{"id":5476,"date":"2023-06-05T23:33:38","date_gmt":"2023-06-05T15:33:38","guid":{"rendered":"https:\/\/aict.nkust.edu.tw\/digitrans\/?p=5476"},"modified":"2024-12-13T20:35:11","modified_gmt":"2024-12-13T12:35:11","slug":"%e5%88%a9%e7%94%a8langchain%e5%af%a6%e4%bd%9cchatpdf%ef%bc%9a%e5%95%8f%e5%80%8b%e5%95%8f%e9%a1%8c%ef%bc%8c%e8%bc%95%e9%ac%86%e6%89%be%e5%87%ba%e6%96%87%e4%bb%b6%e9%87%8d%e9%bb%9e","status":"publish","type":"post","link":"https:\/\/aict.nkust.edu.tw\/digitrans\/?p=5476","title":{"rendered":"\u5229\u7528LangChain\u5be6\u4f5cChatPDF\uff1a\u554f\u500b\u554f\u984c\uff0c\u8f15\u9b06\u627e\u51fa\u6587\u4ef6\u91cd\u9ede"},"content":{"rendered":"\n

2023\/06\/05 | \u00a0AIF \u6280\u8853\u767c\u5c55\u4e2d\u5fc3 \u6797\u828a<\/p>\n\n\n\n

\u88ab\u770b\u4e0d\u5b8c\u7684\u6587\u4ef6\u8ffd\u8457\u8dd1\u7684\u4eba\u6709\u798f\u4e86\uff01\u5728OpenAI\u63a8\u51faChatGPT\u5f8c\uff0c\u8a31\u591a\u795e\u5947\u7684\u5c0f\u5de5\u5177\u4e5f\u96a8\u4e4b\u51fa\u73fe\u2500ChatPDF\u5c31\u662f\u5176\u4e2d\u4e00\u500b\u5f88\u597d\u7684\u61c9\u7528\u3002\u4f7f\u7528\u8005\u53ea\u8981\u4e0a\u50b3\u81ea\u5df1\u7684PDF\u6a94\uff0c\u4e0d\u8ad6\u662f\u8ad6\u6587\u3001\u65b0\u805e\u3001\u7522\u696d\u5831\u544a\u751a\u81f3\u662f\u66f8\u7c4d\uff0c\u90fd\u53ef\u4ee5\u900f\u904e\u63d0\u554f\u7684\u65b9\u5f0f\uff0c\u5feb\u901f\u9032\u884c\u6587\u4ef6\u6458\u8981\u7e3d\u7d50\uff0c\u751a\u81f3\u8acb\u5b83\u63d0\u4f9b\u5efa\u8b70\u3002\u662f\u4e0d\u662f\u8b93\u4eba\u5f88\u5fc3\u52d5\uff1f<\/p>\n\n\n\n

\u76ee\u524dChatPDF\u7684\u514d\u8cbb\u984d\u5ea6\u4e0a\u9650\u662f\u6bcf\u59293\u500b120\u9801\u5167\u7684\u6587\u4ef6\uff0c\u7e3d\u5171\u53ef\u4ee5\u554f50\u500b\u554f\u984c\u3002\u770b\u5230\u9019\u88e1\uff0c\u5982\u679c\u4f60\u6025\u8457\u8d95\u5831\u544a\u7684\u8a71\uff0c\u53ef\u4ee5\u5148\u53bb\u8d95\u9032\u5ea6\u6c92\u95dc\u4fc2\u3002\u4f46\u662f\uff0c\u5982\u679c\u60f3\u77e5\u9053\u5982\u4f55\u6253\u9020\u4e00\u500b\u6587\u4ef6\u5c0d\u8a71\u7684\u5de5\u5177\uff0c\u80af\u5b9a\u8981\u7e7c\u7e8c\u770b\u4e0b\u53bb\u3002\u6211\u5011\u6703\u6559\u5404\u4f4d\u5982\u4f55\u7c21\u55ae\u5229\u7528LangChain\u5be6\u4f5cChatPDF\u7684\u529f\u80fd\uff0c\u4e0d\u7528\u53d7\u6bcf\u65e5\u7684\u63d0\u554f\u4e0a\u9650\uff0c\u8b93\u4eba\u60f3\u554f\u591a\u5c11\u5c31\u554f\u591a\u5c11\uff01<\/p>\n\n\n\n

