Harmful Content Classification on JBB-Behaviors n = 200 (blind)
0.828F1 Scorellm/mistral-7b
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| llm/mistral-7bParameters=7,2 B2026.06 | 0.828 | 0.839 | 0.19 | 0.87 | |
| llm/llama3.1-8bParameters=8,0 B2026.06 | 0.807 | 0.864 | 0.08 | 0.91 | |
| GuardNet-E (final)Parameters=47 M2026.06 | 0.714 | 0.747 | 0.53 | 0.85 | |
| deepset-debertaParameters=184 M2026.06 | 0.701 | 0.65 | 0.5 | 0.81 | |
| llm/phi3-3.8bParameters=3,8 B2026.06 | 0.692 | 0.741 | 0.87 | 1 | |
| llm/qwen2.5-7bParameters=7,6 B2026.06 | 0.428 | 0.802 | 0.03 | 0.28 | |
| jackhhao-jailbreakParameters=184 M2026.06 | 0.137 | 0.574 | 0.09 | 0.08 | |
| protectai-v2Parameters=184 M2026.06 | 0 | 0.6 | 0.01 | 0 |