Conflict Resolution on MCR-BENCH
45.9Accuracy (Neutral)CAAD
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CAADModel Size=3B, Decoding / Distill Mode=CAAD (Ours)2026.06 | 45.9 | 82.8 | 11.8 | 45.5 | 79.03 | |
| StudentModel Size=3B, Decoding / Distill Mode=Std. KD2026.06 | 41 | 96.9 | 0.5 | 40.9 | 100 | |
| TeacherModel Size=8B, Decoding / Distill Mode=CD2026.06 | 11.2 | 51.4 | 15 | 41.8 | 83.96 | |
| TeacherModel Size=8B, Decoding / Distill Mode=Greedy Decode2026.06 | 3.9 | 98.6 | 1.1 | 41.2 | 97.37 | |
| StudentModel Size=3B, Decoding / Distill Mode=Greedy Decode2026.06 | 1.4 | 97.4 | 1 | 34 | 90.65 | |
| StudentModel Size=3B, Decoding / Distill Mode=CD2026.06 | 1 | 56.6 | 7.4 | 26.9 | 87.5 |