RetrievalQA error

Short Course Q&A LangChain: Chat with Your Data
1

In 05_question_answering, when I am trying to run the map_reduce part, I got error:

TypeError: unsupported operand type(s) for +=: ‘OpenAIObject’ and ‘OpenAIObject’

after trying to run

result = qa_chain_mr({“query”: question})

Can somebody help me with this? Thanks.

1 Like
2

Hi You_Wang,

There is indeed a bug in that notebook. I was facing the same issue. The root cause is:

  • ChatOpenAI Output Type: The ChatOpenAI class (from LangChain) returns responses as OpenAIObject instances, which are more complex structures containing additional metadata besides the text.
  • map_reduce Chain Expectation: The map_reduce chain type expects the language model to return plain strings so it can concatenate them using the += operator.

There are a couple of ways to resolve this issue:

1. Switch to a Plain OpenAI LLM Instead of ChatOpenAI

from langchain.llms import OpenAI
llm = OpenAI(model_name=llm_name, temperature=0)

2. Use a Compatible Chain Type with ChatOpenAI

If you prefer to utilize the chat-based capabilities of ChatOpenAI, switch to a chain type that’s compatible with it, such as "refine"

qa_chain_refine = RetrievalQA.from_chain_type(
    llm,
    retriever=vectordb.as_retriever(),
    chain_type="refine"
)

Hope this helps. Happy learning.

Best

Rosa

1 Like
3

Thanks to arosacastillo and You_Wang, I had been searching for this all over the internet for the past two days!

4

Hi @Arpit_Sharma3,

Thanks for your message. Which solution did you use in the end? 1 or 2?

Happy learning

Rosa