Knowledge gap detection on GSM8K
83.6AccuracyGRADEpos
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRADEposBackbone=Llama3.1-8b2026.04 | 83.6 | 89.3 | |
| GRADEposBackbone=Gemma2-9b2026.04 | 77.6 | 82.1 | |
| GRADEpreBackbone=Gemma2-9b2026.04 | 74.5 | 78.3 | |
| P-EntropyBackbone=Gemma2-9b2026.04 | 73.9 | 79.4 | |
| Align-PBackbone=Gemma2-9b2026.04 | 72.7 | 77.6 | |
| GRADEposBackbone=Qwen2.5-7b2026.04 | 71.9 | 77.2 | |
| GRADEpreBackbone=Llama3.1-8b2026.04 | 71.4 | 76.7 | |
| GRADEpreBackbone=Qwen2.5-7b2026.04 | 69.9 | 82.8 | |
| Align-PBackbone=Llama3.1-8b2026.04 | 69.2 | 76.6 | |
| Align-PBackbone=Qwen2.5-7b2026.04 | 64.5 | 66.9 | |
| P-EntropyBackbone=Llama3.1-8b2026.04 | 64.3 | 49.6 | |
| JudgeBackbone=Llama3.1-8b2026.04 | 63.9 | — | |
| P-EntropyBackbone=Qwen2.5-7b2026.04 | 62.7 | 54.1 | |
| ICBackbone=Qwen2.5-7b2026.04 | 60.3 | 63.7 | |
| ICBackbone=Llama3.1-8b2026.04 | 51.8 | 63.9 | |
| JudgeBackbone=Gemma2-9b2026.04 | 51.3 | — | |
| ICBackbone=Gemma2-9b2026.04 | 50.7 | 54.1 | |
| JudgeBackbone=Qwen2.5-7b2026.04 | 47.1 | — |