Refusal Behavior on Refusal
69.3Sorry RateMSRS
Evaluation Results
| Method | Links | |
|---|---|---|
| MSRSBackbone=Mistral-7b-v0.3, Method=MSRS2025.08 | 69.3 | |
| ReFTBackbone=Mistral-7b-v0.3, Method=ReFT2025.08 | 66.8 | |
| CAABackbone=Mistral-7b-v0.3, Method=CAA2025.08 | 66.3 | |
| CS-ReFTBackbone=Mistral-7b-v0.3, Method=CS-ReFT2025.08 | 65.9 | |
| MAT-STEERBackbone=Mistral-7b-v0.3, Method=MAT-STEER2025.08 | 64.4 | |
| ICLBackbone=Mistral-7b-v0.3, Method=ICL2025.08 | 64.2 | |
| Mistral-7b-v0.3Backbone=Mistral-7b-v0.3, Method=Base2025.08 | 63.1 | |
| MSRSBackbone=Llama3-8b-inst., Method=MSRS2025.08 | 52.9 | |
| ICLBackbone=Llama3-8b-inst., Method=ICL2025.08 | 52.1 | |
| CS-ReFTBackbone=Llama3-8b-inst., Method=CS-ReFT2025.08 | 51.4 | |
| CAABackbone=Llama3-8b-inst., Method=CAA2025.08 | 49.3 | |
| Llama3-8b-inst.Backbone=Llama3-8b-inst., Method=Base2025.08 | 49.1 | |
| ReFTBackbone=Llama3-8b-inst., Method=ReFT2025.08 | 45.1 | |
| MSRSBackbone=Qwen2-7b-inst., Method=MSRS2025.08 | 44.5 | |
| MAT-STEERBackbone=Qwen2-7b-inst., Method=MAT-STEER2025.08 | 42.9 | |
| CS-ReFTBackbone=Qwen2-7b-inst., Method=CS-ReFT2025.08 | 42.5 | |
| ReFTBackbone=Qwen2-7b-inst., Method=ReFT2025.08 | 42.1 | |
| MAT-STEERBackbone=Llama3-8b-inst., Method=MAT-STEER2025.08 | 42 | |
| ICLBackbone=Qwen2-7b-inst., Method=ICL2025.08 | 41.3 | |
| CAABackbone=Mistral-7b-v0.3, Method=CAA2025.08 | 40.4 | |
| Qwen2-7b-inst.Backbone=Qwen2-7b-inst., Method=Base2025.08 | 38.4 | |
| ITIBackbone=Llama3-8b-inst., Method=ITI2025.08 | 28 |