Loading history…
MLX
This is r2 (the current revision), saved by jack. Full history
|
MLX
| |
| Developer | Apple machine learning research |
|---|---|
| Released | December 2023 — open source (MIT) |
| Type | Array framework for machine learning on Apple silicon |
| Languages | Python · Swift · C++ |
[ Show / Hide ]
| ||||||||||||||||||||||||||||||||||||||||||||
MLX is Apple's open-source machine-learning framework built specifically for Apple silicon — a NumPy-like array library whose defining trick is the M-series' unified memory: arrays live in memory shared by CPU and GPU, so computation moves between them without copies. Released quietly by Apple's ML research team in late 2023, it became the standard way to run and train models locally on a Mac.
1 Design and ecosystem✎
MLX pairs a familiar NumPy-style API with lazy computation and composable function transformations, in Python, Swift, and C++. Around it grew a practical ecosystem — mlx-lm for running and fine-tuning language models, an active MLX community on Hugging Face publishing quantized models — that turned high-memory Macs, the M3 Ultra Mac Studio above all, into favored local-LLM machines.
2 Position✎
MLX is Apple's most open AI artifact: MIT-licensed, community-driven, and aimed squarely at making the company's unified-memory hardware the best place to run models you keep to yourself. It is the researcher-and-tinkerer counterpart to the productized Apple Intelligence and the Apple Foundation Models underneath it.
Content available under CC BY-SA 4.0