Attribute Steering on HelpSteer
3.89HelpfulnessMSRS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MSRSBackbone=Llama3-8b-inst., Method=MSRS2025.08 | 3.89 | 3.96 | 2.04 | |
| MAT-STEERBackbone=Llama3-8b-inst., Method=MAT-STEER2025.08 | 3.84 | 3.63 | 2.29 | |
| ICLBackbone=Llama3-8b-inst., Method=ICL2025.08 | 3.82 | 3.64 | 2.41 | |
| ITIBackbone=Llama3-8b-inst., Method=ITI2025.08 | 3.82 | 3.21 | 2.06 | |
| MSRSBackbone=Mistral-7b-v0.3, Method=MSRS2025.08 | 3.82 | 3.93 | 2.21 | |
| ReFTBackbone=Mistral-7b-v0.3, Method=ReFT2025.08 | 3.8 | 3.85 | 2.33 | |
| CS-ReFTBackbone=Mistral-7b-v0.3, Method=CS-ReFT2025.08 | 3.79 | 3.89 | 2.29 | |
| ReFTBackbone=Llama3-8b-inst., Method=ReFT2025.08 | 3.78 | 3.83 | 2.38 | |
| CS-ReFTBackbone=Llama3-8b-inst., Method=CS-ReFT2025.08 | 3.78 | 3.86 | 2.01 | |
| MAT-STEERBackbone=Mistral-7b-v0.3, Method=MAT-STEER2025.08 | 3.78 | 3.86 | 2.29 | |
| CAABackbone=Llama3-8b-inst., Method=CAA2025.08 | 3.77 | 3.89 | 2.51 | |
| ICLBackbone=Mistral-7b-v0.3, Method=ICL2025.08 | 3.77 | 3.94 | 2.39 | |
| Llama3-8b-inst.Backbone=Llama3-8b-inst., Method=Base2025.08 | 3.76 | 3.41 | 2.33 | |
| MSRSBackbone=Qwen2-7b-inst., Method=MSRS2025.08 | 3.76 | 3.82 | 2.17 | |
| CAABackbone=Mistral-7b-v0.3, Method=CAA2025.08 | 3.76 | 3.83 | 2.27 | |
| Mistral-7b-v0.3Backbone=Mistral-7b-v0.3, Method=Base2025.08 | 3.75 | 3.92 | 2.36 | |
| CAABackbone=Qwen2-7b-inst., Method=CAA2025.08 | 3.73 | 3.87 | 2.2 | |
| CS-ReFTBackbone=Qwen2-7b-inst., Method=CS-ReFT2025.08 | 3.72 | 3.84 | 2.22 | |
| MAT-STEERBackbone=Qwen2-7b-inst., Method=MAT-STEER2025.08 | 3.7 | 3.77 | 2.25 | |
| ICLBackbone=Qwen2-7b-inst., Method=ICL2025.08 | 3.65 | 3.88 | 2.4 | |
| ReFTBackbone=Qwen2-7b-inst., Method=ReFT2025.08 | 3.63 | 3.78 | 2.38 | |
| Qwen2-7b-inst.Backbone=Qwen2-7b-inst., Method=Base2025.08 | 3.51 | 3.8 | 2.28 |