Value Alignment Text Rewriting on Value Alignment Dataset
87.32Consistency MeanVISA
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| VISAModel=Qwen3-4B2026.03 | 87.32 | 0.107 | 87.33 | 0.1179 | 87.29 | 0.1354 | 77.56 | 0.3049 | 70.75 | 0.3143 | |
| Simple-PromptModel=GPT-4o-Mini2026.03 | 84.06 | 0.1496 | 85.23 | 0.1599 | 81.72 | 0.1979 | 79.86 | 0.3011 | 69.35 | 0.3167 | |
| Simple-PromptModel=GPT-4o2026.03 | 78.31 | 0.1777 | 81.71 | 0.1883 | 71.51 | 0.2666 | 77.17 | 0.2877 | 70.89 | 0.3079 | |
| Rule-Based-InstructModel=GPT-4o-mini2026.03 | 76.52 | 0.1721 | 75.64 | 0.1994 | 78.29 | 0.2042 | 80.33 | 0.304 | 69.77 | 0.3138 | |
| CoT-InstructModel=GPT-4o-mini2026.03 | 68.12 | 0.1947 | 63.86 | 0.2404 | 76.65 | 0.2275 | 80.79 | 0.3044 | 69.93 | 0.3154 | |
| Rule-Based-InstructModel=GPT-4o2026.03 | 67.92 | 0.2349 | 69.71 | 0.2563 | 64.34 | 0.3078 | 77.19 | 0.2893 | 72.14 | 0.3069 | |
| CoT-InstructModel=GPT-4o2026.03 | 65.14 | 0.1821 | 59.08 | 0.2439 | 77.25 | 0.217 | 75.7 | 0.2728 | 74.58 | 0.2968 | |
| Simple-PromptModel=Gemini-3-Pro2026.03 | 61.28 | 0.2174 | 62.3 | 0.2385 | 59.24 | 0.3494 | 70.95 | 0.2551 | 74.59 | 0.287 | |
| CoT-InstructModel=Gemini-3-Pro2026.03 | 58.28 | 0.223 | 55.2 | 0.2559 | 64.44 | 0.2959 | 72.28 | 0.2507 | 74.54 | 0.3092 | |
| Rule-Based-InstructModel=Gemini-3-Pro2026.03 | 49.31 | 0.236 | 49.25 | 0.2681 | 49.42 | 0.2998 | 72.51 | 0.2661 | 75.91 | 0.3083 | |
| VanillaModel=Qwen3-4B2026.03 | 23.4 | 0.1694 | 17.4 | 0.1615 | 35.38 | 0.2821 | 90.81 | 0.2832 | 67.42 | 0.3305 |