Hugging Face's logo

Buckets:
hf-doc-build
/
doc

Files
xet
hf-doc-build/doc / transformers /v4.30.2 /ko /multilingual.html
rtrm's picture
rtrm
download
raw
36.6 kB
<meta charset="utf-8" /><meta http-equiv="content-security-policy" content=""><meta name="hf:doc:metadata" content="{&quot;local&quot;:&quot;multilingual-models-for-inference&quot;,&quot;sections&quot;:[{&quot;local&quot;:&quot;xlm&quot;,&quot;sections&quot;:[{&quot;local&quot;:&quot;xlm-with-language-embeddings&quot;,&quot;title&quot;:&quot;언어 임베딩을 사용하는 XLM&quot;},{&quot;local&quot;:&quot;xlm-without-language-embeddings&quot;,&quot;title&quot;:&quot;언어 임베딩을 사용하지 않는 XLM&quot;}],&quot;title&quot;:&quot;XLM&quot;},{&quot;local&quot;:&quot;bert&quot;,&quot;title&quot;:&quot;BERT&quot;},{&quot;local&quot;:&quot;xlmroberta&quot;,&quot;title&quot;:&quot;XLM-RoBERTa&quot;},{&quot;local&quot;:&quot;m2m100&quot;,&quot;title&quot;:&quot;M2M100&quot;},{&quot;local&quot;:&quot;mbart&quot;,&quot;title&quot;:&quot;MBart&quot;}],&quot;title&quot;:&quot;다국어 모델 추론하기&quot;}" data-svelte="svelte-1phssyn">
 <link rel="modulepreload" href="/docs/transformers/v4.30.2/ko/_app/assets/pages/__layout.svelte-hf-doc-builder.css">
 <link rel="modulepreload" href="/docs/transformers/v4.30.2/ko/_app/start-hf-doc-builder.js">
 <link rel="modulepreload" href="/docs/transformers/v4.30.2/ko/_app/chunks/vendor-hf-doc-builder.js">
 <link rel="modulepreload" href="/docs/transformers/v4.30.2/ko/_app/chunks/paths-hf-doc-builder.js">
 <link rel="modulepreload" href="/docs/transformers/v4.30.2/ko/_app/pages/__layout.svelte-hf-doc-builder.js">
 <link rel="modulepreload" href="/docs/transformers/v4.30.2/ko/_app/pages/multilingual.mdx-hf-doc-builder.js">
 <link rel="modulepreload" href="/docs/transformers/v4.30.2/ko/_app/chunks/IconCopyLink-hf-doc-builder.js">
 <link rel="modulepreload" href="/docs/transformers/v4.30.2/ko/_app/chunks/CodeBlock-hf-doc-builder.js">
 <link rel="modulepreload" href="/docs/transformers/v4.30.2/ko/_app/chunks/DocNotebookDropdown-hf-doc-builder.js">
<h1 class="relative group"><a id="multilingual-models-for-inference" 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="#multilingual-models-for-inference"><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>다국어 모델 추론하기
 </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>🤗 Transformers에는 여러 종류의 다국어(multilingual) 모델이 있으며, 단일 언어(monolingual) 모델과 추론 시 사용법이 다릅니다.
그렇다고 해서 <em>모든</em> 다국어 모델의 사용법이 다른 것은 아닙니다.</p>
<p><a href="https://huggingface.co/bert-base-multilingual-uncased" rel="nofollow">bert-base-multilingual-uncased</a>와 같은 몇몇 모델은 단일 언어 모델처럼 사용할 수 있습니다.
이번 가이드에서 다국어 모델의 추론 시 사용 방법을 알아볼 것입니다.</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>XLM에는 10가지 체크포인트(checkpoint)가 있는데, 이 중 하나만 단일 언어입니다.
