Over-Safety Evaluation on XSTest (XST Metric)
23.2XST Over-SafetyTELLME
Evaluation Results
| Method | Links | |
|---|---|---|
| TELLMEBackbone=Qwen2.5-7B2025.02 | 23.2 | |
| TELLME NT-XentBackbone=Qwen2.5-7B2025.02 | 22.4 | |
| TELLME NT-XentBackbone=Llama-3.1-8B2025.02 | 21.2 | |
| TELLMEBackbone=Llama-3.1-8B2025.02 | 18 | |
| SFTBackbone=Llama-3.1-8B2025.02 | 16.4 | |
| OriginBackbone=Qwen2.5-7B2025.02 | 16 | |
| SFTBackbone=Mistral-7B-v0.32025.02 | 15.6 | |
| OriginBackbone=Mistral-7B-v0.32025.02 | 14.4 | |
| SFTBackbone=Qwen2.5-7B2025.02 | 12 | |
| TELLME NT-XentBackbone=Mistral-7B-v0.32025.02 | 10.8 | |
| TELLMEBackbone=Mistral-7B-v0.32025.02 | 9.2 | |
| OriginBackbone=Llama-3.1-8B2025.02 | 6.4 |