Value Alignment on ValueKaleidoscope VITAL Steerable setting
72.11Value Alignment ScoreVanilla
Evaluation results
| Method | Links | |
|---|---|---|
| VanillaBase Model=Qwen2.5-14B, Steering Method=Vanilla2026.01 | 72.11 | |
| VISPABase Model=ChatGPT, Steering Backbone=LLaMA3-8B2026.01 | 68.7 | |
| VanillaBase Model=LLaMA3-8B, Steering Method=Vanilla2026.01 | 67.71 | |
| VISPABase Model=ChatGPT, Steering Backbone=LLaMA2-7B2026.01 | 67.31 | |
| VanillaBase Model=Qwen2.5-7B, Steering Method=Vanilla2026.01 | 66.68 | |
| VanillaBase Model=ChatGPT, Steering Method=Vanilla2026.01 | 65.6 | |
| VISPABase Model=Qwen2.5-14B, Steering Backbone=LLaMA3-8B2026.01 | 63.89 | |
| VISPABase Model=Qwen2.5-14B, Steering Backbone=LLaMA2-7B2026.01 | 62.17 | |
| VISPABase Model=Qwen2.5-7B, Steering Backbone=LLaMA3-8B2026.01 | 58.58 | |
| ModPluralBase Model=Qwen2.5-14B, Steering Method=ModPlural2026.01 | 58.22 | |
| EthosBase Model=Qwen2.5-7B, Steering Method=Ethos2026.01 | 57.66 | |
| VISPABase Model=Qwen2.5-7B, Steering Backbone=LLaMA2-7B2026.01 | 56.6 | |
| VISPABase Model=LLaMA3-8B, Steering Backbone=LLaMA3-8B2026.01 | 52.63 | |
| VISPABase Model=LLaMA3-8B, Steering Backbone=LLaMA2-7B2026.01 | 50.81 | |
| MoEBase Model=Qwen2.5-7B, Steering Method=MoE2026.01 | 50.64 | |
| EthosBase Model=LLaMA3-8B, Steering Method=Ethos2026.01 | 50.34 | |
| VISPABase Model=Gemma-7B, Steering Backbone=LLaMA3-8B2026.01 | 50.22 | |
| MoEBase Model=Qwen2.5-14B, Steering Method=MoE2026.01 | 49.99 | |
| VISPABase Model=Gemma-7B, Steering Backbone=LLaMA2-7B2026.01 | 49.97 | |
| ModPluralBase Model=Qwen2.5-7B, Steering Method=ModPlural2026.01 | 49.87 | |
| VISPABase Model=LLaMA2-7B, Steering Backbone=LLaMA3-8B2026.01 | 49.57 | |
| VISPABase Model=LLaMA2-7B, Steering Backbone=LLaMA2-7B2026.01 | 49.56 | |
| VanillaBase Model=Gemma-7B, Steering Method=Vanilla2026.01 | 48.54 | |
| EthosBase Model=Qwen2.5-14B, Steering Method=Ethos2026.01 | 48.33 | |
| EthosBase Model=ChatGPT, Steering Method=Ethos2026.01 | 48.02 | |
| ModPluralBase Model=ChatGPT, Steering Method=ModPlural2026.01 | 47 | |
| MoEBase Model=LLaMA3-8B, Steering Method=MoE2026.01 | 45.53 | |
| MoEBase Model=ChatGPT, Steering Method=MoE2026.01 | 44.9 | |
| ModPluralBase Model=Gemma-7B, Steering Method=ModPlural2026.01 | 42.03 | |
| ModPluralBase Model=LLaMA3-8B, Steering Method=ModPlural2026.01 | 41.78 | |
| MoEBase Model=Gemma-7B, Steering Method=MoE2026.01 | 41.74 | |
| VISPABase Model=LLaMA2-13B, Steering Backbone=LLaMA3-8B2026.01 | 41.53 | |
| EthosBase Model=LLaMA2-13B, Steering Method=Ethos2026.01 | 39.6 | |
| EthosBase Model=LLaMA2-7B, Steering Method=Ethos2026.01 | 38.42 | |
| EthosBase Model=Gemma-7B, Steering Method=Ethos2026.01 | 37.75 | |
| MoEBase Model=LLaMA2-7B, Steering Method=MoE2026.01 | 35.48 | |
| MoEBase Model=LLaMA2-13B, Steering Method=MoE2026.01 | 35.23 | |
| ModPluralBase Model=LLaMA2-13B, Steering Method=ModPlural2026.01 | 35.07 | |
| ModPluralBase Model=LLaMA2-7B, Steering Method=ModPlural2026.01 | 34.92 | |
| VanillaBase Model=LLaMA2-7B, Steering Method=Vanilla2026.01 | 34.33 | |
| VISPABase Model=LLaMA2-13B, Steering Backbone=LLaMA2-7B2026.01 | 33.93 | |
| VanillaBase Model=LLaMA2-13B, Steering Method=Vanilla2026.01 | 19.8 |