Buckets:

hf-doc-build
/

doc-dev

Files

xet

hf-doc-build/doc-dev / transformers /pr_31579 /ko /task_summary.html

rtrm

about 1 month ago

download

raw

79.4 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"🤗 Transformers로 할 수 있는 것","local":"what__transformers_can_do","sections":[{"title":"오디오","local":"audio","sections":[{"title":"오디오 분류","local":"audio_classification","sections":[],"depth":3},{"title":"자동 음성 인식","local":"automatic_speech_recognition","sections":[],"depth":3}],"depth":2},{"title":"컴퓨터 비전","local":"computer_vision","sections":[{"title":"이미지 분류","local":"image_classification","sections":[],"depth":3},{"title":"객체 탐지","local":"object_detection","sections":[],"depth":3},{"title":"이미지 분할","local":"image_segmentation","sections":[],"depth":3},{"title":"깊이 추정","local":"depth_estimation","sections":[],"depth":3}],"depth":2},{"title":"자연어처리","local":"natural_language_processing","sections":[{"title":"텍스트 분류","local":"text_classification","sections":[],"depth":3},{"title":"토큰 분류","local":"token_classification","sections":[],"depth":3},{"title":"질의응답","local":"question_answering","sections":[],"depth":3},{"title":"요약","local":"summarization","sections":[],"depth":3},{"title":"번역","local":"translation","sections":[],"depth":3},{"title":"언어 모델링","local":"language_modeling","sections":[],"depth":3}],"depth":2}],"depth":1}">
	<link href="/docs/transformers/pr_31579/ko/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/entry/start.7687ccc8.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/chunks/scheduler.56730f09.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/chunks/singletons.c918839e.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/chunks/index.bdfeb96e.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/chunks/paths.b779440d.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/entry/app.c00676df.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/chunks/index.1f144517.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/nodes/0.23ad8ff0.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/chunks/each.e59479a4.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/nodes/58.789b001c.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/chunks/CodeBlock.738eeccb.js">
	<link rel="modulepreload" href="/docs/transformers/pr_31579/ko/_app/immutable/chunks/EditOnGithub.854793f1.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"🤗 Transformers로 할 수 있는 것","local":"what__transformers_can_do","sections":[{"title":"오디오","local":"audio","sections":[{"title":"오디오 분류","local":"audio_classification","sections":[],"depth":3},{"title":"자동 음성 인식","local":"automatic_speech_recognition","sections":[],"depth":3}],"depth":2},{"title":"컴퓨터 비전","local":"computer_vision","sections":[{"title":"이미지 분류","local":"image_classification","sections":[],"depth":3},{"title":"객체 탐지","local":"object_detection","sections":[],"depth":3},{"title":"이미지 분할","local":"image_segmentation","sections":[],"depth":3},{"title":"깊이 추정","local":"depth_estimation","sections":[],"depth":3}],"depth":2},{"title":"자연어처리","local":"natural_language_processing","sections":[{"title":"텍스트 분류","local":"text_classification","sections":[],"depth":3},{"title":"토큰 분류","local":"token_classification","sections":[],"depth":3},{"title":"질의응답","local":"question_answering","sections":[],"depth":3},{"title":"요약","local":"summarization","sections":[],"depth":3},{"title":"번역","local":"translation","sections":[],"depth":3},{"title":"언어 모델링","local":"language_modeling","sections":[],"depth":3}],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="what__transformers_can_do" 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="#what__transformers_can_do"><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>🤗 Transformers로 할 수 있는 것</span></h1> <p data-svelte-h="svelte-10idx7x">🤗 Transformers는 자연어처리(NLP), 컴퓨터 비전, 오디오 및 음성 처리 작업에 대한 사전훈련된 최첨단 모델 라이브러리입니다.
	이 라이브러리는 트랜스포머 모델뿐만 아니라 컴퓨터 비전 작업을 위한 현대적인 합성곱 신경망과 같은 트랜스포머가 아닌 모델도 포함하고 있습니다.</p> <p data-svelte-h="svelte-rz0m0r">스마트폰, 앱, 텔레비전과 같은 오늘날 가장 인기 있는 소비자 제품을 살펴보면, 딥러닝 기술이 그 뒤에 사용되고 있을 확률이 높습니다.
	스마트폰으로 촬영한 사진에서 배경 객체를 제거하고 싶다면 어떻게 할까요? 이는 파놉틱 세그멘테이션 작업의 예입니다(아직 이게 무엇인지 모른다면, 다음 섹션에서 설명하겠습니다!).</p> <p data-svelte-h="svelte-d62f10">이 페이지는 다양한 음성 및 오디오, 컴퓨터 비전, NLP 작업을 🤗 Transformers 라이브러리를 활용하여 다루는 간단한 예제를 3줄의 코드로 제공합니다.</p> <h2 class="relative group"><a id="audio" 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="#audio"><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>오디오</span></h2> <p data-svelte-h="svelte-13r673m">음성 및 오디오 처리 작업은 다른 모달리티와 약간 다릅니다. 이는 주로 오디오가 연속적인 신호로 입력되기 때문입니다.
	텍스트와 달리 원본 오디오 파형(waveform)은 문장이 단어로 나눠지는 것처럼 깔끔하게 이산적인 묶음으로 나눌 수 없습니다.
	이를 극복하기 위해 원본 오디오 신호는 일정한 간격으로 샘플링됩니다. 해당 간격 내에서 더 많은 샘플을 취할 경우 샘플링률이 높아지며, 오디오는 원본 오디오 소스에 더 가까워집니다.</p> <p data-svelte-h="svelte-qps12j">과거의 접근 방식은 오디오에서 유용한 특징을 추출하기 위해 오디오를 전처리하는 것이었습니다.
	하지만 현재는 원본 오디오 파형을 특성 인코더에 직접 넣어서 오디오 표현(representation)을 추출하는 것이 더 일반적입니다.
	이렇게 하면 전처리 단계가 단순해지고 모델이 가장 중요한 특징을 학습할 수 있습니다.</p> <h3 class="relative group"><a id="audio_classification" 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="#audio_classification"><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>오디오 분류</span></h3> <p data-svelte-h="svelte-18pih3a">오디오 분류는 오디오 데이터에 미리 정의된 클래스 집합의 레이블을 지정하는 작업입니다. 이는 많은 구체적인 응용 프로그램을 포함한 넓은 범주입니다.</p> <p data-svelte-h="svelte-9ctq6o">일부 예시는 다음과 같습니다:</p> <ul data-svelte-h="svelte-5dixf0"><li>음향 장면 분류: 오디오에 장면 레이블(“사무실”, “해변”, “경기장”)을 지정합니다.</li> <li>음향 이벤트 감지: 오디오에 소리 이벤트 레이블(“차 경적”, “고래 울음소리”, “유리 파손”)을 지정합니다.</li> <li>태깅: 여러 가지 소리(새 지저귐, 회의에서의 화자 식별)가 포함된 오디오에 레이블을 지정합니다.</li> <li>음악 분류: 음악에 장르 레이블(“메탈”, “힙합”, “컨트리”)을 지정합니다.</li></ul> <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> pipeline

