Unsupervised Error Diagnosis on GSM8K
0.898AUROCBMC
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BMCModel=Dream-v0-Instruct-7B2026.04 | 0.898 | 0.787 | |
| BMCModel=LLaDA-8B-Instruct2026.04 | 0.893 | 0.943 | |
| Self-ConsistencyModel=LLaDA-8B-Instruct2026.04 | 0.872 | 0.929 | |
| Model ConfidenceModel=LLaDA-8B-Instruct2026.04 | 0.753 | 0.879 | |
| Self-ConsistencyModel=Dream-v0-Instruct-7B2026.04 | 0.684 | 0.498 | |
| Model ConfidenceModel=Dream-v0-Instruct-7B2026.04 | 0.641 | 0.468 | |
| Self-EvaluationModel=Dream-v0-Instruct-7B2026.04 | 0.565 | 0.401 | |
| Self-EvaluationModel=LLaDA-8B-Instruct2026.04 | 0.549 | 0.737 |