Hate Speech Detection on HateXplain (test)
85.13Macro F1 ScoreX-MuTeST with LLM
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| X-MuTeST with LLMFramework=X-MuTeST, Consultant=LLM2026.01 | 85.13 | 86.04 | — | — | — | — | — | 88.27 | — | |
| Muril-Rationale-LIMEBackbone=MuRIL, Explainer=LIME, With-Rationale=true2026.01 | 83.85 | 84.33 | — | — | — | — | — | 87.44 | — | |
| Muril-Rationale-XMuTeSTBackbone=MuRIL, Explainer=X-MuTeST, With-Rationale=true2026.01 | 83.85 | 84.33 | — | — | — | — | — | 87.44 | — | |
| BERT-HateXplain-LIMEBackbone=BERT, Training-Source=HateXplain, Explainer=LIME2026.01 | 82.47 | 83.39 | — | — | — | — | — | 86.47 | — | |
| XLMR-Rationale-LIMEBackbone=XLM-R, Explainer=LIME, With-Rationale=true2026.01 | 82.45 | 83.49 | — | — | — | — | — | 86.73 | — | |
| XLMR-Rationale-XMuTeSTBackbone=XLM-R, Explainer=X-MuTeST, With-Rationale=true2026.01 | 82.45 | 83.49 | — | — | — | — | — | 86.73 | — | |
| Muril-LIMEBackbone=MuRIL, Explainer=LIME2026.01 | 81.44 | 83.07 | — | — | — | — | — | 86.94 | — | |
| Muril-XMuTeSTBackbone=MuRIL, Explainer=X-MuTeST2026.01 | 81.44 | 83.07 | — | — | — | — | — | 86.94 | — | |
| XLMR-LIMEBackbone=XLM-R, Explainer=LIME2026.01 | 80.75 | 82.23 | — | — | — | — | — | 86.09 | — | |
| XLMR-XMuTeSTBackbone=XLM-R, Explainer=X-MuTeST2026.01 | 80.75 | 82.23 | — | — | — | — | — | 86.09 | — | |
| BERT-ILPBackbone=BERT, Explainer=ILP2026.01 | 80.28 | 81.45 | — | — | — | — | — | 79.67 | — | |
| Full modelBase Model=OPT, Early-exit strategy=None2025.11 | 78.62 | — | — | — | — | — | — | — | 12 | |
| PABEEBase Model=OPT, Early-exit strategy=Patience-based2025.11 | 78.55 | — | — | — | — | — | — | — | 9.34 | |
| Prototype-based@L12Base Model=OPT, Early-exit strategy=Static Layer 122025.11 | 77.86 | — | — | — | — | — | — | — | 12 | |
| Full modelBase Model=BERT, Early-exit strategy=None2025.11 | 77.7 | — | — | — | — | — | — | — | 12 | |
| DeeOPTBase Model=OPT, Early-exit strategy=Entropy-based2025.11 | 77.68 | — | — | — | — | — | — | — | 10.25 | |
| HatePrototypesBase Model=OPT, Early-exit strategy=Prototype-based (per-sample)2025.11 | 77.56 | — | — | — | — | — | — | — | 9.47 | |
| PABEEBase Model=BERT, Early-exit strategy=Patience-based2025.11 | 77.14 | — | — | — | — | — | — | — | 9.15 | |
| Prototype-based@L12Base Model=BERT, Early-exit strategy=Static Layer 122025.11 | 77.12 | — | — | — | — | — | — | — | 12 | |
| HatePrototypesBase Model=BERT, Early-exit strategy=Prototype-based (per-sample)2025.11 | 76.64 | — | — | — | — | — | — | — | 9.75 | |
| DeeBERTBase Model=BERT, Early-exit strategy=Entropy-based2025.11 | 76.48 | — | — | — | — | — | — | — | 8.5 | |
| Mistral-24.07Model=Mistral-24.072026.01 | 76.43 | 79.18 | — | — | — | — | — | 84.48 | — | |
| GPT-4oModel=GPT-4o2026.01 | 76.33 | 78.02 | — | — | — | — | — | 82.66 | — | |
| G-RAT2024.12 | 73.85 | — | — | — | — | — | — | — | — | |
| Llama-3.1Model=Llama-3.12026.01 | 73.62 | 75.6 | — | — | — | — | — | 80.86 | — | |
| FR2024.12 | 72.14 | — | — | — | — | — | — | — | — | |
| MGR2024.12 | 71.14 | — | — | — | — | — | — | — | — | |
| MCD2024.12 | 70.37 | — | — | — | — | — | — | — | — | |
| BERT-MRPration.=true, pre-fin.=true2022.11 | 69.9 | 70.4 | 86.2 | 81.5 | 74.8 | 85.4 | — | — | — | |
| GenSPP2024.12 | 69.71 | — | — | — | — | — | — | — | — | |
| BERT-RPration.=true, pre-fin.=true2022.11 | 69.3 | 70.7 | 85.3 | 81.4 | 74.6 | 84.8 | — | — | — | |
| BERT-HateXplainration.=true, pre-fin.=false2022.11 | 68.7 | 69.8 | 85.1 | 80.7 | 74.5 | 76.3 | — | — | — | |
| BERT-MLMration.=false, pre-fin.=true2022.11 | 67.5 | 70 | 85.4 | 79 | 67.7 | 80.9 | — | — | — | |
| BERTration.=false, pre-fin.=false2022.11 | 67.4 | 69 | 84.3 | 76.2 | 70.9 | 75.7 | — | — | — | |
| Yuvion-32BParameters=32B2026.06 | 63.6 | — | — | — | — | — | — | — | — | |
| Qwen1.5-72BParameters=72B2026.06 | 63.4 | — | — | — | — | — | — | — | — | |
| Prototype-based@L1Base Model=BERT, Early-exit strategy=Static Layer 12025.11 | 60.67 | — | — | — | — | — | — | — | 1 | |
| DeepSeek-67BParameters=67B2026.06 | 58.6 | — | — | — | — | — | — | — | — | |
| Yuvion-8BParameters=8B2026.06 | 58.5 | — | — | — | — | — | — | — | — | |
| Qwen2-72BParameters=72B2026.06 | 58.3 | — | — | — | — | — | — | — | — | |
| GPT-3.52026.06 | 55.6 | — | — | — | — | — | — | — | — | |
| Prototype-based@L1Base Model=OPT, Early-exit strategy=Static Layer 12025.11 | 53.26 | — | — | — | — | — | — | — | 1 | |
| Qwen2-8BParameters=8B2026.06 | 51.8 | — | — | — | — | — | — | — | — | |
| Qwen2-32BParameters=32B2026.06 | 50.7 | — | — | — | — | — | — | — | — | |
| xList-HateModel=llama3-1b2026.02 | — | — | 0.519 | — | — | — | -0.015 | — | — | |
| xList-HateModel=llama3-3b2026.02 | — | — | 0.767 | — | — | — | 0.146 | — | — | |
| xList-HateModel=gemma3-4b2026.02 | — | — | 0.806 | — | — | — | 0.187 | — | — | |
| xList-HateModel=mistral-7b2026.02 | — | — | 0.797 | — | — | — | 0.176 | — | — | |
| xList-HateModel=llama3-8b2026.02 | — | — | 0.608 | — | — | — | 0.046 | — | — | |
| xList-HateModel=gemma3-12b2026.02 | — | — | 0.831 | — | — | — | 0.17 | — | — | |
| xList-HateModel=mistral-24b2026.02 | — | — | 0.802 | — | — | — | 0.178 | — | — | |
| xList-HateModel=gemma3-27b2026.02 | — | — | 0.802 | — | — | — | 0.197 | — | — | |
| Zero-shotModel=llama3-1b2026.02 | — | — | 0.534 | — | — | — | — | — | — | |
| Zero-shotModel=llama3-3b2026.02 | — | — | 0.621 | — | — | — | — | — | — | |
| Zero-shotModel=gemma3-4b2026.02 | — | — | 0.619 | — | — | — | — | — | — | |
| Zero-shotModel=mistral-7b2026.02 | — | — | 0.621 | — | — | — | — | — | — | |
| Zero-shotModel=llama3-8b2026.02 | — | — | 0.562 | — | — | — | — | — | — | |
| Zero-shotModel=gemma3-12b2026.02 | — | — | 0.662 | — | — | — | — | — | — | |
| Zero-shotModel=mistral-24b2026.02 | — | — | 0.625 | — | — | — | — | — | — | |
| Zero-shotModel=gemma3-27b2026.02 | — | — | 0.605 | — | — | — | — | — | — |