LangChain\u662f\u4e00\u500b\u6846\u67b6\uff0c\u5e6b\u52a9\u6211\u5011\u5728\u958b\u767c\u61c9\u7528\u7a0b\u5f0f\u6642\u5c0e\u5165\u8a9e\u8a00\u6a21\u578b\u7684\u529f\u80fd\uff0c\u53ef\u4ee5\u66f4\u5feb\u901f\u7684\u5c07\u8cc7\u6599\u4f86\u6e90\u9023\u63a5\u5230\u8a9e\u8a00\u6a21\u578b\uff0c\u4e26\u4f7f\u7528\u8a9e\u8a00\u6a21\u578b\u9032\u884c\u5f8c\u7e8c\u8655\u7406\u3002\u958b\u59cb\u4e4b\u524d\uff0c\u6211\u5011\u5148\u4f86\u770b\u770bChatPDF\u7684\u6548\u679c\uff0c\u7bc4\u4f8b\u4e2d\uff0c\u6211\u5011\u4ee5\u5df4\u9ece\u840a\u96c5\u96c6\u57182022\u5e74\u7684\u5e74\u5ea6\u5831\u544a\u70ba\u4f8b\uff0c\u5c0dChatPDF\u63d0\u51fa\u5e7e\u500b\u7c21\u55ae\u7684\u554f\u984c\uff0c\u5b83\u6703\u4f9d\u7167\u6587\u4ef6\u7684\u5167\u5bb9\u9032\u884c\u56de\u7b54\u3002<\/p>\n\n\n

\n
\"\"
\u5716\u4e00\u3001ChatPDF\u4f7f\u7528\u7bc4\u4f8b<\/figcaption><\/figure>\n<\/div>\n\n\n

\u5229\u7528LLM\u505a\u51fa\u6587\u4ef6\u554f\u7b54<\/h3>\n\n\n\n

\u6587\u4ef6\u554f\u7b54\u7684\u505a\u6cd5\u5927\u81f4\u53ef\u5206\u70ba\u56db\u500b\u6b65\u9a5f\uff0c\u4e14\u6bcf\u4e00\u500b\u6b65\u9a5f\u4e2d\u90fd\u6703\u4f7f\u7528\u5230LLM\uff1a<\/p>\n\n\n\n

1. \u5c07\u6587\u4ef6\u5132\u5b58\u6210vector store<\/strong>
\u7531\u65bcLLM\u8f38\u5165\u7684\u5b57\u6578\u901a\u5e38\u90fd\u6709\u4e0a\u9650\uff0c\u6211\u5011\u5728\u8b80\u53d6PDF\u6a94\u5f8c\u6703\u5c0d\u6587\u5b57\u505a\u5207\u5272\uff0c\u5207\u5206\u6210\u591a\u500b\u5c0f\u7684\u5340\u584a(chunk)\uff0c\u4e26\u4f7f\u7528\u4e00\u500bembedding model\u5c07\u6587\u5b57\u8f49\u63db\u6210\u5411\u91cf\uff0c\u5132\u5b58\u6210vector store\u4ee5\u4f9b\u5f8c\u7e8c\u67e5\u8a62\u3002<\/p>\n\n\n\n

2. \u5728\u5c0d\u8a71\u4e2d\u63d0\u51fa\u554f\u984c<\/strong>
\u554f\u7b54\u7cfb\u7d71\u6703\u8a18\u9304\u804a\u5929\u7684\u904e\u7a0b\uff0c\u7576\u4f7f\u7528\u8005\u63d0\u51fa\u65b0\u7684\u554f\u984c\u6642\uff0c\u6703\u5229\u7528LLM\u5c07\u4e4b\u524d\u7684\u5c0d\u8a71\u4ee5\u53ca\u65b0\u7684\u554f\u984c\uff0c\u6539\u5beb\u6210\u4e00\u500b\u7368\u7acb\u7684\u554f\u53e5\uff0c\u9019\u6a23\u5c31\u4e0d\u7528\u628a\u6574\u500b\u5c0d\u8a71\u7d00\u9304\u90fd\u4e1f\u9032\u53bb\u505a\u67e5\u8a62\u3002<\/p>\n\n\n\n

