Buckets:

hf-doc-build
/

doc-dev

hf-doc-build/doc-dev / transformers /pr_35339 /ko /_app /immutable /nodes /131.fcb294ea.js

rtrm's picture

3 months ago

89.7 kB

	import{s as wn,o as Tn,n as un}from"../chunks/scheduler.bdbef820.js";import{S as Cn,i as bn,g as p,s as t,r,A as fn,h as i,f as l,c as e,j as yn,u as j,x as c,k as _l,y as In,a as n,v as M,d as o,t as h,w as g}from"../chunks/index.33f81d56.js";import{T as _n}from"../chunks/Tip.34194030.js";import{C as x}from"../chunks/CodeBlock.362b34a4.js";import{D as Vn}from"../chunks/DocNotebookDropdown.d5db5928.js";import{H as y,E as kn}from"../chunks/EditOnGithub.a9246e21.js";function Rn(Os){let m,w='이 작업과 호환되는 모든 아키텍처와 체크포인트를 보려면 <a href="https://huggingface.co/tasks/image-to-text" rel="nofollow">작업 페이지</a>를 확인하는 것이 좋습니다.';return{c(){m=p("p"),m.innerHTML=w},l(U){m=i(U,"P",{"data-svelte-h":!0}),c(m)!=="svelte-17cmma8"&&(m.innerHTML=w)},m(U,Ds){n(U,m,Ds)},p:un,d(U){U&&l(m)}}}function Zn(Os){let m,w,U,Ds,T,Ps,u,Ks,C,Vl=`문서 시각적 질의 응답(Document Visual Question Answering)이라고도 하는
	문서 질의 응답(Document Question Answering)은 문서 이미지에 대한 질문에 답변을 주는 태스크입니다.
	이 태스크를 지원하는 모델의 입력은 일반적으로 이미지와 질문의 조합이고, 출력은 자연어로 된 답변입니다. 이러한 모델은 텍스트, 단어의 위치(바운딩 박스), 이미지 등 다양한 모달리티를 활용합니다.`,sa,b,kl="이 가이드는 다음 내용을 설명합니다:",aa,f,Rl='<li><a href="https://huggingface.co/datasets/nielsr/docvqa_1200_examples_donut" rel="nofollow">DocVQA dataset</a>을 사용해 <a href="../model_doc/layoutlmv2">LayoutLMv2</a> 미세 조정하기</li> <li>추론을 위해 미세 조정된 모델을 사용하기</li>',la,J,na,I,Zl=`LayoutLMv2는 토큰의 마지막 은닉층 위에 질의 응답 헤드를 추가해 답변의 시작 토큰과 끝 토큰의 위치를 예측함으로써 문서 질의 응답 태스크를 해결합니다. 즉, 문맥이 주어졌을 때 질문에 답하는 정보를 추출하는 추출형 질의 응답(Extractive question answering)으로 문제를 처리합니다.
	문맥은 OCR 엔진의 출력에서 가져오며, 여기서는 Google의 Tesseract를 사용합니다.`,ta,_,Ql="시작하기 전에 필요한 라이브러리가 모두 설치되어 있는지 확인하세요. LayoutLMv2는 detectron2, torchvision 및 테서랙트를 필요로 합니다.",ea,V,pa,k,ia,R,ca,Z,Al="필요한 라이브러리들을 모두 설치한 후 런타임을 다시 시작합니다.",ra,Q,$l=`커뮤니티에 당신의 모델을 공유하는 것을 권장합니다. Hugging Face 계정에 로그인해서 모델을 🤗 Hub에 업로드하세요.
	프롬프트가 실행되면, 로그인을 위해 토큰을 입력하세요:`,ja,A,Ma,$,Xl="몇 가지 전역 변수를 정의해 보겠습니다.",oa,X,ha,v,ga,N,vl=`이 가이드에서는 🤗 Hub에서 찾을 수 있는 전처리된 DocVQA의 작은 샘플을 사용합니다.
	DocVQA의 전체 데이터 세트를 사용하고 싶다면, <a href="https://rrc.cvc.uab.es/?ch=17" rel="nofollow">DocVQA homepage</a>에 가입 후 다운로드 할 수 있습니다. 전체 데이터 세트를 다운로드 했다면, 이 가이드를 계속 진행하기 위해 <a href="https://huggingface.co/docs/datasets/loading#local-and-remote-files" rel="nofollow">🤗 dataset에 파일을 가져오는 방법</a>을 확인하세요.`,xa,G,ma,E,Nl="보시다시피, 데이터 세트는 이미 훈련 세트와 테스트 세트로 나누어져 있습니다. 무작위로 예제를 살펴보면서 특성을 확인해보세요.",Ua,B,Ja,F,Gl="각 필드가 나타내는 내용은 다음과 같습니다:",da,W,El="<li><code>id</code>: 예제의 id</li> <li><code>image</code>: 문서 이미지를 포함하는 PIL.Image.Image 객체</li> <li><code>query</code>: 질문 문자열 - 여러 언어의 자연어로 된 질문</li> <li><code>answers</code>: 사람이 주석을 단 정답 리스트</li> <li><code>words</code> and <code>bounding_boxes</code>: OCR의 결과값들이며 이 가이드에서는 사용하지 않을 예정</li> <li><code>answer</code>: 다른 모델과 일치하는 답변이며 이 가이드에서는 사용하지 않을 예정</li>",ya,z,Bl=`영어로 된 질문만 남기고 다른 모델에 대한 예측을 포함하는 <code>answer</code> 특성을 삭제하겠습니다.
	그리고 주석 작성자가 제공한 데이터 세트에서 첫 번째 답변을 가져옵니다. 또는 무작위로 샘플을 추출할 수도 있습니다.`,wa,q,Ta,Y,Fl=`이 가이드에서 사용하는 LayoutLMv2 체크포인트는 <code>max_position_embeddings = 512</code>로 훈련되었습니다(이 정보는 <a href="https://huggingface.co/microsoft/layoutlmv2-base-uncased/blob/main/config.json#L18" rel="nofollow">체크포인트의 <code>config.json</code> 파일</a>에서 확인할 수 있습니다).
	바로 예제를 잘라낼 수도 있지만, 긴 문서의 끝에 답변이 있어 잘리는 상황을 피하기 위해 여기서는 임베딩이 512보다 길어질 가능성이 있는 몇 가지 예제를 제거하겠습니다.
	데이터 세트에 있는 대부분의 문서가 긴 경우 슬라이딩 윈도우 방법을 사용할 수 있습니다 - 자세한 내용을 확인하고 싶으면 이 <a href="https://github.com/huggingface/notebooks/blob/main/examples/question_answering.ipynb" rel="nofollow">노트북</a>을 확인하세요.`,ua,S,Ca,H,Wl=`이 시점에서 이 데이터 세트의 OCR 특성도 제거해 보겠습니다. OCR 특성은 다른 모델을 미세 조정하기 위한 것으로, 이 가이드에서 사용하는 모델의 입력 요구 사항과 일치하지 않기 때문에 이 특성을 사용하기 위해서는 일부 처리가 필요합니다.
	대신, 원본 데이터에 <code>LayoutLMv2Processor</code>를 사용하여 OCR 및 토큰화를 모두 수행할 수 있습니다.
	이렇게 하면 모델이 요구하는 입력을 얻을 수 있습니다.
	이미지를 수동으로 처리하려면, <a href="../model_doc/layoutlmv2"><code>LayoutLMv2</code> model documentation</a>에서 모델이 요구하는 입력 포맷을 확인해보세요.`,ba,D,fa,L,zl="마지막으로, 데이터 탐색을 완료하기 위해 이미지 예시를 살펴봅시다.",Ia,O,_a,d,ql='<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/docvqa_example.jpg" alt="DocVQA Image Example"/>',Va,P,ka,K,Yl=`문서 질의 응답 태스크는 멀티모달 태스크이며, 각 모달리티의 입력이 모델의 요구에 맞게 전처리 되었는지 확인해야 합니다.
	이미지 데이터를 처리할 수 있는 이미지 프로세서와 텍스트 데이터를 인코딩할 수 있는 토크나이저를 결합한 <code>LayoutLMv2Processor</code>를 가져오는 것부터 시작해 보겠습니다.`,Ra,ss,Za,as,Qa,ls,Sl=`먼저, 프로세서의 <code>image_processor</code>를 사용해 모델에 대한 문서 이미지를 준비해 보겠습니다.
	기본값으로, 이미지 프로세서는 이미지 크기를 224x224로 조정하고 색상 채널의 순서가 올바른지 확인한 후 단어와 정규화된 바운딩 박스를 얻기 위해 테서랙트를 사용해 OCR를 적용합니다.
	이 튜토리얼에서 우리가 필요한 것과 기본값은 완전히 동일합니다. 이미지 배치에 기본 이미지 처리를 적용하고 OCR의 결과를 변환하는 함수를 작성합니다.`,Aa,ns,$a,ts,Hl="이 전처리를 데이터 세트 전체에 빠르게 적용하려면 <code>map</code>를 사용하세요.",Xa,es,va,ps,Na,is,Dl=`이미지에 OCR을 적용했으면 데이터 세트의 텍스트 부분을 모델에 맞게 인코딩해야 합니다.
	이 인코딩에는 이전 단계에서 가져온 단어와 박스를 토큰 수준의 <code>input_ids</code>, <code>attention_mask</code>, <code>token_type_ids</code> 및 <code>bbox</code>로 변환하는 작업이 포함됩니다.
	텍스트를 전처리하려면 프로세서의 <code>tokenizer</code>가 필요합니다.`,Ga,cs,Ea,rs,Ll="위에서 언급한 전처리 외에도 모델을 위해 레이블을 추가해야 합니다. 🤗 Transformers의 <code>xxxForQuestionAnswering</code> 모델의 경우, 레이블은 <code>start_positions</code>와 <code>end_positions</code>로 구성되며 어떤 토큰이 답변의 시작과 끝에 있는지를 나타냅니다.",Ba,js,Ol="레이블 추가를 위해서, 먼저 더 큰 리스트(단어 리스트)에서 하위 리스트(단어로 분할된 답변)을 찾을 수 있는 헬퍼 함수를 정의합니다.",Fa,Ms,Pl=`이 함수는 <code>words_list</code>와 <code>answer_list</code>, 이렇게 두 리스트를 입력으로 받습니다.
	그런 다음 <code>words_list</code>를 반복하여 <code>words_list</code>의 현재 단어(words_list[i])가 <code>answer_list</code>의 첫 번째 단어(answer_list[0])와 같은지,
	현재 단어에서 시작해 <code>answer_list</code>와 같은 길이만큼의 <code>words_list</code>의 하위 리스트가 <code>answer_list</code>와 일치하는지 확인합니다.
