I am training my WaveNet model using PyTorch on a MacBook with an M1 Pro.
When I train it on the CPU, the loss looks normal, but it’s super slow. The loss is decreasing from around 5.54.
However, if I train it on MPS, the loss explodes. The loss starts from 5.54 and goes up to 50000…
Should I set up something for the MPS?