docs/versions/migrating_chains/retrieval_qa/

[giscus[bot]]

docs/versions/migrating_chains/retrieval_qa/

The RetrievalQA chain performed natural-language question answering over a data source using retrieval-augmented generation.

https://python.langchain.com/docs/versions/migrating_chains/retrieval_qa/

[punsoca]

I don't see "docs" being initialized anywhere? I assume you mean "data" parsed from the Webloader.load() ?

[punsoca]

I have a follow up question - I have a PDF that I converted into chunks and created the qa_chain with the Legacy RetrievalQA and it was working fine. so I am attempting to convert the Legacy version to LCEL as follows:

from doting import load_dotenv
from langchain_openai import ChatOpenAI
from langchain.embeddings import OpenAIEmbeddings, HuggingFaceBgeEmbeddings
from langchain_community.document_loaders import PyPDFLoader   
from langchain_community.vectorstores import FAISS    
from langchain_text_splitters import RecursiveCharacterTextSplitter

# for LCEL
from langchain import hub
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough



# the .env file contains the OPENAI_API_KEY and its value
load_dotenv()

# get LLM
llm =  ChatOpenAI(
    temperature = 0,   
    model="gpt-4o-mini",  
)

# PDF file read and convert into chunks
loader = PyPDFLoader('my_file.pdf)
docs = loader.load_and_split()   
# split the docs further into smaller chunks
text_splitter =  RecursiveCharacterTextSplitter( chunk_size=1000, chunk_overlap=200)
chunks = text_splitter.split_documents(docs)

embeddings = OpenAIEmbeddings()

# passing chunks and embeddings
vectorstore = FAISS.from_documents(chunks, embeddings)

prompt = hub.pull("rlm/rag-prompt"). # pick up rag-prompt from langchain.hub



qa_chain = (
    {
        "context": vectorstore.as_retriever() | chunks, # --- error caught in this line when I ran the code
        "question": RunnablePassthrough(),
    }
    | prompt
    | llm
    | StrOutputParser()
)

And here is the error log:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[X], line XX
     23 qa_chain = (
     24     {
---> 25         "context": vector_store.as_retriever() | chunks,
     26         "question": RunnablePassthrough(),
     27     }
     28     | prompt
     29     | llm
     30     | StrOutputParser()
     31 
     32 )
        

File ~/Documents/0_-_Python_Projects/05_Gen_AI/venv_3_11/lib/python3.11/site-packages/langchain_core/runnables/base.py:577, in Runnable.__or__(self, other)
    567 def __or__(
    568     self,
    569     other: Union[
   (...)
    574     ],
    575 ) -> RunnableSerializable[Input, Other]:
    576     """Compose this Runnable with another object to create a RunnableSequence."""
--> 577     return RunnableSequence(self, coerce_to_runnable(other))

File ~/Documents/0_-_Python_Projects/05_Gen_AI/venv_3_11/lib/python3.11/site-packages/langchain_core/runnables/base.py:5843, in coerce_to_runnable(thing)
   5838 else:
   5839     msg = (
   5840         f"Expected a Runnable, callable or dict."
   5841         f"Instead got an unsupported type: {type(thing)}"
   5842     )
-> 5843     raise TypeError(msg)

TypeError: Expected a Runnable, callable or dict.Instead got an unsupported type: <class 'list'>

The error is complaining about the "chunks", which is of type list created by the PDFLoader.load_and_split() method.

I had to ignore the format_docs(docs) method because your sample is loading file from the web (while mine is reading from a PDF file). Appreciate your kind assistance to fix this issue.

[punsoca]

correction - the first import statement should say from dotenv import load_dotenv (I couldn't edit my above post).

[punsoca]

Issue is resolved!

I was finally able to fix the issue after reading the following langchain link : https://python.langchain.com/docs/how_to/passthrough/

I just needed to revise the chaining as follows - and I did not have to use the create_retrieval_chain nor the create_stuff_documents_chain for my PDF test case:

qa_chain = (
    {
        "context": vector_store.as_retriever(),
        "question": RunnablePassthrough()
    }
    | prompt
    | llm
    | StrOutputParser()

)

Then I pass the question in the invoke statement - the chatbot will keep asking questions until user enters 'q' to quit chat:

while True:
    question =  input("\nStart your question about the document beginning with 'what', 'when', 'where', or 'which' (enter 'q' to quit): ")

    if question.lower().strip() == 'q':
        print("Have a good day!")
        break

    elif question.lower().strip() == '' or not (question.lower().startswith(("what", "when", "where", "which"))):
        continue

    else:
        # invoke LCEL chain to get chat response 
        answer = qa_chain.invoke(question)
        print(f"Question:\n{question}")
        print(f"Answer:\n{answer}")