Hi, I want to use Pinecone database to limit gpt-3.5 knowledge. I found a good example here, which I want to use.
(https://github.com/pinecone-io/examples/blob/master/learn/generation/langchain/handbook/05-langchain-retrieval-augmentation.ipynb)
Help me to finish the code from example with RetrievalQAWithSourcesChain, so that chatbot can memorize the conversation.
GPT4 do not provide working solution.
I found a solution to my problem.
from langchain.memory import ConversationBufferMemory
memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True, input_key="question", output_key="answer")
qa_with_sources = RetrievalQAWithSourcesChain.from_chain_type(
llm=llm,
chain_type="stuff",
retriever=vectorstore.as_retriever(),
memory=memory
)
query = "Who was Benito Mussolini?"
res = qa_with_sources.invoke(query)
print(res)
# To continue the conversation, use the same qa_with_sources instance
query2 = "At what time he was in power?"
res2 = qa_with_sources.invoke(query2)
print(res2)
it is correct…But if I have to add a chain which can generate a multiple versions (e.g five versions) of the user input questions and then fetch the all the documents from retriever for all five versions, get the unique documents, rank them from all the fetched documents. and then do the stuff that you have mentioned.
how to add this into existing logic