how to set is_train in resent_mixup when eval on validation set?

[happywmzzz]

Hi, this is a very nice interpretation of mixup using mxnet, I have a question about the "is_train" in resnet_mixup, I think the mixed data and labels are used only in training process, when eval accuracy the data is not mixed.
But in this code, I don't know where set the "is_train" false when get test acc in training. Thx if providing some more information.