microsoft/MediPhi

microsoft/MediPhi is a collection of 3.8 billion parameter Phi-3.5-mini-instruct based small language models, developed by Microsoft Healthcare & Life Sciences, specialized for medical and clinical natural language processing. This model, MediPhi, unifies five expert models fine-tuned on diverse medical corpora (PubMed, Medical Wikipedia, Medical Guidelines, Medical Coding, clinical documents) using the BreadCrumbs merging technique. It is designed for research in medically adapted language models, particularly in memory/compute constrained and latency-bound environments.