In my experience you just need to be

LLogan M

In my experience, you just need to be super verbose in the description. Or if you wanted, you could even do something like "If user mentions the keyword [TOOL], use this tool"

7 comments

mmat

one more question, if you have any thoughts. I've considered just skipping the agent altogether and using just the index to run the chat.

it involves changing the QA prompt to look like this

Plain Text

QA_PROMPT_TMPL = (
    "You are a helpful AI chatbot named 'AI'. Your job is answer questions from the 'User', based on the provided context information.\n"
    "We have provided context information below. \n"
    "---------------------\n"
    "{context_str}"
    "\n---------------------\n"
    "Given this context information, you will be completing the following chat. Make sure to use information from the chat to respond:\n"
    f"{chat_history_str}"
    "User: {query_str}\n"
    "AI: "
)
QA_PROMPT = QuestionAnswerPrompt(QA_PROMPT_TMPL)

In your opinion do you see any potential issues with this?

LLogan M

If the conversation gets too long, you might have to make some extra calls to summarize the chat history or something 🤔 Or just add some extra processing to cut off the chat history at a certain length

LLogan M

But I think it should work!

mmat

thanks! also this makes me wonder what the langchain chat agent does, since i'm guessing it can't deal with infinite chat lengths either 🤔

LLogan M

Yea it will just error out at a certain point lol but they have different memory modules that deal with the length in different ways

LLogan M

The base ConversationMemoryBuffer doesn't do anything special I think

mmat

ok i think this gives me enough to work with for now. thank you again!

Add a reply

Find answers from the community

In my experience you just need to be