Now that the required task is known though, it doesn't really matter. If AMD understand that, they should have no problem putting engineers on making Pytorch work well.

Considering its importance, it shouldn't be one engineer. It should be 50+.