>>106521872
It's very large. Because when you run with CPU, it uses float32, and when you run with GPU, it uses float16. Plus all those Pytorch libraries and DLLs take up an extra few GB on top of the actual model size.