about the difference of evaluating after training in normal method

[xiaomengxin123]

Excuse me, I know lora a little, I use your method and I find it powerful in training. In my method, I save model as ".pth" after training and I use ".pth" to evaluating model but I find it worse than full params finetuning. So should I also "save_lora_parameters" after training and evaluate it by loading "save_lora_parameters" and ".pth"? Besides, does the parameters of "r" and "alpha" have an effect?