Mathematical Reasoning on GSM8K 512
81.2AccuracyNI Sampling
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| NI SamplingModel=LLaDA 1.52026.04 | 81.2 | 53.56 | 140.6 | |
| ThresholdModel=LLaDA 1.52026.04 | 80.89 | 72.38 | 105.6 | |
| FullModel=LLaDA 1.52026.04 | 80.67 | 512 | 15 | |
| NI SamplingModel=LLaDA-8B-Instruct2026.04 | 76.57 | 51.08 | 147 | |
| ThresholdModel=LLaDA-8B-Instruct2026.04 | 75.28 | 73.29 | 104.4 | |
| FullModel=LLaDA-8B-Instruct2026.04 | 74.83 | 512 | 14.4 |