </span>outputs = model(input_ids, langs=langs)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-docsh6">Lo script <a href="https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-generation/run_generation.py" rel="nofollow">run_generation.py</a> può generare testo tramite i language embeddings usando i checkpoints <code>xlm-clm</code>.</p> <h3 class="relative group"><a id="xlm-senza-language-embeddings" 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="#xlm-senza-language-embeddings"><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>XLM senza language embeddings</span></h3> <p data-svelte-h="svelte-1fsmnuh">I seguenti modelli XLM non richiedono l’utilizzo dei language embeddings per fare inferenza:</p> <ul data-svelte-h="svelte-1r1fxak"><li><code>FacebookAI/xlm-mlm-17-1280</code> (Modellazione mascherata del linguaggio, 17 lingue)</li> <li><code>FacebookAI/xlm-mlm-100-1280</code> (Modellazione mascherata del linguaggio, 100 lingue)</li></ul> <p data-svelte-h="svelte-18famm0">Questi modelli sono utilizzati per rappresentazioni generiche di frasi, a differenza dei precedenti checkpoints XML.</p> <h2 class="relative group"><a id="bert" 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="#bert"><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>BERT</span></h2> <p data-svelte-h="svelte-16x26sb">Il seguente modello BERT può essere usato per compiti multilingue:</p> <ul data-svelte-h="svelte-1ufkrdp"><li><code>google-bert/bert-base-multilingual-uncased</code> (Modellazione mascherata del linguaggio + Previsione della prossima frase, 102 lingue)</li> <li><code>google-bert/bert-base-multilingual-cased</code> (Modellazione mascherata del linguaggio + Previsione della prossima frase, 104 lingue)</li></ul> <p data-svelte-h="svelte-nozyay">Questi modelli non richiedono language embeddings per fare inferenza. Riescono ad identificare il linguaggio dal contesto e inferire di conseguenza.</p> <h2 class="relative group"><a id="xlm-roberta" 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="#xlm-roberta"><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>XLM-RoBERTa</span></h2> <p data-svelte-h="svelte-6sllth">Il seguente modello XLM-RoBERTa può essere usato per compiti multilingue:</p> <ul data-svelte-h="svelte-1h2reme"><li><code>FacebookAI/xlm-roberta-base</code> (Modellazione mascherata del linguaggio, 100 lingue)</li> <li><code>FacebookAI/xlm-roberta-large</code> (Modellazione mascherata del linguaggio, 100 lingue)</li></ul> <p data-svelte-h="svelte-1t17k0a">XLM-RoBERTa è stato addestrato su 2.5TB di dati CommonCrawl appena creati e puliti in 100 lingue. Offre notevoli vantaggi rispetto ai modelli multilingue rilasciati in precedenza, come mBERT o XLM, in compiti come la classificazione, l’etichettatura delle sequenze e la risposta alle domande.</p> <h2 class="relative group"><a id="m2m100" 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="#m2m100"><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>M2M100</span></h2> <p data-svelte-h="svelte-sftpup">Il seguente modello M2M100 può essere usato per compiti multilingue:</p> <ul data-svelte-h="svelte-13d4mst"><li><code>facebook/m2m100_418M</code> (Traduzione)</li> <li><code>facebook/m2m100_1.2B</code> (Traduzione)</li></ul> <p data-svelte-h="svelte-13sr045">In questo esempio, carica il checkpoint <code>facebook/m2m100_418M</code> per tradurre dal cinese all’inglese. Puoi impostare la lingua di partenza nel tokenizer:</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-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> M2M100ForConditionalGeneration, M2M100Tokenizer | |
| <span class="hljs-meta">>>> </span>en_text = <span class="hljs-string">"Do not meddle in the affairs of wizards, for they are subtle and quick to anger."</span> | |
| <span class="hljs-meta">>>> </span>chinese_text = <span class="hljs-string">"不要插手巫師的事務, 因為他們是微妙的, 很快就會發怒."</span> | |
| <span class="hljs-meta">>>> </span>tokenizer = M2M100Tokenizer.from_pretrained(<span class="hljs-string">"facebook/m2m100_418M"</span>, src_lang=<span class="hljs-string">"zh"</span>) | |
| <span class="hljs-meta">>>> </span>model = M2M100ForConditionalGeneration.from_pretrained(<span class="hljs-string">"facebook/m2m100_418M"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1ysq4ea">Applica il tokenizer al testo:</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-meta">>>> </span>encoded_zh = tokenizer(chinese_text, return_tensors=<span class="hljs-string">"pt"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1hcjl5u">M2M100 forza l’id della lingua obiettivo come primo token generato per tradurre nella lingua obiettivo. Imposta il parametro <code>forced_bos_token_id</code> a <code>en</code> nel metodo <code>generate</code> per tradurre in inglese:</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-meta">>>> </span>generated_tokens = model.generate(**encoded_zh, forced_bos_token_id=tokenizer.get_lang_id(<span class="hljs-string">"en"</span>)) | |
