Check out our model release HERE
Meditron is the latest update to our suite of state-of-the-art open Large Medical Language Models trained fromĀ Llama-3.1Ā [8B] and [70B]
---
šÆ OPEN weights released here šĀ https://lnkd.in/eF8q9A94
š¤CO-DESIGNED: from day one with an interdisciplinary team of practicing clinicians from around the world, global humanitarian organizations, AI ethicists, and data scientists
šSTATE OF THE ART: Out-performing GPT4 on the standard medical benchmarks (*see footnote)
šREPRESENTATIVE: Continually pre-trained on an enormous corpus of expert-curated open-access medical literature and global clinical practice guidelines that is specifically inclusive of low-resource and humanitarian settings
š¤š¤ŗ join a diverse global community of clinical experts š©šæāāļøš§š»āāļøšØš¼āāļøš©š½āāļøš§š¾āāļøš©š¼āāļøšØš»āāļø in theĀ MOOVE. Sign up to show your interest here šĀ https://lnkd.in/eQb3KyDc
--
Built with ā¤ļø by an incredible and mostly volunteer collaborative ofĀ
--š¼Students and researchers atĀ EPFL,Ā Yale School of Medicine, andĀ Carnegie Mellon University Africa
--āļøHumanitarian advisors at theĀ International Committee of the Red Cross - ICRCĀ
--š„Clinicians from across the world from Nairobi to Lausanne!
š·ļø Tagged in the comments by our phenomenal team leadĀ Alexandre SallinenĀ
I am extremely lucky and grateful to work with you all ā¤ļøĀ
(*) While these benchmarks are helpful indicators of performance, we certainly canāt rely on results from static, hackable, generic multiple-choice questions from medical exams with a skewed geographic representation.Ā medicineisnotmultiplechoice
We need robust,Ā transparentĀ real worldĀ representativeĀ expert evaluationĀ in massive-scale clinical trials i.e.#jointhemooveĀ š·š¼š¹šæšØšš°šŖš³š¬šæš¦šŗšøšŗš¬šØš¦ā¦Greetings from Kigali!
More models coming soon!