Safety Classification on ToxicChat (in-distribution)
82.2F1 Score (%)ToxicChat-T5-large
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ToxicChat-T5-largeAdded Params (M)=780, Extra LM call=true2026.01 | 82.2 | 0.885 | |
| MULI (logits)Added Params (M)=0.13, Extra LM call=false, Backbone=Llama-3.2-3B2026.01 | 77.8 | 0.829 |