alexgoldberg/hebrew-manuscript-contents-ner

Hebrew manuscript Contents NER

BIO token-classifier (WORK / FOLIO / WORK_AUTHOR) trained on MARC 505 contents fields of Hebrew manuscripts. Part of the MHM (Mapping Hebrew Manuscripts) Pipeline.

Architecture: DictaBERT encoder + 2-layer custom NER head (Linear → ReLU → Dropout → Linear). Custom code in modeling_mhm_ner.py. Load with trust_remote_code=True. Not on HF Inference Providers serverless tier; use locally or on a dedicated endpoint.