	<span class="hljs-meta">>>> </span>classifier = pipeline(task=<span class="hljs-string">"audio-classification"</span>, model=<span class="hljs-string">"superb/hubert-base-superb-er"</span>)
	<span class="hljs-meta">>>> </span>preds = classifier(<span class="hljs-string">"https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac"</span>)
	<span class="hljs-meta">>>> </span>preds = [{<span class="hljs-string">"score"</span>: <span class="hljs-built_in">round</span>(pred[<span class="hljs-string">"score"</span>], <span class="hljs-number">4</span>), <span class="hljs-string">"label"</span>: pred[<span class="hljs-string">"label"</span>]} <span class="hljs-keyword">for</span> pred <span class="hljs-keyword">in</span> preds]
	<span class="hljs-meta">>>> </span>preds
	[{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.4532</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'hap'</span>},
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.3622</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'sad'</span>},
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.0943</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'neu'</span>},
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.0903</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'ang'</span>}]<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="automatic_speech_recognition" 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="#automatic_speech_recognition"><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>자동 음성 인식</span></h3> <p data-svelte-h="svelte-cdyfh4">자동 음성 인식(ASR)은 음성을 텍스트로 변환하는 작업입니다.
	음성은 인간의 자연스러운 의사소통 형태이기 때문에 ASR은 가장 일반적인 오디오 작업 중 하나입니다.
	오늘날 ASR 시스템은 스피커, 전화 및 자동차와 같은 “스마트” 기술 제품에 내장되어 있습니다.
	우리는 가상 비서에게 음악 재생, 알림 설정 및 날씨 정보를 요청할 수 있습니다.</p> <p data-svelte-h="svelte-190cybw">하지만 트랜스포머 아키텍처가 해결하는 데 도움을 준 핵심 도전 과제 중 하나는 양이 데이터 양이 적은 언어(low-resource language)에 대한 것입니다. 대량의 음성 데이터로 사전 훈련한 후 데이터 양이 적은 언어에서 레이블이 지정된 음성 데이터 1시간만으로 모델을 미세 조정하면 이전의 100배 많은 레이블이 지정된 데이터로 훈련된 ASR 시스템보다 훨씬 더 높은 품질의 결과를 얻을 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 class=""><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline

