It looks as though NMT assumes that it is being run on a CUDA-enabled machine – numerous things in init.py throw exceptions if you don’t have CUDA. I understand that Torch can run on ARM-based Macs, but I can’t see what you would have to do to the maze of things in init.py in order to make this work.
Hi Allan,
Maybe try Google Colab or Lightning Studios?
Yeah, I can go that way, but I quite like doing things locally if I can – it makes integrating things easier. There’s M3 support for tensorflow, so it’s a bit annoying that OpenNMT is written in such a way that you can’t use this.