Table 4 Comparison of GatorTronS with existing transformer-based LLMs for semantic textual similarity, natural language inference, and question answering.

	Semantic textual similarity	Natural language inference	Question answering
	2019 n2c2²³	MedNLI²⁴	emrQA Medication²⁵		emrQA Relation²⁵
Transformer	Pearson correlation	Accuracy	F1 score	Exact Match	F1 score	Exact Match
ClinicalBERT	0.879	0.827	0.691	0.241	0.931	0.853
GatorTron, 90B	0.881	0.867	0.718	0.298	0.954	0.903
GatorTronS, 1B	0.853	0.851	0.702	0.288	0.965	0.924
GatorTronS, 5B	0.888	0.882	0.726	0.305	0.968	0.926
GatorTronS, 10B	0.893	0.886	0.728	0.311	0.972	0.929
GatorTronS, 20B	0.898	0.885	0.726	0.307	0.973	0.927

Quick links

Search