Based on the report from the Verge Apple has released a new machine learning framework and model library named MLX, designed to run efficiently on Apple Silicon. This represents a move towards generative AI applications, diverging from its previously conservative approach to AI.
MLX and MLX Data are accessible via open-source platforms like GitHub and PyPI, and were inspired by frameworks such as PyTorch, Jax, and ArrayFire. A key feature of MLX is its shared memory, enabling tasks to run on supported devices without data transfer. It aims to be user-friendly for developers while providing enough power to train advanced AI models.
Awni Hannun from Apple described MLX Data as an efficient and flexible package for data loading, compatible with MLX, PyTorch, and Jax. Despite its long history of integrating AI into its products, Apple has focused on machine learning, avoiding generative AI applications unlike competitors like Microsoft and Google. The company has only recently started exploring foundational models for incorporation into its services.
MLX @ Git-Hub: https://ml-explore.github.io/mlx/build/html/index.html