IRT Parameter Prediction on ELA (5-fold)
0.503Pearson Corr (b, 1PL)Qwen-14B
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Qwen-14BModel size=14B2026.01 | 0.503 | 0.721 | 0.446 | 0.169 | 0.485 | 0.936 | |
| Qwen-1.7BModel size=1.7B2026.01 | 0.409 | 0.766 | 0.238 | 0.191 | 0.41 | 1.041 | |
| Qwen-8BModel size=8B2026.01 | 0.408 | 0.75 | 0.336 | 0.191 | 0.429 | 1.014 | |
| Qwen-4BModel size=4B2026.01 | 0.404 | 0.764 | 0.332 | 0.186 | 0.391 | 1.008 | |
| ModernBERTMode=Fine-tuned baseline2026.01 | 0.239 | 0.868 | 0.061 | 0.231 | 0.132 | 1.251 | |
| FeaturesType=Feature-based model2026.01 | 0.16 | 0.827 | 0.194 | 0.2 | 0.04 | 1.098 |