Knowledge Conflict Resolution on KID-Bench Category C v2
78.1Accuracy (C-Light)Conflict-Aware
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Conflict-AwareBackbone=Gemma-2B, Decoding strategy=greedy decoding2026.04 | 78.1 | 83.6 | 71 | |
| SLBBackbone=Gemma-2B, Decoding strategy=greedy decoding, Method configuration=25%/1.75×2026.04 | 75 | 73.8 | 60.9 | |
| D2L BaselineBackbone=Gemma-2B, Decoding strategy=greedy decoding2026.04 | 57.8 | 55.7 | 46.4 |