Commonsense Reasoning on LMEval
45.6Average ScoreTransformer
Evaluation Results
| Method | Links | |
|---|---|---|
| TransformerLayers=24, Protocol=Zero-shot2025.06 | 45.6 | |
| Log-Linear Gated DeltaNetLayers=21, Log-Linear variation=true, Protocol=Zero-shot2025.06 | 45.5 | |
| Gated DeltaNetLayers=21, Log-Linear variation=false, Protocol=Zero-shot2025.06 | 45 | |
| Log-Linear Mamba-2Layers=21, Log-Linear variation=true, Protocol=Zero-shot2025.06 | 44.9 | |
| Mamba-2Layers=21, Log-Linear variation=false, Protocol=Zero-shot2025.06 | 44.8 | |
| TransformerLayers=21, Protocol=Zero-shot2025.06 | 44 |