Masking is mainly needed for training so we can paralellize our training, but during inference it is not usually needed but it comes up implicitly
Masking is mainly needed for training so we can paralellize our training, but during inference it is not usually needed but it comes up implicitly