Safety Classification on HoliSafe-Bench (ECE, NLL)
8.4ECELlama
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LlamaPrompt Selection Protocol=Mean2026.05 | 8.4 | 0.689 | |
| MiniCPMPrompt Selection Protocol=Mean2026.05 | 14.4 | 0.658 | |
| Qwen3-VL 4BPrompt Selection Protocol=Mean2026.05 | 15.3 | 0.685 | |
| MiniCPMPrompt Selection Protocol=Random2026.05 | 15.3 | 0.73 | |
| InternVLPrompt Selection Protocol=Mean2026.05 | 15.5 | 0.682 | |
| Qwen3-VL 8BPrompt Selection Protocol=Mean2026.05 | 16.6 | 0.727 | |
| GemmaPrompt Selection Protocol=Mean2026.05 | 21.5 | 0.902 | |
| InternVLPrompt Selection Protocol=Random2026.05 | 21.7 | 1.082 | |
| Qwen30BPrompt Selection Protocol=Mean2026.05 | 22.7 | 0.857 | |
| Qwen3-VL 4BPrompt Selection Protocol=Selected2026.05 | 24.1 | 1.338 | |
| Qwen3-VL 8BPrompt Selection Protocol=Random2026.05 | 24.2 | 1.805 | |
| InternVLPrompt Selection Protocol=Selected2026.05 | 24.6 | 1.07 | |
| Qwen3-VL 4BPrompt Selection Protocol=Random2026.05 | 24.8 | 1.716 | |
| Qwen30BPrompt Selection Protocol=Selected2026.05 | 24.8 | 1.656 | |
| LlamaPrompt Selection Protocol=Random2026.05 | 25.3 | 1.234 | |
| Qwen3-VL 8BPrompt Selection Protocol=Selected2026.05 | 26.4 | 1.028 | |
| Qwen30BPrompt Selection Protocol=Random2026.05 | 28.4 | 1.359 | |
| LlamaPrompt Selection Protocol=Selected2026.05 | 28.4 | 1.074 | |
| GemmaPrompt Selection Protocol=Random2026.05 | 34.8 | 2.769 | |
| MiniCPMPrompt Selection Protocol=Selected2026.05 | 37.5 | 1.389 | |
| GemmaPrompt Selection Protocol=Selected2026.05 | 48.2 | 2.292 |