Toxicity Classification on CivilComments Hard subset - demographic identities
86.9Hard AccuracyRobust Filtered
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Robust Filtered2026.06 | 86.9 | 4 | 3.1 | |
| Robust MoE2026.06 | 86.8 | 3.7 | 2.9 | |
| Single Expert2026.06 | 86.7 | 10.5 | 9.7 | |
| Vanilla MoE2026.06 | 86.6 | 10.8 | 10 | |
| MoCaE2026.06 | 86.6 | 10.1 | 9.1 | |
| FGR2026.06 | 85.7 | 6.5 | 5.3 | |
| FGR + Robust2026.06 | 85.3 | 6.5 | 5.7 |