Commonsense Reasoning on HellaSwag (200-sample subset)
67AccuracyLlama
Evaluation Results
| Method | Links | |
|---|---|---|
| LlamaSetting=CPU2026.06 | 67 | |
| LlamaSetting=NPU2026.06 | 60 | |
| LLaDASetting=CPU2026.06 | 51 | |
| LLaDA (Ablation)Setting=CPU2026.06 | 51 | |
| LLaDA (Ablation)Setting=NPU2026.06 | 50 | |
| LLaDA llada.cppSetting=llada.cpp2026.06 | 49.5 | |
| LLaDA (Ablation)Setting=NPU + MBSD + DPPR2026.06 | 49.5 | |
| LLaDA (Ablation)Setting=NPU + MBSD2026.06 | 46 |