Mathematical Reasoning on GSM-Symbolic (Accuracy, Constraint Satisfaction Rate)
49.6AccuracyLLaDA-8B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LLaDA-8BDecoding Strategy=Greedy, Constraint Setting=Base2026.07 | 49.6 | 81.5 | |
| LLaDA-8BDecoding Strategy=Greedy, Constraint Setting=Ours2026.07 | 49.2 | — | |
| LLaDA-8BDecoding Strategy=Sampling, Constraint Setting=Ours2026.07 | 49.2 | — | |
| LLaDA-8BDecoding Strategy=Sampling, Constraint Setting=Base2026.07 | 46.9 | 80.1 | |
| Dream-7BDecoding Strategy=Greedy, Constraint Setting=Ours2026.07 | 40.2 | — | |
| Dream-7BDecoding Strategy=Greedy, Constraint Setting=Base2026.07 | 39.9 | 80.2 | |
| Dream-7BDecoding Strategy=Sampling, Constraint Setting=Ours2026.07 | 38 | — | |
| Dream-7BDecoding Strategy=Sampling, Constraint Setting=Base2026.07 | 15 | 54.6 |