Can I use vectorstore with LLMChain?

[vashat]

Hi!

Trying to build a chat with openai chatgpt that can make use of info from my own documents. If I use LLMChain the chat behaves exactly like in openai web interface, I get the same high quality answers. However there seams no way of implementing LLmChain with vectorstores so I can get it to include my documents?

If I try to use ConversationalRetrievalChain instead I can use vectorstores and retrieve info from my docs but the chat quality is bad, it ignores my prompts like when I prompt it to impersonate a historical figure (it starts saying that it is an AI model after just some questions and that it can't impersonate).

Is there a way I can both have a chat that behaves exactly like onchat.openai.com but also can make use of local documents?

[sebastianseck]

retriever = vectorstore.as_retriever(search_kwargs=dict(k=1))
memory = VectorStoreRetrieverMemory(retriever=retriever)
LLMChain(llm=llm, prompt=prompt, verbose=True, memory=memory)

[vashat]

retriever = vectorstore.as_retriever(search_kwargs=dict(k=1))
memory = VectorStoreRetrieverMemory(retriever=retriever)
LLMChain(llm=llm, prompt=prompt, verbose=True, memory=memory)

Thank you! That solved it!

[nikhildevnani58]

Hi, i have a requirement where i need to use Multi Prompt Chain to select a prompt out of 2 on the basis of given query and after selecting that particular prompt i need to extract the relevant document for that query from the vector data base. The query if to compare 2 report on the similar categories. With the help of Multi prompt Chain i am able to select a particular prompt but stuck at point how can we extract relevant documents for that query and use for Comparision.

[davidmarcemps]

I'd need something similar to that. Any ideas on how to use a vectorstore as an LLMChain to be part of a MultiPromptChain?

[nikhildevnani58]

I am looking for the same thing only but didn't get any solution yet.

[magallardo]

How do you use the above solution to indicate the retriever to replace a placeholder in the prompt?