List your company
ResearchFormerly WizWand

Content Injection scenario

99.5ASRSINQ
-3.9822.88549.7576.615May 14, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SINQModel=Qwen2.5 7B, Type=DIOB2026.0599.5——
FP4Model=Llama3.1 8B, Type=Zero-shot2026.0598.9——
HQQModel=Qwen2.5 7B, Type=DIOB2026.0598.9——
FP4Model=Qwen2.5 7B, Type=Zero-shot2026.0598.7——
AWQModel=Qwen2.5 7B, Type=DDOB2026.0598.5——
HQQModel=Llama3.1 8B, Type=DIOB2026.0598.3——
GPTQ 4-bitModel=Qwen2.5 7B, Type=DDOB2026.0598.3——
GGUF Q4_0Model=Qwen2.5 7B, Type=Zero-shot2026.0597.9——
SINQModel=Llama3.1 8B, Type=DIOB2026.0597.7——
NF4Model=Llama3.1 8B, Type=Zero-shot2026.0596.8——
NF4Model=Qwen2.5 7B, Type=Zero-shot2026.0596.5——
GPTQ 4-bitModel=Llama3.1 8B, Type=DDOB2026.0596.3——
AWQModel=Llama3.1 8B, Type=DDOB2026.0596.3——
GPTQ 8-bitModel=Qwen2.5 7B, Type=DDOB2026.0595.9——
LLM.int8()Model=Llama3.1 8B, Type=Zero-shot2026.0595.8——
GGUF Q4_0Model=Llama3.1 8B, Type=Zero-shot2026.0595.7——
GGUF IQ4_XSModel=Llama3.1 8B, Type=DDOB2026.0595.6——
LLM.int8()Model=Qwen2.5 7B, Type=Zero-shot2026.0595.5——
GGUF Q4_K_MModel=Llama3.1 8B, Type=DIOB2026.0595.5——
GPTQ 8-bitModel=Llama3.1 8B, Type=DDOB2026.0594.9——
GGUF Q4_K_MModel=Qwen2.5 7B, Type=DIOB2026.0594.3——
AutoRoundModel=Qwen2.5 7B, Type=DDOB2026.0590.3——
HQQModel=Mistral 7B, Type=DIOB2026.0585——
SINQModel=Mistral 7B, Type=DIOB2026.0585——
LLM.int8()Model=Mistral 7B, Type=Zero-shot2026.0583.6——
GPTQ 8-bitModel=Mistral 7B, Type=DDOB2026.0582.7——
GPTQ 4-bitModel=Mistral 7B, Type=DDOB2026.0581.8——
NF4Model=Mistral 7B, Type=Zero-shot2026.0581.3——
AWQModel=Mistral 7B, Type=DDOB2026.0579.8——
AutoRoundModel=Mistral 7B, Type=DDOB2026.0564.7——
GGUF IQ4_XSModel=Qwen2.5 7B, Type=DDOB2026.0542.9——
FP4Model=Mistral 7B, Type=Zero-shot2026.0535.5——
AutoRoundModel=Llama3.1 8B, Type=DDOB2026.0519.4——
GGUF Q4_K_MModel=Mistral 7B, Type=DIOB2026.057.4——
GGUF Q4_0Model=Mistral 7B, Type=Zero-shot2026.057.3——
GGUF IQ4_XSModel=Mistral 7B, Type=DDOB2026.055.5——
AttackedModel=Mistral 7B, Precision=BF162026.054.2——
AttackedModel=Llama3.1 8B, Precision=BF162026.052.5——
AttackedModel=Qwen2.5 7B, Precision=BF162026.051.9——
OriginalModel=Qwen2.5 7B, Precision=BF162026.050.1——
OriginalModel=Llama3.1 8B, Precision=BF162026.050——
OriginalModel=Mistral 7B, Precision=BF162026.050——
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Unpruned, Sparsity Ratio=-2025.10—0.10
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Mag., Sparsity Ratio=20%2025.10—94.30
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=SparseGPT, Sparsity Ratio=20%2025.10—9.10
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=SparseGPT, Sparsity Ratio=2:42025.10—4.70
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=SparseGPT, Sparsity Ratio=50%2025.10—340
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Wanda, Sparsity Ratio=20%2025.10—63.50
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Wanda, Sparsity Ratio=2:42025.10—83.510
Llama3.1-8BModel=Llama3.1-8B, Pruning Method=Wanda, Sparsity Ratio=50%2025.10—98.80
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Unpruned, Sparsity Ratio=-2025.10—0.90
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Mag., Sparsity Ratio=20%2025.10—61.50
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=SparseGPT, Sparsity Ratio=20%2025.10—1110
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=SparseGPT, Sparsity Ratio=2:42025.10—90
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=SparseGPT, Sparsity Ratio=50%2025.10—53.20
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Wanda, Sparsity Ratio=20%2025.10—27.30
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Wanda, Sparsity Ratio=2:42025.10—62.510
OLMo-2-7BModel=OLMo-2-7B, Pruning Method=Wanda, Sparsity Ratio=50%2025.10—96.610
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Unpruned, Sparsity Ratio=-2025.10—0.10
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Mag., Sparsity Ratio=20%2025.10—92.20
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=SparseGPT, Sparsity Ratio=20%2025.10—24.90
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=SparseGPT, Sparsity Ratio=2:42025.10—34.710
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=SparseGPT, Sparsity Ratio=50%2025.10—62.10
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Wanda, Sparsity Ratio=20%2025.10—75.50
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Wanda, Sparsity Ratio=2:42025.10—81.910
Qwen2.5-7BModel=Qwen2.5-7B, Pruning Method=Wanda, Sparsity Ratio=50%2025.10—99.50