나머지 체크포인트 9개는 언어 임베딩을 사용하는 체크포인트와 그렇지 않은 체크포인트의 두 가지 범주로 나눌 수 있습니다.</p>
<h3 class="relative group"><a id="xlm-with-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-with-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>언어 임베딩을 사용하는 XLM
 </span></h3>
<p>다음 XLM 모델은 추론 시에 언어 임베딩을 사용합니다:</p>
<ul><li><code>xlm-mlm-ende-1024</code> (마스킹된 언어 모델링, 영어-독일어)</li>
<li><code>xlm-mlm-enfr-1024</code> (마스킹된 언어 모델링, 영어-프랑스어)</li>
<li><code>xlm-mlm-enro-1024</code> (마스킹된 언어 모델링, 영어-루마니아어)</li>
<li><code>xlm-mlm-xnli15-1024</code> (마스킹된 언어 모델링, XNLI 데이터 세트에서 제공하는 15개 국어)</li>
<li><code>xlm-mlm-tlm-xnli15-1024</code> (마스킹된 언어 모델링 + 번역, XNLI 데이터 세트에서 제공하는 15개 국어)</li>
<li><code>xlm-clm-enfr-1024</code> (Causal language modeling, 영어-프랑스어)</li>
<li><code>xlm-clm-ende-1024</code> (Causal language modeling, 영어-독일어)</li></ul>
<p>언어 임베딩은 모델에 전달된 <code>input_ids</code>와 동일한 shape의 텐서로 표현됩니다.
이러한 텐서의 값은 사용된 언어에 따라 다르며 토크나이저의 <code>lang2id</code> 및 <code>id2lang</code> 속성에 의해 식별됩니다.</p>
<p>다음 예제에서는 <code>xlm-clm-enfr-1024</code> 체크포인트(코잘 언어 모델링(causal language modeling), 영어-프랑스어)를 가져옵니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">import</span> torch
<span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> XLMTokenizer, XLMWithLMHeadModel
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = XLMTokenizer.from_pretrained(<span class="hljs-string">&quot;xlm-clm-enfr-1024&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = XLMWithLMHeadModel.from_pretrained(<span class="hljs-string">&quot;xlm-clm-enfr-1024&quot;</span>)<!-- HTML_TAG_END --></pre></div>
<p>토크나이저의 <code>lang2id</code> 속성은 모델의 언어와 해당 ID를 표시합니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-built_in">print</span>(tokenizer.lang2id)
{<span class="hljs-string">&#x27;en&#x27;</span>: <span class="hljs-number">0</span>, <span class="hljs-string">&#x27;fr&#x27;</span>: <span class="hljs-number">1</span>}<!-- HTML_TAG_END --></pre></div>
<p>다음으로, 예제 입력을 만듭니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span>input_ids = torch.tensor([tokenizer.encode(<span class="hljs-string">&quot;Wikipedia was used to&quot;</span>)]) <span class="hljs-comment"># 배치 크기는 1입니다</span><!-- HTML_TAG_END --></pre></div>
<p>언어 ID를 <code>&quot;en&quot;</code>으로 설정해 언어 임베딩을 정의합니다.
언어 임베딩은 영어의 언어 ID인 <code>0</code>으로 채워진 텐서입니다.
이 텐서는 <code>input_ids</code>와 같은 크기여야 합니다. </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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span>language_id = tokenizer.lang2id[<span class="hljs-string">&quot;en&quot;</span>] <span class="hljs-comment"># 0</span>
<span class="hljs-meta">&gt;&gt;&gt; </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">&gt;&gt;&gt; </span><span class="hljs-comment"># (batch_size, sequence_length) shape의 텐서가 되도록 만듭니다.</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>langs = langs.view(<span class="hljs-number">1</span>, -<span class="hljs-number">1</span>) <span class="hljs-comment"># 이제 [1, sequence_length] shape이 되었습니다(배치 크기는 1입니다)</span><!-- HTML_TAG_END --></pre></div>
<p>이제 <code>input_ids</code>와 언어 임베딩을 모델로 전달합니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span>outputs = model(input_ids, langs=langs)<!-- HTML_TAG_END --></pre></div>
<p><a href="https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-generation/run_generation.py" rel="nofollow">run_generation.py</a> 스크립트로 <code>xlm-clm</code> 체크포인트를 사용해 텍스트와 언어 임베딩을 생성할 수 있습니다.</p>
<h3 class="relative group"><a id="xlm-without-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-without-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>언어 임베딩을 사용하지 않는 XLM
 </span></h3>
<p>다음 XLM 모델은 추론 시에 언어 임베딩이 필요하지 않습니다:</p>
<ul><li><code>xlm-mlm-17-1280</code> (마스킹된 언어 모델링, 17개 국어)</li>
<li><code>xlm-mlm-100-1280</code> (마스킹된 언어 모델링, 100개 국어)</li></ul>
<p>이전의 XLM 체크포인트와 달리 이 모델은 일반 문장 표현에 사용됩니다.</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>다음 BERT 모델은 다국어 태스크에 사용할 수 있습니다:</p>
<ul><li><code>bert-base-multilingual-uncased</code> (마스킹된 언어 모델링 + 다음 문장 예측, 102개 국어)</li>
<li><code>bert-base-multilingual-cased</code> (마스킹된 언어 모델링 + 다음 문장 예측, 104개 국어)</li></ul>
<p>이러한 모델은 추론 시에 언어 임베딩이 필요하지 않습니다.
