Modular open-sources its Mojo AI programming language’s core components

AI

Modular Inc. today open-sourced the core components of Mojo, a programming language designed for writing artificial intelligence software. 

The code is available under a customized version of the Apache 2 license. Modular added modifications designed to make Mojo easier to combine with software distributed under GPL2, another popular open-source license. The highest-profile project that uses GPL2 is the Linux kernel.

Palo Alto, California-based Modular is backed by $130 million in funding from Alphabet Inc.’s GV fund and other investors. The company made its Mojo programming language generally available last August. Since then, it has built up a user base of more than 175,000 developers and 50,000 organizations.

AI models are often written in multiple programming languages. Developers implement a neural network’s simplest components in Python, a language that is easy to learn but provides relatively slow performance. The rest of the code is typically written in C++, which is both faster than Python and considerably more difficult to use.

Modular positions Mojo as a more convenient alternative. It provides a relatively easy-to-use syntax that is similar to Python, but runs code up to thousands of times faster. The result is that developers can write speedy AI models without learning a complex technology such as C++.

The Mojo components that Modular open-sourced today form the language’s so-called standard library. This is the part of a programming language that includes its core syntax elements and other essential features. Mojo’s standard library includes features for tasks such as optimizing AI hyperparameters, the settings that determine how a neural network processes data.

The company says that it’s not only making its standard library code publicly accessible but will also enable external developers to contribute improvements. Additionally, it will release new versions of Mojo’s compiler on a daily basis. The goal is to let developers test their contributions to the standard library on the latest version of the compiler. 

Modular uses Mojo to power a commercial AI platform, MAX, that it debuted late last year. The platform includes tools that ease tasks such as running AI applications in Kubernetes environments. the company detailed today that it plans to make some of MAX’s components available under an open-source license down the line.

In conjunction, it debuted a new version of the platform with several feature additions. According to Modular, one of the enhancements will make it easier to mix Mojo and Python code in the same AI project. The company has also added performance improvements designed to speed up AI inference. 

Image: Unsplash