	<span class="hljs-meta">>>> </span>transcriber = pipeline(task=<span class="hljs-string">"automatic-speech-recognition"</span>, model=<span class="hljs-string">"openai/whisper-small"</span>)
	<span class="hljs-meta">>>> </span>transcriber(<span class="hljs-string">"https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac"</span>)
	{<span class="hljs-string">'text'</span>: <span class="hljs-string">' I have a dream that one day this nation will rise up and live out the true meaning of its creed.'</span>}<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="computer_vision" 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="#computer_vision"><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>컴퓨터 비전</span></h2> <p data-svelte-h="svelte-lp1ekz">컴퓨터 비전 작업 중 가장 초기의 성공적인 작업 중 하나는 <a href="glossary#convolution">합성곱 신경망(CNN)</a>을 사용하여 우편번호 숫자 이미지를 인식하는 것이었습니다. 이미지는 픽셀로 구성되어 있으며 각 픽셀은 숫자 값으로 표현됩니다. 이로써 이미지를 픽셀 값의 행렬로 나타내는 것이 쉬워집니다. 특정한 픽셀 값의 조합은 이미지의 색상을 의미합니다.</p> <p data-svelte-h="svelte-uo6axu">컴퓨터 비전 작업은 일반적으로 다음 두 가지 방법으로 접근 가능합니다:</p> <ol data-svelte-h="svelte-imx589"><li><p>합성곱을 사용하여 이미지의 낮은 수준 특징에서 높은 수준의 추상적인 요소까지 계층적으로 학습합니다.</p></li> <li><p>이미지를 패치로 나누고 트랜스포머를 사용하여 점진적으로 각 이미지 패치가 서로 어떠한 방식으로 연관되어 이미지를 형성하는지 학습합니다. <code>CNN</code>에서 선호하는 상향식 접근법과는 달리, 이 방식은 흐릿한 이미지로 초안을 그리고 점진적으로 선명한 이미지로 만들어가는 것과 유사합니다.</p></li></ol> <h3 class="relative group"><a id="image_classification" 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="#image_classification"><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>이미지 분류</span></h3> <p data-svelte-h="svelte-1o1g4f9">이미지 분류는 한 개의 전체 이미지에 미리 정의된 클래스 집합의 레이블을 지정하는 작업입니다.</p> <p data-svelte-h="svelte-14qqsoc">대부분의 분류 작업과 마찬가지로, 이미지 분류에는 다양한 실용적인 용도가 있으며, 일부 예시는 다음과 같습니다:</p> <ul data-svelte-h="svelte-1ong27"><li>의료: 질병을 감지하거나 환자 건강을 모니터링하기 위해 의료 이미지에 레이블을 지정합니다.</li> <li>환경: 위성 이미지를 분류하여 산림 벌채를 감시하고 야생 지역 관리를 위한 정보를 제공하거나 산불을 감지합니다.</li> <li>농업: 작물 이미지를 분류하여 식물 건강을 확인하거나 위성 이미지를 분류하여 토지 이용 관찰에 사용합니다.</li> <li>생태학: 동물이나 식물 종 이미지를 분류하여 야생 동물 개체군을 조사하거나 멸종 위기에 처한 종을 추적합니다.</li></ul> <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> pipeline

	<span class="hljs-meta">>>> </span>classifier = pipeline(task=<span class="hljs-string">"image-classification"</span>)
	<span class="hljs-meta">>>> </span>preds = classifier(
	<span class="hljs-meta">... </span> <span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"</span>
	<span class="hljs-meta">... </span>)
	<span class="hljs-meta">>>> </span>preds = [{<span class="hljs-string">"score"</span>: <span class="hljs-built_in">round</span>(pred[<span class="hljs-string">"score"</span>], <span class="hljs-number">4</span>), <span class="hljs-string">"label"</span>: pred[<span class="hljs-string">"label"</span>]} <span class="hljs-keyword">for</span> pred <span class="hljs-keyword">in</span> preds]
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(*preds, sep=<span class="hljs-string">"\n"</span>)
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.4335</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'lynx, catamount'</span>}
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.0348</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'cougar, puma, catamount, mountain lion, painter, panther, Felis concolor'</span>}
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.0324</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'snow leopard, ounce, Panthera uncia'</span>}
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.0239</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'Egyptian cat'</span>}
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.0229</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'tiger cat'</span>}<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="object_detection" 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="#object_detection"><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>객체 탐지</span></h3> <p data-svelte-h="svelte-10dx22t">이미지 분류와 달리 객체 탐지는 이미지 내에서 여러 객체를 식별하고 바운딩 박스로 정의된 객체의 위치를 파악합니다.</p> <p data-svelte-h="svelte-1bbfjj6">객체 탐지의 몇 가지 응용 예시는 다음과 같습니다:</p> <ul data-svelte-h="svelte-doxpjs"><li>자율 주행 차량: 다른 차량, 보행자 및 신호등과 같은 일상적인 교통 객체를 감지합니다.</li> <li>원격 감지: 재난 모니터링, 도시 계획 및 기상 예측 등을 수행합니다.</li> <li>결함 탐지: 건물의 균열이나 구조적 손상, 제조 결함 등을 탐지합니다.</li></ul> <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> pipeline