문맥에서 언어를 식별하고, 식별된 언어로 추론합니다.</p>
<h2 class="relative group"><a id="xlmroberta" 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="#xlmroberta"><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>다음 XLM-RoBERTa 또한 다국어 다국어 태스크에 사용할 수 있습니다:</p>
<ul><li><code>xlm-roberta-base</code> (마스킹된 언어 모델링, 100개 국어)</li>
<li><code>xlm-roberta-large</code> (마스킹된 언어 모델링, 100개 국어)</li></ul>
<p>XLM-RoBERTa는 100개 국어에 대해 새로 생성되고 정제된 2.5TB 규모의 CommonCrawl 데이터로 학습되었습니다.
이전에 공개된 mBERT나 XLM과 같은 다국어 모델에 비해 분류, 시퀀스 라벨링, 질의 응답과 같은 다운스트림(downstream) 작업에서 이점이 있습니다.</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>다음 M2M100 모델 또한 다국어 다국어 태스크에 사용할 수 있습니다:</p>
<ul><li><code>facebook/m2m100_418M</code> (번역)</li>
<li><code>facebook/m2m100_1.2B</code> (번역)</li></ul>
<p>이 예제에서는 <code>facebook/m2m100_418M</code> 체크포인트를 가져와서 중국어를 영어로 번역합니다.
토크나이저에서 번역 대상 언어(source language)를 설정할 수 있습니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> M2M100ForConditionalGeneration, M2M100Tokenizer
<span class="hljs-meta">&gt;&gt;&gt; </span>en_text = <span class="hljs-string">&quot;Do not meddle in the affairs of wizards, for they are subtle and quick to anger.&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>chinese_text = <span class="hljs-string">&quot;不要插手巫師的事務, 因為他們是微妙的, 很快就會發怒.&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = M2M100Tokenizer.from_pretrained(<span class="hljs-string">&quot;facebook/m2m100_418M&quot;</span>, src_lang=<span class="hljs-string">&quot;zh&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = M2M100ForConditionalGeneration.from_pretrained(<span class="hljs-string">&quot;facebook/m2m100_418M&quot;</span>)<!-- HTML_TAG_END --></pre></div>
<p>문장을 토큰화합니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span>encoded_zh = tokenizer(chinese_text, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)<!-- HTML_TAG_END --></pre></div>
<p>M2M100은 번역을 진행하기 위해 첫 번째로 생성되는 토큰은 번역할 언어(target language) ID로 강제 지정합니다.
