[muxr]

zendnn is for CPUs. Pytorch ROCm version doesn't require you to change a single line of code. So all the same semantics like `device = "cuda"` work the same when using AMD GPUs.