List your company

Long-context language understanding on RULER 0 shot v1 (test)

94.71CWE ScoreVanilla
84.93487.47290.0192.548Nov 3, 2025

Evaluation results

MethodLinks
VanillaModel=Llama 3.1 8B, Compression Ratio (CR)=12025.1194.7192.111,00057.2255.7259.9
GEARModel=Llama 3.1 8B, Compression Ratio (CR)=52025.1194.0191.911,00056.8255.7259.7
VanillaModel=Mistral NeMo 12B, Compression Ratio (CR)=12025.1193.2183.011,00036.2222.4152.8
kvtc16xModel=Llama 3.1 8B, Compression Ratio (CR)=18-202025.1192.4189.011,00057.2255.2259.2
VanillaModel=MN-Minitron 8B, Compression Ratio (CR)=12025.1192.4186.211,00062.0264.9258.1
kvtc16xModel=Mistral NeMo 12B, Compression Ratio (CR)=192025.1190.9182.111,00033.8222.9152.2
kvtc16xModel=MN-Minitron 8B, Compression Ratio (CR)=192025.1185.3186.311,00062.2264.7257.3