Mathematical Reasoning on GSM8K (Accuracy, Recall)
81AccuracyFull
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FullBackbone=LLaMA-3.1-8B-Instruct2026.06 | 81 | 100 | |
| RaBitQCacheBackbone=LLaMA-3.1-8B-Instruct2026.06 | 77 | 88.8 | |
| DSBackbone=LLaMA-3.1-8B-Instruct2026.06 | 77 | 72.7 | |
| SparQBackbone=LLaMA-3.1-8B-Instruct2026.06 | 77 | 60.5 | |
| QuestBackbone=LLaMA-3.1-8B-Instruct2026.06 | 70 | 81.3 | |
| BaselineSparsity=0%, Calib. data=-, Shot=8-shot2025.11 | 57.6 | — | |
| PuzzleMoESparsity=25%, Calib. data=C4, Shot=8-shot2025.11 | 55.4 | — | |
| PuzzleMoESparsity=50%, Calib. data=C4, Shot=8-shot2025.11 | 51.7 | — | |
| NAEESparsity=25%, Calib. data=Math, Shot=8-shot2025.11 | 48.7 | — | |
| NAEESparsity=25%, Calib. data=C4, Shot=8-shot2025.11 | 41.5 | — | |
| NAEESparsity=50%, Calib. data=Math, Shot=8-shot2025.11 | 38.8 | — | |
| WandaSparsity=50%, Calib. data=C4, Shot=8-shot2025.11 | 32.2 | — | |
| NAEESparsity=50%, Calib. data=C4, Shot=8-shot2025.11 | 28.6 | — |