List your company
ResearchFormerly WizWand

Multi-task Language Understanding on MMLU (MMLU Performance (MM))

65MMLU Performance (MM)Baseline
47.73652.21856.761.182Jun 2, 2026
Updated 5mo ago

Evaluation results

MethodLinks
BaselineBackbone=Llama-3.1-8B-Instruct2026.0665
NoiseBackbone=Llama-3.1-8B-Instruct2026.0664.6
MeanBackbone=Llama-3.1-8B-Instruct2026.0663.9
EMBERBackbone=Llama-3.1-8B-Instruct2026.0663.6
RMUBackbone=Llama-3.1-8B-Instruct2026.0662.2
RMU + EMBERBackbone=Llama-3.1-8B-Instruct2026.0662
SNMF + EMBERBackbone=Llama-3.1-8B-Instruct2026.0660.4
CRISP + EMBERBackbone=Llama-3.1-8B-Instruct2026.0660.4
CRISPBackbone=Llama-3.1-8B-Instruct2026.0659.3
SNMFBackbone=Llama-3.1-8B-Instruct2026.0658.7
NoiseBackbone=Gemma-2-2B-it2026.0654.8
MeanBackbone=Gemma-2-2B-it2026.0654.2
BaselineBackbone=Gemma-2-2B-it2026.0654.2
EMBERBackbone=Gemma-2-2B-it2026.0653.7
CRISP + EMBERBackbone=Gemma-2-2B-it2026.0653.2
CRISPBackbone=Gemma-2-2B-it2026.0652.6
RMU + EMBERBackbone=Gemma-2-2B-it2026.0651.8
SNMF + EMBERBackbone=Gemma-2-2B-it2026.0651
RMUBackbone=Gemma-2-2B-it2026.0650.3
SNMFBackbone=Gemma-2-2B-it2026.0648.4