	이 조건이 참이라면 일치하는 항목을 발견했음을 의미하며, 함수는 일치 항목, 시작 인덱스(idx) 및 종료 인덱스(idx + len(answer_list) - 1)를 기록합니다. 일치하는 항목이 두 개 이상 발견되면 함수는 첫 번째 항목만 반환합니다. 일치하는 항목이 없다면 함수는 (<code>None</code>, 0, 0)을 반환합니다.`,Wa,os,za,hs,Kl="이 함수가 어떻게 정답의 위치를 찾는지 설명하기 위해 다음 예제에서 함수를 사용해 보겠습니다:",qa,gs,Ya,xs,sn="한편, 위 예제가 인코딩되면 다음과 같이 표시됩니다:",Sa,ms,Ha,Us,an="이제 인코딩된 입력에서 정답의 위치를 찾아야 합니다.",Da,Js,ln="<li><code>token_type_ids</code>는 어떤 토큰이 질문에 속하는지, 그리고 어떤 토큰이 문서의 단어에 포함되는지를 알려줍니다.</li> <li><code>tokenizer.cls_token_id</code> 입력의 시작 부분에 있는 특수 토큰을 찾는 데 도움을 줍니다.</li> <li><code>word_ids</code>는 원본 <code>words</code>에서 찾은 답변을 전체 인코딩된 입력의 동일한 답과 일치시키고 인코딩된 입력에서 답변의 시작/끝 위치를 결정합니다.</li>",La,ds,nn="위 내용들을 염두에 두고 데이터 세트 예제의 배치를 인코딩하는 함수를 만들어 보겠습니다:",Oa,ys,Pa,ws,tn="이제 이 전처리 함수가 있으니 전체 데이터 세트를 인코딩할 수 있습니다:",Ka,Ts,sl,us,en="인코딩된 데이터 세트의 특성이 어떻게 생겼는지 확인해 보겠습니다:",al,Cs,ll,bs,nl,fs,pn=`문서 질의 응답을 평가하려면 상당한 양의 후처리가 필요합니다. 시간이 너무 많이 걸리지 않도록 이 가이드에서는 평가 단계를 생략합니다.
	<a href="/docs/transformers/pr_35339/ko/main_classes/trainer#transformers.Trainer">Trainer</a>가 훈련 과정에서 평가 손실(evaluation loss)을 계속 계산하기 때문에 모델의 성능을 대략적으로 알 수 있습니다.
	추출적(Extractive) 질의 응답은 보통 F1/exact match 방법을 사용해 평가됩니다.
	직접 구현해보고 싶으시다면, Hugging Face course의 <a href="https://huggingface.co/course/chapter7/7?fw=pt#postprocessing" rel="nofollow">Question Answering chapter</a>을 참고하세요.`,tl,Is,el,_s,cn=`축하합니다! 이 가이드의 가장 어려운 부분을 성공적으로 처리했으니 이제 나만의 모델을 훈련할 준비가 되었습니다.
	훈련은 다음과 같은 단계로 이루어져 있습니다:`,pl,Vs,rn='<li>전처리에서의 동일한 체크포인트를 사용하기 위해 <a href="/docs/transformers/pr_35339/ko/model_doc/auto#transformers.AutoModelForDocumentQuestionAnswering">AutoModelForDocumentQuestionAnswering</a>으로 모델을 가져옵니다.</li> <li><a href="/docs/transformers/pr_35339/ko/main_classes/trainer#transformers.TrainingArguments">TrainingArguments</a>로 훈련 하이퍼파라미터를 정합니다.</li> <li>예제를 배치 처리하는 함수를 정의합니다. 여기서는 <a href="/docs/transformers/pr_35339/ko/main_classes/data_collator#transformers.DefaultDataCollator">DefaultDataCollator</a>가 적당합니다.</li> <li>모델, 데이터 세트, 데이터 콜레이터(Data collator)와 함께 <a href="/docs/transformers/pr_35339/ko/main_classes/trainer#transformers.Trainer">Trainer</a>에 훈련 인수들을 전달합니다.</li> <li><a href="/docs/transformers/pr_35339/ko/main_classes/trainer#transformers.Trainer.train">train()</a>을 호출해서 모델을 미세 조정합니다.</li>',il,ks,cl,Rs,jn=`<a href="/docs/transformers/pr_35339/ko/main_classes/trainer#transformers.TrainingArguments">TrainingArguments</a>에서 <code>output_dir</code>을 사용하여 모델을 저장할 위치를 지정하고, 적절한 하이퍼파라미터를 설정합니다.
	모델을 커뮤니티와 공유하려면 <code>push_to_hub</code>를 <code>True</code>로 설정하세요 (모델을 업로드하려면 Hugging Face에 로그인해야 합니다).
	이 경우 <code>output_dir</code>은 모델의 체크포인트를 푸시할 레포지토리의 이름이 됩니다.`,rl,Zs,jl,Qs,Mn="간단한 데이터 콜레이터를 정의하여 예제를 함께 배치합니다.",Ml,As,ol,$s,on='마지막으로, 모든 것을 한 곳에 모아 <a href="/docs/transformers/pr_35339/ko/main_classes/trainer#transformers.Trainer.train">train()</a>을 호출합니다:',hl,Xs,gl,vs,hn="최종 모델을 🤗 Hub에 추가하려면, 모델 카드를 생성하고 <code>push_to_hub</code>를 호출합니다:",xl,Ns,ml,Gs,Ul,Es,gn=`이제 LayoutLMv2 모델을 미세 조정하고 🤗 Hub에 업로드했으니 추론에도 사용할 수 있습니다.
	추론을 위해 미세 조정된 모델을 사용해 보는 가장 간단한 방법은 <code>Pipeline</code>을 사용하는 것 입니다.`,Jl,Bs,xn="예를 들어 보겠습니다:",dl,Fs,yl,Ws,mn="그 다음, 모델로 문서 질의 응답을 하기 위해 파이프라인을 인스턴스화하고 이미지 + 질문 조합을 전달합니다.",wl,zs,Tl,qs,Un="원한다면 파이프라인의 결과를 수동으로 복제할 수도 있습니다:",ul,Ys,Jn="<li>이미지와 질문을 가져와 모델의 프로세서를 사용해 모델에 맞게 준비합니다.</li> <li>모델을 통해 결과 또는 전처리를 전달합니다.</li> <li>모델은 어떤 토큰이 답변의 시작에 있는지, 어떤 토큰이 답변이 끝에 있는지를 나타내는 <code>start_logits</code>와 <code>end_logits</code>를 반환합니다. 둘 다 (batch_size, sequence_length) 형태를 갖습니다.</li> <li><code>start_logits</code>와 <code>end_logits</code>의 마지막 차원을 최대로 만드는 값을 찾아 예상 <code>start_idx</code>와 <code>end_idx</code>를 얻습니다.</li> <li>토크나이저로 답변을 디코딩합니다.</li>",Cl,Ss,bl,Hs,fl,Ls,Il;return T=new y({props:{title:"문서 질의 응답(Document Question Answering)",local:"document_question_answering",headingTag:"h1"}}),u=new Vn({props:{classNames:"absolute z-10 right-0 top-0",options:[{label:"Mixed",value:"https://colab.research.google.com/github/huggingface/notebooks/blob/main/transformers_doc/ko/document_question_answering.ipynb"},{label:"PyTorch",value:"https://colab.research.google.com/github/huggingface/notebooks/blob/main/transformers_doc/ko/pytorch/document_question_answering.ipynb"},{label:"TensorFlow",value:"https://colab.research.google.com/github/huggingface/notebooks/blob/main/transformers_doc/ko/tensorflow/document_question_answering.ipynb"},{label:"Mixed",value:"https://studiolab.sagemaker.aws/import/github/huggingface/notebooks/blob/main/transformers_doc/ko/document_question_answering.ipynb"},{label:"PyTorch",value:"https://studiolab.sagemaker.aws/import/github/huggingface/notebooks/blob/main/transformers_doc/ko/pytorch/document_question_answering.ipynb"},{label:"TensorFlow",value:"https://studiolab.sagemaker.aws/import/github/huggingface/notebooks/blob/main/transformers_doc/ko/tensorflow/document_question_answering.ipynb"}]}}),J=new _n({props:{$$slots:{default:[Rn]},$$scope:{ctx:Os}}}),V=new x({props:{code:"cGlwJTIwaW5zdGFsbCUyMC1xJTIwdHJhbnNmb3JtZXJzJTIwZGF0YXNldHM=",highlighted:"pip install -q transformers datasets",wrap:!1}}),k=new x({props:{code:"cGlwJTIwaW5zdGFsbCUyMCdnaXQlMkJodHRwcyUzQSUyRiUyRmdpdGh1Yi5jb20lMkZmYWNlYm9va3Jlc2VhcmNoJTJGZGV0ZWN0cm9uMi5naXQnJTBBcGlwJTIwaW5zdGFsbCUyMHRvcmNodmlzaW9u",highlighted:`pip install <span class="hljs-string">'git+https://github.com/facebookresearch/detectron2.git'</span>
	pip install torchvision`,wrap:!1}}),R=new x({props:{code:"c3VkbyUyMGFwdCUyMGluc3RhbGwlMjB0ZXNzZXJhY3Qtb2NyJTBBcGlwJTIwaW5zdGFsbCUyMC1xJTIwcHl0ZXNzZXJhY3Q=",highlighted:`sudo apt install tesseract-ocr
	pip install -q pytesseract`,wrap:!1}}),A=new x({props:{code:"ZnJvbSUyMGh1Z2dpbmdmYWNlX2h1YiUyMGltcG9ydCUyMG5vdGVib29rX2xvZ2luJTBBJTBBbm90ZWJvb2tfbG9naW4oKQ==",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> huggingface_hub <span class="hljs-keyword">import</span> notebook_login

