How the pytorch freeze network in some layers, only the rest of the training?

4

I am wondering whether to set .eval() for those frozen layers since it may still update its running mean and running var.