Mathematical Reasoning on GSM8K (Accuracy, NFE Reduction, TPS Speedup)
8.25NFE ReductionSpiffy
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SpiffyBase Model=SDAR-8B-Chat-b322025.09 | 8.25 | 6.28 | 88 | |
| Dynamic UnmaskingBase Model=SDAR-8B-Chat-b322025.09 | 6.71 | 5.67 | 88 | |
| SpiffyBase Model=LLaDA-8B-Instruct2025.09 | 4.97 | 3.55 | 79 | |
| SpiffyBase Model=Dream-7B-Instruct2025.09 | 3.76 | 2.56 | 79 | |
| Dynamic UnmaskingBase Model=LLaDA-8B-Instruct2025.09 | 3.31 | 3.2 | 79 | |
| Dynamic UnmaskingBase Model=Dream-7B-Instruct2025.09 | 2.35 | 2.28 | 78 | |
| BaselineBase Model=LLaDA-8B-Instruct2025.09 | 1 | 1 | 79 | |
| BaselineBase Model=Dream-7B-Instruct2025.09 | 1 | 1 | 79 | |
| BaselineBase Model=SDAR-8B-Chat-b322025.09 | 1 | 1 | 90 |