finetuning loss drop suddenly after 1 epoch

[anyili]

tried to finetune the model using alpaca data set. used same parameters in the github. What I found is the loss drops suddenly after every 1 epoch. And after 3 epochs, it seems the fine-tuned model has poor performance.
Did anyone have the same problem?

[zhisbug]

@anyili this is expected. Keep up your training!