Request for CPU-only or ONNX support to enable lightweight deployment

[tavallaie]

Hi, I really appreciate the lightweight design of the Perth model!
However, I noticed that despite its small size, it currently requires PyTorch with GPU support, which brings in heavy dependencies like CUDA.

Would it be possible to provide a version compatible with CPU-only PyTorch or even export the model to ONNX format? This would greatly help in deploying the model on resource-constrained environments such as embedded systems, low-power devices, or even directly in web browsers via ONNX.js or similar runtimes.

Thanks for your great work!

P.S. also, maybe this approach, solve #6