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

[SpandanMadan]

Try this. Reduce your learning rate drastically. Try viewing your gradients wrt input and see if there’s any place where they are blowing up(going to inf). If yes,see why this happens.