[LLM] Synonyms 和 Langchain 实现 RAG 检索服务

[hailiang-wang]

通过 Synonyms 结合大语言模型、LangChain 实现 RAG 服务，

示例程序：
https://github.com/hailiang-wang/llm-get-started/tree/master/003_rag_langchain

结果比对：Synonyms 中文近义词 Embeddings 比 HuggingFaceEmbeddings 更擅长处理中文

https://blog.csdn.net/samurais/article/details/148340086

[littleone01]

I have received your email, thanks a lot!

[hailiang-wang]

install embeddings:

pip install -U embeddings-zh 

embeddings-zh is using Synonyms underneath.

now, build a qa model.

def get_qa_model(docs, ollama_model):
    '''
    Get a QA Model
    '''
    from embeddings_zh import EmbeddingsZh

    logger.info("[get_qa_model] ollama_model %s", ollama_model) # ollama_model, e.g. "deepseek-r1:14b"
    # Create vector store and retriever
    logger.info("🔍 Creating embeddings and setting up the retriever, doc len %d ..." % len(docs))
    # print(docs[0])
    text_splitter = SemanticChunker(EmbeddingsZh())
    documents = text_splitter.split_documents(docs)


    # Instantiate the embedding model
    embedder = EmbeddingsZh()

    vector = InMemoryVectorStore(embedder) # https://python.langchain.com/docs/concepts/vectorstores/
    # print("documents", documents[0])
    vector.add_documents(documents=documents)

    retriever = vector.as_retriever(search_type="similarity", search_kwargs={"k": 3})

    # Define the LLM and the prompt
    llm = OllamaLLM(model=ollama_model)
    prompt = """
    1. Use the following pieces of context to answer the question at the end.
    2. If you don't know the answer, just say that "I don't know" but don't make up an answer on your own.\n
    3. Keep the answer crisp and limited to 3,4 sentences.
    Context: {context}
    Question: {question}
    Helpful Answer:"""
    QA_CHAIN_PROMPT = PromptTemplate.from_template(prompt)

    # Define the document and combination chains
    llm_chain = LLMChain(llm=llm, prompt=QA_CHAIN_PROMPT, verbose=True)
    document_prompt = PromptTemplate(
        input_variables=["page_content", "source"],
        template="Context:\ncontent:{page_content}\nsource:{source}",
    )
    combine_documents_chain = StuffDocumentsChain(
        llm_chain=llm_chain,
        document_variable_name="context",
        document_prompt=document_prompt,
        verbose=True
    )

    qa = RetrievalQA(
        combine_documents_chain=combine_documents_chain,
        retriever=retriever,
        verbose=True,
        return_source_documents=True
    )

    return qa

full example, checkout
https://github.com/hailiang-wang/llm-get-started/tree/master/003_rag_langchain