Common-sense Reasoning on Common-sense Reasoning Benchmarks Zero-shot
80.1BoolQ AccuracyDOT-MoE
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DOT-MoEBase Model=Qwen2.5 7B, Active Params=4.76B2026.06 | 80.1 | 95 | 74.2 | 66.2 | 50.3 | 67.9 | 72.3 | |
| DOT-MoEBase Model=LLaMA-2 7B, Active Params=3.49B2026.06 | 66.4 | 89.6 | 67.2 | 57.9 | 34 | 53.9 | 61.5 | |
| DISP-LLMBase Model=Qwen2.5 7B, Active Params=4.91B2026.06 | 65.7 | 94.1 | 70.2 | 62.8 | 45.9 | 61.3 | 66.7 | |
| DOT-MoEBase Model=Qwen2.5 7B, Active Params=3.34B2026.06 | 64.3 | 93.9 | 68.5 | 58.7 | 39.6 | 55 | 63.3 | |
| DOT-MoEBase Model=LLaMA-3 8B, Active Params=3.80B2026.06 | 62.3 | 90.9 | 65 | 56 | 34.1 | 50.1 | 59.8 | |
| CMoEBase Model=Qwen2.5 7B, Active Params=4.76B, Setting=E8A2S22026.06 | 60.3 | 87 | 62.8 | 51.5 | 28.4 | 42.7 | 55.5 | |
| DISP-LLMBase Model=LLaMA-2 7B, Active Params=3.57B2026.06 | 60 | 83.5 | 65.2 | 57 | 29.4 | 49.5 | 57.4 | |
| DISP-LLMBase Model=LLaMA-3 8B, Active Params=3.84B2026.06 | 51 | 84.5 | 60.6 | 53.9 | 26.3 | 40.1 | 52.7 | |
| DISP-LLMBase Model=Qwen2.5 7B, Active Params=3.67B2026.06 | 46.8 | 84.8 | 63.5 | 57.2 | 35.1 | 46.4 | 55.6 | |
| CMoEBase Model=LLaMA-2 7B, Active Params=3.49B, Setting=E8A1S12026.06 | 46.1 | 65.3 | 52.8 | 48.7 | 23.8 | 30.1 | 44.5 | |
| CMoEBase Model=Qwen2.5 7B, Active Params=3.34B, Setting=E8A1S12026.06 | 43.1 | 41.8 | 54.2 | 50.4 | 25 | 27.1 | 40.3 | |
| ShortGPTBase Model=Qwen2.5 7B, Active Params=4.82B2026.06 | 42.2 | 55.5 | 59.4 | 51.7 | 27.5 | 35.9 | 45.4 | |
| CMoEBase Model=LLaMA-3 8B, Active Params=3.80B, Setting=E8A1S12026.06 | 41 | 51.1 | 53.9 | 51 | 25.3 | 28.6 | 41.8 | |
| SliceGPTBase Model=LLaMA-2 7B, Active Params=3.73B2026.06 | 38.3 | 73.4 | 55.2 | 52.8 | 24.5 | 33 | 46.2 | |
| SliceGPTBase Model=Qwen2.5 7B, Active Params=4.75B2026.06 | 38 | 80.6 | 56.5 | 53.5 | 22.8 | 33.5 | 47.5 | |
| SliceGPTBase Model=LLaMA-3 8B, Active Params=3.90B2026.06 | 37.8 | 62.4 | 53.3 | 49.6 | 22.1 | 27.6 | 42.1 |