How to use RetrievalQA with UnstructuredExcelLoader?

[dokato]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from langchain.chains import RetrievalQA
from langchain_openai import AzureChatOpenAI
from langchain_community.document_loaders import UnstructuredExcelLoader

loader = UnstructuredExcelLoader("sixnations.xlsx", mode="elements")
docs = loader.load()

print(len(docs))
llm = AzureChatOpenAI()

chain = RetrievalQA(llm=llm, documents=docs)

Description

I'm trying to use UnstructuredExcelLoader with RetrievalQA. However, I got an error like this?

ValidationError: 4 validation errors for RetrievalQA
combine_documents_chain
  Field required [type=missing, input_value={'llm': AzureChatOpenAI(c...\nItaly\nWales\n\n\n')]}, input_type=dict]
    For further information visit https://errors.pydantic.dev/2.9/v/missing
retriever
  Field required [type=missing, input_value={'llm': AzureChatOpenAI(c...\nItaly\nWales\n\n\n')]}, input_type=dict]
    For further information visit https://errors.pydantic.dev/2.9/v/missing
llm
  Extra inputs are not permitted [type=extra_forbidden, input_value=AzureChatOpenAI(client=<o...openai_api_type='azure'), input_type=AzureChatOpenAI]
    For further information visit https://errors.pydantic.dev/2.9/v/extra_forbidden
documents
  Extra inputs are not permitted [type=extra_forbidden, input_value=[Document(metadata={'sour...d\nItaly\nWales\n\n\n')], input_type=list]
    For further information visit https://errors.pydantic.dev/2.9/v/extra_forbidden

How do I interact with my loaded spreadsheet? It doesn't have to me necessarily via RetrievalQA chain.

System Info

langchain==0.3.0
langchain-community==0.3.0
langchain-core==0.3.2
langchain-openai==0.2.0
langchain-text-splitters==0.3.0

python 3.12.6

[dosubot[bot]]

To use UnstructuredExcelLoader with RetrievalQA in LangChain, you need to set up a retriever and not pass the documents directly to the RetrievalQA chain. Here’s how you can do it:

1. Install the necessary packages:

   %pip install --upgrade --quiet langchain-community unstructured openpyxl

2. Load the Excel file using UnstructuredExcelLoader:

   from langchain_community.document_loaders import UnstructuredExcelLoader
   
   loader = UnstructuredExcelLoader("sixnations.xlsx", mode="elements")
   docs = loader.load()

3. Set up the RetrievalQA chain:

   from langchain.chains import RetrievalQA
   from langchain.vectorstores import FAISS
   from langchain.embeddings import OpenAIEmbeddings
   
   # Assuming you have OpenAI API key set up in your environment
   embeddings = OpenAIEmbeddings()
   vectorstore = FAISS.from_documents(docs, embeddings)
   retriever = vectorstore.as_retriever()
   
   qa_chain = RetrievalQA(retriever=retriever)

4. Interact with the RetrievalQA chain:

   query = "Who won the Six Nations in 2020?"
   answer = qa_chain.run(query)
   print(answer)

If you want to interact with your loaded spreadsheet without using the RetrievalQA chain, you can directly work with the docs object returned by the UnstructuredExcelLoader. For example, you can print the content of the documents or process them as needed:

for doc in docs:
    print(doc.page_content)

This will give you access to the raw text content of the Excel file, which you can then manipulate or analyze as required ^[1].

To continue talking to Dosu, mention @dosu.

[dosubot[bot]]

To resolve the AttributeError: 'AzureChatOpenAI' object has no attribute 'prompt', you need to use the ChatPromptTemplate to handle prompts correctly. Here’s how you can set it up:

1. Ensure environment variables are set:

   import os
   
   if not os.getenv("AZURE_OPENAI_ENDPOINT"):
       raise ValueError("Please set the environment variable AZURE_OPENAI_ENDPOINT")
   
   if not os.getenv("AZURE_OPENAI_API_KEY"):
       raise ValueError("Please set the environment variable AZURE_OPENAI_API_KEY")
   
   if not os.getenv("AZURE_CHAT_DEPLOYMENT"):
       raise ValueError("Please set the environment variable AZURE_CHAT_DEPLOYMENT")
   
   api_version = os.getenv("OPENAI_API_VERSION", "2023-05-15")

2. Set up the prompt using ChatPromptTemplate:

   from langchain_core.prompts import ChatPromptTemplate
   from langchain_openai import AzureChatOpenAI
   
   template = """Answer the question based only on the following context:
   {context}
   Question: {question}
   """
   
   _prompt = ChatPromptTemplate.from_template(template)
   _model = AzureChatOpenAI(
       deployment_name=os.environ["AZURE_CHAT_DEPLOYMENT"],
       api_version=api_version,
   )

3. Set up the RetrievalQA chain:

   from langchain.chains import create_retrieval_chain, create_history_aware_retriever
   from langchain.vectorstores import FAISS
   from langchain.embeddings import OpenAIEmbeddings
   
   # Load the Excel file
   from langchain_community.document_loaders import UnstructuredExcelLoader
   loader = UnstructuredExcelLoader("sixnations.xlsx", mode="elements")
   docs = loader.load()
   
