This release contains three trained models (checkpoints) related to the "Stylometry for Real-World Expert Coders: a Zero-shot Approach" paper, respectively: - MLAllVocaBSoftAtt referees to the soft attention model trained with infoNCE loss without B.P.E, with bounding. - SoftAtt64k104AuthClass referees to the soft-attention model trained in a classification setup with 64k B.P.E. token, with bounding. - SelfAtt64k104authClass referees to the self-attention model