Everything seems to be very Cuda centric, has anyone tried training or other ml projects using AMD GPUs? How was it?