3. \u641c\u5c0b\u6587\u4ef6\u4e2d\u8207\u554f\u984c\u6700\u76f8\u95dc\u7684\u90e8\u5206<\/strong>
\u4f7f\u7528\u525b\u525b\u7684embedding model\u5c07\u554f\u984c\u672c\u8eab\u8f49\u63db\u6210\u5411\u91cf\uff0c\u5c31\u53ef\u4ee5\u67e5\u8a62vector store\u4e2d\u8207\u554f\u984c\u6700\u76f8\u95dc\u7684\u90e8\u5206\u6709\u54ea\u4e9b\u6bb5\u843d\u3002<\/p>\n\n\n\n

4. \u91dd\u5c0d\u554f\u984c\u4e26\u6839\u64da\u76f8\u95dc\u6587\u4ef6\u9032\u884c\u56de\u7b54<\/strong>
\u53d6\u51fa\u6700\u76f8\u95dc\u7684\u5167\u5bb9\u4e4b\u5f8c\uff0c\u8b93LLM\u95b1\u8b80\u554f\u984c\u4ee5\u53ca\u76f8\u95dc\u6587\u4ef6\u7684\u5167\u5bb9\uff0c\u6700\u5f8c\u7d66\u51fa\u7b54\u6848\u3002<\/p>\n\n\n

\n
\"\"
\u5716\u4e8c\u3001\u6587\u4ef6\u554f\u7b54\u7684\u6d41\u7a0b<\/figcaption><\/figure>\n<\/div>\n\n\n

\u5be6\u4f5c\u89e3\u6790<\/h3>\n\n\n\n

\u63a5\u4e0b\u4f86\u6211\u5011\u6703\u4f7f\u7528LangChain\u5be6\u4f5c\u51fa\u6587\u4ef6\u554f\u7b54\u7684\u529f\u80fd\uff0c\u7fd2\u6163\u4f7f\u7528\u4ecb\u9762\u7684\u4eba\u53ef\u4ee5\u5148\u4f7f\u7528ChatPDF\u5c31\u597d\uff0c\u4f46\u5982\u679c\u60f3\u8981\u66f4\u4e86\u89e3\u6d41\u7a0b\u548c\u7a0b\u5f0f\u78bc\u7684\u4eba\u53ef\u4ee5\u7e7c\u7e8c\u5f80\u4e0b\u3002<\/p>\n\n\n\n

\u9996\u5148\uff0c\u5c07\u76f8\u95dc\u5957\u4ef6\u5b89\u88dd\u4e26import \u9032\u4f86<\/p>\n\n\n\n

<\/td>!pip install langchain pypdf sentence_transformers chromadb tiktoken openai<\/td><\/tr>
<\/td><\/td><\/tr>
<\/td>from langchain.embeddings.openai import OpenAIEmbeddings<\/td><\/tr>
<\/td>from langchain.document_loaders import TextLoader, PyPDFLoader<\/td><\/tr>
<\/td>from langchain.text_splitter import RecursiveCharacterTextSplitter<\/td><\/tr>
<\/td>from langchain.vectorstores import Chroma<\/td><\/tr>
<\/td>from langchain.chat_models import ChatOpenAI<\/td><\/tr>
<\/td>from langchain.chains import ConversationalRetrievalChain<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

\u8a2d\u5b9a\u4f60\u7684OpenAI API key(https:\/\/platform.openai.com\/account\/api-keys)<\/p>\n\n\n\n

<\/td>import os<\/td><\/tr>
<\/td>os.environ[“OPENAI_API_KEY”] = “sk-“<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

\u8b80\u53d6PDF\u6a94\u6848\u3002\u9019\u88e1\u6211\u5011\u4f7f\u7528LangChain\u63d0\u4f9b\u7684\u8b80\u6a94\u65b9\u5f0f\u3002<\/p>\n\n\n\n

