LlamaIndex
Log in

Find answers from the community

Updated 6 months ago

Hi everyone,

At a glance
WWellington Santana
·
Hi everyone,

I'm currently working on optimizing a function calling prompt for better performance with GPT, but I'm hitting a snag with obtaining accurate results. Despite reviewing available materials, I've found limited guidance on enhancing this specific aspect. Here's a succinct overview:

Goal: Improve function selection based on user queries and context, acknowledging that multiple functions may need to be called for a comprehensive answer.

Available Tools:

GetColumnsConfigTool - Fetches board columns' IDs and names.
GetWorkItemsDeliveryStatusTool - Retrieves work items' delivery status.
GetWorkItensTool - Acquires work items by keys or IDs.
Issue: For the question, "What are the delivery dates for tasks that are in progress?", I anticipated calls to both GetWorkItensTool and GetDeliveryStatusForWorkItems. However, only GetWorkItensTool is being selected.

Query: Does anyone have insights on refining function selection within prompts? Are there models superior to GPT-4-turbo for this use case, or techniques to ensure broader coverage and accuracy?

Appreciate any advice or resources that could shed light on this challenge.
T
W
2 comments
TTeemu
You should be able to call multiple. Is that your full prompt? Maybe try more detailed tool descriptions and emphasize the fact you have to sometimes use multiple tools. I've never really had that issue before with GPT-4, it's usually good at it with minimal prompting 🤔
WWellington Santana
My prompt complete

You are an agent responsible for managing various software delivery data.

  1. Consider the user questions and understand which tool should be used, if one is sufficient, or if you need more than one.
  1. Return all tools that need to answer all user questions.
Add a reply
Sign up and join the conversation on Discord