List your company
ResearchFormerly WizWand

Over Refusal on Over Refusal scenario

100ASRHQQ
-3.79223.15450.177.046May 14, 2026
Updated 5mo ago

Evaluation results

MethodLinks
HQQModel=Qwen2.5 7B, Type=DIOB2026.05100——
SINQModel=Qwen2.5 7B, Type=DIOB2026.0599.9——
AWQModel=Qwen2.5 7B, Type=DDOB2026.0599.5——
GGUF Q4_0Model=Qwen2.5 7B, Type=Zero-shot2026.0598.3——
NF4Model=Qwen2.5 7B, Type=Zero-shot2026.0595.8——
LLM.int8()Model=Qwen2.5 7B, Type=Zero-shot2026.0595——
GPTQ 4-bitModel=Qwen2.5 7B, Type=DDOB2026.0594.3——
FP4Model=Qwen2.5 7B, Type=Zero-shot2026.0593.4——
GGUF Q4_K_MModel=Qwen2.5 7B, Type=DIOB2026.0592.9——
HQQModel=Llama3.1 8B, Type=DIOB2026.0592.9——
SINQModel=Llama3.1 8B, Type=DIOB2026.0589.6——
LLM.int8()Model=Llama3.1 8B, Type=Zero-shot2026.0589.1——
GGUF IQ4_XSModel=Llama3.1 8B, Type=DDOB2026.0585.1——
FP4Model=Llama3.1 8B, Type=Zero-shot2026.0584.4——
GGUF Q4_K_MModel=Llama3.1 8B, Type=DIOB2026.0581.9——
GPTQ 8-bitModel=Qwen2.5 7B, Type=DDOB2026.0578.3——
AWQModel=Llama3.1 8B, Type=DDOB2026.0574.3——
NF4Model=Llama3.1 8B, Type=Zero-shot2026.0573.7——
GGUF Q4_0Model=Llama3.1 8B, Type=Zero-shot2026.0570.4——
GPTQ 4-bitModel=Llama3.1 8B, Type=DDOB2026.0568.8——
GPTQ 8-bitModel=Llama3.1 8B, Type=DDOB2026.0566.9——
AWQModel=Mistral 7B, Type=DDOB2026.0549.6——
GPTQ 4-bitModel=Mistral 7B, Type=DDOB2026.0548.7——
LLM.int8()Model=Mistral 7B, Type=Zero-shot2026.0548.3——
NF4Model=Mistral 7B, Type=Zero-shot2026.0547.7——
SINQModel=Mistral 7B, Type=DIOB2026.0543——
AutoRoundModel=Mistral 7B, Type=DDOB2026.0541.7——
AutoRoundModel=Qwen2.5 7B, Type=DDOB2026.0537.9——
HQQModel=Mistral 7B, Type=DIOB2026.0533.7——
FP4Model=Mistral 7B, Type=Zero-shot2026.0527.7——
GPTQ 8-bitModel=Mistral 7B, Type=DDOB2026.0525.1——
GGUF Q4_K_MModel=Mistral 7B, Type=DIOB2026.0513.5——
AutoRoundModel=Llama3.1 8B, Type=DDOB2026.0510.6——
GGUF IQ4_XSModel=Qwen2.5 7B, Type=DDOB2026.057.7——
GGUF IQ4_XSModel=Mistral 7B, Type=DDOB2026.055.3——
AttackedModel=Mistral 7B, Precision=BF162026.054.3——
GGUF Q4_0Model=Mistral 7B, Type=Zero-shot2026.054.3——
AttackedModel=Llama3.1 8B, Precision=BF162026.051.2——
OriginalModel=Qwen2.5 7B, Precision=BF162026.050.8——
OriginalModel=Mistral 7B, Precision=BF162026.050.8——
AttackedModel=Qwen2.5 7B, Precision=BF162026.050.3——
OriginalModel=Llama3.1 8B, Precision=BF162026.050.2——
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Unpruned, Sparsity Ratio=-2025.10—0.50.5
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Mag., Sparsity Ratio=20%2025.10—95.50.3
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=SparseGPT, Sparsity Ratio=20%2025.10—70.40.3
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=SparseGPT, Sparsity Ratio=2:42025.10—21.42.9
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=SparseGPT, Sparsity Ratio=50%2025.10—78.32.7
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Wanda, Sparsity Ratio=20%2025.10—930.8
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Wanda, Sparsity Ratio=2:42025.10—63.21.8
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Wanda, Sparsity Ratio=50%2025.10—97.32.9
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Unpruned, Sparsity Ratio=-2025.10—2.12.5
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Mag., Sparsity Ratio=20%2025.10—92.72.1
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=SparseGPT, Sparsity Ratio=20%2025.10—78.82.6
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=SparseGPT, Sparsity Ratio=2:42025.10—47.98.7
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=SparseGPT, Sparsity Ratio=50%2025.10—98.74.6
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Wanda, Sparsity Ratio=20%2025.10—91.12.1
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Wanda, Sparsity Ratio=2:42025.10—78.76.9
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Wanda, Sparsity Ratio=50%2025.10—97.24.1
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Unpruned, Sparsity Ratio=-2025.10—1.10.4
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Mag., Sparsity Ratio=20%2025.10—93.90.3
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=SparseGPT, Sparsity Ratio=20%2025.10—51.30.9
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=SparseGPT, Sparsity Ratio=2:42025.10—40.92.8
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=SparseGPT, Sparsity Ratio=50%2025.10—67.81.1
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Wanda, Sparsity Ratio=20%2025.10—93.70.8
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Wanda, Sparsity Ratio=2:42025.10—96.34.1
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Wanda, Sparsity Ratio=50%2025.10—98.41.7