Adding Mistake on ZebraLogicBench (ZLB)
83.8AOCFaithfulness Only
Evaluation Results
| Method | Links | |
|---|---|---|
| Faithfulness OnlyBackbone=Qwen3-14B2026.02 | 83.8 | |
| REMULBackbone=Qwen3-14B2026.02 | 82.6 | |
| MAT-SteerBackbone=Qwen3-14B2026.02 | 80.8 | |
| OriginalBackbone=Qwen3-14B2026.02 | 79.3 | |
| Balanced RewardsBackbone=Qwen3-14B2026.02 | 79 | |
| Correctness OnlyBackbone=Qwen3-14B2026.02 | 77.8 | |
| Hint OptimizedBackbone=Qwen3-14B2026.02 | 77.1 |