<\/td># \u8b80\u53d6\u6a94\u6848<\/td><\/tr>
<\/td>file_path = “\/content\/LOREAL_2022_Annual_Report.pdf”<\/td><\/tr>
<\/td>loader = file_path.endswith(“.pdf”) and PyPDFLoader(file_path) or TextLoader(file_path)<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

\u7531\u65bc\u8a9e\u8a00\u6a21\u578b\u4e00\u6b21\u80fd\u8b80\u53d6\u7684\u6587\u5b57\u6578\u91cf\u6709\u9650\uff0c\u6211\u5011\u9700\u8981\u5c07\u6587\u4ef6\u5207\u5206\u6210\u591a\u500b\u5c0f\u5340\u584a\u3002<\/p>\n\n\n\n

<\/td># \u9078\u64c7 splitter \u4e26\u5c07\u6587\u5b57\u5207\u5206\u6210\u591a\u500b chunk<\/td><\/tr>
<\/td>splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=0)<\/td><\/tr>
<\/td>texts = loader.load_and_split(splitter)<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

\u9078\u64c7\u4e00\u500b\u6b32\u4f7f\u7528\u7684embedding\u6a21\u578b (\u4e5f\u53ef\u4ee5\u4f7f\u7528HuggingFace\u4e0a\u5176\u4ed6\u7684\u6a21\u578b)\uff0c\u4e26\u900f\u904eChroma (\u4e00\u500b\u5c08\u9580\u70baLLM embedding database\u8a2d\u8a08\u7684\u5de5\u5177) \u5c07\u525b\u525b\u7684\u6587\u5b57\u8f49\u63db\u6210\u5411\u91cf\uff0c\u6700\u5f8c\u5132\u5b58\u6210\u4e00\u500b\u672c\u5730\u7684vector store\u3002<\/p>\n\n\n\n

<\/td># \u5efa\u7acb\u672c\u5730 db<\/td><\/tr>
<\/td>embeddings = OpenAIEmbeddings()<\/td><\/tr>
<\/td>vectorstore = Chroma.from_documents(texts, embeddings)<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

\u6700\u5f8c\u5efa\u7acb\u5c0d\u8a71chain\uff0c\u9019\u88e1\u6211\u5011\u4f7f\u7528LangChain\u7684ConversationalRetrievalChain\uff0c\u6bcf\u6b21\u90fd\u6839\u64da\u4f7f\u7528\u8005\u63d0\u51fa\u7684\u554f\u984c (question) \u4ee5\u53ca\u5c0d\u8a71\u7d00\u9304 (chat_history) \u5c0d\u6587\u4ef6\u9032\u884c\u641c\u7d22\u3002\u5176\u4e2d\uff0c\u6240\u4f7f\u7528\u7684LLM\u4e5f\u53ef\u4ee5\u66ff\u63db\u6210\u5176\u4ed6\u6a21\u578b\u3002<\/p>\n\n\n\n

<\/td># \u5c0d\u8a71 chain<\/td><\/tr>
<\/td>qa = ConversationalRetrievalChain.from_llm(ChatOpenAI(temperature=0), vectorstore.as_retriever())<\/td><\/tr>
<\/td>query = input(‘\\nQ: ‘)<\/td><\/tr>
<\/td>result = qa({“question”: query, ‘chat_history’: []})<\/td><\/tr>
<\/td>print(‘A:’, result[‘answer’])<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

\u5982\u679c\u6211\u5011\u60f3\u8981\u4f7f\u7528\u5230\u5c0d\u8a71\u7684\u529f\u80fd\uff0c\u5fc5\u9808\u6bcf\u6b21\u5c07\u5c0d\u8a71\u7d00\u9304\u5132\u5b58\u8d77\u4f86\u4e26\u4ea4\u7d66\u5c0d\u8a71chain\uff0c\u5728\u9019\u88e1\u6211\u5011\u4f7f\u7528\u4e00\u500bwhile\u8ff4\u5708\u9032\u884c\u5c0d\u8a71\uff0c\u9019\u6a23\u4e00\u4f86\u4e5f\u53ef\u4ee5\u64c1\u6709\u8ffd\u554f\u7684\u529f\u80fd\u3002<\/p>\n\n\n\n