| <span class="hljs-meta">>>> </span>tokenizer.batch_decode(generated_tokens, skip_special_tokens=<span class="hljs-literal">True</span>) | |
| <span class="hljs-string">'Do not interfere with the matters of the witches, because they are delicate and will soon be angry.'</span><!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="mbart" 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="#mbart"><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>MBart</span></h2> <p data-svelte-h="svelte-hqe28c">Il seguente modello MBart può essere usato per compiti multilingue:</p> <ul data-svelte-h="svelte-15l5ps7"><li><code>facebook/mbart-large-50-one-to-many-mmt</code> (Traduzione automatica multilingue uno-a-molti, 50 lingue)</li> <li><code>facebook/mbart-large-50-many-to-many-mmt</code> (Traduzione automatica multilingue molti-a-molti, 50 lingue)</li> <li><code>facebook/mbart-large-50-many-to-one-mmt</code> (Traduzione automatica multilingue molti-a-uno, 50 lingue)</li> <li><code>facebook/mbart-large-50</code> (Traduzione multilingue, 50 lingue)</li> <li><code>facebook/mbart-large-cc25</code></li></ul> <p data-svelte-h="svelte-niv51s">In questo esempio, carica il checkpoint <code>facebook/mbart-large-50-many-to-many-mmt</code> per tradurre dal finlandese all’inglese. Puoi impostare la lingua di partenza nel tokenizer:</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-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, AutoModelForSeq2SeqLM | |
| <span class="hljs-meta">>>> </span>en_text = <span class="hljs-string">"Do not meddle in the affairs of wizards, for they are subtle and quick to anger."</span> | |
| <span class="hljs-meta">>>> </span>fi_text = <span class="hljs-string">"Älä sekaannu velhojen asioihin, sillä ne ovat hienovaraisia ja nopeasti vihaisia."</span> | |
| <span class="hljs-meta">>>> </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"facebook/mbart-large-50-many-to-many-mmt"</span>, src_lang=<span class="hljs-string">"fi_FI"</span>) | |
| <span class="hljs-meta">>>> </span>model = AutoModelForSeq2SeqLM.from_pretrained(<span class="hljs-string">"facebook/mbart-large-50-many-to-many-mmt"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-5lihjd">Applica il tokenizer sul testo:</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-meta">>>> </span>encoded_en = tokenizer(en_text, return_tensors=<span class="hljs-string">"pt"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-21vlhj">MBart forza l’id della lingua obiettivo come primo token generato per tradurre nella lingua obiettivo. Imposta il parametro <code>forced_bos_token_id</code> a <code>en</code> nel metodo <code>generate</code> per tradurre in inglese:</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-meta">>>> </span>generated_tokens = model.generate(**encoded_en, forced_bos_token_id=tokenizer.lang_code_to_id(<span class="hljs-string">"en_XX"</span>)) | |
| <span class="hljs-meta">>>> </span>tokenizer.batch_decode(generated_tokens, skip_special_tokens=<span class="hljs-literal">True</span>) | |
| <span class="hljs-string">"Don't interfere with the wizard's affairs, because they are subtle, will soon get angry."</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-m0bu83">Se stai usando il checkpoint <code>facebook/mbart-large-50-many-to-one-mmt</code>, non hai bisogno di forzare l’id della lingua obiettivo come primo token generato altrimenti l’uso è lo stesso.</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/transformers/blob/main/docs/source/it/multilingual.md" 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_728xau = { | |
| assets: "/docs/transformers/pr_33913/it", | |
| base: "/docs/transformers/pr_33913/it", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/transformers/pr_33913/it/_app/immutable/entry/start.cef8f86b.js"), | |
| import("/docs/transformers/pr_33913/it/_app/immutable/entry/app.d399d34f.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 16], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 39.9 kB
- Xet hash:
- 451b2e34e72e92c503457a8924716b0829cb6b928d718e05bca9986b6ad05f8f
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.