	<span class="hljs-meta">>>> </span>notebook_login()`,wrap:!1}}),X=new x({props:{code:"bW9kZWxfY2hlY2twb2ludCUyMCUzRCUyMCUyMm1pY3Jvc29mdCUyRmxheW91dGxtdjItYmFzZS11bmNhc2VkJTIyJTBBYmF0Y2hfc2l6ZSUyMCUzRCUyMDQ=",highlighted:`<span class="hljs-meta">>>> </span>model_checkpoint = <span class="hljs-string">"microsoft/layoutlmv2-base-uncased"</span>
	<span class="hljs-meta">>>> </span>batch_size = <span class="hljs-number">4</span>`,wrap:!1}}),v=new y({props:{title:"데이터 불러오기",local:"load-the-data",headingTag:"h2"}}),G=new x({props:{code:"ZnJvbSUyMGRhdGFzZXRzJTIwaW1wb3J0JTIwbG9hZF9kYXRhc2V0JTBBJTBBZGF0YXNldCUyMCUzRCUyMGxvYWRfZGF0YXNldCglMjJuaWVsc3IlMkZkb2N2cWFfMTIwMF9leGFtcGxlcyUyMiklMEFkYXRhc2V0",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset

	<span class="hljs-meta">>>> </span>dataset = load_dataset(<span class="hljs-string">"nielsr/docvqa_1200_examples"</span>)
	<span class="hljs-meta">>>> </span>dataset
	DatasetDict({
	train: Dataset({
	features: [<span class="hljs-string">'id'</span>, <span class="hljs-string">'image'</span>, <span class="hljs-string">'query'</span>, <span class="hljs-string">'answers'</span>, <span class="hljs-string">'words'</span>, <span class="hljs-string">'bounding_boxes'</span>, <span class="hljs-string">'answer'</span>],
	num_rows: <span class="hljs-number">1000</span>
	})
	test: Dataset({
	features: [<span class="hljs-string">'id'</span>, <span class="hljs-string">'image'</span>, <span class="hljs-string">'query'</span>, <span class="hljs-string">'answers'</span>, <span class="hljs-string">'words'</span>, <span class="hljs-string">'bounding_boxes'</span>, <span class="hljs-string">'answer'</span>],
	num_rows: <span class="hljs-number">200</span>
	})
	})`,wrap:!1}}),B=new x({props:{code:"ZGF0YXNldCU1QiUyMnRyYWluJTIyJTVELmZlYXR1cmVz",highlighted:'<span class="hljs-meta">>>> </span>dataset[<span class="hljs-string">"train"</span>].features',wrap:!1}}),q=new x({props:{code:"dXBkYXRlZF9kYXRhc2V0JTIwJTNEJTIwZGF0YXNldC5tYXAobGFtYmRhJTIwZXhhbXBsZSUzQSUyMCU3QiUyMnF1ZXN0aW9uJTIyJTNBJTIwZXhhbXBsZSU1QiUyMnF1ZXJ5JTIyJTVEJTVCJTIyZW4lMjIlNUQlN0QlMkMlMjByZW1vdmVfY29sdW1ucyUzRCU1QiUyMnF1ZXJ5JTIyJTVEKSUwQXVwZGF0ZWRfZGF0YXNldCUyMCUzRCUyMHVwZGF0ZWRfZGF0YXNldC5tYXAoJTBBJTIwJTIwJTIwJTIwbGFtYmRhJTIwZXhhbXBsZSUzQSUyMCU3QiUyMmFuc3dlciUyMiUzQSUyMGV4YW1wbGUlNUIlMjJhbnN3ZXJzJTIyJTVEJTVCMCU1RCU3RCUyQyUyMHJlbW92ZV9jb2x1bW5zJTNEJTVCJTIyYW5zd2VyJTIyJTJDJTIwJTIyYW5zd2VycyUyMiU1RCUwQSk=",highlighted:`<span class="hljs-meta">>>> </span>updated_dataset = dataset.<span class="hljs-built_in">map</span>(<span class="hljs-keyword">lambda</span> example: {<span class="hljs-string">"question"</span>: example[<span class="hljs-string">"query"</span>][<span class="hljs-string">"en"</span>]}, remove_columns=[<span class="hljs-string">"query"</span>])
	<span class="hljs-meta">>>> </span>updated_dataset = updated_dataset.<span class="hljs-built_in">map</span>(
	<span class="hljs-meta">... </span> <span class="hljs-keyword">lambda</span> example: {<span class="hljs-string">"answer"</span>: example[<span class="hljs-string">"answers"</span>][<span class="hljs-number">0</span>]}, remove_columns=[<span class="hljs-string">"answer"</span>, <span class="hljs-string">"answers"</span>]
	<span class="hljs-meta">... </span>)`,wrap:!1}}),S=new x({props:{code:"dXBkYXRlZF9kYXRhc2V0JTIwJTNEJTIwdXBkYXRlZF9kYXRhc2V0LmZpbHRlcihsYW1iZGElMjB4JTNBJTIwbGVuKHglNUIlMjJ3b3JkcyUyMiU1RCklMjAlMkIlMjBsZW4oeCU1QiUyMnF1ZXN0aW9uJTIyJTVELnNwbGl0KCkpJTIwJTNDJTIwNTEyKQ==",highlighted:'<span class="hljs-meta">>>> </span>updated_dataset = updated_dataset.<span class="hljs-built_in">filter</span>(<span class="hljs-keyword">lambda</span> x: <span class="hljs-built_in">len</span>(x[<span class="hljs-string">"words"</span>]) + <span class="hljs-built_in">len</span>(x[<span class="hljs-string">"question"</span>].split()) < <span class="hljs-number">512</span>)',wrap:!1}}),D=new x({props:{code:"dXBkYXRlZF9kYXRhc2V0JTIwJTNEJTIwdXBkYXRlZF9kYXRhc2V0LnJlbW92ZV9jb2x1bW5zKCUyMndvcmRzJTIyKSUwQXVwZGF0ZWRfZGF0YXNldCUyMCUzRCUyMHVwZGF0ZWRfZGF0YXNldC5yZW1vdmVfY29sdW1ucyglMjJib3VuZGluZ19ib3hlcyUyMik=",highlighted:`<span class="hljs-meta">>>> </span>updated_dataset = updated_dataset.remove_columns(<span class="hljs-string">"words"</span>)
	<span class="hljs-meta">>>> </span>updated_dataset = updated_dataset.remove_columns(<span class="hljs-string">"bounding_boxes"</span>)`,wrap:!1}}),O=new x({props:{code:"dXBkYXRlZF9kYXRhc2V0JTVCJTIydHJhaW4lMjIlNUQlNUIxMSU1RCU1QiUyMmltYWdlJTIyJTVE",highlighted:'<span class="hljs-meta">>>> </span>updated_dataset[<span class="hljs-string">"train"</span>][<span class="hljs-number">11</span>][<span class="hljs-string">"image"</span>]',wrap:!1}}),P=new y({props:{title:"데이터 전처리",local:"preprocess-the-data",headingTag:"h2"}}),ss=new x({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMEF1dG9Qcm9jZXNzb3IlMEElMEFwcm9jZXNzb3IlMjAlM0QlMjBBdXRvUHJvY2Vzc29yLmZyb21fcHJldHJhaW5lZChtb2RlbF9jaGVja3BvaW50KQ==",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor

	<span class="hljs-meta">>>> </span>processor = AutoProcessor.from_pretrained(model_checkpoint)`,wrap:!1}}),as=new y({props:{title:"문서 이미지 전처리",local:"preprocessing-document-images",headingTag:"h3"}}),ns=new x({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span>image_processor = processor.image_processor


	<span class="hljs-meta">>>> </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">get_ocr_words_and_boxes</span>(<span class="hljs-params">examples</span>):
	<span class="hljs-meta">... </span> images = [image.convert(<span class="hljs-string">"RGB"</span>) <span class="hljs-keyword">for</span> image <span class="hljs-keyword">in</span> examples[<span class="hljs-string">"image"</span>]]
	<span class="hljs-meta">... </span> encoded_inputs = image_processor(images)

	<span class="hljs-meta">... </span> examples[<span class="hljs-string">"image"</span>] = encoded_inputs.pixel_values
	<span class="hljs-meta">... </span> examples[<span class="hljs-string">"words"</span>] = encoded_inputs.words
	<span class="hljs-meta">... </span> examples[<span class="hljs-string">"boxes"</span>] = encoded_inputs.boxes

