Models / bio-nlp / pubmedbert-base-embeddings / benchmarks

bio-nlp/pubmedbert-base-embeddings

Visibility: PUBLIC

Model card FilesBenchmarks

Group by

BC5CDR chemical NER bio-nlp/bc5cdr-chemical-ner

Drawn from: bc5cdr

atlas-n100-v1

Metric	Value	95% CI	n	Rank	Best in catalogue	Source	Overlap	Board
Chemical entity F1 (token probe) ↑ higher is better · probe	0.805	[0.762, 0.845]	100	11 / 16	0.889	reproduced	overlap Train/test overlap declared by the evaluator: PubMed abstracts	Board
Chemical entity precision (token probe) ↑ higher is better · probe	0.797	[0.742, 0.851]	100	11 / 16	0.870	reproduced	overlap Train/test overlap declared by the evaluator: PubMed abstracts	Board
Chemical entity recall (token probe) ↑ higher is better · probe	0.812	[0.769, 0.854]	100	11 / 16	0.914	reproduced	overlap Train/test overlap declared by the evaluator: PubMed abstracts	Board

Clinical entity recognition (MACCROBAT) bio-nlp/clinical-entity-recognition

Drawn from: MACCROBAT_biomedical_ner

atlas-n100-v1

Metric	Value	95% CI	n	Rank	Best in catalogue	Source	Overlap	Board
Entity F1, 3 types (token probe) ↑ higher is better · probe	0.489	[0.414, 0.562]	100	8 / 16	0.724	reproduced	overlap Train/test overlap declared by the evaluator: PubMed Central full text	Board
Entity precision, 3 types (token probe) ↑ higher is better · probe	0.464	[0.388, 0.537]	100	8 / 16	0.709	reproduced	overlap Train/test overlap declared by the evaluator: PubMed Central full text	Board
Entity recall, 3 types (token probe) ↑ higher is better · probe	0.517	[0.437, 0.597]	100	10 / 16	0.739	reproduced	overlap Train/test overlap declared by the evaluator: PubMed Central full text	Board
Entity macro F1, 3 types (token probe) ↑ higher is better · probe · secondary	0.491	[0.408, 0.556]	100	—	0.728	reproduced	overlap Train/test overlap declared by the evaluator: PubMed Central full text	Board

PubMedQA accuracy bio-nlp/pubmedqa-accuracy

Drawn from: PubMedQA

atlas-full-v1

Metric	Value	95% CI	n	Rank	Best in catalogue	Source	Overlap	Board
Accuracy (logistic probe) ↑ higher is better · probe	0.516	[0.480, 0.558]	500	14 / 17	0.576	reproduced	overlap Train/test overlap declared by the evaluator: PubMed abstracts	Board
Macro F1 (logistic probe) ↑ higher is better · probe	0.362	[0.318, 0.408]	500	12 / 17	0.444	reproduced	overlap Train/test overlap declared by the evaluator: PubMed abstracts	Board

Spanish occupation entity recognition (ProfNER) bio-nlp/spanish-occupation-entity-recognition

Drawn from: ProfNER_corpus_NER

atlas-n100-v1

Metric	Value	95% CI	n	Rank	Best in catalogue	Source	Overlap	Board
Entity F1, PROFESION + SITUACION_LABORAL (token probe) ↑ higher is better · probe	0.259	[0.077, 0.413]	100	7 / 16	0.630	reproduced	none declared	Board
Entity precision, PROFESION + SITUACION_LABORAL (token probe) ↑ higher is better · probe	0.636	[0.429, 1.000]	100	8 / 16	1.000	reproduced	none declared	Board
Entity recall, PROFESION + SITUACION_LABORAL (token probe) ↑ higher is better · probe	0.163	[0.042, 0.300]	100	8 / 16	0.535	reproduced	none declared	Board
Entity macro F1, PROFESION + SITUACION_LABORAL (token probe) ↑ higher is better · probe · secondary	0.137	[0.048, 0.400]	100	—	0.329	reproduced	none declared	Board