Query Planning using Open Source LLMs and Function Calling

Author(s): Vatsal Saglani

Originally published on Towards AI.

Image from ChatGPT

The past year I’ve had a lot of hands-on experience working with the GPT models via the API. I developed multiple applications for various use cases and had a great time innovating with them and breathing life into various different frameworks and processes using these models.

OpenAI, with its GPT models, provides a couple of unique features like JSON mode and Function Calling, which were (probably still are) hard to compete against. Yes, there are many open-source models (even closed ones like Claude) that do a decent job at generating JSON output, classifying functions, and then generating parameters in a multi-step manner. But most of them struggle when it comes to generating huge JSON objects or single-step function classification and parameter generation.

But this didn’t stop me from trying out all the open-source models as and when they dropped.

After a lot of disappointing trials and errors (and mostly errors) a ray of hope emerged when the Mistral-7B and Mistral-7B-Instruct models were dropped. When I tried out the Mistral-7B-Instruct model, I was quite shocked to see how quickly it was generating text, and when I saw the outputs, it felt like using the early version of the GPT-3.5 and GPT-3.5-Turbo (but… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI