Save model for resuming training

I implemented a GAN model and because I need to train it for 500 epochs, I’ve saved the result of each 10 epochs for both models:{
            'epoch': epoch + 1,
            'gen_state_dict': gen.state_dict(),
            'disc_state_dict': disc.state_dict(),
            'gen_optim': opt_gen.state_dict(),
            'disc_optim': opt_disc.state_dict(),
            }, os.path.join("", 'gan_epoch-{}.pt'.format(epoch + 1)))

and I load it:

disc = Discriminator(in_channels=3).to(device)
gen = Generator(in_channels=3).to(device)

checkpoint = torch.load("/content/drive/MyDrive/Epochs/")




The code works well but I am wondering if the results will be correct, I have noticed that the training becomes faster, before saving the models, one epoch takes 20 minutes, now it takes only 8 minutes also the discriminator loss increases a lot from to is this normal?

Did you make sure the same datasets are loaded and contain the same number of samples?
I would be suspicious, if the training suddenly sees a 2x speedup without changing anything else.

Thank you for your reply.
Yes, I’m sure the same dataset is loaded, I ran it for the second time, and now it took 20 minutes, I’m really confused why this happened