	<span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> examples`,wrap:!1}}),es=new x({props:{code:"ZGF0YXNldF93aXRoX29jciUyMCUzRCUyMHVwZGF0ZWRfZGF0YXNldC5tYXAoZ2V0X29jcl93b3Jkc19hbmRfYm94ZXMlMkMlMjBiYXRjaGVkJTNEVHJ1ZSUyQyUyMGJhdGNoX3NpemUlM0QyKQ==",highlighted:'<span class="hljs-meta">>>> </span>dataset_with_ocr = updated_dataset.<span class="hljs-built_in">map</span>(get_ocr_words_and_boxes, batched=<span class="hljs-literal">True</span>, batch_size=<span class="hljs-number">2</span>)',wrap:!1}}),ps=new y({props:{title:"텍스트 데이터 전처리",local:"preprocessing-text-data",headingTag:"h3"}}),cs=new x({props:{code:"dG9rZW5pemVyJTIwJTNEJTIwcHJvY2Vzc29yLnRva2VuaXplcg==",highlighted:'<span class="hljs-meta">>>> </span>tokenizer = processor.tokenizer',wrap:!1}}),os=new x({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">subfinder</span>(<span class="hljs-params">words_list, answer_list</span>):
	<span class="hljs-meta">... </span> matches = []
	<span class="hljs-meta">... </span> start_indices = []
	<span class="hljs-meta">... </span> end_indices = []
	<span class="hljs-meta">... </span> <span class="hljs-keyword">for</span> idx, i <span class="hljs-keyword">in</span> <span class="hljs-built_in">enumerate</span>(<span class="hljs-built_in">range</span>(<span class="hljs-built_in">len</span>(words_list))):
	<span class="hljs-meta">... </span> <span class="hljs-keyword">if</span> words_list[i] == answer_list[<span class="hljs-number">0</span>] <span class="hljs-keyword">and</span> words_list[i : i + <span class="hljs-built_in">len</span>(answer_list)] == answer_list:
	<span class="hljs-meta">... </span> matches.append(answer_list)
	<span class="hljs-meta">... </span> start_indices.append(idx)
	<span class="hljs-meta">... </span> end_indices.append(idx + <span class="hljs-built_in">len</span>(answer_list) - <span class="hljs-number">1</span>)
	<span class="hljs-meta">... </span> <span class="hljs-keyword">if</span> matches:
	<span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> matches[<span class="hljs-number">0</span>], start_indices[<span class="hljs-number">0</span>], end_indices[<span class="hljs-number">0</span>]
	<span class="hljs-meta">... </span> <span class="hljs-keyword">else</span>:
	<span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> <span class="hljs-literal">None</span>, <span class="hljs-number">0</span>, <span class="hljs-number">0</span>`,wrap:!1}}),gs=new x({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span>example = dataset_with_ocr[<span class="hljs-string">"train"</span>][<span class="hljs-number">1</span>]
	<span class="hljs-meta">>>> </span>words = [word.lower() <span class="hljs-keyword">for</span> word <span class="hljs-keyword">in</span> example[<span class="hljs-string">"words"</span>]]
	<span class="hljs-meta">>>> </span><span class="hljs-keyword">match</span>, word_idx_start, word_idx_end = subfinder(words, example[<span class="hljs-string">"answer"</span>].lower().split())
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(<span class="hljs-string">"Question: "</span>, example[<span class="hljs-string">"question"</span>])
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(<span class="hljs-string">"Words:"</span>, words)
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(<span class="hljs-string">"Answer: "</span>, example[<span class="hljs-string">"answer"</span>])
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(<span class="hljs-string">"start_index"</span>, word_idx_start)
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(<span class="hljs-string">"end_index"</span>, word_idx_end)
	Question: Who <span class="hljs-keyword">is</span> <span class="hljs-keyword">in</span> cc <span class="hljs-keyword">in</span> this letter?
	Words: [<span class="hljs-string">'wie'</span>, <span class="hljs-string">'baw'</span>, <span class="hljs-string">'brown'</span>, <span class="hljs-string">'&'</span>, <span class="hljs-string">'williamson'</span>, <span class="hljs-string">'tobacco'</span>, <span class="hljs-string">'corporation'</span>, <span class="hljs-string">'research'</span>, <span class="hljs-string">'&'</span>, <span class="hljs-string">'development'</span>, <span class="hljs-string">'internal'</span>, <span class="hljs-string">'correspondence'</span>, <span class="hljs-string">'to:'</span>, <span class="hljs-string">'r.'</span>, <span class="hljs-string">'h.'</span>, <span class="hljs-string">'honeycutt'</span>, <span class="hljs-string">'ce:'</span>, <span class="hljs-string">'t.f.'</span>, <span class="hljs-string">'riehl'</span>, <span class="hljs-string">'from:'</span>, <span class="hljs-string">'.'</span>, <span class="hljs-string">'c.j.'</span>, <span class="hljs-string">'cook'</span>, <span class="hljs-string">'date:'</span>, <span class="hljs-string">'may'</span>, <span class="hljs-string">'8,'</span>, <span class="hljs-string">'1995'</span>, <span class="hljs-string">'subject:'</span>, <span class="hljs-string">'review'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'existing'</span>, <span class="hljs-string">'brainstorming'</span>, <span class="hljs-string">'ideas/483'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'major'</span>, <span class="hljs-string">'function'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'product'</span>, <span class="hljs-string">'innovation'</span>, <span class="hljs-string">'graup'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'to'</span>, <span class="hljs-string">'develop'</span>, <span class="hljs-string">'marketable'</span>, <span class="hljs-string">'nove!'</span>, <span class="hljs-string">'products'</span>, <span class="hljs-string">'that'</span>, <span class="hljs-string">'would'</span>, <span class="hljs-string">'be'</span>, <span class="hljs-string">'profitable'</span>, <span class="hljs-string">'to'</span>, <span class="hljs-string">'manufacture'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'sell.'</span>, <span class="hljs-string">'novel'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'defined'</span>, <span class="hljs-string">'as:'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'a'</span>, <span class="hljs-string">'new'</span>, <span class="hljs-string">'kind,'</span>, <span class="hljs-string">'or'</span>, <span class="hljs-string">'different'</span>, <span class="hljs-string">'from'</span>, <span class="hljs-string">'anything'</span>, <span class="hljs-string">'seen'</span>, <span class="hljs-string">'or'</span>, <span class="hljs-string">'known'</span>, <span class="hljs-string">'before.'</span>, <span class="hljs-string">'innovation'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'defined'</span>, <span class="hljs-string">'as:'</span>, <span class="hljs-string">'something'</span>, <span class="hljs-string">'new'</span>, <span class="hljs-string">'or'</span>, <span class="hljs-string">'different'</span>, <span class="hljs-string">'introduced;'</span>, <span class="hljs-string">'act'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'innovating;'</span>, <span class="hljs-string">'introduction'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'new'</span>, <span class="hljs-string">'things'</span>, <span class="hljs-string">'or'</span>, <span class="hljs-string">'methods.'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'products'</span>, <span class="hljs-string">'may'</span>, <span class="hljs-string">'incorporate'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'latest'</span>, <span class="hljs-string">'technologies,'</span>, <span class="hljs-string">'materials'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'know-how'</span>, <span class="hljs-string">'available'</span>, <span class="hljs-string">'to'</span>, <span class="hljs-string">'give'</span>, <span class="hljs-string">'then'</span>, <span class="hljs-string">'a'</span>, <span class="hljs-string">'unique'</span>, <span class="hljs-string">'taste'</span>, <span class="hljs-string">'or'</span>, <span class="hljs-string">'look.'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'first'</span>, <span class="hljs-string">'task'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'product'</span>, <span class="hljs-string">'innovation'</span>, <span class="hljs-string">'group'</span>, <span class="hljs-string">'was'</span>, <span class="hljs-string">'to'</span>, <span class="hljs-string">'assemble,'</span>, <span class="hljs-string">'review'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'categorize'</span>, <span class="hljs-string">'a'</span>, <span class="hljs-string">'list'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'existing'</span>, <span class="hljs-string">'brainstorming'</span>, <span class="hljs-string">'ideas.'</span>, <span class="hljs-string">'ideas'</span>, <span class="hljs-string">'were'</span>, <span class="hljs-string">'grouped'</span>, <span class="hljs-string">'into'</span>, <span class="hljs-string">'two'</span>, <span class="hljs-string">'major'</span>, <span class="hljs-string">'categories'</span>, <span class="hljs-string">'labeled'</span>, <span class="hljs-string">'appearance'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'taste/aroma.'</span>, <span class="hljs-string">'these'</span>, <span class="hljs-string">'categories'</span>, <span class="hljs-string">'are'</span>, <span class="hljs-string">'used'</span>, <span class="hljs-string">'for'</span>, <span class="hljs-string">'novel'</span>, <span class="hljs-string">'products'</span>, <span class="hljs-string">'that'</span>, <span class="hljs-string">'may'</span>, <span class="hljs-string">'differ'</span>, <span class="hljs-string">'from'</span>, <span class="hljs-string">'a'</span>, <span class="hljs-string">'visual'</span>, <span class="hljs-string">'and/or'</span>, <span class="hljs-string">'taste/aroma'</span>, <span class="hljs-string">'point'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'view'</span>, <span class="hljs-string">'compared'</span>, <span class="hljs-string">'to'</span>, <span class="hljs-string">'canventional'</span>, <span class="hljs-string">'cigarettes.'</span>, <span class="hljs-string">'other'</span>, <span class="hljs-string">'categories'</span>, <span class="hljs-string">'include'</span>, <span class="hljs-string">'a'</span>, <span class="hljs-string">'combination'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'the'</span>, <span class="hljs-string">'above,'</span>, <span class="hljs-string">'filters,'</span>, <span class="hljs-string">'packaging'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'brand'</span>, <span class="hljs-string">'extensions.'</span>, <span class="hljs-string">'appearance'</span>, <span class="hljs-string">'this'</span>, <span class="hljs-string">'category'</span>, <span class="hljs-string">'is'</span>, <span class="hljs-string">'used'</span>, <span class="hljs-string">'for'</span>, <span class="hljs-string">'novel'</span>, <span class="hljs-string">'cigarette'</span>, <span class="hljs-string">'constructions'</span>, <span class="hljs-string">'that'</span>, <span class="hljs-string">'yield'</span>, <span class="hljs-string">'visually'</span>, <span class="hljs-string">'different'</span>, <span class="hljs-string">'products'</span>, <span class="hljs-string">'with'</span>, <span class="hljs-string">'minimal'</span>, <span class="hljs-string">'changes'</span>, <span class="hljs-string">'in'</span>, <span class="hljs-string">'smoke'</span>, <span class="hljs-string">'chemistry'</span>, <span class="hljs-string">'two'</span>, <span class="hljs-string">'cigarettes'</span>, <span class="hljs-string">'in'</span>, <span class="hljs-string">'cne.'</span>, <span class="hljs-string">'emulti-plug'</span>, <span class="hljs-string">'te'</span>, <span class="hljs-string">'build'</span>, <span class="hljs-string">'yaur'</span>, <span class="hljs-string">'awn'</span>, <span class="hljs-string">'cigarette.'</span>, <span class="hljs-string">'eswitchable'</span>, <span class="hljs-string">'menthol'</span>, <span class="hljs-string">'or'</span>, <span class="hljs-string">'non'</span>, <span class="hljs-string">'menthol'</span>, <span class="hljs-string">'cigarette.'</span>, <span class="hljs-string">'cigarettes'</span>, <span class="hljs-string">'with'</span>, <span class="hljs-string">'interspaced'</span>, <span class="hljs-string">'perforations'</span>, <span class="hljs-string">'to'</span>, <span class="hljs-string">'enable'</span>, <span class="hljs-string">'smoker'</span>, <span class="hljs-string">'to'</span>, <span class="hljs-string">'separate'</span>, <span class="hljs-string">'unburned'</span>, <span class="hljs-string">'section'</span>, <span class="hljs-string">'for'</span>, <span class="hljs-string">'future'</span>, <span class="hljs-string">'smoking.'</span>, <span class="hljs-string">'«short'</span>, <span class="hljs-string">'cigarette,'</span>, <span class="hljs-string">'tobacco'</span>, <span class="hljs-string">'section'</span>, <span class="hljs-string">'30'</span>, <span class="hljs-string">'mm.'</span>, <span class="hljs-string">'«extremely'</span>, <span class="hljs-string">'fast'</span>, <span class="hljs-string">'buming'</span>, <span class="hljs-string">'cigarette.'</span>, <span class="hljs-string">'«novel'</span>, <span class="hljs-string">'cigarette'</span>, <span class="hljs-string">'constructions'</span>, <span class="hljs-string">'that'</span>, <span class="hljs-string">'permit'</span>, <span class="hljs-string">'a'</span>, <span class="hljs-string">'significant'</span>, <span class="hljs-string">'reduction'</span>, <span class="hljs-string">'iretobacco'</span>, <span class="hljs-string">'weight'</span>, <span class="hljs-string">'while'</span>, <span class="hljs-string">'maintaining'</span>, <span class="hljs-string">'smoking'</span>, <span class="hljs-string">'mechanics'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'visual'</span>, <span class="hljs-string">'characteristics.'</span>, <span class="hljs-string">'higher'</span>, <span class="hljs-string">'basis'</span>, <span class="hljs-string">'weight'</span>, <span class="hljs-string">'paper:'</span>, <span class="hljs-string">'potential'</span>, <span class="hljs-string">'reduction'</span>, <span class="hljs-string">'in'</span>, <span class="hljs-string">'tobacco'</span>, <span class="hljs-string">'weight.'</span>, <span class="hljs-string">'«more'</span>, <span class="hljs-string">'rigid'</span>, <span class="hljs-string">'tobacco'</span>, <span class="hljs-string">'column;'</span>, <span class="hljs-string">'stiffing'</span>, <span class="hljs-string">'agent'</span>, <span class="hljs-string">'for'</span>, <span class="hljs-string">'tobacco;'</span>, <span class="hljs-string">'e.g.'</span>, <span class="hljs-string">'starch'</span>, <span class="hljs-string">'colored'</span>, <span class="hljs-string">'tow'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'cigarette'</span>, <span class="hljs-string">'papers;'</span>, <span class="hljs-string">'seasonal'</span>, <span class="hljs-string">'promotions,'</span>, <span class="hljs-string">'e.g.'</span>, <span class="hljs-string">'pastel'</span>, <span class="hljs-string">'colored'</span>, <span class="hljs-string">'cigarettes'</span>, <span class="hljs-string">'for'</span>, <span class="hljs-string">'easter'</span>, <span class="hljs-string">'or'</span>, <span class="hljs-string">'in'</span>, <span class="hljs-string">'an'</span>, <span class="hljs-string">'ebony'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'ivory'</span>, <span class="hljs-string">'brand'</span>, <span class="hljs-string">'containing'</span>, <span class="hljs-string">'a'</span>, <span class="hljs-string">'mixture'</span>, <span class="hljs-string">'of'</span>, <span class="hljs-string">'all'</span>, <span class="hljs-string">'black'</span>, <span class="hljs-string">'(black'</span>, <span class="hljs-string">'paper'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'tow)'</span>, <span class="hljs-string">'and'</span>, <span class="hljs-string">'ail'</span>, <span class="hljs-string">'white'</span>, <span class="hljs-string">'cigarettes.'</span>, <span class="hljs-string">'499150498'</span>]
	Answer: T.F. Riehl
	start_index <span class="hljs-number">17</span>
	end_index <span class="hljs-number">18</span>`,wrap:!1}}),ms=new x({props:{code:"ZW5jb2RpbmclMjAlM0QlMjB0b2tlbml6ZXIoZXhhbXBsZSU1QiUyMnF1ZXN0aW9uJTIyJTVEJTJDJTIwZXhhbXBsZSU1QiUyMndvcmRzJTIyJTVEJTJDJTIwZXhhbXBsZSU1QiUyMmJveGVzJTIyJTVEKSUwQXRva2VuaXplci5kZWNvZGUoZW5jb2RpbmclNUIlMjJpbnB1dF9pZHMlMjIlNUQp",highlighted:`<span class="hljs-meta">>>> </span>encoding = tokenizer(example[<span class="hljs-string">"question"</span>], example[<span class="hljs-string">"words"</span>], example[<span class="hljs-string">"boxes"</span>])
	<span class="hljs-meta">>>> </span>tokenizer.decode(encoding[<span class="hljs-string">"input_ids"</span>])
	[CLS] who <span class="hljs-keyword">is</span> <span class="hljs-keyword">in</span> cc <span class="hljs-keyword">in</span> this letter? [SEP] wie baw brown & williamson tobacco corporation research & development ...`,wrap:!1}}),ys=new x({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">encode_dataset</span>(<span class="hljs-params">examples, max_length=<span class="hljs-number">512</span></span>):
	<span class="hljs-meta">... </span> questions = examples[<span class="hljs-string">"question"</span>]
	<span class="hljs-meta">... </span> words = examples[<span class="hljs-string">"words"</span>]
	<span class="hljs-meta">... </span> boxes = examples[<span class="hljs-string">"boxes"</span>]
	<span class="hljs-meta">... </span> answers = examples[<span class="hljs-string">"answer"</span>]

