Apple-Ki framework MLX: In the future support for Nvidias Cuda

Anyone with Apple’s useful machine learning framework MLX So far, had to work for a Mac with as many M-chip power as possible. Fans of tools such as LM studiowith which local LLMs can be carried out, swear on it under MacOS and pull accordingly MLX-optimized variants of Llama, Qwen or Mistral, which should better take advantage of Apple Socs and their unified memory. MLX could now spread on other platforms when it comes to some developers. As part of the Github project behind MLX On a Cuda-Backend filed. Means: Nvidia-based systems could use MLX in the future.

Develop on the Mac, perform on the cluster

However, the project is still at the beginning. If there is a product maturity, it would be conceivable that MLX-optimized models Comparatively simple for the more powerful Nvidia GPUs could be, for example as part of the training. In addition, MLX models could at least theoretically spread faster because Cuda has a much larger community behind.

MLX is specifically optimized for Apple Silicon and its GPU interface metal. In the case of Nvidia GPUs, Metal to Cuda (stands for Compute Unified Device Architecture) would then be changed. MLX comes with bindings for Swift, C and C ++. The main developer behind the Cuda baking with the user name zcbenz comes from Japan. According to his information, Apple finances parts of the work.

Developer: Apple involved in the project

“There are mainly two reasons for a Cuda baking,” says the developer. “Firstly, Cuda Unified Memory supports (…) Second, Nvidia hardware is widespread in the scientific field and on large computing clusters (” massive computations “, writes Zcbenz in the original).” It is therefore useful if you could write and test code locally on your Mac in order to then exercise it on supercomputers. “That would be a good developer experience.”

It remains to be seen how quickly the project moves forward. It started in spring. This week Zcbenz announced that the Cuda baking was still in development and not officially published. “There will be more information” if things are in a better form “.