PyTorch usability on AMD hardware?

As of today, not easy to build, nor as effective as tensorflow benchmarks suggest. I would not recommend it. I’m keeping an eye open for next months upcoming ameliorations (maybe ?)