How do I make all hyperparameters, learnable, for example, number of epochs, learning rate, normalization mean, std, image size, random crop size, make all of these learnable.
No hyperparameters in my training.
How do I make every hyperparameter a nn.Parameter()
So that computer figures these things out on its own.
Exactly, have a look to the state of the art posts here. The thing is to find the metric that will indicates if your training is good and minimize this value. You can apply these concepts to your problem.