The AI community is abuzz with the launch of Gorilla, a revolutionary language model that brings an unprecedented level of accuracy and functionality to invoking APIs using natural language queries. 

Gorilla is an open-source project, licensed under Apache 2.0, and is the result of extensive fine-tuning on Falcon and MPT created by Shishir Patil who is currently a Phd Student in ML systems at UC Berkeley. 

Unlike its predecessors, including the highly acclaimed GPT-4, Gorilla truly stands out by significantly reducing hallucinations and incorrect syntax when interacting with over 1,600 APIs and counting. This achievement is made possible by Gorilla’s capability to parse the Abstract Syntax Tree (AST) when writing code, resulting in semantically and syntactically correct API invocations.

One of the key features that make Gorilla exceptional is its compatibility with commercial use, allowing developers to incorporate it into their projects without any obligations. It’s available for use in various environments, including Google Colab and Command Line Interface (CLI) through pip installation of “gorilla-cli.”

APIBench, a meticulously curated collection of APIs, accompanies Gorilla. This extensive library facilitates easy training and expands the available resources for Gorilla to call upon. Furthermore, the project actively encourages API developers to join the cause and contribute their APIs for inclusion in the ever-growing API store.

For developers and language model enthusiasts, Gorilla represents a remarkable advancement in leveraging large language models for practical purposes. With its capabilities surpassing even the highly-touted GPT-4, Gorilla has emerged as a game-changer in the domain of API integration and usage.

GitHub Repository: https://github.com/ShishirPatil/gorilla

Read the Paper: arXiv