Adversarial Safety on HarmBench
49.12AccuracyCRL-Token
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CRL-TokenBackbone=Gemma 2 2B, Intervention Layer=212026.02 | 49.12 | 7.66 | |
| Gemma 2 2BBackbone=Gemma 2 2B2026.02 | 41.46 | — |
| Method | Links | ||
|---|---|---|---|
| CRL-TokenBackbone=Gemma 2 2B, Intervention Layer=212026.02 | 49.12 | 7.66 | |
| Gemma 2 2BBackbone=Gemma 2 2B2026.02 | 41.46 | — |