EPFLiGHT/Apertus-70B-MeditronFO
Apertus-70B-MeditronFO is a 70 billion parameter medical specialist large language model developed by EPFLiGHT, fine-tuned from Apertus-70B-Instruct. It is part of the Fully Open Meditron family, emphasizing an end-to-end auditable pipeline for clinical LLMs with open weights, data, and training. This model achieves 54.56% medical accuracy on a five-benchmark average, outperforming its base model, and is optimized for medicine-related tasks with a 32768 token context length.
Loading preview...
Apertus-70B-MeditronFO: A Fully Open Medical LLM
Apertus-70B-MeditronFO is a 70 billion parameter medical specialist LLM developed by EPFLiGHT, building upon the Apertus-70B-Instruct base model. It is a flagship model within the Fully Open Meditron family, distinguished by its commitment to an end-to-end auditable pipeline for clinical LLMs, encompassing open weights, open data, an open training recipe, and a clinician-vetted corpus construction.
Key Capabilities & Performance
- Superior Medical Accuracy: Achieves 54.56% on a five-benchmark average (MedMCQA, MedQA, PubMedQA, MedXpertQA, HealthBench), significantly surpassing the 48.35% of its base model, Apertus-70B-Instruct.
- Clinician-Validated Preference: Preferred over Apertus-70B-Instruct in 89.7% of comparisons on AutoMOOVE, an LLM-judge evaluation based on 516 clinician-written vignettes.
- Extensive Medical Training: Fine-tuned on the Fully Open Meditron Corpus, which includes 533,888 examples of curated QA, synthetic QA, guidelines QA, and synthetic vignettes, with answers generated and validated by clinicians.
- Auditable Training: Features a transparent training recipe, including two-stage n-gram decontamination against evaluation benchmarks.
Intended Use & Limitations
This model is specialized for medicine-related tasks and is intended as an assistive tool. Users should be aware that generated content may not always be factually accurate or free from biases, and critical evaluation of all outputs is essential. It should not be used as a definitive source of medical information.