\u672c\u6559\u7a0b\u5c06\u5e26\u4f60\u4ece\u96f6\u5f00\u59cb\uff0c\u4f7f\u7528 LangChain \u6784\u5efa\u4e00\u4e2a\u5b8c\u6574\u7684\u95ee\u7b54\u7cfb\u7edf\uff08Q&A\uff09\u3002<\/p>\n
## \u9879\u76ee\u6982\u8ff0
\n\u6211\u4eec\u8981\u6784\u5efa\u4e00\u4e2a\u667a\u80fd\u95ee\u7b54\u7cfb\u7edf\uff0c\u5b83\u80fd\u591f\uff1a\u63a5\u53d7\u7528\u6237\u95ee\u9898\u3001\u4ece\u77e5\u8bc6\u5e93\u4e2d\u68c0\u7d22\u76f8\u5173\u6587\u6863\u3001\u4f7f\u7528 LLM \u751f\u6210\u51c6\u786e\u7684\u56de\u7b54\u3001\u652f\u6301\u591a\u79cd\u6587\u6863\u683c\u5f0f\u3002<\/p>\n
## \u51c6\u5907\u5de5\u4f5c
\n“`bash
\npip install langchain langchain-openai langchain-community langchain-chroma python-dotenv
\n“`<\/p>\n
## \u7b2c\u4e00\u6b65\uff1a\u521b\u5efa\u6587\u6863\u52a0\u8f7d\u5668
\n“`python
\nfrom langchain_core.documents import Document
\nfrom pathlib import Path<\/p>\n
def load_text_files(directory: str) -> list[Document]:
\n documents = []
\n for file_path in Path(directory).rglob(“*.txt”):
\n with open(file_path, ‘r’, encoding=’utf-8′) as f:
\n content = f.read()
\n doc = Document(
\n page_content=content,
\n metadata={“source”: str(file_path)}
\n )
\n documents.append(doc)
\n return documents
\n“`<\/p>\n
## \u7b2c\u4e8c\u6b65\uff1a\u6587\u6863\u5206\u5757
\n“`python
\nfrom langchain_text_splitters import RecursiveCharacterTextSplitter<\/p>\n
def split_documents(documents, chunk_size=1000, chunk_overlap=200):
\n text_splitter = RecursiveCharacterTextSplitter(
\n chunk_size=chunk_size,
\n chunk_overlap=chunk_overlap,
\n separators=[“nn”, “n”, “\u3002”, “\uff01”, “\uff1f”, “\uff0c”, “\u3001”, ” “]
\n )
\n chunks = text_splitter.split_documents(documents)
\n return chunks
\n“`<\/p>\n
## \u7b2c\u4e09\u6b65\uff1a\u521b\u5efa\u5411\u91cf\u5b58\u50a8
\n“`python
\nfrom langchain_openai import OpenAIEmbeddings
\nfrom langchain_chroma import Chroma<\/p>\n
def create_vectorstore(documents, persist_directory=”.\/chroma_db”):
\n embeddings = OpenAIEmbeddings(model=”text-embedding-3-small”)
\n vectorstore = Chroma.from_documents(
\n documents=documents,
\n embedding=embeddings,
\n persist_directory=persist_directory
\n )
\n return vectorstore
\n“`<\/p>\n
## \u7b2c\u56db\u6b65\uff1a\u521b\u5efa\u95ee\u7b54\u94fe
\n“`python
\nfrom langchain.chains import create_retrieval_chain
\nfrom langchain.chains.combine_documents import create_stuff_documents_chain
\nfrom langchain_openai import ChatOpenAI
\nfrom langchain_core.prompts import ChatPromptTemplate<\/p>\n
def create_qa_chain(retriever, model_name=”gpt-4o-mini”):
\n llm = ChatOpenAI(model=model_name, temperature=0)<\/p>\n
template = “””
\n\u4f60\u662f\u4e00\u4e2a\u4e13\u4e1a\u7684\u95ee\u7b54\u52a9\u624b\u3002\u8bf7\u6839\u636e\u4ee5\u4e0b\u4e0a\u4e0b\u6587\u4fe1\u606f\u56de\u7b54\u7528\u6237\u7684\u95ee\u9898\u3002<\/p>\n
\u4e0a\u4e0b\u6587\u4fe1\u606f:
\n{context}<\/p>\n
\u7528\u6237\u95ee\u9898: {question}<\/p>\n
\u8981\u6c42:
\n1. \u53ea\u4f7f\u7528\u4e0a\u4e0b\u6587\u4e2d\u7684\u4fe1\u606f\u56de\u7b54
\n2. \u5982\u679c\u4e0a\u4e0b\u6587\u4e2d\u6ca1\u6709\u7b54\u6848\uff0c\u8bf7\u660e\u786e\u8bf4\u660e”\u6211\u65e0\u6cd5\u4ece\u63d0\u4f9b\u7684\u4fe1\u606f\u4e2d\u627e\u5230\u7b54\u6848”
