---
title: README
emoji: 📈
colorFrom: indigo
colorTo: yellow
sdk: static
pinned: false
thumbnail: >-
https://cdn-avatars.huggingface.co/v1/production/uploads/66ea90952fef8317d8fba8ec/SgCp3hhhVu3HCuHt8sYn5.png
---
## About Us
Meet **LiGHT**, the [Laboratory for Intelligent Global Health & Humanitarian Response Technologies](https://www.light-laboratory.org/), based at EPFL (School of Computer Science), Ariadne Labs (Harvard), and the Koita Centre for Digital Health at [Ashoka University](https://www.ashoka.edu.in/).
We build clinically grounded, auditable AI for medicine — co-designed and evaluated with clinicians, humanitarian responders, and data scientists worldwide.
---
## 🆕 Fully Open Meditron
Our latest release, **Fully Open Meditron**, is the first end-to-end auditable pipeline for clinical LLMs — open weights, open data, open training recipe, and clinician-vetted corpus construction.
**Apertus-70B-MeditronFO** establishes a new state of the art among fully open medical LLMs, and **Gemma-3-27B-MeditronFO** outperforms MedGemma on HealthBench Hard despite being trained from a fully open pipeline.
👉 Browse the [**MeditronFO Collection**](https://huggingface.co/collections/EPFLiGHT/meditronfo) — six models across three families (Apertus, OLMo-2, EuroLLM, Gemma-3), plus the [Fully Open Meditron Corpus](https://huggingface.co/datasets/EPFLiGHT/fully-open-meditron) (~601k clinician-audited examples).
---
## Previous Work
- [**Meditron-3**](https://openreview.net/pdf?id=ZcD35zKujO) — open medical LLM suite spanning Llama-3, Phi-4, Qwen2.5, and Gemma-2 bases. See the [Meditron-3 Collection](https://huggingface.co/collections/EPFLiGHT/meditron-3).
- [**MEDITRON**](https://arxiv.org/pdf/2311.16079) — original 7B/70B medical specialist with curated clinical guidelines.
---
#### 🌐 Stay connected
- **Prof:** [Annie Hartley](https://www.linkedin.com/in/annie-hartley-324832210/)
- **Lab:** [light-laboratory.org](https://www.light-laboratory.org/)