Find answers from the community

Updated 5 months ago

Hey Is it possible using a langchain

At a glance

The post asks if it is possible to use a LangChain vectorstore agent that takes a GPT vector store index as input. The comments indicate that this is possible, and a community member provides a link to documentation on how to use GPT index with LangChain. Another community member tries the approach and shares their experience, noting that they obtained better results by changing the agent to "zero-shot-react-description" and using an LLMPredictor with a GPTSimpleVectorIndex. They ask how to combine these methods, and a community member suggests defining the GPTSimpleVectorIndex as a Tool abstraction for an agent. The discussion focuses on prompt engineering to get the desired results.

Useful resources
Hey! Is it possible using a langchain vectorstore agent which takes as input a GPT vector store index?
j
A
7 comments
Now i will try it. But can I say that you’re like a magician?😂 you’re great, thx, hope it works in my code. For what I can I support you!
@jerryjliu0 i tried it and changing the agent to “zero-shot-react-description” I’m obtaining better results based on the text. I’m taking inspiration from this one: https://langchain.readthedocs.io/en/latest/modules/agents/examples/agent_vectorstore.html

My goal is to use a router (unfortunately using GPTTreeIndex is impossible cause I can’t summarize all my docs, they are too much and I would lose too details) for searching and combine/refine the answer based on the texts.

However, when I use LLMPredictor and a query on the GPTSimpleVectorIndex i get the best answer! (Is it due to the prompt engineering?)

How can I combine all this methods? Sorry, it’s a tricky question
@AndreaSel93 given the notebook example, did you try to define GPTSimpleVectorIndex as a Tool abstraction for an agent/
I basically followed the your 🙂 maybe i didnt get it, sorry, what do you mean for Tool abstraction? Defining the GPTSimpleVectorIndex and then passing it to the tool? If you mean this, yes I did just as indicated in the link you provided.
Ah I see. Then it might be a matter of prompt engineering on the agent side (esp tuning the text description) to get the results that you'd want
Add a reply
Sign up and join the conversation on Discord