3 ms·

yeah i mean that is exactly why distillation works. if you just were one hotting it would be the same as training on same dataset