List your company

Backdoor Detection on AlpacaLlama

0.5293F1 ScoreLLMScan
-0.0211720.1217390.264650.407561Dec 22, 2025

Evaluation results

MethodLinks
LLMScanDefense Paradigm=Inference-time behavior analysis2025.120.5293——95.6721.63
ONIONDefense Paradigm=Data-level filtering2025.12010——
PEFTGuardDefense Paradigm=Weight-level inspection2025.120——1000