ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs

The paper introduces ToolLLM, a framework for enabling tool-use capabilities in open-source large language models (LLMs). They create a dataset called ToolBench using ChatGPT and develop a depth-first search-based decision tree (DFSDT) to enhance the planning and reasoning capabilities of LLMs. They also develop an automatic evaluator called ToolEval and fine-tune LLaMA on ToolBench to obtain ToolLLaMA, which demonstrates a remarkable ability to execute complex instructions. They provide a publicly available code, trained models, and a demo.

PODCASTS: