Toxicity Reduction on HolisticBias
100.7HolisticBias RatioPhi-3-Mini
Evaluation Results
| Method | Links | |
|---|---|---|
| Phi-3-Minireranking=none2026.02 | 100.7 | |
| FARO-cfbackbone=Phi-3-Mini, reranking=FARO reward model2026.02 | 99.5 | |
| FARO-dpbackbone=Phi-3-Mini, reranking=FARO reward model2026.02 | 94.7 | |
| Qwen-2.5-1.5Breranking=none2026.02 | 94.5 | |
| FARO-cfbackbone=Qwen-2.5-1.5B, reranking=FARO reward model2026.02 | 93.5 | |
| FARO-eobackbone=Phi-3-Mini, reranking=FARO reward model2026.02 | 92.5 | |
| FARO-eobackbone=Qwen-2.5-1.5B, reranking=FARO reward model2026.02 | 91.1 | |
| FARO-dpbackbone=Qwen-2.5-1.5B, reranking=FARO reward model2026.02 | 90.6 | |
| Gemma-2-2b-itreranking=none2026.02 | 71.2 | |
| FARO-dpbackbone=Gemma-2-2b-it, reranking=FARO reward model2026.02 | 46.2 | |
| FARO-cfbackbone=Gemma-2-2b-it, reranking=FARO reward model2026.02 | 35.6 | |
| FARO-eobackbone=Gemma-2-2b-it, reranking=FARO reward model2026.02 | 19.3 |