Implement PyTorch inverted dropout from a supplied mask during training while returning inputs unchanged in evaluation mode.
