Commonsense Reasoning on Commonsense Reasoning Benchmarks (HellaSwag, WinoGrande, BoolQ)
73.6HellaSwagQUEST
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| QUESTBudget=256, Model=Trado-8B-Instruct2026.04 | 73.6 | 63.2 | 75 | 70.6 | |
| LoSABudget=128, Model=Trado-8B-Instruct2026.04 | 70.8 | 66 | 72.92 | 69.91 | |
| LoSABudget=256, Model=Trado-8B-Instruct2026.04 | 70.4 | 65.2 | 73.96 | 69.85 | |
| DenseBudget=–, Model=Trado-8B-Instruct2026.04 | 69.6 | 63.6 | 72.92 | 68.71 | |
| QUESTBudget=128, Model=Trado-8B-Instruct2026.04 | 67.6 | 66.8 | 73.96 | 69.45 |