List your company

Adversarial Detection on Prefilling

99.4DSRSALO
-3.97622.86249.776.538May 2, 2026

Evaluation results

MethodLinks
SALOTarget Model=Qwen2.5-7B-Instruct, Params=<20M, Grad=false2026.0599.4
SALOTarget Model=Mistral-7B-Instruct-v0.3, Params=<20M, Grad=false2026.0599.4
SALOTarget Model=Llama-3.1-8B-Instruct, Params=<20M, Grad=false2026.0598.8
Linear ProbeTarget Model=Qwen2.5-7B-Instruct, Params=<1M, Grad=false2026.0596.5
Linear ProbeTarget Model=Llama-3.1-8B-Instruct, Params=<1M, Grad=false2026.0587.3
Linear ProbeTarget Model=Mistral-7B-Instruct-v0.3, Params=<1M, Grad=false2026.0568.1
Smooth LLMTarget Model=Llama-3.1-8B-Instruct, Grad=false2026.0566.5
Smooth LLMTarget Model=Qwen2.5-7B-Instruct, Grad=false2026.0557.9
PPL FilterTarget Model=Llama-3.1-8B-Instruct, Grad=false2026.0543.7
Smooth LLMTarget Model=Mistral-7B-Instruct-v0.3, Grad=false2026.0542.9
No Defense (1-ASR)Target Model=Llama-3.1-8B-Instruct2026.0539.8
PPL FilterTarget Model=Qwen2.5-7B-Instruct, Grad=false2026.0539.2
No Defense (1-ASR)Target Model=Qwen2.5-7B-Instruct2026.0524.8
No Defense (1-ASR)Target Model=Mistral-7B-Instruct-v0.32026.0515
GradSafeTarget Model=Llama-3.1-8B-Instruct, Grad=true2026.0512
GradSafeTarget Model=Mistral-7B-Instruct-v0.3, Grad=true2026.051.7
PPL FilterTarget Model=Mistral-7B-Instruct-v0.3, Grad=false2026.050
GradSafeTarget Model=Qwen2.5-7B-Instruct, Grad=true2026.050