Wrong indexes when used for NER

[montis96]

Hi,

I've executed the fine tuning of the model for NER. The problem is that when I try to use the model to find entities from a text, 90% of cases, the starting index of the entity predicted is one number before the real index of the word in the text.

Is it possible that I did something wrong during the fine tuning process or maybe it is a problem of the tokenizer?

Thank you