Question Answering on CoQA (PR-AUC)
60PR-AUCDissimilarity
Evaluation Results
| Method | Links | |
|---|---|---|
| DissimilarityStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 60 | |
| DissimilarityStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 57.8 | |
| CocoaPPLStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 57.7 | |
| CocoaMSPStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 54.8 | |
| CocoaPPLStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 54.8 | |
| CocoaMSPStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 50.9 | |
| EccentricityStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 50.5 | |
| EigVecDissimilarityStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 49.7 | |
| EccentricityStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 45.9 | |
| EigVecDissimilarityStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 44.3 | |
| Degree MatrixStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 38.5 | |
| Semantic EntropyStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 36.5 | |
| SARStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 34.7 | |
| SARStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 32.9 | |
| Degree MatrixStrategy=beam-search, Backbone=Gemma 3 4B base2025.12 | 32.8 | |
| Semantic EntropyStrategy=multinomial-sampling, Backbone=Gemma 3 4B base2025.12 | 30.1 |