Commonsense Reasoning on HellaSwag (Accuracy, Normalized Accuracy)
44.3AccuracyAIRAhybrid-D
Evaluation results
| Method | Links | ||
|---|---|---|---|
| AIRAhybrid-DShot=0-shot, Architecture Variant=Stacked2026.05 | 44.3 | 57.6 | |
| Nemotron-2 Approx.Shot=0-shot2026.05 | 44.2 | 57.2 | |
| AIRAhybrid-BShot=0-shot, Architecture Variant=Stretched2026.05 | 44.2 | 57.6 | |
| AIRAhybrid-DShot=0-shot, Architecture Variant=Stretched2026.05 | 44.1 | 57.8 | |
| AIRAhybrid-EShot=0-shot, Architecture Variant=Stretched2026.05 | 44.1 | 57.1 | |
| AIRAhybrid-AShot=0-shot, Architecture Variant=Stretched2026.05 | 44 | 56.6 | |
| AIRAhybrid-BShot=0-shot, Architecture Variant=Stacked2026.05 | 44 | 57.1 | |
| Composer (2Mb-M-3A)Shot=0-shot2026.05 | 43.9 | 56.9 | |
| AIRAhybrid-EShot=0-shot, Architecture Variant=Stacked2026.05 | 43.5 | 56.5 | |
| Nemotron-H Approx.Shot=0-shot2026.05 | 43.4 | 56.3 | |
| AIRAhybrid-CShot=0-shot, Architecture Variant=Stretched2026.05 | 43.3 | 56.1 | |
| Mamba (Mb + M)Shot=0-shot2026.05 | 43.2 | 55.8 |