Combining loss for multiple dataloaders

Thanks @goku . I looked that approach, but it doesnt really work, becuase the batch size of 2 datasets is different in my case