Science Question Answering on SciQ (Acc, Acc Norm)
94.1AccuracyDense
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DenseBase Model=DeepSeek-7B, Sparsity=Dense2025.06 | 94.1 | — | |
| DenseBase Model=LLaMA-2-7B, Sparsity=Dense2025.06 | 94 | — | |
| SparsegptBase Model=DeepSeek-7B, Sparsity=2:42025.06 | 92.3 | — | |
| SparsegptBase Model=LLaMA-2-7B, Sparsity=2:42025.06 | 91.1 | — | |
| Pruner-ZBase Model=LLaMA-2-7B, Sparsity=2:42025.06 | 91.1 | — | |
| MaskProBase Model=LLaMA-2-7B, Sparsity=2:42025.06 | 91.1 | — | |
| SoftmaxQuantization=4-bit2025.04 | 90.9 | 86.9 | |
| MaskProBase Model=DeepSeek-7B, Sparsity=2:42025.06 | 90.9 | — | |
| SoftmaxQuantization=8-bit2025.04 | 90.3 | 86 | |
| Pruner-ZBase Model=DeepSeek-7B, Sparsity=2:42025.06 | 90.2 | — | |
| SoftpickQuantization=4-bit2025.04 | 86.9 | 79.4 | |
| SoftpickQuantization=8-bit2025.04 | 86.9 | 79.9 | |
| MOE+Lngram2026.05 | 71 | — | |
| MOE+Lngram-23Lconfiguration=23L2026.05 | 70.8 | — | |
| MOE+Engram2026.05 | 69.8 | — | |
| MOE2026.05 | 68.3 | — |