Bias QA on BBQ Ambig
85.04AccuracyCRL-Token
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CRL-TokenBackbone=LLaMA-3.1 8B, Type=bias, Layer=7, Evaluation Protocol=1-shot2026.02 | 85.04 | 1.07 | |
| LLaMA-3.1 8BEvaluation Protocol=1-shot2026.02 | 83.97 | — | |
| CRL-TokenBackbone=Gemma 2 2B, Intervention Layer=52026.02 | 65.86 | 5.69 | |
| Gemma 2 2BBackbone=Gemma 2 2B2026.02 | 60.17 | — |