영어로 번역하기 위해 <code>generate</code> 메소드에서 <code>forced_bos_token_id</code>를 <code>en</code>으로 설정합니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span>generated_tokens = model.generate(**encoded_zh, forced_bos_token_id=tokenizer.get_lang_id(<span class="hljs-string">&quot;en&quot;</span>))
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer.batch_decode(generated_tokens, skip_special_tokens=<span class="hljs-literal">True</span>)
<span class="hljs-string">&#x27;Do not interfere with the matters of the witches, because they are delicate and will soon be angry.&#x27;</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>다음 MBart 모델 또한 다국어 태스크에 사용할 수 있습니다:</p>
<ul><li><code>facebook/mbart-large-50-one-to-many-mmt</code> (일대다 다국어 번역, 50개 국어)</li>
<li><code>facebook/mbart-large-50-many-to-many-mmt</code> (다대다 다국어 번역, 50개 국어)</li>
<li><code>facebook/mbart-large-50-many-to-one-mmt</code> (다대일 다국어 번역, 50개 국어)</li>
<li><code>facebook/mbart-large-50</code> (다국어 번역, 50개 국어)</li>
<li><code>facebook/mbart-large-cc25</code></li></ul>
<p>이 예제에서는 핀란드어를 영어로 번역하기 위해 <code>facebook/mbart-large-50-many-to-many-mmt</code> 체크포인트를 가져옵니다.
토크나이저에서 번역 대상 언어(source language)를 설정할 수 있습니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoTokenizer, AutoModelForSeq2SeqLM
<span class="hljs-meta">&gt;&gt;&gt; </span>en_text = <span class="hljs-string">&quot;Do not meddle in the affairs of wizards, for they are subtle and quick to anger.&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>fi_text = <span class="hljs-string">&quot;Älä sekaannu velhojen asioihin, sillä ne ovat hienovaraisia ja nopeasti vihaisia.&quot;</span>
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">&quot;facebook/mbart-large-50-many-to-many-mmt&quot;</span>, src_lang=<span class="hljs-string">&quot;fi_FI&quot;</span>)
<span class="hljs-meta">&gt;&gt;&gt; </span>model = AutoModelForSeq2SeqLM.from_pretrained(<span class="hljs-string">&quot;facebook/mbart-large-50-many-to-many-mmt&quot;</span>)<!-- HTML_TAG_END --></pre></div>
<p>문장을 토큰화합니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span>encoded_en = tokenizer(en_text, return_tensors=<span class="hljs-string">&quot;pt&quot;</span>)<!-- HTML_TAG_END --></pre></div>
<p>MBart는 번역을 진행하기 위해 첫 번째로 생성되는 토큰은 번역할 언어(target language) ID로 강제 지정합니다.
영어로 번역하기 위해 <code>generate</code> 메소드에서 <code>forced_bos_token_id</code>를 <code>en</code>으로 설정합니다:</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><!-- HTML_TAG_START --><span class="hljs-meta">&gt;&gt;&gt; </span>generated_tokens = model.generate(**encoded_en, forced_bos_token_id=tokenizer.lang_code_to_id(<span class="hljs-string">&quot;en_XX&quot;</span>))
<span class="hljs-meta">&gt;&gt;&gt; </span>tokenizer.batch_decode(generated_tokens, skip_special_tokens=<span class="hljs-literal">True</span>)
<span class="hljs-string">&quot;Don&#x27;t interfere with the wizard&#x27;s affairs, because they are subtle, will soon get angry.&quot;</span><!-- HTML_TAG_END --></pre></div>
<p><code>facebook/mbart-large-50-many-to-one-mmt</code> 체크포인트를 사용하고 있다면, 첫 번째로 생성되는 토큰을 번역할 언어(target language) ID로 강제 지정할 필요는 없습니다.</p>
<script type="module" data-hydrate="1c3vo10">
 import { start } from "/docs/transformers/v4.30.2/ko/_app/start-hf-doc-builder.js";
 start({
 target: document.querySelector('[data-hydrate="1c3vo10"]').parentNode,
 paths: {"base":"/docs/transformers/v4.30.2/ko","assets":"/docs/transformers/v4.30.2/ko"},
 session: {},
 route: false,
 spa: false,
 trailing_slash: "never",
 hydrate: {
 status: 200,
 error: null,
 nodes: [
 import("/docs/transformers/v4.30.2/ko/_app/pages/__layout.svelte-hf-doc-builder.js"),
 import("/docs/transformers/v4.30.2/ko/_app/pages/multilingual.mdx-hf-doc-builder.js")
 ],
 params: {}
 }
 });
 </script>

Xet Storage Details

Size:
36.6 kB
·
Xet hash:
5b84134d2c4d7c9074086fc80c99ee447d120711eafd3a33f905f0d069bd5e47

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