Rank/Neuron/Module Dropout

[sangoi-exe]

Please, I would like to request, if it's technically feasible, the implementation of dropouts.

I wasn't familiar with OneTrainer until last week and used to use Kohya, but among all the differences, what attracted me the most was the backup feature of OneTrainer, allowing you to stop and restart training at any time.

However, so far the only thing I'm kind of missing are the dropouts that I was used to, and which help a lot to avoid overfitting and 'challenge' the neural network a bit more to find better solutions.