Mathematical Reasoning on GSM8K (Acc., Steps, TPF)
79.68AccuracyKLASS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| KLASSBackbone=Dream2026.07 | 79.68 | 155.96 | 1.64 | |
| KLASS + TACGBackbone=Dream2026.07 | 78.62 | 91.94 | 2.78 | |
| Conf. + TACGBackbone=LLaDA2026.07 | 77.33 | 72.29 | 3.78 | |
| d3LLM†Backbone=Dream2026.07 | 77.2 | — | 4.8 | |
| KLASS + TACGBackbone=LLaDA2026.07 | 76.42 | 106.66 | 2.51 | |
| KLASSBackbone=LLaDA2026.07 | 76.12 | 98.51 | 2.74 | |
| Conf.Backbone=LLaDA2026.07 | 75.4 | 74.26 | 3.68 | |
| d3LLM†Backbone=LLaDA2026.07 | 74.2 | — | 6.95 | |
| Conf. + TACGBackbone=Dream2026.07 | 73.31 | 69.01 | 3.71 | |
| Conf.Backbone=Dream2026.07 | 73.16 | 74.63 | 3.43 |