<\/td># \u5c0d\u8a71 chain<\/td><\/tr>
<\/td>qa = ConversationalRetrievalChain.from_llm(ChatOpenAI(temperature=0), vectorstore.as_retriever())<\/td><\/tr>
<\/td>chat_history = []<\/td><\/tr>
<\/td>while True:<\/td><\/tr>
<\/td>query = input(‘\\nQ: ‘)<\/td><\/tr>
<\/td>if not query:<\/td><\/tr>
<\/td>break<\/td><\/tr>
<\/td>result = qa({“question”: query + ‘ (\u7528\u7e41\u9ad4\u4e2d\u6587\u56de\u7b54)’, “chat_history”: chat_history})<\/td><\/tr>
<\/td>print(‘A:’, result[‘answer’])<\/td><\/tr>
<\/td>chat_history.append((query, result[‘answer’]))<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n
\n
\"\"
\u5716\u4e09\u3001\u5be6\u969b\u4f7f\u7528\u7d50\u679c<\/figcaption><\/figure>\n<\/div>\n\n\n

\u7d50\u8a9e<\/h3>\n\n\n\n

\u4ee5\u4e0a\uff0c\u6211\u5011\u5229\u7528LangChain\u7c21\u55ae\u5be6\u4f5c\u51fa\u4e86ChatPDF\u7684\u529f\u80fd\uff0c\u8b93\u6211\u5011\u80fd\u4ee5\u804a\u5929\u7684\u5f62\u5f0f\uff0c\u5feb\u901f\u4e86\u89e3\u6587\u4ef6\u5167\u5bb9\uff0c\u4ee5\u5f8c\u5c31\u4e0d\u7528\u6015\u4f86\u4e0d\u53ca\u770b\u6587\u4ef6\u4e86\u5566\uff0c\u90a3\u9019\u6b21\u5148\u5beb\u5230\u9019\uff0c\u6211\u8981\u53bb\u8d95\u5831\u544a\u4e86: )<\/p>\n\n\n\n

Reference<\/h3>\n\n\n\n

ChatPDF\uff1ahttps:\/\/www.chatpdf.com\/<\/p>\n\n\n\n

LangChain\uff1ahttps:\/\/python.langchain.com\/en\/latest\/index.html<\/p>\n\n\n\n

L’Or\u00e9al Paris annul report\uff1ahttps:\/\/www.loreal-finance.com\/eng\/annual-report<\/p>\n\n\n\n

GPT-4 Tutorial: How to Chat With Multiple PDF Files\uff1ahttps:\/\/www.youtube.com\/watch?v=Ix9WIZpArm0<\/p>\n\n\n\n

\u8cc7\u6599\u4f86\u6e90:https:\/\/edge.aif.tw\/express-langchain-chatpdf\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

2023\/06\/05 | \u00a0AIF \u6280\u8853\u767c\u5c55\u4e2d\u5fc3 \u6797\u828a \u88ab\u770b\u4e0d\u5b8c\u7684\u6587\u4ef6\u8ffd\u8457\u8dd1\u7684\u4eba\u6709\u798f\u4e86\uff01\u5728OpenAI\u63a8\u51faChatGPT\u5f8c\uff0c\u8a31\u591a\u795e\u5947\u7684\u5c0f\u5de5\u5177\u4e5f\u96a8\u4e4b\u51fa\u73fe\u2500ChatPDF\u5c31\u662f\u5176\u4e2d\u4e00\u500b\u5f88…<\/p>\n","protected":false},"author":9,"featured_media":5477,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[579,4],"tags":[26,40],"class_list":["post-5476","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-579","category-industry-news","tag-ai","tag-40"],"gutentor_comment":0,"jetpack_featured_media_url":"https:\/\/i0.wp.com\/aict.nkust.edu.tw\/digitrans\/wp-content\/uploads\/2024\/03\/1.jpg?fit=1890%2C720&ssl=1","jetpack-related-posts":[],"jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/5476","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/users\/9"}],"replies":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5476"}],"version-history":[{"count":1,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/5476\/revisions"}],"predecessor-version":[{"id":5478,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/posts\/5476\/revisions\/5478"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=\/wp\/v2\/media\/5477"}],"wp:attachment":[{"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5476"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5476"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aict.nkust.edu.tw\/digitrans\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5476"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}