Knowledge gap detection on HQA
81.5AccuracyGRADEpre
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRADEpreBackbone=Gemma2-9b2026.04 | 81.5 | 0.868 | |
| GRADEposBackbone=Gemma2-9b2026.04 | 80.5 | 0.882 | |
| GRADEposBackbone=Qwen2.5-7b2026.04 | 78.6 | 0.866 | |
| GRADEposBackbone=Llama3.1-8b2026.04 | 77.8 | 0.857 | |
| Align-PBackbone=Gemma2-9b2026.04 | 76.4 | 0.855 | |
| GRADEpreBackbone=Llama3.1-8b2026.04 | 75.4 | 0.81 | |
| Align-PBackbone=Qwen2.5-7b2026.04 | 75.2 | 0.745 | |
| ICBackbone=Llama3.1-8b2026.04 | 74.7 | 0.651 | |
| Align-PBackbone=Llama3.1-8b2026.04 | 73.7 | 0.806 | |
| GRADEpreBackbone=Qwen2.5-7b2026.04 | 71.5 | 0.778 | |
| ICBackbone=Qwen2.5-7b2026.04 | 61.2 | 0.602 | |
| P-EntropyBackbone=Llama3.1-8b2026.04 | 59.6 | 0.607 | |
| JudgeBackbone=Gemma2-9b2026.04 | 57.3 | — | |
| ICBackbone=Gemma2-9b2026.04 | 56.2 | 0.655 | |
| P-EntropyBackbone=Gemma2-9b2026.04 | 54.7 | 0.538 | |
| P-EntropyBackbone=Qwen2.5-7b2026.04 | 54.3 | 0.541 | |
| JudgeBackbone=Llama3.1-8b2026.04 | 53.6 | — | |
| JudgeBackbone=Qwen2.5-7b2026.04 | 53.6 | — |