Apple’s machine learning framework arrives: here’s MLX

Apple recently released a machine learning framework and libraries to allow developers to create ML models that can run on Apple Silicon chips, with perhaps the ultimate goal of bringing generative AI applications to its systems. MLX, this is the name of the new framework, and MLX data libraries are available in open source form on GitHub and PyPI.

MLX is inspired by other frameworks such as PyTorch, Jax, and ArrayFire, with the difference of running on a shared memory system. Concretely, this translates into MLX activities that can run on supported devices without transferring data. MLX is designed to provide a simple user experience for developers, but without sacrificing power, with the ability to train AI models such as Llama by Meta or Stable Diffusion.

Apple has not announced or publicized the MLX release in any particular way: Apple ML researcher Aouni Hanoun reported on this in a post on your!)”

In recent years, Apple has shown a somewhat conservative approach to artificial intelligence and, in fact, has often been reluctant to use these labels formally and publicly in its communications, preferring the term “machine learning” to them. Although it has already incorporated many machine learning features into its applications, it has not yet offered the public a generative AI service or application like ChatGPT, Bard, Midjourney, Dall-E, or Claude.

More technical information for MLX is Available on GitHub.