MEDITRON-70B: Scaling Medical Pretraining for Large Language Models

Benchmark Model Rank Results
few-shot-learning-on-medconceptsqaepfl-llm/meditron-70b#7Accuracy: 25.262
few-shot-learning-on-medconceptsqaepfl-llm/meditron-7b#11Accuracy: 23.787
multiple-choice-question-answering-mcqa-on-21Meditron-70B (CoT + SC)#11Dev Set (Acc-%): 66.0
question-answering-on-medqa-usmleMeditron-70B (CoT + SC)#6Accuracy: 70.2
question-answering-on-medqa-usmleLLAMA-2 (70B SC CoT)#8Accuracy: 61.5
question-answering-on-medqa-usmleLLAMA-2 (70B)#10Accuracy: 59.2
question-answering-on-pubmedqaMeditron-70B (CoT + SC)#1Accuracy: 81.6
zero-shot-learning-on-medconceptsqaepfl-llm/meditron-7b#5Accuracy: 25.751
zero-shot-learning-on-medconceptsqaepfl-llm/meditron-70b#7Accuracy: 25.360