How to set training as False in dropout during eval
qibin-google opened this issue · 0 comments
qibin-google commented
I don't understand how setting self.training = False turns training off in dropout. self.training is passed as an argument when building dropout layer. After the whole network is built, changing self.training should not change the network.