	<span class="hljs-meta">>>> </span>detector = pipeline(task=<span class="hljs-string">"object-detection"</span>)
	<span class="hljs-meta">>>> </span>preds = detector(
	<span class="hljs-meta">... </span> <span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"</span>
	<span class="hljs-meta">... </span>)
	<span class="hljs-meta">>>> </span>preds = [{<span class="hljs-string">"score"</span>: <span class="hljs-built_in">round</span>(pred[<span class="hljs-string">"score"</span>], <span class="hljs-number">4</span>), <span class="hljs-string">"label"</span>: pred[<span class="hljs-string">"label"</span>], <span class="hljs-string">"box"</span>: pred[<span class="hljs-string">"box"</span>]} <span class="hljs-keyword">for</span> pred <span class="hljs-keyword">in</span> preds]
	<span class="hljs-meta">>>> </span>preds
	[{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.9865</span>,
	<span class="hljs-string">'label'</span>: <span class="hljs-string">'cat'</span>,
	<span class="hljs-string">'box'</span>: {<span class="hljs-string">'xmin'</span>: <span class="hljs-number">178</span>, <span class="hljs-string">'ymin'</span>: <span class="hljs-number">154</span>, <span class="hljs-string">'xmax'</span>: <span class="hljs-number">882</span>, <span class="hljs-string">'ymax'</span>: <span class="hljs-number">598</span>}}]<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="image_segmentation" 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="#image_segmentation"><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>이미지 분할</span></h3> <p data-svelte-h="svelte-uvfxg5">이미지 분할은 픽셀 차원의 작업으로, 이미지 내의 모든 픽셀을 클래스에 할당합니다. 이는 객체 탐지와 다릅니다. 객체 탐지는 바운딩 박스를 사용하여 이미지 내의 객체를 레이블링하고 예측하는 반면, 분할은 더 세분화된 작업입니다. 분할은 픽셀 수준에서 객체를 감지할 수 있습니다.</p> <p data-svelte-h="svelte-1hwst5e">이미지 분할에는 여러 유형이 있습니다:</p> <ul data-svelte-h="svelte-bh6x2l"><li>인스턴스 분할: 개체의 클래스를 레이블링하는 것 외에도, 개체의 각 구분된 인스턴스에도 레이블을 지정합니다 (“개-1”, “개-2” 등).</li> <li>파놉틱 분할: 의미적 분할과 인스턴스 분할의 조합입니다. 각 픽셀을 의미적 클래스로 레이블링하는 <strong>동시에</strong> 개체의 각각 구분된 인스턴스로도 레이블을 지정합니다.</li></ul> <p data-svelte-h="svelte-1spkl8f">분할 작업은 자율 주행 차량에서 유용하며, 주변 환경의 픽셀 수준 지도를 생성하여 보행자와 다른 차량 주변에서 안전하게 탐색할 수 있습니다. 또한 의료 영상에서도 유용합니다. 분할 작업이 픽셀 수준에서 객체를 감지할 수 있기 때문에 비정상적인 세포나 장기의 특징을 식별하는 데 도움이 될 수 있습니다. 이미지 분할은 의류 가상 시착이나 카메라를 통해 실제 세계에 가상 개체를 덧씌워 증강 현실 경험을 만드는 등 전자 상거래 분야에서도 사용될 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 class=""><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline

	<span class="hljs-meta">>>> </span>segmenter = pipeline(task=<span class="hljs-string">"image-segmentation"</span>)
	<span class="hljs-meta">>>> </span>preds = segmenter(
	<span class="hljs-meta">... </span> <span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"</span>
	<span class="hljs-meta">... </span>)
	<span class="hljs-meta">>>> </span>preds = [{<span class="hljs-string">"score"</span>: <span class="hljs-built_in">round</span>(pred[<span class="hljs-string">"score"</span>], <span class="hljs-number">4</span>), <span class="hljs-string">"label"</span>: pred[<span class="hljs-string">"label"</span>]} <span class="hljs-keyword">for</span> pred <span class="hljs-keyword">in</span> preds]
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(*preds, sep=<span class="hljs-string">"\n"</span>)
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.9879</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'LABEL_184'</span>}
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.9973</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'snow'</span>}
	{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.9972</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'cat'</span>}<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="depth_estimation" 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="#depth_estimation"><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>깊이 추정</span></h3> <p data-svelte-h="svelte-1exjil8">깊이 추정은 카메라로부터 이미지 내부의 각 픽셀의 거리를 예측합니다. 이 컴퓨터 비전 작업은 특히 장면 이해와 재구성에 중요합니다. 예를 들어, 자율 주행 차량은 보행자, 교통 표지판 및 다른 차량과 같은 객체와의 거리를 이해하여 장애물과 충돌을 피해야 합니다. 깊이 정보는 또한 2D 이미지에서 3D 표현을 구성하는 데 도움이 되며 생물학적 구조나 건물의 고품질 3D 표현을 생성하는 데 사용될 수 있습니다.</p> <p data-svelte-h="svelte-skr9n9">깊이 추정에는 두 가지 접근 방식이 있습니다:</p> <ul data-svelte-h="svelte-urw9j1"><li>스테레오: 약간 다른 각도에서 촬영된 동일한 이미지 두 장을 비교하여 깊이를 추정합니다.</li> <li>단안: 단일 이미지에서 깊이를 추정합니다.</li></ul> <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> pipeline