	<span class="hljs-meta">... </span> <span class="hljs-comment"># 예제 배치를 인코딩하고 start_positions와 end_positions를 초기화합니다</span>
	<span class="hljs-meta">... </span> encoding = tokenizer(questions, words, boxes, max_length=max_length, padding=<span class="hljs-string">"max_length"</span>, truncation=<span class="hljs-literal">True</span>)
	<span class="hljs-meta">... </span> start_positions = []
	<span class="hljs-meta">... </span> end_positions = []

	<span class="hljs-meta">... </span> <span class="hljs-comment"># 배치의 예제를 반복합니다</span>
	<span class="hljs-meta">... </span> <span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> <span class="hljs-built_in">range</span>(<span class="hljs-built_in">len</span>(questions)):
	<span class="hljs-meta">... </span> cls_index = encoding[<span class="hljs-string">"input_ids"</span>][i].index(tokenizer.cls_token_id)

	<span class="hljs-meta">... </span> <span class="hljs-comment"># 예제의 words에서 답변의 위치를 찾습니다</span>
	<span class="hljs-meta">... </span> words_example = [word.lower() <span class="hljs-keyword">for</span> word <span class="hljs-keyword">in</span> words[i]]
	<span class="hljs-meta">... </span> answer = answers[i]
	<span class="hljs-meta">... </span> <span class="hljs-keyword">match</span>, word_idx_start, word_idx_end = subfinder(words_example, answer.lower().split())

	<span class="hljs-meta">... </span> <span class="hljs-keyword">if</span> <span class="hljs-keyword">match</span>:
	<span class="hljs-meta">... </span> <span class="hljs-comment"># 일치하는 항목을 발견하면, \`token_type_ids\`를 사용해 인코딩에서 단어가 시작하는 위치를 찾습니다</span>
	<span class="hljs-meta">... </span> token_type_ids = encoding[<span class="hljs-string">"token_type_ids"</span>][i]
	<span class="hljs-meta">... </span> token_start_index = <span class="hljs-number">0</span>
	<span class="hljs-meta">... </span> <span class="hljs-keyword">while</span> token_type_ids[token_start_index] != <span class="hljs-number">1</span>:
	<span class="hljs-meta">... </span> token_start_index += <span class="hljs-number">1</span>

	<span class="hljs-meta">... </span> token_end_index = <span class="hljs-built_in">len</span>(encoding[<span class="hljs-string">"input_ids"</span>][i]) - <span class="hljs-number">1</span>
	<span class="hljs-meta">... </span> <span class="hljs-keyword">while</span> token_type_ids[token_end_index] != <span class="hljs-number">1</span>:
	<span class="hljs-meta">... </span> token_end_index -= <span class="hljs-number">1</span>

	<span class="hljs-meta">... </span> word_ids = encoding.word_ids(i)[token_start_index : token_end_index + <span class="hljs-number">1</span>]
	<span class="hljs-meta">... </span> start_position = cls_index
	<span class="hljs-meta">... </span> end_position = cls_index

	<span class="hljs-meta">... </span> <span class="hljs-comment"># words의 답변 위치와 일치할 때까지 word_ids를 반복하고 \`token_start_index\`를 늘립니다</span>
	<span class="hljs-meta">... </span> <span class="hljs-comment"># 일치하면 \`token_start_index\`를 인코딩에서 답변의 \`start_position\`으로 저장합니다</span>
	<span class="hljs-meta">... </span> <span class="hljs-keyword">for</span> <span class="hljs-built_in">id</span> <span class="hljs-keyword">in</span> word_ids:
	<span class="hljs-meta">... </span> <span class="hljs-keyword">if</span> <span class="hljs-built_in">id</span> == word_idx_start:
	<span class="hljs-meta">... </span> start_position = token_start_index
	<span class="hljs-meta">... </span> <span class="hljs-keyword">else</span>:
	<span class="hljs-meta">... </span> token_start_index += <span class="hljs-number">1</span>

	<span class="hljs-meta">... </span> <span class="hljs-comment"># 비슷하게, 끝에서 시작해 \`word_ids\`를 반복하며 답변의 \`end_position\`을 찾습니다</span>
	<span class="hljs-meta">... </span> <span class="hljs-keyword">for</span> <span class="hljs-built_in">id</span> <span class="hljs-keyword">in</span> word_ids[::-<span class="hljs-number">1</span>]:
	<span class="hljs-meta">... </span> <span class="hljs-keyword">if</span> <span class="hljs-built_in">id</span> == word_idx_end:
	<span class="hljs-meta">... </span> end_position = token_end_index
	<span class="hljs-meta">... </span> <span class="hljs-keyword">else</span>:
	<span class="hljs-meta">... </span> token_end_index -= <span class="hljs-number">1</span>

	<span class="hljs-meta">... </span> start_positions.append(start_position)
	<span class="hljs-meta">... </span> end_positions.append(end_position)

	<span class="hljs-meta">... </span> <span class="hljs-keyword">else</span>:
	<span class="hljs-meta">... </span> start_positions.append(cls_index)
	<span class="hljs-meta">... </span> end_positions.append(cls_index)

	<span class="hljs-meta">... </span> encoding[<span class="hljs-string">"image"</span>] = examples[<span class="hljs-string">"image"</span>]
	<span class="hljs-meta">... </span> encoding[<span class="hljs-string">"start_positions"</span>] = start_positions
	<span class="hljs-meta">... </span> encoding[<span class="hljs-string">"end_positions"</span>] = end_positions

	<span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> encoding`,wrap:!1}}),Ts=new x({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span>encoded_train_dataset = dataset_with_ocr[<span class="hljs-string">"train"</span>].<span class="hljs-built_in">map</span>(
	<span class="hljs-meta">... </span> encode_dataset, batched=<span class="hljs-literal">True</span>, batch_size=<span class="hljs-number">2</span>, remove_columns=dataset_with_ocr[<span class="hljs-string">"train"</span>].column_names
	<span class="hljs-meta">... </span>)
	<span class="hljs-meta">>>> </span>encoded_test_dataset = dataset_with_ocr[<span class="hljs-string">"test"</span>].<span class="hljs-built_in">map</span>(
	<span class="hljs-meta">... </span> encode_dataset, batched=<span class="hljs-literal">True</span>, batch_size=<span class="hljs-number">2</span>, remove_columns=dataset_with_ocr[<span class="hljs-string">"test"</span>].column_names
	<span class="hljs-meta">... </span>)`,wrap:!1}}),Cs=new x({props:{code:"ZW5jb2RlZF90cmFpbl9kYXRhc2V0LmZlYXR1cmVz",highlighted:`<span class="hljs-meta">>>> </span>encoded_train_dataset.features
	{<span class="hljs-string">'image'</span>: <span class="hljs-type">Sequence</span>(feature=<span class="hljs-type">Sequence</span>(feature=<span class="hljs-type">Sequence</span>(feature=Value(dtype=<span class="hljs-string">'uint8'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), length=-<span class="hljs-number">1</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), length=-<span class="hljs-number">1</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), length=-<span class="hljs-number">1</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>),
	<span class="hljs-string">'input_ids'</span>: <span class="hljs-type">Sequence</span>(feature=Value(dtype=<span class="hljs-string">'int32'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), length=-<span class="hljs-number">1</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>),
	<span class="hljs-string">'token_type_ids'</span>: <span class="hljs-type">Sequence</span>(feature=Value(dtype=<span class="hljs-string">'int8'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), length=-<span class="hljs-number">1</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>),
	<span class="hljs-string">'attention_mask'</span>: <span class="hljs-type">Sequence</span>(feature=Value(dtype=<span class="hljs-string">'int8'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), length=-<span class="hljs-number">1</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>),
	<span class="hljs-string">'bbox'</span>: <span class="hljs-type">Sequence</span>(feature=<span class="hljs-type">Sequence</span>(feature=Value(dtype=<span class="hljs-string">'int64'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), length=-<span class="hljs-number">1</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>), length=-<span class="hljs-number">1</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>),
	<span class="hljs-string">'start_positions'</span>: Value(dtype=<span class="hljs-string">'int64'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>),
	<span class="hljs-string">'end_positions'</span>: Value(dtype=<span class="hljs-string">'int64'</span>, <span class="hljs-built_in">id</span>=<span class="hljs-literal">None</span>)}`,wrap:!1}}),bs=new y({props:{title:"평가",local:"evaluation",headingTag:"h2"}}),Is=new y({props:{title:"훈련",local:"train",headingTag:"h2"}}),ks=new x({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMEF1dG9Nb2RlbEZvckRvY3VtZW50UXVlc3Rpb25BbnN3ZXJpbmclMEElMEFtb2RlbCUyMCUzRCUyMEF1dG9Nb2RlbEZvckRvY3VtZW50UXVlc3Rpb25BbnN3ZXJpbmcuZnJvbV9wcmV0cmFpbmVkKG1vZGVsX2NoZWNrcG9pbnQp",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForDocumentQuestionAnswering

