On September 22, Hugging Face described two changes for people who run AI on their own Macs. Its Transformers library can now run some GGUF files, a common format for smaller, quantized local models. The company also said Jun Kim, creator of the oMLX project, has joined to work on Apple’s MLX ecosystem.
Why another way to open a model file matters
A model file is a bit like a video file: you need compatible software to use it. Many local apps already run GGUF through llama.cpp, including Ollama and LM Studio. The new Transformers path lets developers use selected GGUF checkpoints in Python and PyTorch for experiments, evaluation, and local serving. Hugging Face still recommends llama.cpp when efficient local inference is the main goal.
The fast packed path currently targets Apple silicon and supported Qwen3.5 and Qwen3.8 text models. It needs recent compatible software; general GGUF support does not mean every file or computer gets the same speed. Hugging Face’s performance chart also compares measurements with different timing boundaries, so it is not a clean universal speed ranking.
Where MLX fits
MLX is a separate framework designed for Apple silicon. Hugging Face says it will fund ongoing work on oMLX and hopes to make it easier to bring new Transformers model definitions into MLX tools. That is a development direction, not an announcement that every Transformers model runs in MLX today. For a first local chat, our setup guide explains where Ollama, llama.cpp, MLX-LM, model files, and your computer fit together.