Knowledge Combination on KID-Bench Category B v2
68AccuracyD2L Baseline
Evaluation results
| Method | Links | |
|---|---|---|
| D2L BaselineBackbone=Gemma-2B, Decoding strategy=greedy decoding2026.04 | 68 | |
| SLBBackbone=Gemma-2B, Decoding strategy=greedy decoding, Method configuration=25%/1.75×2026.04 | 68 | |
| Conflict-AwareBackbone=Gemma-2B, Decoding strategy=greedy decoding2026.04 | 64 |