Original filenames info in the predictions file

Hi

Thank you for your work!

Would it be possible to include information relative to the original filenames of the evaluation dataset in the predictions_eval.json file? I.e. to associate a given list of mentions in the predictions file to the original filename (through a key or something)?