	<span class="hljs-meta">>>> </span>depth_estimator = pipeline(task=<span class="hljs-string">"depth-estimation"</span>)
	<span class="hljs-meta">>>> </span>preds = depth_estimator(
	<span class="hljs-meta">... </span> <span class="hljs-string">"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg"</span>
	<span class="hljs-meta">... </span>)<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="natural_language_processing" 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="#natural_language_processing"><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>자연어처리</span></h2> <p data-svelte-h="svelte-juznik">텍스트는 인간이 의사 소통하는 자연스러운 방식 중 하나이기 때문에 자연어처리 역시 가장 일반적인 작업 유형 중 하나입니다. 모델이 인식하는 형식으로 텍스트를 변환하려면 토큰화해야 합니다. 이는 텍스트 시퀀스를 개별 단어 또는 하위 단어(토큰)로 분할한 다음 이러한 토큰을 숫자로 변환하는 것을 의미합니다. 결과적으로 텍스트 시퀀스를 숫자 시퀀스로 표현할 수 있으며, 숫자 시퀀스를 다양한 자연어처리 작업을 해결하기 위한 모델에 입력할 수 있습니다!</p> <h3 class="relative group"><a id="text_classification" 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="#text_classification"><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>텍스트 분류</span></h3> <p data-svelte-h="svelte-1i4o3md">다른 모달리티에서의 분류 작업과 마찬가지로 텍스트 분류는 미리 정의된 클래스 집합에서 텍스트 시퀀스(문장 수준, 단락 또는 문서 등)에 레이블을 지정합니다. 텍스트 분류에는 다양한 실용적인 응용 사례가 있으며, 일부 예시는 다음과 같습니다:</p> <ul data-svelte-h="svelte-5w2u0i"><li>감성 분석: 텍스트를 <code>긍정</code> 또는 <code>부정</code>과 같은 어떤 극성에 따라 레이블링하여 정치, 금융, 마케팅과 같은 분야에서 의사 결정에 정보를 제공하고 지원할 수 있습니다.</li> <li>콘텐츠 분류: 텍스트를 주제에 따라 레이블링(날씨, 스포츠, 금융 등)하여 뉴스 및 소셜 미디어 피드에서 정보를 구성하고 필터링하는 데 도움이 될 수 있습니다.</li></ul> <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> pipeline

	<span class="hljs-meta">>>> </span>classifier = pipeline(task=<span class="hljs-string">"sentiment-analysis"</span>)
	<span class="hljs-meta">>>> </span>preds = classifier(<span class="hljs-string">"Hugging Face is the best thing since sliced bread!"</span>)
	<span class="hljs-meta">>>> </span>preds = [{<span class="hljs-string">"score"</span>: <span class="hljs-built_in">round</span>(pred[<span class="hljs-string">"score"</span>], <span class="hljs-number">4</span>), <span class="hljs-string">"label"</span>: pred[<span class="hljs-string">"label"</span>]} <span class="hljs-keyword">for</span> pred <span class="hljs-keyword">in</span> preds]
	<span class="hljs-meta">>>> </span>preds
	[{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.9991</span>, <span class="hljs-string">'label'</span>: <span class="hljs-string">'POSITIVE'</span>}]<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="token_classification" 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="#token_classification"><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>토큰 분류</span></h3> <p data-svelte-h="svelte-1hgqh5j">모든 자연어처리 작업에서는 텍스트가 개별 단어나 하위 단어로 분리되어 전처리됩니다. 분리된 단어를 <a href="/glossary#token">토큰</a>이라고 합니다. 토큰 분류는 각 토큰에 미리 정의된 클래스 집합의 레이블을 할당합니다.</p> <p data-svelte-h="svelte-16tsf9c">토큰 분류의 두 가지 일반적인 유형은 다음과 같습니다:</p> <ul data-svelte-h="svelte-1wq0vai"><li>개체명 인식 (NER): 토큰을 조직, 인물, 위치 또는 날짜와 같은 개체 범주에 따라 레이블링합니다. NER은 특히 유전체학적인 환경에서 유전자, 단백질 및 약물 이름에 레이블을 지정하는 데 널리 사용됩니다.</li> <li>품사 태깅 (POS): 명사, 동사, 형용사와 같은 품사에 따라 토큰에 레이블을 할당합니다. POS는 번역 시스템이 동일한 단어가 문법적으로 어떻게 다른지 이해하는 데 도움이 됩니다 (명사로 사용되는 “bank(은행)“과 동사로 사용되는 “bank(예금을 예치하다)“과 같은 경우).</li></ul> <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> pipeline