\n3. \u56de\u7b54\u8981\u51c6\u786e\u3001\u7b80\u6d01\u3001\u6709\u6761\u7406
\n4. \u4f7f\u7528\u4e2d\u6587\u56de\u7b54<\/p>\n
\u56de\u7b54:
\n“””
\n prompt = ChatPromptTemplate.from_template(template)
\n combine_docs_chain = create_stuff_documents_chain(llm=llm, prompt=prompt)
\n retrieval_chain = create_retrieval_chain(retriever, combine_docs_chain)
\n return retrieval_chain
\n“`<\/p>\n
## \u7b2c\u4e94\u6b65\uff1a\u6574\u5408\u6240\u6709\u7ec4\u4ef6
\n“`python
\nimport os
\nfrom dotenv import load_dotenv
\nload_dotenv()<\/p>\n
DATA_DIR = “.\/data\/knowledge_base”
\nVECTOR_DB_DIR = “.\/chroma_db”
\nMODEL_NAME = “gpt-4o-mini”<\/p>\n
# \u52a0\u8f7d\u6587\u6863
\ndocuments = load_text_files(DATA_DIR)
\nchunks = split_documents(documents)<\/p>\n
# \u521b\u5efa\u5411\u91cf\u5b58\u50a8
\nvectorstore = create_vectorstore(chunks, VECTOR_DB_DIR)
\nretriever = vectorstore.as_retriever(search_kwargs={“k”: 3})<\/p>\n
# \u521b\u5efa\u95ee\u7b54\u94fe
\nqa_chain = create_qa_chain(retriever, MODEL_NAME)<\/p>\n
# \u4ea4\u4e92\u5f0f\u95ee\u7b54
\nwhile True:
\n question = input(“\u4f60: “).strip()
\n if question.lower() in [‘quit’, ‘exit’, ‘\u9000\u51fa’]:
\n break
\n result = qa_chain.invoke({“query”: question})
\n print(f”\u52a9\u624b: {result[‘answer’]}”)
\n“`<\/p>\n","protected":false},"excerpt":{"rendered":"
\u672c\u6559\u7a0b\u5c06\u5e26\u4f60\u4ece\u96f6\u5f00\u59cb\uff0c\u4f7f\u7528 LangChain \u6784\u5efa\u4e00\u4e2a\u5b8c\u6574\u7684\u95ee\u7b54\u7cfb\u7edf\uff08Q&A\uff09\u3002 ## \u9879\u76ee\u6982\u8ff0 \u6211 […]<\/p>\n","protected":false},"author":0,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"zakra_page_container_layout":"customizer","zakra_page_sidebar_layout":"customizer","zakra_remove_content_margin":false,"zakra_sidebar":"customizer","zakra_transparent_header":"customizer","zakra_logo":0,"zakra_main_header_style":"default","zakra_menu_item_color":"","zakra_menu_item_hover_color":"","zakra_menu_item_active_color":"","zakra_menu_active_style":"","zakra_page_header":true,"footnotes":""},"categories":[7],"tags":[],"class_list":["post-15","post","type-post","status-publish","format-standard","hentry","category-langchain"],"_links":{"self":[{"href":"https:\/\/gmrea.com\/index.php?rest_route=\/wp\/v2\/posts\/15","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/gmrea.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/gmrea.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/gmrea.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=15"}],"version-history":[{"count":0,"href":"https:\/\/gmrea.com\/index.php?rest_route=\/wp\/v2\/posts\/15\/revisions"}],"wp:attachment":[{"href":"https:\/\/gmrea.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=15"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/gmrea.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=15"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/gmrea.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=15"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}