why my loss value is fluctuating

[leechangdong]

Hello I found the project interesting and am adding my custom data to try out the new finetuning
I'm using LORA and 7B models due to gpu memory
However, I'm currently training and the loss value is jagged instead of decreasing.

I am attaching my training script
If anyone has any idea what the problem is, I would appreciate a response. Thank you.