	<span class="hljs-meta">>>> </span>model = AutoModelForDocumentQuestionAnswering.from_pretrained(model_checkpoint)`,wrap:!1}}),Zs=new x({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> TrainingArguments

	<span class="hljs-meta">>>> </span><span class="hljs-comment"># 본인의 레포지토리 ID로 바꾸세요</span>
	<span class="hljs-meta">>>> </span>repo_id = <span class="hljs-string">"MariaK/layoutlmv2-base-uncased_finetuned_docvqa"</span>

	<span class="hljs-meta">>>> </span>training_args = TrainingArguments(
	<span class="hljs-meta">... </span> output_dir=repo_id,
	<span class="hljs-meta">... </span> per_device_train_batch_size=<span class="hljs-number">4</span>,
	<span class="hljs-meta">... </span> num_train_epochs=<span class="hljs-number">20</span>,
	<span class="hljs-meta">... </span> save_steps=<span class="hljs-number">200</span>,
	<span class="hljs-meta">... </span> logging_steps=<span class="hljs-number">50</span>,
	<span class="hljs-meta">... </span> eval_strategy=<span class="hljs-string">"steps"</span>,
	<span class="hljs-meta">... </span> learning_rate=<span class="hljs-number">5e-5</span>,
	<span class="hljs-meta">... </span> save_total_limit=<span class="hljs-number">2</span>,
	<span class="hljs-meta">... </span> remove_unused_columns=<span class="hljs-literal">False</span>,
	<span class="hljs-meta">... </span> push_to_hub=<span class="hljs-literal">True</span>,
	<span class="hljs-meta">... </span>)`,wrap:!1}}),As=new x({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMERlZmF1bHREYXRhQ29sbGF0b3IlMEElMEFkYXRhX2NvbGxhdG9yJTIwJTNEJTIwRGVmYXVsdERhdGFDb2xsYXRvcigp",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> DefaultDataCollator

	<span class="hljs-meta">>>> </span>data_collator = DefaultDataCollator()`,wrap:!1}}),Xs=new x({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMFRyYWluZXIlMEElMEF0cmFpbmVyJTIwJTNEJTIwVHJhaW5lciglMEElMjAlMjAlMjAlMjBtb2RlbCUzRG1vZGVsJTJDJTBBJTIwJTIwJTIwJTIwYXJncyUzRHRyYWluaW5nX2FyZ3MlMkMlMEElMjAlMjAlMjAlMjBkYXRhX2NvbGxhdG9yJTNEZGF0YV9jb2xsYXRvciUyQyUwQSUyMCUyMCUyMCUyMHRyYWluX2RhdGFzZXQlM0RlbmNvZGVkX3RyYWluX2RhdGFzZXQlMkMlMEElMjAlMjAlMjAlMjBldmFsX2RhdGFzZXQlM0RlbmNvZGVkX3Rlc3RfZGF0YXNldCUyQyUwQSUyMCUyMCUyMCUyMHByb2Nlc3NpbmdfY2xhc3MlM0Rwcm9jZXNzb3IlMkMlMEEpJTBBJTBBdHJhaW5lci50cmFpbigp",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Trainer

	<span class="hljs-meta">>>> </span>trainer = Trainer(
	<span class="hljs-meta">... </span> model=model,
	<span class="hljs-meta">... </span> args=training_args,
	<span class="hljs-meta">... </span> data_collator=data_collator,
	<span class="hljs-meta">... </span> train_dataset=encoded_train_dataset,
	<span class="hljs-meta">... </span> eval_dataset=encoded_test_dataset,
	<span class="hljs-meta">... </span> processing_class=processor,
	<span class="hljs-meta">... </span>)

	<span class="hljs-meta">>>> </span>trainer.train()`,wrap:!1}}),Ns=new x({props:{code:"dHJhaW5lci5jcmVhdGVfbW9kZWxfY2FyZCgpJTBBdHJhaW5lci5wdXNoX3RvX2h1Yigp",highlighted:`<span class="hljs-meta">>>> </span>trainer.create_model_card()
	<span class="hljs-meta">>>> </span>trainer.push_to_hub()`,wrap:!1}}),Gs=new y({props:{title:"추론",local:"inference",headingTag:"h2"}}),Fs=new x({props:{code:"ZXhhbXBsZSUyMCUzRCUyMGRhdGFzZXQlNUIlMjJ0ZXN0JTIyJTVEJTVCMiU1RCUwQXF1ZXN0aW9uJTIwJTNEJTIwZXhhbXBsZSU1QiUyMnF1ZXJ5JTIyJTVEJTVCJTIyZW4lMjIlNUQlMEFpbWFnZSUyMCUzRCUyMGV4YW1wbGUlNUIlMjJpbWFnZSUyMiU1RCUwQXByaW50KHF1ZXN0aW9uKSUwQXByaW50KGV4YW1wbGUlNUIlMjJhbnN3ZXJzJTIyJTVEKQ==",highlighted:`<span class="hljs-meta">>>> </span>example = dataset[<span class="hljs-string">"test"</span>][<span class="hljs-number">2</span>]
	<span class="hljs-meta">>>> </span>question = example[<span class="hljs-string">"query"</span>][<span class="hljs-string">"en"</span>]
	<span class="hljs-meta">>>> </span>image = example[<span class="hljs-string">"image"</span>]
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(question)
	<span class="hljs-meta">>>> </span><span class="hljs-built_in">print</span>(example[<span class="hljs-string">"answers"</span>])
	<span class="hljs-string">'Who is ‘presiding’ TRRF GENERAL SESSION (PART 1)?'</span>
	[<span class="hljs-string">'TRRF Vice President'</span>, <span class="hljs-string">'lee a. waller'</span>]`,wrap:!1}}),zs=new x({props:{code:"ZnJvbSUyMHRyYW5zZm9ybWVycyUyMGltcG9ydCUyMHBpcGVsaW5lJTBBJTBBcWFfcGlwZWxpbmUlMjAlM0QlMjBwaXBlbGluZSglMjJkb2N1bWVudC1xdWVzdGlvbi1hbnN3ZXJpbmclMjIlMkMlMjBtb2RlbCUzRCUyMk1hcmlhSyUyRmxheW91dGxtdjItYmFzZS11bmNhc2VkX2ZpbmV0dW5lZF9kb2N2cWElMjIpJTBBcWFfcGlwZWxpbmUoaW1hZ2UlMkMlMjBxdWVzdGlvbik=",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline

	<span class="hljs-meta">>>> </span>qa_pipeline = pipeline(<span class="hljs-string">"document-question-answering"</span>, model=<span class="hljs-string">"MariaK/layoutlmv2-base-uncased_finetuned_docvqa"</span>)
	<span class="hljs-meta">>>> </span>qa_pipeline(image, question)
	[{<span class="hljs-string">'score'</span>: <span class="hljs-number">0.9949808120727539</span>,
	<span class="hljs-string">'answer'</span>: <span class="hljs-string">'Lee A. Waller'</span>,
	<span class="hljs-string">'start'</span>: <span class="hljs-number">55</span>,
	<span class="hljs-string">'end'</span>: <span class="hljs-number">57</span>}]`,wrap:!1}}),Ss=new x({props:{code:"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",highlighted:`<span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch
	<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor
	<span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForDocumentQuestionAnswering

	<span class="hljs-meta">>>> </span>processor = AutoProcessor.from_pretrained(<span class="hljs-string">"MariaK/layoutlmv2-base-uncased_finetuned_docvqa"</span>)
	<span class="hljs-meta">>>> </span>model = AutoModelForDocumentQuestionAnswering.from_pretrained(<span class="hljs-string">"MariaK/layoutlmv2-base-uncased_finetuned_docvqa"</span>)

	<span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad():
	<span class="hljs-meta">... </span> encoding = processor(image.convert(<span class="hljs-string">"RGB"</span>), question, return_tensors=<span class="hljs-string">"pt"</span>)
	<span class="hljs-meta">... </span> outputs = model(**encoding)
	<span class="hljs-meta">... </span> start_logits = outputs.start_logits
	<span class="hljs-meta">... </span> end_logits = outputs.end_logits
	<span class="hljs-meta">... </span> predicted_start_idx = start_logits.argmax(-<span class="hljs-number">1</span>).item()
	<span class="hljs-meta">... </span> predicted_end_idx = end_logits.argmax(-<span class="hljs-number">1</span>).item()

