Performance Prediction on MMLU 14k (test)
1.07MAEDISCO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DISCOSelection=High PDS, Prediction=Sig. + RF2025.10 | 1.07 | 0.987 | |
| DISCOSelection=High JSD, Prediction=Sig. + kNN2025.10 | 1.14 | 0.975 | |
| DISCOSelection=High JSD, Prediction=Sig. + RF2025.10 | 1.3 | 0.987 | |
| DISCOSelection=High PDS, Prediction=Sig. + kNN2025.10 | 1.31 | 0.972 | |
| Model signatureSelection=Random, Prediction=Sig. + RF2025.10 | 1.81 | 0.933 | |
| Model signatureSelection=Random, Prediction=Sig. + kNN2025.10 | 1.82 | 0.912 | |
| tinyBenchmarksSelection=Anchor-corr, Prediction=gp-IRT2025.10 | 2.08 | 0.927 | |
| MetabenchSelection=Best for val., Prediction=ability-IRT2025.10 | 2.08 | 0.904 | |
| tinyBenchmarksSelection=Random, Prediction=gp-IRT2025.10 | 2.79 | 0.922 | |
| tinyBenchmarksSelection=Anchor-IRT, Prediction=gp-IRT2025.10 | 3.25 | 0.922 | |
| BaselineSelection=Random, Prediction=Direct eval.2025.10 | 3.45 | 0.916 |