Knowledge gap detection on MATH
89.1Accuracy (Knowledge Gap)GRADEpos
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRADEposBackbone=Llama3.1-8b2026.04 | 89.1 | 96.2 | |
| GRADEposBackbone=Qwen2.5-7b2026.04 | 87.9 | 92.1 | |
| GRADEposBackbone=Gemma2-9b2026.04 | 87.1 | 93.4 | |
| GRADEpreBackbone=Qwen2.5-7b2026.04 | 80.4 | 87.9 | |
| GRADEpreBackbone=Gemma2-9b2026.04 | 79.4 | 81.3 | |
| GRADEpreBackbone=Llama3.1-8b2026.04 | 77 | 84.1 | |
| Align-PBackbone=Llama3.1-8b2026.04 | 75.3 | 60.7 | |
| Align-PBackbone=Qwen2.5-7b2026.04 | 74.7 | 79.7 | |
| ICBackbone=Gemma2-9b2026.04 | 71.5 | 75.7 | |
| P-EntropyBackbone=Qwen2.5-7b2026.04 | 70.2 | 54.2 | |
| P-EntropyBackbone=Gemma2-9b2026.04 | 69.1 | 75.1 | |
| ICBackbone=Qwen2.5-7b2026.04 | 67.6 | 72.2 | |
| JudgeBackbone=Qwen2.5-7b2026.04 | 66.2 | — | |
| P-EntropyBackbone=Llama3.1-8b2026.04 | 66.1 | 70.4 | |
| Align-PBackbone=Gemma2-9b2026.04 | 61.3 | 63.6 | |
| ICBackbone=Llama3.1-8b2026.04 | 57.5 | 62.7 | |
| JudgeBackbone=Llama3.1-8b2026.04 | 50.1 | — | |
| JudgeBackbone=Gemma2-9b2026.04 | 40 | — |