Baseline model trained on the T2KNOW corpus (Huntington's disease abstracts, 40 UMLS-derived labels, nested and same-span multi-label entities). Three training seeds are provided in seed_12345/, seed_23456/ and seed_34567/.
microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltextcode/ (Zenodo DOI above)The checkpoint uses the custom BertForSpanClassification head from the release code (code/flat_baselines/modeling.py); it cannot be loaded with a stock AutoModel head. Decode with logit_threshold=0.5 and start_confidence=0.9 (decoding.json), which were selected on the validation split.
Trained on the T2KNOW v1.0 document-disjoint train split (validation split for model selection). The v1.0 sentence-level split files contained duplicated sentence records; the corrected release v1.0.1 removes them without changing any annotation. No source text is included in this repository.
Overlap-aware scores from the official T2KNOW evaluator (partial span matches count half) on the de-duplicated T2KNOW v1.0.1 test split (151 documents, 1,413 sentences, 12,579 entities). Predictions are the archived outputs of these checkpoints.
| Seed | Precision | Recall | F1 |
|---|---|---|---|
| 12345 | 0.5739 | 0.7488 | 0.6498 |
| 23456 | 0.5782 | 0.7558 | 0.6552 |
| 34567 | 0.5907 | 0.7466 | 0.6595 |
Mean F1 over the three seeds: 0.6548 (population std 0.0040).
Scores measure agreement with the T2KNOW annotation layer (expert-checked, single annotator, see the paper), not semantic correctness. Very-low-support labels are unreliable.
MIT. The base model microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext is MIT-licensed (Hugging Face model metadata). T2KNOW annotations used for training are MIT-licensed; no third-party source text is included.
Repository: https://huggingface.co/edugredu/t2know-biomedbert
seed_12345/model.safetensors 436 MBseed_23456/model.safetensors 436 MBseed_34567/model.safetensors 436 MBmicrosoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext