Python Gets Its Mojo Working for AI

04/07/2023

A new superset of Python, called Mojo, is being developed to address Python’s performance issues in the artificial intelligence (AI) field. Currently, developers rely on other performant languages like C or Rust for specific sections of code in AI projects, which adds complexity and makes debugging challenging. Mojo aims to be a strict superset of Python, providing full compatibility with the Python ecosystem while offering predictable, low-level performance and control. It also allows for the deployment of subsets of code to accelerators, providing the advantages of systems-level code without the need for C or C++ code below Python. Mojo is being developed by a team led by Chris Lattner, who previously worked on LLVM and created “syntax sugar” for Swift.

The development team claims that Mojo unlocks the features of any device in a way that Python cannot, with benchmarks showing performance gains of up to 3500x. However, these numbers may vary depending on the hardware used. Mojo is currently in development and will be open-sourced, although a release date has not been announced yet. The Mojo playground allows users to explore Mojo code, and the documentation provides detailed insights into the language and its features. While Python is expected to remain a leading language in AI and machine learning, Mojo, and similar solutions aim to bridge the gap and enhance its performance.