Honesty-Helpfulness Alignment Evaluation on HoneSet (full)
92.1Win RateAlignX
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AlignXBackbone=DeepSeek-7B, Evaluation Protocol=Fine-Tuning + MoCaE2026.02 | 92.1 | 27.95 | 93.25 | 52.47 | |
| AlignXBackbone=Mistral-7B, Evaluation Protocol=Fine-Tuning + MoCaE2026.02 | 86.45 | 29.3 | 89.71 | 48.62 | |
| AlignXBackbone=LLaMA-2-7B, Evaluation Protocol=Fine-Tuning + MoCaE2026.02 | 84.72 | 30.89 | 87.63 | 47.15 | |
| AlignXBackbone=Gemma-7B, Evaluation Protocol=Fine-Tuning + MoCaE2026.02 | 82.38 | 31.95 | 85.44 | 45.62 |