Detoxification on SafeNLP (test)
84SimilarityLlama-2-7b
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Llama-2-7bEvaluation Mode=White-box2024.01 | 84 | 76.9 | |
| Falcon-7bEvaluation Mode=White-box2024.01 | 66 | 58.9 | |
| GPT-3.5-TurboEvaluation Mode=White-box2024.01 | 53 | 3.36 | |
| Mistral-7bEvaluation Mode=White-box2024.01 | 48 | 33.1 | |
| GPT-2-XLEvaluation Mode=White-box2024.01 | 46 | 28.18 | |
| Falcon-7b + CPEvaluation Mode=White-box, Backbone=Falcon-7b2024.01 | 46 | 36.6 | |
| Mistral-7b + CPEvaluation Mode=White-box, Backbone=Mistral-7b2024.01 | 40 | 4.3 | |
| GPT-2Evaluation Mode=White-box2024.01 | 36 | 28.94 | |
| CHRTEvaluation Mode=White-box, Backbone=GPT-22024.01 | 34 | 25.7 | |
| Distill-GPT-2Evaluation Mode=White-box2024.01 | 24 | 30.4 | |
| Model ArithmeticEvaluation Mode=White-box, Backbone=Mistral-7b2024.01 | 24 | 12.2 | |
| Llama-2-7b + CPEvaluation Mode=White-box, Backbone=Llama-2-7b2024.01 | 24 | 11.4 | |
| CHRTEvaluation Mode=White-box, Backbone=Mistral-7b2024.01 | 22 | 13.6 |