	<span class="hljs-meta">>>> </span>classifier = pipeline(task=<span class="hljs-string">"ner"</span>)
	<span class="hljs-meta">>>> </span>preds = classifier(<span class="hljs-string">"Hugging Face is a French company based in New York City."</span>)
	<span class="hljs-meta">>>> </span>preds = [
	<span class="hljs-meta">... </span> {
	<span class="hljs-meta">... </span> <span class="hljs-string">"entity"</span>: pred[<span class="hljs-string">"entity"</span>],
	<span class="hljs-meta">... </span> <span class="hljs-string">"score"</span>: <span class="hljs-built_in">round</span>(pred[<span class="hljs-string">"score"</span>], <span class="hljs-number">4</span>),
	<span class="hljs-meta">... </span> <span class="hljs-string">"index"</span>: pred[<span class="hljs-string">"index"</span>],
	<span class="hljs-meta">... </span> <span class="hljs-string">"word"</span>: pred[<span class="hljs-string">"word"</span>],
	<span class="hljs-meta">... </span> <span class="hljs-string">"start"</span>: pred[<span class="hljs-string">"start"</span>],
	<span class="hljs-meta">... </span> <span class="hljs-string">"end"</span>: pred[<span class="hljs-string">"end"</span>],
	<span class="hljs-meta">... </span> }
	<span class="hljs-meta">... </span> <span class="hljs-keyword">for</span> pred <span class="hljs-keyword">in</span> preds
	<span class="hljs-meta">... </span>]
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(*preds, sep=<span class="hljs-string">"\n"</span>)
	{<span class="hljs-string">'entity'</span>: <span class="hljs-string">'I-ORG'</span>, <span class="hljs-string">'score'</span>: <span class="hljs-number">0.9968</span>, <span class="hljs-string">'index'</span>: <span class="hljs-number">1</span>, <span class="hljs-string">'word'</span>: <span class="hljs-string">'Hu'</span>, <span class="hljs-string">'start'</span>: <span class="hljs-number">0</span>, <span class="hljs-string">'end'</span>: <span class="hljs-number">2</span>}
	{<span class="hljs-string">'entity'</span>: <span class="hljs-string">'I-ORG'</span>, <span class="hljs-string">'score'</span>: <span class="hljs-number">0.9293</span>, <span class="hljs-string">'index'</span>: <span class="hljs-number">2</span>, <span class="hljs-string">'word'</span>: <span class="hljs-string">'##gging'</span>, <span class="hljs-string">'start'</span>: <span class="hljs-number">2</span>, <span class="hljs-string">'end'</span>: <span class="hljs-number">7</span>}
	{<span class="hljs-string">'entity'</span>: <span class="hljs-string">'I-ORG'</span>, <span class="hljs-string">'score'</span>: <span class="hljs-number">0.9763</span>, <span class="hljs-string">'index'</span>: <span class="hljs-number">3</span>, <span class="hljs-string">'word'</span>: <span class="hljs-string">'Face'</span>, <span class="hljs-string">'start'</span>: <span class="hljs-number">8</span>, <span class="hljs-string">'end'</span>: <span class="hljs-number">12</span>}
	{<span class="hljs-string">'entity'</span>: <span class="hljs-string">'I-MISC'</span>, <span class="hljs-string">'score'</span>: <span class="hljs-number">0.9983</span>, <span class="hljs-string">'index'</span>: <span class="hljs-number">6</span>, <span class="hljs-string">'word'</span>: <span class="hljs-string">'French'</span>, <span class="hljs-string">'start'</span>: <span class="hljs-number">18</span>, <span class="hljs-string">'end'</span>: <span class="hljs-number">24</span>}
	{<span class="hljs-string">'entity'</span>: <span class="hljs-string">'I-LOC'</span>, <span class="hljs-string">'score'</span>: <span class="hljs-number">0.999</span>, <span class="hljs-string">'index'</span>: <span class="hljs-number">10</span>, <span class="hljs-string">'word'</span>: <span class="hljs-string">'New'</span>, <span class="hljs-string">'start'</span>: <span class="hljs-number">42</span>, <span class="hljs-string">'end'</span>: <span class="hljs-number">45</span>}
	{<span class="hljs-string">'entity'</span>: <span class="hljs-string">'I-LOC'</span>, <span class="hljs-string">'score'</span>: <span class="hljs-number">0.9987</span>, <span class="hljs-string">'index'</span>: <span class="hljs-number">11</span>, <span class="hljs-string">'word'</span>: <span class="hljs-string">'York'</span>, <span class="hljs-string">'start'</span>: <span class="hljs-number">46</span>, <span class="hljs-string">'end'</span>: <span class="hljs-number">50</span>}
	{<span class="hljs-string">'entity'</span>: <span class="hljs-string">'I-LOC'</span>, <span class="hljs-string">'score'</span>: <span class="hljs-number">0.9992</span>, <span class="hljs-string">'index'</span>: <span class="hljs-number">12</span>, <span class="hljs-string">'word'</span>: <span class="hljs-string">'City'</span>, <span class="hljs-string">'start'</span>: <span class="hljs-number">51</span>, <span class="hljs-string">'end'</span>: <span class="hljs-number">55</span>}<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="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="#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>질의응답</span></h3> <p data-svelte-h="svelte-1bb9mft">질의응답은 또 하나의 토큰 차원의 작업으로, 문맥이 있을 때(개방형 도메인)와 문맥이 없을 때(폐쇄형 도메인) 질문에 대한 답변을 반환합니다. 이 작업은 가상 비서에게 식당이 영업 중인지와 같은 질문을 할 때마다 발생할 수 있습니다. 고객 지원 또는 기술 지원을 제공하거나 검색 엔진이 요청한 정보를 검색하는 데 도움을 줄 수 있습니다.</p> <p data-svelte-h="svelte-17o3plg">질문 답변에는 일반적으로 두 가지 유형이 있습니다:</p> <ul data-svelte-h="svelte-1ayvag1"><li>추출형: 질문과 문맥이 주어졌을 때, 모델이 주어진 문맥의 일부에서 가져온 텍스트의 범위를 답변으로 합니다.</li> <li>생성형: 질문과 문맥이 주어졌을 때, 주어진 문맥을 통해 답변을 생성합니다. 이 접근 방식은 <code>QuestionAnsweringPipeline</code> 대신 <code>Text2TextGenerationPipeline</code>을 통해 처리됩니다.</li></ul> <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> pipeline

