Q&A chain without vectorstore

[SomaKorada07]

Hi Community,

I want to use LangChain's chain capability to build a Q&A chain but I would be storing my chunks along with embeddings in a local CSV file and using semantic search to retrieve similar docs to user query. All the chains related to this like RetrievalQA are using vectorstore for storing and retrieval which is not the case for me.
Also, the main reason for me to have a chain is I would be creating an agent with this chain and other tools (actionable ones).

Thanks in advance!

Regards,
Soma

[dosubot[bot]]

Answer generated by a 🤖

Answer

I understand that you're looking to use LangChain's chain capability to build a Q&A chain, but instead of using a vectorstore for storing and retrieval, you want to store your chunks along with embeddings in a local CSV file. You also plan to use semantic search to retrieve similar documents to the user's query.

One potential solution to your issue is to create a custom chain that doesn't use a vectorstore. You can achieve this by subclassing the BaseRetrievalQA class and overriding the _get_docs method to use your own method of retrieving documents. Here's how you can do it:

from langchain.chains.retrieval_qa.base import BaseRetrievalQA
from langchain.schema import Document
from typing import List, Dict, Any
from langchain.callbacks.manager import CallbackManagerForChainRun

class CSVFileQA(BaseRetrievalQA):
    """Chain for question-answering against a CSV file."""

    csv_file_path: str
    """Path to the CSV file."""

    def _get_docs(
        self,
        question: str,
        *,
        run_manager: CallbackManagerForChainRun,
    ) -> List[Document]:
        """Get docs."""
        # Implement your own method of retrieving documents from the CSV file here.
        # This is just a placeholder implementation.
        docs = []
        return docs

In this code, you'll need to replace the placeholder implementation of _get_docs with your own method of retrieving documents from the CSV file. The question parameter is the user's query, and the method should return a list of Document objects that are similar to the query. The Document class is defined in langchain.schema, and it represents a document that can be used in a chain.

This is just one potential solution and there might be other ways to achieve your goal. If you find this solution helpful and think it might be useful to others, I encourage you to make a pull request to update the LangChain documentation. This way, other users who might encounter the same issue can benefit from your contribution.

Sources

• libs/langchain/langchain/chains/retrieval_qa/base.py

---

This response is meant to be useful, save you time, and share context. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

[SomaKorada07]

thank you! I was able to implement the above custom chain. However, when I am adding this chain (converted to a tool) along with other tools, this Q&A tool is never getting called. Can someone please help me with this. Thanks in advance!

[devstein]

How many tools is the model choosing between?

[SomaKorada07]

for now i am just trying a Q&A tool (like the above one) with another 3 actionable tools. my main task is to make the agent differentiate a user query between choosing Q&A tool and one of the actionable tools.

[SomaKorada07]

@devstein even when I am using only Q&A tool, it does not seem to work! but when i use the run method of the CSVFileQA chain, I do get results. I do not get any results (the Q&A tool is not getting called) through an agent when that's the only tool given to the agent too. It always is giving generic answer and not calling the tool. can you please help on this.

[devstein]

Can you share your code?

[SomaKorada07]

@devstein using CONVERSATIONAL_REACT_DESCRIPTION Agent type seemed to have worked better in my case.