LLM-judge evaluation on AXBENCH
92.5Concept ScoreSPLIT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SPLITModel=Gemma-2-9B-IT, Form=Local Weight2026.02 | 92.5 | 81.125 | |
| SPLITModel=Gemma-2-9B-IT, Form=Vector2026.02 | 92.5 | 82.375 | |
| SPLITModel=Qwen-2.5-7B-IT, Form=Vector2026.02 | 90.625 | 82.5 | |
| REPSModel=Gemma-2-9B-IT, Form=Local Weight2026.02 | 88.75 | 81.81 | |
| SPLITModel=Gemma-2-9B-IT, Form=LoRA2026.02 | 88.75 | 82.06 | |
| SFTModel=Gemma-2-9B-IT, Form=LoRA2026.02 | 88.125 | 75.94 | |
| REPSModel=Gemma-2-9B-IT, Form=LoRA2026.02 | 86.875 | 82.625 | |
| SPLITModel=Qwen-2.5-7B-IT, Form=LoRA2026.02 | 86.875 | 81.81 | |
| REPSModel=Qwen-2.5-7B-IT, Form=Vector2026.02 | 86.875 | 82.06 | |
| SFTModel=Gemma-2-9B-IT, Form=Vector2026.02 | 85 | 72.435 | |
| REPSModel=Gemma-2-9B-IT, Form=Vector2026.02 | 85 | 77.75 | |
| SPLITModel=Qwen-2.5-7B-IT, Form=Local Weight2026.02 | 85 | 71.625 | |
| REPSModel=Qwen-2.5-7B-IT, Form=Local Weight2026.02 | 84.375 | 70.815 | |
| SFTModel=Gemma-2-9B-IT, Form=Local Weight2026.02 | 83.125 | 72.69 | |
| SFTModel=Qwen-2.5-7B-IT, Form=Vector2026.02 | 78.75 | 79 | |
| SFTModel=Qwen-2.5-7B-IT, Form=Local Weight2026.02 | 76.875 | 56.435 | |
| SFTModel=Qwen-2.5-7B-IT, Form=LoRA2026.02 | 74.375 | 65.875 | |
| REPSModel=Qwen-2.5-7B-IT, Form=LoRA2026.02 | 74.375 | 70.065 | |
| DiffMeanModel=Gemma-2-9B-IT, Form=Vector2026.02 | 58.125 | 52.75 | |
| DiffMeanModel=Qwen-2.5-7B-IT, Form=Vector2026.02 | 47.5 | 49.75 | |
| VanillaModel=Gemma-2-9B-IT, Form=Vanilla2026.02 | 23.75 | 24.75 | |
| VanillaModel=Qwen-2.5-7B-IT, Form=Vanilla2026.02 | 22.5 | 23.565 |