MedScribe SOAP Training Data — 712 Curated Samples
Training, validation, and test splits for fine-tuning
google/medgemma-4b-it to
generate concise clinical SOAP notes.
Used to train the MedScribe SOAP LoRA adapter.
Dataset Description
712 medical encounter transcript → SOAP note pairs designed to teach a
language model to produce concise clinical shorthand rather than verbose
textbook prose.
Each sample consists of:
