Sudden Increase in Loss value while finetuning

[nikithakriz]

I am finetuning tapas in custom dataset. After creating the dataset while training there is an sudden increase in loss value. After debugging i understood that some of the files creating this issue. But i didn't get the exact logic behind the issue and why that files creating the issue. They are more common like other files. After removing this files from teh dataset the training started with normal loss value.

Issue in Loss Value:
Epoch: 0
Loss: 9963.3525390625