--- license: mit tags: [scientific-documents, figure-references] --- # figrefs models Trained models for finding the sentences that discuss each figure of a research paper (code: https://github.com/rcharan05/figrefs). | Folder | Model | Role | |---|---|---| | SPAN_ES | ModernBERT-large span tagger, early stopping | Recommended single model | | C0L | DeBERTa-v3-large cross-encoder | Ensemble (with SPAN) | | SPAN | ModernBERT-large span tagger | Ensemble (with C0L) | Trained on FigSum+ GS1 (15 biomedical articles). Use with `demo/run_local.py` from the code repository.