Harmlessness on Toxigen
100Toxigen (%)SafeRLHF Specialist
Evaluation Results
| Method | Links | |
|---|---|---|
| SafeRLHF Specialist2025.05 | 100 | |
| SafeRLHF Specialist2025.05 | 100 | |
| Standard UniformSampling=Uniform2025.05 | 99.99 | |
| Standard2025.05 | 99.99 | |
| AMA Resamplingoptimization_strategy=resampling2025.05 | 99.99 | |
| Model Averaging2025.05 | 99.96 | |
| Standard Uniform2025.05 | 99.93 | |
| Standard2025.05 | 99.91 | |
| AMA ResamplingStrategy=Resampling2025.05 | 99.8 | |
| AMA Reweightingoptimization_strategy=reweighting2025.05 | 99.74 | |
| Model Averaging2025.05 | 96.7 | |
| AMA ReweightingStrategy=Reweighting2025.05 | 95.91 | |
| CodeUltraFeedback Specialist2025.05 | 76.31 | |
| Chatbot Arena 2024 Specialist2025.05 | 69.77 | |
| Zephyr-7b-sft-full2025.05 | 51.6 | |
| Zephyr-7b-sft-full2025.05 | 51.6 | |
| UltraFeedback Specialist2025.05 | 36.1 |