Researchers from UC Berkeley and Microsoft Research have developed Gorilla, a 7 billion-parameter model that excels in API calls and can be run on consumer GPUs. Large Language Models (LLMs) like GPT-4 have difficulty generating accurate input arguments for API calls. Gorilla is fully open-source and aims to fulfill the unfulfilled potential of LLMs in effectively using tools through API calls.

source update: A Fully OpenSource LLM Tuned For API Calls – Towards AI