	<span class="hljs-meta">>>> </span>processor.tokenizer.decode(encoding.input_ids.squeeze()[predicted_start_idx : predicted_end_idx + <span class="hljs-number">1</span>])
	<span class="hljs-string">'lee a. waller'</span>`,wrap:!1}}),Hs=new kn({props:{source:"https://github.com/huggingface/transformers/blob/main/docs/source/ko/tasks/document_question_answering.md"}}),{c(){m=p("meta"),w=t(),U=p("p"),Ds=t(),r(T.$$.fragment),Ps=t(),r(u.$$.fragment),Ks=t(),C=p("p"),C.textContent=Vl,sa=t(),b=p("p"),b.textContent=kl,aa=t(),f=p("ul"),f.innerHTML=Rl,la=t(),r(J.$$.fragment),na=t(),I=p("p"),I.textContent=Zl,ta=t(),_=p("p"),_.textContent=Ql,ea=t(),r(V.$$.fragment),pa=t(),r(k.$$.fragment),ia=t(),r(R.$$.fragment),ca=t(),Z=p("p"),Z.textContent=Al,ra=t(),Q=p("p"),Q.textContent=$l,ja=t(),r(A.$$.fragment),Ma=t(),$=p("p"),$.textContent=Xl,oa=t(),r(X.$$.fragment),ha=t(),r(v.$$.fragment),ga=t(),N=p("p"),N.innerHTML=vl,xa=t(),r(G.$$.fragment),ma=t(),E=p("p"),E.textContent=Nl,Ua=t(),r(B.$$.fragment),Ja=t(),F=p("p"),F.textContent=Gl,da=t(),W=p("ul"),W.innerHTML=El,ya=t(),z=p("p"),z.innerHTML=Bl,wa=t(),r(q.$$.fragment),Ta=t(),Y=p("p"),Y.innerHTML=Fl,ua=t(),r(S.$$.fragment),Ca=t(),H=p("p"),H.innerHTML=Wl,ba=t(),r(D.$$.fragment),fa=t(),L=p("p"),L.textContent=zl,Ia=t(),r(O.$$.fragment),_a=t(),d=p("div"),d.innerHTML=ql,Va=t(),r(P.$$.fragment),ka=t(),K=p("p"),K.innerHTML=Yl,Ra=t(),r(ss.$$.fragment),Za=t(),r(as.$$.fragment),Qa=t(),ls=p("p"),ls.innerHTML=Sl,Aa=t(),r(ns.$$.fragment),$a=t(),ts=p("p"),ts.innerHTML=Hl,Xa=t(),r(es.$$.fragment),va=t(),r(ps.$$.fragment),Na=t(),is=p("p"),is.innerHTML=Dl,Ga=t(),r(cs.$$.fragment),Ea=t(),rs=p("p"),rs.innerHTML=Ll,Ba=t(),js=p("p"),js.textContent=Ol,Fa=t(),Ms=p("p"),Ms.innerHTML=Pl,Wa=t(),r(os.$$.fragment),za=t(),hs=p("p"),hs.textContent=Kl,qa=t(),r(gs.$$.fragment),Ya=t(),xs=p("p"),xs.textContent=sn,Sa=t(),r(ms.$$.fragment),Ha=t(),Us=p("p"),Us.textContent=an,Da=t(),Js=p("ul"),Js.innerHTML=ln,La=t(),ds=p("p"),ds.textContent=nn,Oa=t(),r(ys.$$.fragment),Pa=t(),ws=p("p"),ws.textContent=tn,Ka=t(),r(Ts.$$.fragment),sl=t(),us=p("p"),us.textContent=en,al=t(),r(Cs.$$.fragment),ll=t(),r(bs.$$.fragment),nl=t(),fs=p("p"),fs.innerHTML=pn,tl=t(),r(Is.$$.fragment),el=t(),_s=p("p"),_s.textContent=cn,pl=t(),Vs=p("ul"),Vs.innerHTML=rn,il=t(),r(ks.$$.fragment),cl=t(),Rs=p("p"),Rs.innerHTML=jn,rl=t(),r(Zs.$$.fragment),jl=t(),Qs=p("p"),Qs.textContent=Mn,Ml=t(),r(As.$$.fragment),ol=t(),$s=p("p"),$s.innerHTML=on,hl=t(),r(Xs.$$.fragment),gl=t(),vs=p("p"),vs.innerHTML=hn,xl=t(),r(Ns.$$.fragment),ml=t(),r(Gs.$$.fragment),Ul=t(),Es=p("p"),Es.innerHTML=gn,Jl=t(),Bs=p("p"),Bs.textContent=xn,dl=t(),r(Fs.$$.fragment),yl=t(),Ws=p("p"),Ws.textContent=mn,wl=t(),r(zs.$$.fragment),Tl=t(),qs=p("p"),qs.textContent=Un,ul=t(),Ys=p("ol"),Ys.innerHTML=Jn,Cl=t(),r(Ss.$$.fragment),bl=t(),r(Hs.$$.fragment),fl=t(),Ls=p("p"),this.h()},l(s){const a=fn("svelte-u9bgzb",document.head);m=i(a,"META",{name:!0,content:!0}),a.forEach(l),w=e(s),U=i(s,"P",{}),yn(U).forEach(l),Ds=e(s),j(T.$$.fragment,s),Ps=e(s),j(u.$$.fragment,s),Ks=e(s),C=i(s,"P",{"data-svelte-h":!0}),c(C)!=="svelte-oqnn9h"&&(C.textContent=Vl),sa=e(s),b=i(s,"P",{"data-svelte-h":!0}),c(b)!=="svelte-k9bbb9"&&(b.textContent=kl),aa=e(s),f=i(s,"UL",{"data-svelte-h":!0}),c(f)!=="svelte-1t9y1pd"&&(f.innerHTML=Rl),la=e(s),j(J.$$.fragment,s),na=e(s),I=i(s,"P",{"data-svelte-h":!0}),c(I)!=="svelte-1t45j8o"&&(I.textContent=Zl),ta=e(s),_=i(s,"P",{"data-svelte-h":!0}),c(_)!=="svelte-ddzvb0"&&(_.textContent=Ql),ea=e(s),j(V.$$.fragment,s),pa=e(s),j(k.$$.fragment,s),ia=e(s),j(R.$$.fragment,s),ca=e(s),Z=i(s,"P",{"data-svelte-h":!0}),c(Z)!=="svelte-1uvmu64"&&(Z.textContent=Al),ra=e(s),Q=i(s,"P",{"data-svelte-h":!0}),c(Q)!=="svelte-xyvcw8"&&(Q.textContent=$l),ja=e(s),j(A.$$.fragment,s),Ma=e(s),$=i(s,"P",{"data-svelte-h":!0}),c($)!=="svelte-13bifrs"&&($.textContent=Xl),oa=e(s),j(X.$$.fragment,s),ha=e(s),j(v.$$.fragment,s),ga=e(s),N=i(s,"P",{"data-svelte-h":!0}),c(N)!=="svelte-12y9xdp"&&(N.innerHTML=vl),xa=e(s),j(G.$$.fragment,s),ma=e(s),E=i(s,"P",{"data-svelte-h":!0}),c(E)!=="svelte-9rg4tz"&&(E.textContent=Nl),Ua=e(s),j(B.$$.fragment,s),Ja=e(s),F=i(s,"P",{"data-svelte-h":!0}),c(F)!=="svelte-zq0ej4"&&(F.textContent=Gl),da=e(s),W=i(s,"UL",{"data-svelte-h":!0}),c(W)!=="svelte-g2ws24"&&(W.innerHTML=El),ya=e(s),z=i(s,"P",{"data-svelte-h":!0}),c(z)!=="svelte-xeapog"&&(z.innerHTML=Bl),wa=e(s),j(q.$$.fragment,s),Ta=e(s),Y=i(s,"P",{"data-svelte-h":!0}),c(Y)!=="svelte-m4zkx9"&&(Y.innerHTML=Fl),ua=e(s),j(S.$$.fragment,s),Ca=e(s),H=i(s,"P",{"data-svelte-h":!0}),c(H)!=="svelte-1xnig6m"&&(H.innerHTML=Wl),ba=e(s),j(D.$$.fragment,s),fa=e(s),L=i(s,"P",{"data-svelte-h":!0}),c(L)!=="svelte-lp4iy7"&&(L.textContent=zl),Ia=e(s),j(O.$$.fragment,s),_a=e(s),d=i(s,"DIV",{class:!0,"data-svelte-h":!0}),c(d)!=="svelte-q63tj1"&&(d.innerHTML=ql),Va=e(s),j(P.$$.fragment,s),ka=e(s),K=i(s,"P",{"data-svelte-h":!0}),c(K)!=="svelte-1y93l9d"&&(K.innerHTML=Yl),Ra=e(s),j(ss.$$.fragment,s),Za=e(s),j(as.$$.fragment,s),Qa=e(s),ls=i(s,"P",{"data-svelte-h":!0}),c(ls)!=="svelte-1kcsbqg"&&(ls.innerHTML=Sl),Aa=e(s),j(ns.$$.fragment,s),$a=e(s),ts=i(s,"P",{"data-svelte-h":!0}),c(ts)!=="svelte-1rksgl2"&&(ts.innerHTML=Hl),Xa=e(s),j(es.$$.fragment,s),va=e(s),j(ps.$$.fragment,s),Na=e(s),is=i(s,"P",{"data-svelte-h":!0}),c(is)!=="svelte-1l26czp"&&(is.innerHTML=Dl),Ga=e(s),j(cs.$$.fragment,s),Ea=e(s),rs=i(s,"P",{"data-svelte-h":!0}),c(rs)!=="svelte-1yiqpne"&&(rs.innerHTML=Ll),Ba=e(s),js=i(s,"P",{"data-svelte-h":!0}),c(js)!=="svelte-rkm9nf"&&(js.textContent=Ol),Fa=e(s),Ms=i(s,"P",{"data-svelte-h":!0}),c(Ms)!=="svelte-1pxhw5r"&&(Ms.innerHTML=Pl),Wa=e(s),j(os.$$.fragment,s),za=e(s),hs=i(s,"P",{"data-svelte-h":!0}),c(hs)!=="svelte-mdy9jm"&&(hs.textContent=Kl),qa=e(s),j(gs.$$.fragment,s),Ya=e(s),xs=i(s,"P",{"data-svelte-h":!0}),c(xs)!=="svelte-1gff4qz"&&(xs.textContent=sn),Sa=e(s),j(ms.$$.fragment,s),Ha=e(s),Us=i(s,"P",{"data-svelte-h":!0}),c(Us)!=="svelte-zzn8kk"&&(Us.textContent=an),Da=e(s),Js=i(s,"UL",{"data-svelte-h":!0}),c(Js)!=="svelte-f0s3pn"&&(Js.innerHTML=ln),La=e(s),ds=i(s,"P",{"data-svelte-h":!0}),c(ds)!=="svelte-1h1oq5v"&&(ds.textContent=nn),Oa=e(s),j(ys.$$.fragment,s),Pa=e(s),ws=i(s,"P",{"data-svelte-h":!0}),c(ws)!=="svelte-10ovv0a"&&(ws.textContent=tn),Ka=e(s),j(Ts.$$.fragment,s),sl=e(s),us=i(s,"P",{"data-svelte-h":!0}),c(us)!=="svelte-1cmvv07"&&(us.textContent=en),al=e(s),j(Cs.$$.fragment,s),ll=e(s),j(bs.$$.fragment,s),nl=e(s),fs=i(s,"P",{"data-svelte-h":!0}),c(fs)!=="svelte-1glr4v9"&&(fs.innerHTML=pn),tl=e(s),j(Is.$$.fragment,s),el=e(s),_s=i(s,"P",{"data-svelte-h":!0}),c(_s)!=="svelte-1r96dak"&&(_s.textContent=cn),pl=e(s),Vs=i(s,"UL",{"data-svelte-h":!0}),c(Vs)!=="svelte-1jvarfp"&&(Vs.innerHTML=rn),il=e(s),j(ks.$$.fragment,s),cl=e(s),Rs=i(s,"P",{"data-svelte-h":!0}),c(Rs)!=="svelte-1lx4pzf"&&(Rs.innerHTML=jn),rl=e(s),j(Zs.$$.fragment,s),jl=e(s),Qs=i(s,"P",{"data-svelte-h":!0}),c(Qs)!=="svelte-1hdarm6"&&(Qs.textContent=Mn),Ml=e(s),j(As.$$.fragment,s),ol=e(s),$s=i(s,"P",{"data-svelte-h":!0}),c($s)!=="svelte-evzyvq"&&($s.innerHTML=on),hl=e(s),j(Xs.$$.fragment,s),gl=e(s),vs=i(s,"P",{"data-svelte-h":!0}),c(vs)!=="svelte-1gum9w7"&&(vs.innerHTML=hn),xl=e(s),j(Ns.$$.fragment,s),ml=e(s),j(Gs.$$.fragment,s),Ul=e(s),Es=i(s,"P",{"data-svelte-h":!0}),c(Es)!=="svelte-1flysn8"&&(Es.innerHTML=gn),Jl=e(s),Bs=i(s,"P",{"data-svelte-h":!0}),c(Bs)!=="svelte-1dcjn75"&&(Bs.textContent=xn),dl=e(s),j(Fs.$$.fragment,s),yl=e(s),Ws=i(s,"P",{"data-svelte-h":!0}),c(Ws)!=="svelte-gaec9h"&&(Ws.textContent=mn),wl=e(s),j(zs.$$.fragment,s),Tl=e(s),qs=i(s,"P",{"data-svelte-h":!0}),c(qs)!=="svelte-4epvs7"&&(qs.textContent=Un),ul=e(s),Ys=i(s,"OL",{"data-svelte-h":!0}),c(Ys)!=="svelte-10zqnq5"&&(Ys.innerHTML=Jn),Cl=e(s),j(Ss.$$.fragment,s),bl=e(s),j(Hs.$$.fragment,s),fl=e(s),Ls=i(s,"P",{}),yn(Ls).forEach(l),this.h()},h(){_l(m,"name","hf:doc:metadata"),_l(m,"content",Qn),_l(d,"class","flex justify-center")},m(s,a){In(document.head,m),n(s,w,a),n(s,U,a),n(s,Ds,a),M(T,s,a),n(s,Ps,a),M(u,s,a),n(s,Ks,a),n(s,C,a),n(s,sa,a),n(s,b,a),n(s,aa,a),n(s,f,a),n(s,la,a),M(J,s,a),n(s,na,a),n(s,I,a),n(s,ta,a),n(s,_,a),n(s,ea,a),M(V,s,a),n(s,pa,a),M(k,s,a),n(s,ia,a),M(R,s,a),n(s,ca,a),n(s,Z,a),n(s,ra,a),n(s,Q,a),n(s,ja,a),M(A,s,a),n(s,Ma,a),n(s,$,a),n(s,oa,a),M(X,s,a),n(s,ha,a),M(v,s,a),n(s,ga,a),n(s,N,a),n(s,xa,a),M(G,s,a),n(s,ma,a),n(s,E,a),n(s,Ua,a),M(B,s,a),n(s,Ja,a),n(s,F,a),n(s,da,a),n(s,W,a),n(s,ya,a),n(s,z,a),n(s,wa,a),M(q,s,a),n(s,Ta,a),n(s,Y,a),n(s,ua,a),M(S,s,a),n(s,Ca,a),n(s,H,a),n(s,ba,a),M(D,s,a),n(s,fa,a),n(s,L,a),n(s,Ia,a),M(O,s,a),n(s,_a,a),n(s,d,a),n(s,Va,a),M(P,s,a),n(s,ka,a),n(s,K,a),n(s,Ra,a),M(ss,s,a),n(s,Za,a),M(as,s,a),n(s,Qa,a),n(s,ls,a),n(s,Aa,a),M(ns,s,a),n(s,$a,a),n(s,ts,a),n(s,Xa,a),M(es,s,a),n(s,va,a),M(ps,s,a),n(s,Na,a),n(s,is,a),n(s,Ga,a),M(cs,s,a),n(s,Ea,a),n(s,rs,a),n(s,Ba,a),n(s,js,a),n(s,Fa,a),n(s,Ms,a),n(s,Wa,a),M(os,s,a),n(s,za,a),n(s,hs,a),n(s,qa,a),M(gs,s,a),n(s,Ya,a),n(s,xs,a),n(s,Sa,a),M(ms,s,a),n(s,Ha,a),n(s,Us,a),n(s,Da,a),n(s,Js,a),n(s,La,a),n(s,ds,a),n(s,Oa,a),M(ys,s,a),n(s,Pa,a),n(s,ws,a),n(s,Ka,a),M(Ts,s,a),n(s,sl,a),n(s,us,a),n(s,al,a),M(Cs,s,a),n(s,ll,a),M(bs,s,a),n(s,nl,a),n(s,fs,a),n(s,tl,a),M(Is,s,a),n(s,el,a),n(s,_s,a),n(s,pl,a),n(s,Vs,a),n(s,il,a),M(ks,s,a),n(s,cl,a),n(s,Rs,a),n(s,rl,a),M(Zs,s,a),n(s,jl,a),n(s,Qs,a),n(s,Ml,a),M(As,s,a),n(s,ol,a),n(s,$s,a),n(s,hl,a),M(Xs,s,a),n(s,gl,a),n(s,vs,a),n(s,xl,a),M(Ns,s,a),n(s,ml,a),M(Gs,s,a),n(s,Ul,a),n(s,Es,a),n(s,Jl,a),n(s,Bs,a),n(s,dl,a),M(Fs,s,a),n(s,yl,a),n(s,Ws,a),n(s,wl,a),M(zs,s,a),n(s,Tl,a),n(s,qs,a),n(s,ul,a),n(s,Ys,a),n(s,Cl,a),M(Ss,s,a),n(s,bl,a),M(Hs,s,a),n(s,fl,a),n(s,Ls,a),Il=!0},p(s,[a]){const dn={};a&2&&(dn.$$scope={dirty:a,ctx:s}),J.$set(dn)},i(s){Il\|\|(o(T.$$.fragment,s),o(u.$$.fragment,s),o(J.$$.fragment,s),o(V.$$.fragment,s),o(k.$$.fragment,s),o(R.$$.fragment,s),o(A.$$.fragment,s),o(X.$$.fragment,s),o(v.$$.fragment,s),o(G.$$.fragment,s),o(B.$$.fragment,s),o(q.$$.fragment,s),o(S.$$.fragment,s),o(D.$$.fragment,s),o(O.$$.fragment,s),o(P.$$.fragment,s),o(ss.$$.fragment,s),o(as.$$.fragment,s),o(ns.$$.fragment,s),o(es.$$.fragment,s),o(ps.$$.fragment,s),o(cs.$$.fragment,s),o(os.$$.fragment,s),o(gs.$$.fragment,s),o(ms.$$.fragment,s),o(ys.$$.fragment,s),o(Ts.$$.fragment,s),o(Cs.$$.fragment,s),o(bs.$$.fragment,s),o(Is.$$.fragment,s),o(ks.$$.fragment,s),o(Zs.$$.fragment,s),o(As.$$.fragment,s),o(Xs.$$.fragment,s),o(Ns.$$.fragment,s),o(Gs.$$.fragment,s),o(Fs.$$.fragment,s),o(zs.$$.fragment,s),o(Ss.$$.fragment,s),o(Hs.$$.fragment,s),Il=!0)},o(s){h(T.$$.fragment,s),h(u.$$.fragment,s),h(J.$$.fragment,s),h(V.$$.fragment,s),h(k.$$.fragment,s),h(R.$$.fragment,s),h(A.$$.fragment,s),h(X.$$.fragment,s),h(v.$$.fragment,s),h(G.$$.fragment,s),h(B.$$.fragment,s),h(q.$$.fragment,s),h(S.$$.fragment,s),h(D.$$.fragment,s),h(O.$$.fragment,s),h(P.$$.fragment,s),h(ss.$$.fragment,s),h(as.$$.fragment,s),h(ns.$$.fragment,s),h(es.$$.fragment,s),h(ps.$$.fragment,s),h(cs.$$.fragment,s),h(os.$$.fragment,s),h(gs.$$.fragment,s),h(ms.$$.fragment,s),h(ys.$$.fragment,s),h(Ts.$$.fragment,s),h(Cs.$$.fragment,s),h(bs.$$.fragment,s),h(Is.$$.fragment,s),h(ks.$$.fragment,s),h(Zs.$$.fragment,s),h(As.$$.fragment,s),h(Xs.$$.fragment,s),h(Ns.$$.fragment,s),h(Gs.$$.fragment,s),h(Fs.$$.fragment,s),h(zs.$$.fragment,s),h(Ss.$$.fragment,s),h(Hs.$$.fragment,s),Il=!1},d(s){s&&(l(w),l(U),l(Ds),l(Ps),l(Ks),l(C),l(sa),l(b),l(aa),l(f),l(la),l(na),l(I),l(ta),l(_),l(ea),l(pa),l(ia),l(ca),l(Z),l(ra),l(Q),l(ja),l(Ma),l($),l(oa),l(ha),l(ga),l(N),l(xa),l(ma),l(E),l(Ua),l(Ja),l(F),l(da),l(W),l(ya),l(z),l(wa),l(Ta),l(Y),l(ua),l(Ca),l(H),l(ba),l(fa),l(L),l(Ia),l(_a),l(d),l(Va),l(ka),l(K),l(Ra),l(Za),l(Qa),l(ls),l(Aa),l($a),l(ts),l(Xa),l(va),l(Na),l(is),l(Ga),l(Ea),l(rs),l(Ba),l(js),l(Fa),l(Ms),l(Wa),l(za),l(hs),l(qa),l(Ya),l(xs),l(Sa),l(Ha),l(Us),l(Da),l(Js),l(La),l(ds),l(Oa),l(Pa),l(ws),l(Ka),l(sl),l(us),l(al),l(ll),l(nl),l(fs),l(tl),l(el),l(_s),l(pl),l(Vs),l(il),l(cl),l(Rs),l(rl),l(jl),l(Qs),l(Ml),l(ol),l($s),l(hl),l(gl),l(vs),l(xl),l(ml),l(Ul),l(Es),l(Jl),l(Bs),l(dl),l(yl),l(Ws),l(wl),l(Tl),l(qs),l(ul),l(Ys),l(Cl),l(bl),l(fl),l(Ls)),l(m),g(T,s),g(u,s),g(J,s),g(V,s),g(k,s),g(R,s),g(A,s),g(X,s),g(v,s),g(G,s),g(B,s),g(q,s),g(S,s),g(D,s),g(O,s),g(P,s),g(ss,s),g(as,s),g(ns,s),g(es,s),g(ps,s),g(cs,s),g(os,s),g(gs,s),g(ms,s),g(ys,s),g(Ts,s),g(Cs,s),g(bs,s),g(Is,s),g(ks,s),g(Zs,s),g(As,s),g(Xs,s),g(Ns,s),g(Gs,s),g(Fs,s),g(zs,s),g(Ss,s),g(Hs,s)}}}const Qn='{"title":"문서 질의 응답(Document Question Answering)","local":"document_question_answering","sections":[{"title":"데이터 불러오기","local":"load-the-data","sections":[],"depth":2},{"title":"데이터 전처리","local":"preprocess-the-data","sections":[{"title":"문서 이미지 전처리","local":"preprocessing-document-images","sections":[],"depth":3},{"title":"텍스트 데이터 전처리","local":"preprocessing-text-data","sections":[],"depth":3}],"depth":2},{"title":"평가","local":"evaluation","sections":[],"depth":2},{"title":"훈련","local":"train","sections":[],"depth":2},{"title":"추론","local":"inference","sections":[],"depth":2}],"depth":1}';function An(Os){return Tn(()=>{new URLSearchParams(window.location.search).get("fw")}),[]}class Bn extends Cn{constructor(m){super(),bn(this,m,An,Zn,wn,{})}}export{Bn as component};

Xet Storage Details

Size:: 89.7 kB
Xet hash:: e2d33a2d3b6e2e640ce5cc375516e5137013cb16a752cfe4c5f371737d42f694

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