LangChain 集成 #

用户提问 → LangChain
              ↓
  1. Embedding 模型将问题转为向量
              ↓
  2. Easysearch 向量检索（kNN）找到相关文档
              ↓
  3. 将相关文档 + 问题发送给 LLM
              ↓
  4. LLM 生成基于上下文的回答
              ↓
用户得到答案

pip install langchain langchain-community elasticsearch

from elasticsearch import Elasticsearch

es = Elasticsearch(
    hosts=["https://localhost:9200"],
    basic_auth=("admin", "your-password"),
    verify_certs=False  # 自签名证书时使用
)

# 验证连接
print(es.info())

from langchain_community.embeddings import HuggingFaceEmbeddings

embeddings = HuggingFaceEmbeddings(
    model_name="BAAI/bge-base-zh-v1.5"  # 中文 embedding 模型
)

from langchain_community.vectorstores import ElasticsearchStore

vector_store = ElasticsearchStore(
    es_connection=es,
    index_name="langchain-docs",
    embedding=embeddings,
)

from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.document_loaders import TextLoader

# 加载文档
loader = TextLoader("knowledge_base.txt", encoding="utf-8")
documents = loader.load()

# 分块
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=500,
    chunk_overlap=50
)
docs = text_splitter.split_documents(documents)

# 写入 Easysearch
vector_store.add_documents(docs)

results = vector_store.similarity_search(
    query="Easysearch 如何配置安全？",
    k=5
)
for doc in results:
    print(doc.page_content[:100])

from langchain.chains import RetrievalQA
from langchain_community.llms import Ollama  # 或使用其他 LLM

# 初始化 LLM
llm = Ollama(model="qwen2.5")  # 本地部署的模型

# 构建 RAG 链
retriever = vector_store.as_retriever(search_kwargs={"k": 5})
qa_chain = RetrievalQA.from_chain_type(
    llm=llm,
    chain_type="stuff",
    retriever=retriever,
    return_source_documents=True
)

# 提问
response = qa_chain.invoke({"query": "Easysearch 的安全功能有哪些？"})
print(response["result"])

vector_store = ElasticsearchStore(
    es_connection=es,
    index_name="langchain-docs",
    embedding=embeddings,
    strategy=ElasticsearchStore.ApproxRetrievalStrategy(
        hybrid=True  # 启用混合搜索
    )
)