	<span class="hljs-meta">>>> </span>question_answerer = pipeline(task=<span class="hljs-string">"question-answering"</span>)
	<span class="hljs-meta">>>> </span>preds = question_answerer(
	<span class="hljs-meta">... </span> question=<span class="hljs-string">"What is the name of the repository?"</span>,
	<span class="hljs-meta">... </span> context=<span class="hljs-string">"The name of the repository is huggingface/transformers"</span>,
	<span class="hljs-meta">... </span>)
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(
	<span class="hljs-meta">... </span> <span class="hljs-string">f"score: <span class="hljs-subst">{<span class="hljs-built_in">round</span>(preds[<span class="hljs-string">'score'</span>], <span class="hljs-number">4</span>)}</span>, start: <span class="hljs-subst">{preds[<span class="hljs-string">'start'</span>]}</span>, end: <span class="hljs-subst">{preds[<span class="hljs-string">'end'</span>]}</span>, answer: <span class="hljs-subst">{preds[<span class="hljs-string">'answer'</span>]}</span>"</span>
	<span class="hljs-meta">... </span>)
	score: <span class="hljs-number">0.9327</span>, start: <span class="hljs-number">30</span>, end: <span class="hljs-number">54</span>, answer: huggingface/transformers<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="summarization" 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="#summarization"><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>요약</span></h3> <p data-svelte-h="svelte-11gr80y">요약은 원본 문서의 의미를 최대한 보존하면서 긴 문서를 짧은 문서로 만드는 작업입니다. 요약은 <code>sequence-to-sequence</code> 작업입니다. 입력보다 짧은 텍스트 시퀀스를 출력합니다. 요약 작업은 독자가 장문 문서들의 주요 포인트를 빠르게 이해하는 데 도움을 줄 수 있습니다. 입법안, 법률 및 금융 문서, 특허 및 과학 논문은 요약 작업이 독자의 시간을 절약하고 독서 보조 도구로 사용될 수 있는 몇 가지 예시입니다.</p> <p data-svelte-h="svelte-fmz0j8">질문 답변과 마찬가지로 요약에는 두 가지 유형이 있습니다:</p> <ul data-svelte-h="svelte-mxtzg1"><li>추출형: 원본 텍스트에서 가장 중요한 문장을 식별하고 추출합니다.</li> <li>생성형: 원본 텍스트에서 목표 요약을 생성합니다. 입력 문서에 없는 새로운 단어를 포함할 수도 있습니다. <code>SummarizationPipeline</code>은 생성형 접근 방식을 사용합니다.</li></ul> <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> pipeline

	<span class="hljs-meta">>>> </span>summarizer = pipeline(task=<span class="hljs-string">"summarization"</span>)
	<span class="hljs-meta">>>> </span>summarizer(
	<span class="hljs-meta">... </span> <span class="hljs-string">"In this work, we presented the Transformer, the first sequence transduction model based entirely on attention, replacing the recurrent layers most commonly used in encoder-decoder architectures with multi-headed self-attention. For translation tasks, the Transformer can be trained significantly faster than architectures based on recurrent or convolutional layers. On both WMT 2014 English-to-German and WMT 2014 English-to-French translation tasks, we achieve a new state of the art. In the former task our best model outperforms even all previously reported ensembles."</span>
	<span class="hljs-meta">... </span>)
	[{<span class="hljs-string">'summary_text'</span>: <span class="hljs-string">' The Transformer is the first sequence transduction model based entirely on attention . It replaces the recurrent layers most commonly used in encoder-decoder architectures with multi-headed self-attention . For translation tasks, the Transformer can be trained significantly faster than architectures based on recurrent or convolutional layers .'</span>}]<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="translation" 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="#translation"><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>번역</span></h3> <p data-svelte-h="svelte-173rkq6">번역은 한 언어로 된 텍스트 시퀀스를 다른 언어로 변환하는 작업입니다. 이는 서로 다른 배경을 가진 사람들이 서로 소통하는 데 도움을 주는 중요한 역할을 합니다. 더 넓은 대중에게 콘텐츠를 번역하여 전달하거나, 새로운 언어를 배우는 데 도움이 되는 학습 도구가 될 수도 있습니다. 요약과 마찬가지로, 번역은 <code>sequence-to-sequence</code> 작업입니다. 즉, 모델은 입력 시퀀스를 받아서 출력이 되는 목표 시퀀스를 반환합니다.</p> <p data-svelte-h="svelte-4yqe8s">초기의 번역 모델은 대부분 단일 언어로 이루어져 있었지만, 최근에는 많은 언어 쌍 간에 번역을 수행할 수 있는 다중 언어 모델에 대한 관심이 높아지고 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 class=""><!-- HTML_TAG_START --><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline

