Novel Knowledge Recall on KID-Bench Category A v2
97.1AccuracyConflict-Aware
Evaluation results
| Method | Links | |
|---|---|---|
| Conflict-AwareBackbone=Gemma-2B, Decoding strategy=greedy decoding2026.04 | 97.1 | |
| D2L BaselineBackbone=Gemma-2B, Decoding strategy=greedy decoding2026.04 | 96.7 | |
| SLBBackbone=Gemma-2B, Decoding strategy=greedy decoding, Method configuration=25%/1.75×2026.04 | 95.5 |