Record 016 · Runtimes · 2023
67%23.16.U
MLX
Apple
“Apple's array framework that made local LLMs and diffusion native on M-series.”
MLX (and mlx-lm) uses unified memory so a 128GB Mac Studio is a model host, not a GPU-rental client. The local-Mac stack.
Opened first-class local ML on Apple silicon.
- Unified-memory arrays
- mlx-lm inference
- Fine-tune on Mac
NumPy for the unified memory Mac.
