Behavioral Alignment Steering on Six Behavioral Traits (Persona Dataset) (test)
99.6Conscientious (Accuracy)INNSTEER
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| INNSTEERModel Backbone=LLaMA3-8B2026.06 | 99.6 | 12.76 | 98.81 | 9.41 | 99.23 | 11.99 | 97.73 | 12.41 | 98.47 | 10.8 | 97.47 | 9.69 | |
| INNSTEERModel Backbone=Qwen2.5-3B2026.06 | 98.07 | 4.97 | 83.37 | 4.84 | 93.19 | 5.04 | 96.07 | 7.04 | 89.67 | 7.43 | 76.59 | 6.62 | |
| INNSTEERModel Backbone=Qwen2.5-7B2026.06 | 96.43 | 17.74 | 95.4 | 12.14 | 93.74 | 16.28 | 95.97 | 10.71 | 94.16 | 14.24 | 97.61 | 13.01 | |
| INNSTEERModel Backbone=Qwen2.5-32B2026.06 | 96.01 | 4.94 | 99.48 | 4.77 | 94.75 | 5.54 | 98.41 | 4.17 | 99.13 | 5.39 | 85.8 | 5.45 | |
| INNSTEERModel Backbone=LLaMA3-3B2026.06 | 95.64 | 26.46 | 97.92 | 20.87 | 95.7 | 25.42 | 90.72 | 30.41 | 93.54 | 25.85 | 86.93 | 24.6 | |
| BaseModel Backbone=Qwen2.5-32B2026.06 | 95.57 | 3.45 | 60.77 | 3.38 | 82.59 | 3.76 | 97.89 | 3.88 | 91.97 | 3.81 | 67.28 | 3.69 | |
| BaseModel Backbone=Qwen2.5-7B2026.06 | 88.08 | 15.3 | 84.39 | 9.8 | 83.28 | 10.94 | 94.43 | 11.82 | 87.43 | 10.61 | 73.8 | 9.5 | |
| ITIModel Backbone=LLaMA3-3B2026.06 | 83.34 | 26.9 | 82.74 | 21.61 | 78.43 | 25.49 | 69.04 | 30.36 | 78.43 | 26.91 | 68.5 | 25.34 | |
| BaseModel Backbone=LLaMA3-8B2026.06 | 82.06 | 12.79 | 68.93 | 9.46 | 74.17 | 12.2 | 81.86 | 12.32 | 81.22 | 11.08 | 77.45 | 9.55 | |
| ODEModel Backbone=LLaMA3-3B2026.06 | 81.13 | 26.85 | 74.85 | 21.44 | 78.42 | 25.36 | 70.77 | 30.4 | 90.98 | 26.73 | 76.7 | 25.34 | |
| PoEModel Backbone=LLaMA3-3B2026.06 | 77.82 | 26.84 | 64.68 | 21.57 | 75.56 | 24.94 | 64.33 | 30.17 | 64.1 | 26.25 | 68.75 | 25.04 | |
| PoDModel Backbone=LLaMA3-3B2026.06 | 77.81 | 26.84 | 64.34 | 21.57 | 75.51 | 24.93 | 64.24 | 30.17 | 63.5 | 30.17 | 68.7 | 25.04 | |
| CAAModel Backbone=LLaMA3-3B2026.06 | 77.16 | 26.89 | 64.9 | 21.56 | 76 | 25.34 | 66.61 | 30.29 | 87.08 | 26.59 | 71.01 | 25.18 | |
| LAcTModel Backbone=LLaMA3-3B2026.06 | 75.07 | 26.9 | 61.27 | 21.5 | 74.41 | 25.26 | 65.83 | 30.3 | 82.6 | 26.58 | 69.26 | 25.17 | |
| TFModel Backbone=LLaMA3-3B2026.06 | 73.53 | 26.95 | 56.03 | 21.89 | 73.87 | 25.75 | 65.93 | 30.8 | 70.65 | 27.21 | 66.98 | 25.87 | |
| BaseModel Backbone=LLaMA3-3B2026.06 | 72.51 | 26.82 | 55.68 | 21.4 | 70.96 | 25.13 | 59.77 | 30.25 | 75.56 | 26.38 | 65.89 | 25 | |
| RECModel Backbone=LLaMA3-3B2026.06 | 72.51 | 26.83 | 55.69 | 21.4 | 72.97 | 25.13 | 59.78 | 30.25 | 75.58 | 26.38 | 66.89 | 26 | |
| HPRModel Backbone=LLaMA3-3B2026.06 | 70.92 | 26.91 | 73.45 | 21.91 | 74.59 | 25.52 | 70.69 | 30.4 | 78.21 | 22.22 | 65.4 | 25.56 | |
| BaseModel Backbone=Qwen2.5-3B2026.06 | 70.2 | 4.95 | 59.74 | 4.91 | 62.49 | 4.42 | 57.02 | 4.6 | 75.5 | 5.42 | 60.6 | 4.35 |