Apple quietly open-sources MLX to run models on Mac chips
Apple just quietly dropped MLX and MLX Data, a new machine learning framework tailored specifically for Apple Silicon. Inspired by PyTorch and Jax, it takes advantage of unified memory so CPUs and GPUs can process tasks without moving data, packing enough punch to train and run models like Llama and Stable Diffusion locally. Why it matters: Apple is finally giving developers the tools to turn MacBooks into capable generative AI workstations.
MLX is a new open-source machine learning framework created by Apple's machine learning research team, built specifically to squeeze maximum performance out of Apple Silicon. Alongside it, Apple also dropped MLX Data, a companion deep learning library for efficient data loading.
Basically, it gives developers a way to train and run modern generative AI models directly on a Mac without jumping through hoops.
What you can do with it
MLX is designed to bring serious model-training and inference capabilities right to Apple Silicon hardware.
According to Computerworld, the framework packs enough muscle to train and run hefty open-source models like Meta's Llama as well as Stable Diffusion for image generation.
Its companion package, MLX Data, handles the data ingestion side of the equation.
Apple machine learning researcher Awni Hannun noted on X that MLX Data is completely framework-agnostic. That means you can use it seamlessly with MLX, PyTorch, or Jax without having to rewrite your data pipelines.
How it works under the hood
The design of MLX borrows familiar ideas from battle-tested frameworks like PyTorch, Jax, and ArrayFire, making it feel intuitive to anyone who has already built deep learning projects.
The standout difference is how it handles memory.
Because Apple Silicon uses a unified memory architecture, MLX allows tasks running across supported devices, currently CPUs and GPUs, to operate on the same pool of memory without moving data back and forth. Eliminating that data transfer bottleneck gives it a huge efficiency advantage.
Both MLX and MLX Data are fully open-source and already available for developers on GitHub and PyPI.
Why it matters
If you are a developer working on an M-series Mac and want to experiment with local models, MLX is absolutely worth checking out today. Instead of wrestling with generic toolchains that treat Apple chips as an afterthought, you now have a framework engineered around their unified memory.
That said, MLX is brand-new and will not replace the massive CUDA or PyTorch ecosystem overnight. We see it as Apple finally turning the Mac into a credible local generative AI sandbox, giving developers a reason to keep their workflows on Apple Silicon instead of immediately renting cloud GPUs.
Sources (1)
- theverge.com(opens in a new tab)Originaltheverge.com