Clinical-T5-Large

Clinical-T5-Large is a generative clinical language model from Columbia University, published at EMNLP 2022 Findings. It takes T5-large and continues pretraining on MIMIC clinical text, producing a text-to-text model tuned for the vocabulary and phrasing of real hospital notes. The authors evaluated it on clinical document classification, concept extraction, and natural language inference tasks.

A 2024 follow-up study found clinical T5 models give only marginal gains over general models and generalize worse across clinical domains, so it is best treated as a research baseline rather than a deployment-ready clinical tool. Checkpoints are public on Hugging Face (luqh/ClinicalT5-large) and PhysioNet.