List your company
ResearchFormerly WizWand

Commonsense Reasoning on OBQA (first-token accuracy)

91.4First-Token AccuracyPhi-4-14B
36.69650.89865.179.302May 21, 2025
Updated 7mo ago

Evaluation results

MethodLinks
Phi-4-14BStrategy=Prompt Engineering2025.0591.4
Phi-4-14BStrategy=Prefilling2025.0590
Gemma-2-9BStrategy=Prefilling2025.0589.8
Gemma-2-9BStrategy=Prompt Engineering2025.0587.6
Ministral-8BStrategy=Prefilling2025.0585.2
Llama-3.1-8BStrategy=Prefilling2025.0584.8
Qwen-2-7BStrategy=Prefilling2025.0584.8
Qwen-2-7BStrategy=Prompt Engineering2025.0584.2
Phi-4-14BStrategy=FTP2025.0584
Qwen-2-7BStrategy=FTP2025.0583.8
Ministral-8BStrategy=FTP2025.0582.4
Mistral-Nemo-12BStrategy=Prefilling2025.0580.8
Mistral-Nemo-12BStrategy=FTP2025.0580.6
Llama-3.1-8BStrategy=Prompt Engineering2025.0580.4
Ministral-8BStrategy=Prompt Engineering2025.0576.4
Zephyr-7BStrategy=FTP2025.0573
Llama-3.1-8BStrategy=FTP2025.0571.2
Gemma-7BStrategy=Prefilling2025.0571.2
Zephyr-7BStrategy=Prefilling2025.0570.8
Mistral-Nemo-12BStrategy=Prompt Engineering2025.0569.8
Zephyr-7BStrategy=Prompt Engineering2025.0567.6
Gemma-7BStrategy=Prompt Engineering2025.0563.8
Gemma-7BStrategy=FTP2025.0540.6
Gemma-2-9BStrategy=FTP2025.0538.8