LLM Alignment on Base Model Evaluation Set
79.93Win RateAlignX
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AlignXBackbone=DeepSeek-7B2026.02 | 79.93 | 35.88 | 74.91 | 39.65 | |
| AlignXBackbone=Mistral-7B2026.02 | 78.65 | 36.95 | 72.45 | 38.72 | |
| AlignXBackbone=Gemma-7B2026.02 | 75.8 | 38.1 | 69.85 | 35.85 | |
| AlignXBackbone=LLaMA-2-7B2026.02 | 37.45 | 40.2 | 39.6 | 12.28 | |
| TrinityXBackbone=LLaMA-2-7B2026.02 | 36.75 | 41.03 | 40.66 | 12.12 | |
| H3Fusion2026.02 | 13.79 | 42 | 18.82 | -3.13 |