   # Set up the embeddings and vector store
   embeddings = OpenAIEmbeddings()
   vectorstore = FAISS.from_documents(docs, embeddings)
   retriever = vectorstore.as_retriever()
   
   # Create history-aware retriever
   history_aware_retriever = create_history_aware_retriever(
       _model, retriever, _prompt
   )
   
   # Define system prompt for QA
   system_prompt = (
       "You are an assistant for question-answering tasks. "
       "Use the following pieces of retrieved context to answer "
       "the question. If you don't know the answer, say that you "
       "don't know. Use three sentences maximum and keep the "
       "answer concise."
       "\n\n"
       "{context}"
   )
   
   qa_prompt = ChatPromptTemplate.from_messages(
       [
           ("system", system_prompt),
           ("placeholder", "{chat_history}"),
           ("human", "{input}"),
       ]
   )
   
   # Create the QA chain
   qa_chain = create_stuff_documents_chain(_model, qa_prompt)
   convo_qa_chain = create_retrieval_chain(history_aware_retriever, qa_chain)
   
   # Use the chain
   query = "Who won the Six Nations in 2020?"
   answer = convo_qa_chain.invoke({"input": query, "chat_history": []})
   print(answer)

By following these steps, you ensure that the prompts are correctly set up and avoid the AttributeError ^[1][2][3][4].

To continue talking to Dosu, mention @dosu.

[dokato]

@dosu it results in:

ValueError: Expected `input` to be a prompt variable, but got ['context', 'question']

[dosubot[bot]]

To resolve the ValueError: Expected 'input' to be a prompt variable, but got ['context', 'question'], ensure that the input_variables in your PromptTemplate match the expected variables in your prompt. Here’s how you can set it up correctly:

1. Define the QA_PROMPT with the correct input_variables:

   from langchain_core.prompts import PromptTemplate
   
   prompt_template = """Use the following pieces of context to answer the question at the end. If you don't know the answer, just say that you don't know, don't try to make up an answer.
   
   {context}
   
   Question: {question}
   Helpful Answer:"""
   QA_PROMPT = PromptTemplate(
       template=prompt_template, input_variables=["context", "question"]
   )

2. Ensure you provide the necessary context and question variables when using the prompt:

   from langchain_openai import AzureChatOpenAI
   
   response = AzureChatOpenAI(
       prompt=QA_PROMPT,
       context="Your context here",
       question="Your question here"
   )

3. Set up the RetrievalQA chain correctly:

   from langchain.chains import create_retrieval_chain, create_stuff_documents_chain
   from langchain.vectorstores import FAISS
   from langchain.embeddings import OpenAIEmbeddings
   from langchain_community.document_loaders import UnstructuredExcelLoader
   
   # Load the Excel file
   loader = UnstructuredExcelLoader("sixnations.xlsx", mode="elements")
   docs = loader.load()
   
   # Set up the embeddings and vector store
   embeddings = OpenAIEmbeddings()
   vectorstore = FAISS.from_documents(docs, embeddings)
   retriever = vectorstore.as_retriever()
   
   # Create the combine_documents_chain
   combine_docs_chain = create_stuff_documents_chain(
       llm=AzureChatOpenAI(),
       prompt=QA_PROMPT
   )
   
   # Create the retrieval chain
   retrieval_chain = create_retrieval_chain(retriever, combine_docs_chain)
   
   # Use the chain
   query = "Who won the Six Nations in 2020?"
   answer = retrieval_chain.invoke({"context": "Your context here", "question": query})
   print(answer)

This setup ensures that the input variables are correctly mapped and passed through the chain, resolving the ValueError ^[1][2][3].

To continue talking to Dosu, mention @dosu.

[dokato]

@dosu this fails with:

     59     retrieval_docs: Runnable[dict, RetrieverOutput] = retriever
     60 else:
---> 61     retrieval_docs = (lambda x: x["input"]) | retriever
     63 retrieval_chain = (
     64     RunnablePassthrough.assign(
     65         context=retrieval_docs.with_config(run_name="retrieve_documents"),
     66     ).assign(answer=combine_docs_chain)
     67 ).with_config(run_name="retrieval_chain")
     69 return retrieval_chain

KeyError: 'input'

[dosubot[bot]]

To resolve the KeyError: 'input' when setting up the retrieval chain, ensure that the input data structure matches the expected format. Specifically, the AgentInput class should have an input field, and the data passed to the agent should include this field.

Here is the relevant part of the code that defines the AgentInput class and sets up the agent:

class AgentInput(BaseModel):
    input: str
    chat_history: List[Tuple[str, str]] = Field(
        ..., extra={"widget": {"type": "chat", "input": "input", "output": "output"}}
    )

agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True).with_types(
    input_type=AgentInput
)

Make sure that when you call the agent, you provide the input field in the data structure. For example:

data = {
    "input": "Your query here",
    "chat_history": [("previous user message", "previous AI response")]
}

result = agent_executor(data)

This ensures that the input field is present and correctly passed to the agent, preventing the KeyError: 'input' ^[1].

To continue talking to Dosu, mention @dosu.