[olliej]

This seems to be really short training - I though modern NN took hours to train (for prod scale), even with hardware acceleration?