Physical Commonsense Reasoning on PIQA (5-shot)
77.8AccuracyMoE 128e8a1s
Evaluation Results
| Method | Links | |
|---|---|---|
| MoE 128e8a1sModel=MoE 128e8a1s, Training Steps=100k2026.05 | 77.8 | |
| MoE 128e8a4g1sModel=MoE 128e8a4g1s, Training Steps=100k2026.05 | 77.7 | |
| DenseModel=Dense, Training Steps=100k2026.05 | 72.6 |