May I ask if any techniques were used during the training process of this model?

[jungyin]

Hello, I am trying to collect my dataset again and train the model, but during the training process, I tried using 1e-5 LR and after 100 epochs of training, the performance, especially on HRV, was very poor. As shown in the figure below

However, when I trained the model using the pre trained, pure dataset, the performance was instantly very good, as shown in the following figure

Then, I tried using the pure dataset and trained again, but it was still very poor. So, I want to know, may I have done something wrong? I am currently using MAE as the loss calculator