List your company

Benign Prompt Filtering on GSM8K

0False Positive RatePerplexity Filter
-2.816.13553.9Apr 1, 2026

Evaluation results

MethodLinks
Perplexity FilterBase Model=Qwen2.5-7B-Instruct2026.040
GradientCuffBase Model=Qwen2.5-7B-Instruct2026.040
Prompt GuardBase Model=Qwen2.5-7B-Instruct2026.040
Llama GuardBase Model=Qwen2.5-7B-Instruct, Guardrail Stage/Mode=Pre2026.040
Llama GuardBase Model=Qwen2.5-7B-Instruct, Guardrail Stage/Mode=Post2026.040
SelfDefendBase Model=Qwen2.5-7B-Instruct, Guardrail Stage/Mode=Direct2026.040
WildGuardBase Model=Qwen2.5-7B-Instruct, Guardrail Stage/Mode=Pre2026.040
WildGuardBase Model=Qwen2.5-7B-Instruct, Guardrail Stage/Mode=Post2026.040
SelfGraderBase Model=Qwen2.5-7B-Instruct2026.040
SelfDefendBase Model=Qwen2.5-7B-Instruct, Guardrail Stage/Mode=Intent2026.0470