	<span class="hljs-meta">>>> </span>text = <span class="hljs-string">"translate English to French: Hugging Face is a community-based open-source platform for machine learning."</span>
	<span class="hljs-meta">>>> </span>translator = pipeline(task=<span class="hljs-string">"translation"</span>, model=<span class="hljs-string">"google-t5/t5-small"</span>)
	<span class="hljs-meta">>>> </span>translator(text)
	[{<span class="hljs-string">'translation_text'</span>: <span class="hljs-string">"Hugging Face est une tribune communautaire de l'apprentissage des machines."</span>}]<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="language_modeling" 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="#language_modeling"><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 모델링</span></h3> <p data-svelte-h="svelte-1i9u0s6">언어 모델링은 텍스트 시퀀스에서 단어를 예측하는 작업입니다. 사전 훈련된 언어 모델은 많은 다른 하위 작업에 따라 미세 조정될 수 있기 때문에 매우 인기 있는 자연어처리 작업이 되었습니다. 최근에는 제로 샷(zero-shot) 또는 퓨 샷(few-shot) 학습이 가능한 대규모 언어 모델(Large Language Models, LLM)에 대한 많은 관심이 발생하고 있습니다. 이는 모델이 명시적으로 훈련되지 않은 작업도 해결할 수 있다는 것을 의미합니다! 언어 모델은 유창하고 설득력 있는 텍스트를 생성하는 데 사용될 수 있지만, 텍스트가 항상 정확하지는 않을 수 있으므로 주의가 필요합니다.</p> <p data-svelte-h="svelte-1pvuplr">언어 모델링에는 두 가지 유형이 있습니다:</p> <ul><li><p data-svelte-h="svelte-bmp22e">인과적 언어 모델링: 이 모델의 목적은 시퀀스에서 다음 토큰을 예측하는 것이며, 미래 토큰이 마스킹 됩니다.</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> pipeline

	<span class="hljs-meta">>>> </span>prompt = <span class="hljs-string">"Hugging Face is a community-based open-source platform for machine learning."</span>
	<span class="hljs-meta">>>> </span>generator = pipeline(task=<span class="hljs-string">"text-generation"</span>)
	<span class="hljs-meta">>>> </span>generator(prompt) <span class="hljs-comment"># doctest: +SKIP</span><!-- HTML_TAG_END --></pre></div></li> <li><p data-svelte-h="svelte-axjjse">마스킹된 언어 모델링: 이 모델의 목적은 시퀀스 내의 마스킹된 토큰을 예측하는 것이며, 시퀀스 내의 모든 토큰에 대한 접근이 제공됩니다.</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>text = <span class="hljs-string">"Hugging Face is a community-based open-source <mask> for machine learning."</span>
	<span class="hljs-meta">>>> </span>fill_mask = pipeline(task=<span class="hljs-string">"fill-mask"</span>)
	<span class="hljs-meta">>>> </span>preds = fill_mask(text, top_k=<span class="hljs-number">1</span>)
	<span class="hljs-meta">>>> </span>preds = [
	<span class="hljs-meta">... </span> {
	<span class="hljs-meta">... </span> <span class="hljs-string">"score"</span>: <span class="hljs-built_in">round</span>(pred[<span class="hljs-string">"score"</span>], <span class="hljs-number">4</span>),
	<span class="hljs-meta">... </span> <span class="hljs-string">"token"</span>: pred[<span class="hljs-string">"token"</span>],
	<span class="hljs-meta">... </span> <span class="hljs-string">"token_str"</span>: pred[<span class="hljs-string">"token_str"</span>],
	<span class="hljs-meta">... </span> <span class="hljs-string">"sequence"</span>: pred[<span class="hljs-string">"sequence"</span>],
	<span class="hljs-meta">... </span> }
	<span class="hljs-meta">... </span> <span class="hljs-keyword">for</span> pred <span class="hljs-keyword">in</span> preds
	<span class="hljs-meta">... </span>]
	<span class="hljs-meta">>>> </span>preds
	[{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.2236</span>,
	<span class="hljs-string">'token'</span>: <span class="hljs-number">1761</span>,
	<span class="hljs-string">'token_str'</span>: <span class="hljs-string">' platform'</span>,
	<span class="hljs-string">'sequence'</span>: <span class="hljs-string">'Hugging Face is a community-based open-source platform for machine learning.'</span>}]<!-- HTML_TAG_END --></pre></div></li></ul> <p data-svelte-h="svelte-sjudlh">이 페이지를 통해 각 모달리티의 다양한 작업 유형과 각 작업의 실용적 중요성에 대해 추가적인 배경 정보를 얻으셨기를 바랍니다. 다음 <a href="tasks_explained">섹션</a>에서는 🤗 Transformer가 이러한 작업을 해결하는 <strong>방법</strong>에 대해 알아보실 수 있습니다.</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/ko/task_summary.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_5r6mt7 = {
	assets: "/docs/transformers/pr_31579/ko",
	base: "/docs/transformers/pr_31579/ko",
	env: {}
	};

	const element = document.currentScript.parentElement;

	const data = [null,null];

	Promise.all([
	import("/docs/transformers/pr_31579/ko/_app/immutable/entry/start.7687ccc8.js"),
	import("/docs/transformers/pr_31579/ko/_app/immutable/entry/app.c00676df.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 58],
	data,
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 79.4 kB
Xet hash:: 7482a813ca29ea190906ee6cef5ed352bbf9d40fcf2832012a2a981de10066d6

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