Commonsense Reasoning Accuracy on HellaSwag
50.95AccuracyFIXED_E=128
Evaluation Results
| Method | Links | |
|---|---|---|
| FIXED_E=128Experts=1282026.05 | 50.95 | |
| EMO (Stage 5)E=64→1282026.05 | 49.86 | |
| FIXED_E=32Experts=322026.05 | 47.14 | |
| FIXED_E=16Experts=162026.05 | 44.59 | |
| EMO (Stage 4)E=32→642026.05 | 44.33 | |
| EMO (Stage 3)E=16→322026.05 | 41.93 | |
| EMO (Stage 2)E=8→162026.05 | 39.81 | |
| EMO (Stage 1)E=82026.05 | 38.88 |