Inconsistency in saving and loading weights

After training a model for several epochs, i save the model and will load it then for further evaluation. During inference (after loading) the model does not perform as well as it did during training.
I am setting the model to eval() mode and i have

    torch.backends.cudnn.deterministic = True

in both parts of my code. Any ideas what i might be doing wrong?

Can you show the difference in weights between before saving the model and after loading the model?
And how you load and save the model?