Language Understanding (Multi-choice QA) on MMLU moral scenarios
2.82NFE SpeedupSpiffy
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SpiffyBase Model=LLaDA-8B-Instruct2025.09 | 2.82 | 2.07 | 38 | |
| Dynamic UnmaskingBase Model=LLaDA-8B-Instruct2025.09 | 1.62 | 1.6 | 36 | |
| LLaDA-8B-InstructMode=Baseline2025.09 | 1 | 1 | 35 |