Why is my token classification model performing poorly on custom data?

I fine-tuned a token classification model on my own dataset, but the accuracy is low. What could be the possible reasons?