Bad results with pre-trained blank models

[TANGTANG-BOY]

Hello, I have recently been using this program of yours for learning, but I have noticed that when I use the blank pre-trained model for task fine-tuning I get poor results, not up to what is shown in your results, I haven't changed any of the code, but it seems like there is overfitting when I change the learning rate, the results are as follows: