List your company
ResearchFormerly WizWand

Adversarial Detection on GCG

100DSRPPL Filter
14.40836.62958.8581.071May 2, 2026
Updated 6mo ago

Evaluation results

MethodLinks
PPL FilterTarget Model=Qwen2.5-7B-Instruct, Grad=false2026.05100
PPL FilterTarget Model=Mistral-7B-Instruct-v0.3, Grad=false2026.05100
PPL FilterTarget Model=Llama-3.1-8B-Instruct, Grad=false2026.05100
SALOTarget Model=Mistral-7B-Instruct-v0.3, Params=<20M, Grad=false2026.0596.5
Smooth LLMTarget Model=Qwen2.5-7B-Instruct, Grad=false2026.0594.6
No Defense (1-ASR)Target Model=Llama-3.1-8B-Instruct2026.0588.3
SALOTarget Model=Qwen2.5-7B-Instruct, Params=<20M, Grad=false2026.0588.3
SALOTarget Model=Llama-3.1-8B-Instruct, Params=<20M, Grad=false2026.0585.4
Linear ProbeTarget Model=Qwen2.5-7B-Instruct, Params=<1M, Grad=false2026.0582.9
Smooth LLMTarget Model=Llama-3.1-8B-Instruct, Grad=false2026.0580.6
GradSafeTarget Model=Llama-3.1-8B-Instruct, Grad=true2026.0576.4
No Defense (1-ASR)Target Model=Qwen2.5-7B-Instruct2026.0571.2
Linear ProbeTarget Model=Mistral-7B-Instruct-v0.3, Params=<1M, Grad=false2026.0571
Linear ProbeTarget Model=Llama-3.1-8B-Instruct, Params=<1M, Grad=false2026.0564.2
Smooth LLMTarget Model=Mistral-7B-Instruct-v0.3, Grad=false2026.0555.8
No Defense (1-ASR)Target Model=Mistral-7B-Instruct-v0.32026.0544.6
GradSafeTarget Model=Mistral-7B-Instruct-v0.3, Grad=true2026.0536.7
GradSafeTarget Model=Qwen2.5-7B-Instruct, Grad=true2026.0517.7