Modular Inc. today announced the general availability of its Mojo programming language, which promises to speed up artificial intelligence applications by multiple orders of magnitude. 

The language is becoming available for download as part of a software development kit. The SDK includes not only Mojo but also supporting components designed to ease developers’ work. It’s launching a few weeks after Modular raised $100 million in funding from General Catalyst, Alphabet Inc.’s GV and other institutional investors.

Developers usually write AI software in the Python programming language, which provides a syntax that is not only simple to understand but also concise. That means complex tasks can be accomplished using relatively few lines of code. But alongside its benefits, Python has a major shortcoming: It’s relatively slow, which affects the performance of AI applications written in the language.

Modular developed its Mojo language as a faster alternative to Python. According to the startup, a program written in Mojo can run up to 68,000 times faster than if it were implemented using Python. That value proposition has drawn significant interest: The startup disclosed today that more than 120,000 developers had signed up to try Mojo ahead of today’s launch.

There are several other, more established programming languages that are likewise significantly faster than Python. However, they’re also more difficult to learn and use, which complicates developers’ work. A core selling point of Mojo is that its syntax is similar to Python in many respects, which means writing AI code is relatively simple.

One major contributor to Python’s ease of use is that it offers a large number of packages. Those are prepackaged software building blocks written by the open-source community. Instead of building an AI application from scratch, developers can assemble it from a combination of open-source Python packages and custom code, which saves time.

Modular says that Mojo can be used together with many existing Python packages. As a result, developers don’t have to start with a blank canvas when writing a new application.

Mojo also promises to simplify AI projects in other ways. Most notably, it removes the need to use multiple programming languages in the same application.

In enterprise AI projects that prioritize performance, developers often combine Python with a faster language such as C++ or Nvidia Corp.’s CUDA. Python is used to implement the relatively simple components of an AI application. The faster programming, in turn, is used to write performance-sensitive components that must run as fast as possible.

Writing an application in multiple programming languages makes the code more difficult to read for developers, which can reduce productivity. It also makes the code trickier to troubleshoot when errors emerge. Debugging tools are often not well-equipped to troubleshoot software written in multiple languages.

Because of its speed, Mojo can be used to write both the performance-sensitive components of an AI application and the rest of its code base. Modular says this feature removes the need to use multiple programming languages and thereby eases developers’ work. 

“Mojo combines the usability of Python that researchers love with the systems programming features (types, user-defined zero-cost abstractions, memory safety, auto-tuning, compile-time metaprogramming) that AI developers usually get from C, C++, or CUDA,” Modular product manager Pramod Ramarao wrote in a blog post. 

Mojo is becoming available today as part of an SDK that combines the language with several supporting components. Those components are designed to reduce the amount of effort involved in writing Mojo-powered applications. 

The first component of the SDK is a shell, or command line tool, that developers can use to manage the formatting of their code and create documentation that explains how it works. They can also use the shell to turn the raw code into a functioning program.

The tool is available alongside an extension for Visual Studio Code, Microsoft Corp.’s widely-used code editor. The extension displays autocomplete suggestions in the editor’s interface. It also offers several other productivity features, including the ability to quickly access explanations about the Mojo syntax from within the Visual Studio interface.

The third component of Modular’s new SDK is an integration with Jupyter Notebook. That’s a popular, open-source coding tool often used for AI development tasks. Jupyter Notebook allows developers to write code and then run it to test how it works in a single interface, which speeds up prototyping.

Modular says its SDK is currently available on Linux with support for central processing units. In the coming months, Modular plans to add support for Windows and macOS along with the ability to run Mojo code on graphics cards. In the longer term, the startup plans to extend the SDK with several additional features, including a debugging tool for fixing code errors. 

Image: Unsplash