Mathematical Reasoning on GSM8K CoT LLaDA-8B-Instruct (test)
80.51ScoreAnchor-based History-stable Decoding
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Anchor-based History-stable DecodingBase Model=LLaDA-8B-Instruct2026.04 | 80.51 | 3.41 | 75.06 | |
| Fast_dLLMBase Model=LLaDA-8B-Instruct2026.04 | 80.29 | 3.41 | 75.1 | |
| KLASSBase Model=LLaDA-8B-Instruct2026.04 | 80.06 | 1.99 | 128.64 | |
| PC-samplerBase Model=LLaDA-8B-Instruct2026.04 | 79.98 | 1 | 256 | |
| SaberBase Model=LLaDA-8B-Instruct2026.04 | 79.61 | 3.35 | 76.49 | |
| LLaDABase Model=LLaDA-8B-Instruct2026.04 | 79.53 | 1 | 256 |