Factual Knowledge Evaluation on HHEM, PopQA, TriviaQA Suite
96.22HHEM AccuracySEAL
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SEALModel=Qwen3, Size=8B2026.04 | 96.22 | 58.54 | 70.35 | 75.04 | |
| PRISMModel=Qwen3, Size=8B2026.04 | 95.13 | 60.83 | 71.22 | 75.73 | |
| PRISMModel=Qwen3, Size=4B2026.04 | 95.03 | 59.54 | 68.9 | 74.49 | |
| NOVAModel=Qwen3, Size=4B2026.04 | 94.83 | 58.89 | 69.34 | 74.35 | |
| SFTModel=Qwen3, Size=8B2026.04 | 94.83 | 54.9 | 69.22 | 72.98 | |
| NOVAModel=Qwen3, Size=8B2026.04 | 94.83 | 59.11 | 70.72 | 74.89 | |
| SFTModel=Qwen3, Size=4B2026.04 | 94.53 | 58.54 | 68.95 | 74.01 | |
| SEALModel=Qwen3, Size=4B2026.04 | 92.25 | 57.76 | 68.87 | 72.96 | |
| PRISMModel=Llama 3.1, Size=8B2026.04 | 89.76 | 59.75 | 69.23 | 72.91 | |
| SEALModel=Llama 3.1, Size=8B2026.04 | 88.57 | 57.76 | 67.94 | 71.42 | |
| SFTModel=Llama 3.1, Size=8B2026.04 | 88.12 | 58.18 | 68.1 | 71.47 | |
| NOVAModel=Llama 3.1, Size=8B2026.04 | 84.1 | 50.89 | 66.47 | 67.15 |