OLMoE: Open Mixture-of-Experts Language Models

Benchmark Model Rank Results
question-answering-on-copa-zero-shotOLMoE-1B-7B#5Accuracy: 78.0
common-sense-reasoning-on-arc-easyOLMoE-1B-7B (5-shot)#10Accuracy: 84.2
instruction-following-on-ifevalOLMoE-1B-7B-Instruct#23Prompt-level loose-accuracy: 48.1
multi-task-language-understanding-on-mmluOLMoE-1B-7B (5-shot)#58Average (%): 54.1
question-answering-on-piqaOLMoE-1B-7B (5-shot)#64Accuracy: 79.8
sentence-completion-on-hellaswagOLMoE-1B-7B (5-shot)#76Accuracy: 80.0
common-sense-reasoning-on-winograndeOLMoE-1B-7B (5-shot)#78Accuracy: 70.2
arithmetic-reasoning-on-gsm8kOLMoE-1B-7B-Instruct (8-shot)#245Accuracy: 45.5