Language Model Detoxification on Human Evaluation 50 generations (test)
0.49Detoxification CountLM-Steer
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LM-Steerbaseline_comparison=GPT-22023.05 | 0.49 | 0.42 | 0.64 | |
| LM-Steerbaseline_comparison=DExperts2023.05 | 0.48 | 0.5 | 0.64 | |
| DExpertsbaseline_comparison=LM-Steer2023.05 | 0.39 | 0.46 | 0.23 | |
| LM-Steerbaseline_comparison=LoRA2023.05 | 0.38 | 0.42 | 0.36 | |
| GPT-2baseline_comparison=LM-Steer2023.05 | 0.38 | 0.43 | 0.42 | |
| LoRAbaseline_comparison=LM-Steer2023.05 | 0.23 | 0.2 | 0.25 |