Project proposal: Language Models Can Teach Themselves to Use Tools

hello tomer , me ( Ofir ) and Julia found a nice topic for the course project. We will explore function calling in Large Language Models (LLMs), focusing on how models can decide when and how to invoke external tools (such as calculators, weather APIs, or file search) during their response generation. The core idea is based on the "Toolformer" paper, which introduces a method for enabling LLMs to use external tools through self-supervised learning, without requiring human-labeled data.

paper ref : https://arxiv.org/pdf/2302.04761 code at :https://github.com/lucidrains/toolformer-pytorch