Question Answering on ARC Challenge (PR-AUC)
0.636PR-AUCEccentricity
Evaluation Results
| Method | Links | |
|---|---|---|
| EccentricityStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 0.636 | |
| EigVecDissimilarityStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 0.621 | |
| CocoaPPLStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 0.61 | |
| CocoaMSPStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 0.606 | |
| DissimilarityStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 0.604 | |
| CocoaMSPStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 0.595 | |
| CocoaPPLStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 0.595 | |
| EigVecDissimilarityStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 0.562 | |
| EccentricityStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 0.549 | |
| DissimilarityStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 0.545 | |
| Degree MatrixStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 0.473 | |
| Semantic EntropyStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 0.454 | |
| Semantic EntropyStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 0.431 | |
| Degree MatrixStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 0.419 | |
| SARStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 0.269 | |
| SARStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 0.264 |