GatorTron

GatorTron is a clinical-note-extraction-oriented language model family (base/medium/large, 345M-8.9B parameters), pretrained using a BERT architecture (via NVIDIA's Megatron package) on a dataset of 82B words of de-identified clinical notes from the University of Florida Health System, 6.1B words from PubMed CC0, 2.5B words from WikiText, and 0.5B words of de-identified clinical notes from MIMIC-III. It is a joint effort between the University of Florida and NVIDIA, validated in a clinical NLP human-evaluation study (UF News, 2023), and matches the PRD's own named example of a "clinical-note extraction model" directly.

Licence note: license: apache-2.0, confirmed directly against the raw Hugging Face model card YAML frontmatter (not derived metadata), matching this registry's Geneformer entry's own verification method.

Registry note on model_card_uri: this points to the model's standard Hugging Face model card, not a GA4GH Genomic AI Model Card v0.6.0-formatted document. No model has yet published a card in the GA4GH v0.6.0 format, since it is a newly drafted specification.