Vocabulary Difficulty Prediction on KVL BEA Shared Task Open Track 2026 (test)
0.321Chinese RMSEStatistical Optimum
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Statistical OptimumType=Simulation2026.05 | 0.321 | 0.304 | 0.205 | 0.277 | |
| finetuned_llms_plusType=Submission2026.05 | 0.63 | 0.726 | 0.742 | 0.699 | |
| open_maxType=Submission2026.05 | 0.631 | 0.723 | 0.743 | 0.699 | |
| finetuned_llmsType=Submission2026.05 | 0.64 | 0.731 | 0.76 | 0.71 | |
| ≤32B LLM AverageType=Average Ensemble, Size constraint=≤32B2026.05 | 0.645 | 0.743 | 0.767 | 0.719 | |
| ≤14B LLM AverageType=Average Ensemble, Size constraint=≤14B2026.05 | 0.662 | 0.77 | 0.804 | 0.745 | |
| GLM-4-32BModel Family=GLM-4, Size=32B2026.05 | 0.678 | 0.769 | 0.805 | 0.751 | |
| Qwen2.5-32BModel Family=Qwen2.5, Size=32B2026.05 | 0.678 | 0.777 | 0.799 | 0.752 | |
| Ministral-3-14BModel Family=Ministral-3, Size=14B2026.05 | 0.681 | 0.781 | 0.799 | 0.753 | |
| ≤9B LLM AverageType=Average Ensemble, Size constraint=≤9B2026.05 | 0.683 | 0.791 | 0.835 | 0.769 | |
| Open-Track BaselineBackbone=fine-tuned XLM-RoBERTa2026.05 | 1.034 | 1.166 | 1.198 | 1.133 |