List your company

Over-refusal evaluation on Poisoned GPT4-LLM subset

35.7Informative Refusal RateQCB
-1.0128.51918.0527.581Jun 28, 2026

Evaluation results

MethodLinks
QCBModel=Gemma-2B, Quantization=INT82026.0635.7
QCBModel=Gemma-2B, Quantization=NF42026.0634.2
QCBModel=Gemma-2B, Quantization=FP42026.0631.3
QCBModel=Phi-2-2.7B, Quantization=NF42026.0630.2
QCBModel=Phi-2-2.7B, Quantization=INT82026.0626.1
QCBModel=Phi-2-2.7B, Quantization=FP42026.0623.8
QCBModel=LLaMA3-8B, Quantization=FP42026.0612.6
QCBModel=LLaMA3-8B, Quantization=NF42026.0612.3
QCBModel=LLaMA3-8B, Quantization=INT82026.0611
CleanModel=LLaMA3-8B, Quantization=INT82026.061.07
CleanModel=Phi-2-2.7B, Quantization=NF42026.061
QuantGuardModel=LLaMA3-8B, Quantization=INT82026.061
QuantGuardModel=LLaMA3-8B, Quantization=NF42026.061
CleanModel=Phi-2-2.7B, Quantization=INT82026.060.86
QuantGuardModel=Gemma-2B, Quantization=INT82026.060.86
QuantGuardModel=Phi-2-2.7B, Quantization=FP42026.060.8
QuantGuardModel=Gemma-2B, Quantization=FP42026.060.8
QuantGuardModel=Gemma-2B, Quantization=NF42026.060.8
CleanModel=LLaMA3-8B, Quantization=FP42026.060.73
CleanModel=LLaMA3-8B, Quantization=NF42026.060.73
CleanModel=Phi-2-2.7B, Quantization=FP42026.060.67
QuantGuardModel=Phi-2-2.7B, Quantization=INT82026.060.67
QuantGuardModel=Phi-2-2.7B, Quantization=NF42026.060.67
QuantGuardModel=LLaMA3-8B, Quantization=FP42026.060.6
CleanModel=Gemma-2B, Quantization=INT82026.060.4
CleanModel=Gemma-2B, Quantization=FP42026.060.4
CleanModel=Gemma-2B, Quantization=NF42026.060.4