problem about train process

[lingcon01]

Thank you for your excellent work. However, I encountered the following issue during training: I am using your code as well as the dataset released with TargetDiff. The loss for positions remains very high and does not decrease. Could you please let me know what might be causing this problem?