Loss increase when load pth

I train my model and loss decrease to 2.0+. But when I load the pth and training continue, the loss increase to 3.0 - 4.0+, I need retrain the model that loss decreate to 2.0+. But I reload pth again, the effect is same.

How are you testing the accuracy after loading?
If you are using the validation set, did you make sure to call model.eval() before?
Are you also restoring the optimizer for the training after reloading?

I fix I train dataloader not use shuffle.