Zero-shot Common Sense Reasoning on SIQA
41.91Accuracy (Zero-shot)Gated DeltaNet-H2
Evaluation results
| Method | Links | |
|---|---|---|
| Gated DeltaNet-H2Setting=Zero-shot, Model Size=1.3B parameters2025.10 | 41.91 | |
| Nirvana (Ours)Setting=Zero-shot, Model Size=1.3B parameters2025.10 | 41.62 | |
| Nirvana-noTriggerSetting=Zero-shot, Model Size=1.3B parameters2025.10 | 41.5 | |
| Gated DeltaNet-H1Setting=Zero-shot, Model Size=1.3B parameters2025.10 | 41.4 | |
| RetNetSetting=Zero-shot, Model Size=1.3B parameters2025.10 | 40.78 | |
| Transformer++Setting=Zero-shot, Model Size=1.3B parameters2025.10 | 40.66 | |
| HGRN2Setting=Zero-shot, Model Size=1.3B parameters2025.10 | 40.63 | |
| Gated DeltaNetSetting=Zero-shot, Model Size=1.3B parameters2025.10 | 40.63 | |
| DeltaNetSetting=Zero-shot, Model Size=1.3B parameters2025.10 | 40.22 | |
| Mamba2Setting=Zero-shot, Model Size=1.3B parameters2025.10 | 40.2 | |
| SambaSetting=Zero-shot, Model Size=1.3B parameters2025.10 | 39.96 | |
| MambaSetting=Zero-shot, Model Size=1.3B parameters2025.10 | 37.76 |