Commonsense Reasoning on HellaSwag (200 samples)
0.67Log-likelihood ScoreGemma 3 12B IT (Base)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Gemma 3 12B IT (Base)Mode=Base, Quantization=4-bit NF4, Max sequence length=512, Zero-shot=true2026.04 | 0.67 | — | |
| Brainstacks (Routed)Mode=Routed, Quantization=4-bit NF4, Max sequence length=512, Zero-shot=true, Routing mechanism=meta-router selective gating, Chat floor=0.20, Gating=sigmoid2026.04 | 0.65 | 0.02 |