General Reflection on AI and Fragmentation

[vricosti]

First of all, a huge thank you for your work. I stumbled upon it by accident after having issues trying to use the original Python version. I also found a Swift fork that seems to be based on your work initially. I'm new to the AI "game," but I already see fragmentation regarding OS/hardware. This implementation is pure CPU, and if I want a faster version, I would need a GPU version, and generally, CUDA is preferred or CoreML in the Apple world. Similarly, if we use CoreML, I understand that we need to convert the models into a specific format. Since Swift is available on Windows and Linux, I was almost wondering if it wouldn't be simpler to implement CoreML on these platforms and use the Swift version, ...
Is implementing a GPU version complicated and time-consuming, and if someone does it, will it work on an AMD card, for example?
Will the GPU version be much faster on a macintel with a card like a 5700XT, for example?
Sorry for all these questions.