Text Classification on CivilComments (ECE, Brier, AUROC, F1)
0.092ECESemantic Softmax
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Semantic SoftmaxModel=Phi-4-mini2026.05 | 0.092 | 0.498 | 83.5 | 45.1 | |
| Semantic SoftmaxModel=Qwen-3-1.7B2026.05 | 0.108 | 0.517 | 88.2 | 45.1 | |
| StandardModel=Phi-4-mini2026.05 | 0.395 | 0.542 | 81.2 | 42.1 | |
| StandardModel=Qwen-3-1.7B2026.05 | 0.482 | 0.571 | 78.4 | 41.2 |