List your company
ResearchFormerly WizWand

Hate-speech detection on OLID (test)

12AvgExitFull model
0.563.536.59.47Nov 9, 2025
Updated 7mo ago

Evaluation results

MethodLinks
Full modelBase Model=BERT, Early-exit strategy=None2025.111279.97
Prototype-based@L12Base Model=BERT, Early-exit strategy=Static Layer 122025.111279.91
Full modelBase Model=OPT, Early-exit strategy=None2025.111280.77
Prototype-based@L12Base Model=OPT, Early-exit strategy=Static Layer 122025.111281.19
DeeBERTBase Model=BERT, Early-exit strategy=Entropy-based2025.1110.8179.57
HatePrototypesBase Model=BERT, Early-exit strategy=Prototype-based (per-sample)2025.1110.7179.89
PABEEBase Model=OPT, Early-exit strategy=Patience-based2025.119.0246.34
HatePrototypesBase Model=OPT, Early-exit strategy=Prototype-based (per-sample)2025.118.8581.11
DeeOPTBase Model=OPT, Early-exit strategy=Entropy-based2025.118.8372.44
PABEEBase Model=BERT, Early-exit strategy=Patience-based2025.118.6251.21
Prototype-based@L1Base Model=BERT, Early-exit strategy=Static Layer 12025.11148.98
Prototype-based@L1Base Model=OPT, Early-exit strategy=Static Layer 12025.11154.55