A Q not related to the new version do

AAndreaSel93

A Q not related to the new version: do the QA and REFINE prompts (also the similarity top_k) work with langchain agents using llama index as tool? or with GPTIndexChatMemory?

8 comments

LLogan M

You should be able to set the prompts in the query_kwargs of each tool I think!

LLogan M

same with the similarity_top_k

AAndreaSel93

This code works well, but being directly from langchain I don't think similarity and prompt work here:

tools = [
    Tool(
        name = "GPT Index",
        func=lambda q: str(index.query(q)),
        description="useful for ...",
        return_direct=True,

    ),
]
# set Logging to DEBUG for more detailed outputs
memory = ConversationBufferMemory(memory_key="chat_history")
llm=OpenAI(temperature=0)
agent_chain = initialize_agent(tools, llm, agent="conversational-react-description", memory=memory)
agent_chain.run(input="hi, i am bob")

AAndreaSel93

.

OTHER SOLUTION BELOW:

memory = GPTIndexChatMemory(
    index=index,
    memory_key="chat_history", 
    query_kwargs={"response_mode": "compact",
                  "similarity_top_k": 3},
    # return_source returns source nodes instead of querying index
    return_source=True,
    # return_messages returns context in message format
    return_messages=True
)
llm = OpenAIChat(temperature=0)
# llm=OpenAI(temperature=0)
agent_chain = initialize_agent([], llm, agent="conversational-react-description", memory=memory)
agent_chain.run(input="hi, i am bob")

This works fine with that query ("hi, i am bob"), but when I ask for index content I got this error:

Out[21]: 'Agent stopped due to max iterations.'

AAndreaSel93

I can't figure out why😩

LLogan M

Right, so where you do index.query() you can set the prompts and top_k 💪

LLogan M

Yea that one is a little weird 🤔

AAndreaSel93

🤦🏻‍♂️🤦🏻‍♂️🤦🏻‍♂️ don’t even seen!

Anyway, do you think are there any differences in performance? Which one do you suggest?

Add a reply

Find answers from the community

A Q not related to the new version do