Clinical Camel 70B

Clinical Camel was released in May 2023 by researchers at the University of Toronto and Vector Institute as an open expert-level medical language model. Fine-tuned from LLaMA-2 with QLoRA, it introduced dialogue-based knowledge encoding, synthesizing 100,000 multi-step dialogues from 20,000 open-access clinical articles to teach the model medical reasoning.\n\nThe authors reported state-of-the-art results among open medical models at the time, surpassing GPT-3.5 on medical QA benchmarks including USMLE-style exams, PubMedQA, MedQA, and MedMCQA. They stress it is a research artifact requiring rigorous human evaluation before any clinical use.