Hi! I am using core.Function to call BatchNorm symbol of MXNet. However, I cannot input moving_mean and moving_var to the symbol because it appears that core.Function only accepts trainable parameters. I hope core.Function could accept auxiliary states as well so that I could get acceleration from symbols such as BatchNorm.
Hi! I am using
core.Function
to callBatchNorm
symbol of MXNet. However, I cannot inputmoving_mean
andmoving_var
to the symbol because it appears thatcore.Function
only accepts trainable parameters. I hopecore.Function
could accept auxiliary states as well so that I could get acceleration from symbols such asBatchNorm
.