General Capability Harmlessness on General LLM Task Benchmark
62Average AccuracyKinGuard
Evaluation Results
| Method | Links | |
|---|---|---|
| KinGuardBackbone Model=Qwen2.5, Fingerprinting Method=KinGuard2026.01 | 62 | |
| KinGuardBackbone Model=LLaMA3, Fingerprinting Method=KinGuard2026.01 | 61.3 | |
| IF-SFTBackbone Model=Qwen2.5, Fingerprinting Method=IF-SFT2026.01 | 61 | |
| ProFlingoBackbone Model=Qwen2.5, Fingerprinting Method=ProFlingo2026.01 | 60.9 | |
| IF-SFTBackbone Model=LLaMA3, Fingerprinting Method=IF-SFT2026.01 | 60.4 | |
| Chain&HashBackbone Model=Qwen2.5, Fingerprinting Method=Chain&Hash2026.01 | 59.8 | |
| ProFlingoBackbone Model=LLaMA3, Fingerprinting Method=ProFlingo2026.01 | 59.6 | |
| Chain&HashBackbone Model=LLaMA3, Fingerprinting Method=Chain&Hash2026.01 | 59.4 | |
| KinGuardBackbone Model=LLaMA2, Fingerprinting Method=KinGuard2026.01 | 58.8 | |
| IF-SFTBackbone Model=LLaMA2, Fingerprinting Method=IF-SFT2026.01 | 55.9 | |
| ProFlingoBackbone Model=LLaMA2, Fingerprinting Method=ProFlingo2026.01 | 55.9 | |
| Chain&HashBackbone Model=LLaMA2, Fingerprinting Method=Chain&Hash2026.01 | 55.8 |