Have AMD said anything about whether the NPU will be OK for a little bit of training as well, or just inference?