Efficiency Evaluation of Factual Representation Extraction on Corpus of 11,427 facts
0.305Total Processing Time (s)CLARE
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| CLAREBackbone=GPT2-XL2026.03 | 0.305 | 2.61 | 6,400 | 973,000 | 7.016 | 2.596 | |
| CLAREBackbone=Qwen2.52026.03 | 0.305 | 3 | 14,340 | 2,130,000 | 29.981 | 2.919 | |
| CLAREBackbone=Llama32026.03 | 0.305 | 3.007 | 16,380 | 1,960,000 | 31.539 | 2.91 | |
| CLAREBackbone=GPT-J2026.03 | 0.305 | 2.502 | 16,380 | 1,480,000 | 23.905 | 2.864 | |
| CLAREBackbone=Mistral2026.03 | 0.306 | 2.722 | 16,380 | 1,770,000 | 28.478 | 2.861 | |
| CLAREBackbone=Qwen22026.03 | 0.307 | 2.827 | 14,340 | 2,130,000 | 29.981 | 2.919 | |
| CLAREBackbone=Llama3.22026.03 | 0.307 | 2.521 | 12,290 | 1,050,000 | 12.995 | 2.881 | |
| GradSimBackbone=GPT-J2026.03 | 0.763 | — | 24,200,000,000 | — | 68.474 | — | |
| GradSimBackbone=Llama3.22026.03 | 0.774 | — | 12,900,000,000 | — | 37.445 | — | |
| GradSimBackbone=GPT2-XL2026.03 | 0.796 | — | 6,230,000,000 | — | 18.212 | — | |
| GradSimBackbone=Mistral2026.03 | 0.833 | — | 29,000,000,000 | — | 81.483 | — | |
| GradSimBackbone=Qwen22026.03 | 0.868 | — | 30,500,000,000 | — | 87.513 | — | |
| GradSimBackbone=Qwen2.52026.03 | 0.915 | — | 30,500,000,000 | — | 87.513 | — | |
| GradSimBackbone=Llama32026.03 | 0.917 | — | 32,100,000,000 | — | 91.765 | — |