[Python] What is the relationship between Intel Extension for PyTorch and PyTorch XPU versions?

A while ago, I was training a deep learning model on a computer without an NVIDIA GPU but with an Intel GPU. I only used the CPU for training, which was painfully slow. It suddenly occurred to me: can the PyTorch framework leverage Intel GPUs for acceleration? I came across the Intel Extension for PyTorch and the PyTorch XPU version. I’m quite confused about the relationship between these two—could someone clarify this for me?

I’ve gone through the relevant documentation, but perhaps due to a lack of background knowledge or not reading carefully enough, I still couldn’t clearly understand the relationship between them.

Continue reading...