I was wondering how could I load the weights and the optimizer state of an already trained model into another with a similar architecture but with different "loss_weights&q