I’m stepping the world of Langchain. I have tried several approaches to build/learn a basic ChatBot application.

However, I’m still not clear on the differences between RunnableSerializable vs RetrievalQA . I noticed they are both defined in different packages as well langchain.chains and langchain_core

Both of the below codes produce the same responses on my data!

Here’s my sample code using RetrievalQA

#Time taken: 48 sec    
retrieval_qa = RetrievalQA.from_chain_type(
        llm=llm,
        chain_type="stuff",
        retriever=db.as_retriever(search_kwargs={'k': 10}),
        return_source_documents=True,
        chain_type_kwargs={"prompt": prompt},
        verbose=True )

And using RunnableSerializable

#Time taken: 69 sec
retrieval_chain = (
            {   "context": db.as_retriever(search_kwargs={'k': 10})
                , "question": RunnablePassthrough()
            }
            | prompt
            | llm
            | StrOutputParser()
    )

My questions here are:

1. What is the performance impact of one over the other?
2. How do I set verbose=True for debugging purposes in RunnableSerializable
3. Can I do return_source_documents in RunnableSerializable
4. What are some of the use cases where I would prefer one other the other?

Sys Config for benchmarking performance:

• Macbook Pro : 8GB – M1 Chip
• LLM : Llama3 (via Ollama)
• Embedding : “mxbai-embed-large”
• Langchain : 0.2

References:

• https://api.python.langchain.com/en/latest/runnables/langchain_core.runnables.base.RunnableSerializable.html

• https://api.python.langchain.com/en/latest/chains/langchain.chains.retrieval_qa.base.RetrievalQA.html