Training issue

[Akhillalpp]

During training, if testset evaluation is removed, then model returns a different result. That is if we Remove lines

print("Expert 1 - Test") dice_test = eval(test_loader, criterion_seg, model, n_classes=n_classes)

from the below code, training result is different.

        if t % 10 == 0 or t > 45:
        print("Epoch", t, "/", iterations)
        print("Validation")
        dice = eval(val_loader, criterion_seg, model, dice_s=True, n_classes=n_classes)
        print("Expert 1 - Test")
        dice_test = eval(test_loader, criterion_seg, model, n_classes=n_classes)

Is it implies the evaluation affects the model parameters? That's why we are getting a different result?
model.eval() is given in eval function. But still results are different. Why?

[Deepthi-vr-code]

Found same issue. Can anyone clarify on this