Harmful Query Evaluation on AdvBench
0Harmful RatePolicyAlign
Evaluation Results
| Method | Links | |
|---|---|---|
| PolicyAlignBackbone=Qwen3-4B2026.06 | 0 | |
| ICLBackbone=Qwen3-4B2026.06 | 0.38 | |
| SFTBackbone=Qwen3-4B2026.06 | 0.77 | |
| BaseBackbone=Qwen3-4B2026.06 | 2.31 |
| Method | Links | |
|---|---|---|
| PolicyAlignBackbone=Qwen3-4B2026.06 | 0 | |
| ICLBackbone=Qwen3-4B2026.06 | 0.38 | |
| SFTBackbone=Qwen3-4B2026.06 | 0.77 | |
| BaseBackbone=Qwen3-4B2026.06 | 2.31 |