Quality Estimation on WMT en-de 22
0.722Pearson RCOMET Kiwi
Evaluation Results
| Method | Links | |
|---|---|---|
| COMET KiwiBackbone=TransformerBase, Evaluation Setting=Eval Self2026.01 | 0.722 | |
| COMET KiwiBackbone=DeltaLM, Evaluation Setting=Eval Self2026.01 | 0.634 | |
| Sigmoid HeadBackbone=TransformerBase, Evaluation Setting=Eval Self2026.01 | 0.586 | |
| Sigmoid HeadBackbone=DeltaLM, Evaluation Setting=Eval Self2026.01 | 0.515 | |
| Boosted ProbBackbone=TransformerBase, Evaluation Setting=Eval Self2026.01 | 0.367 | |
| COMET KiwiEvaluation Mode=Eval Others2026.01 | 0.365 | |
| Boosted ProbBackbone=DeltaLM, Evaluation Setting=Eval Self2026.01 | 0.291 | |
| DeltaLMQE Head=Sigmoid Head, Evaluation Mode=Eval Others2026.01 | 0.201 | |
| Softmax HeadBackbone=TransformerBase, Evaluation Setting=Eval Self2026.01 | 0.199 | |
| TransformerBaseQE Head=Sigmoid Head, Evaluation Mode=Eval Others2026.01 | 0.166 | |
| Softmax HeadBackbone=DeltaLM, Evaluation Setting=Eval Self2026.01 | 0.165 | |
| DeltaLMQE Head=Boosted Prob, Evaluation Mode=Eval Others2026.01 | 0.141 | |
| TransformerBaseQE Head=Boosted Prob, Evaluation Mode=Eval Others2026.01 | 0.131 | |
| DeltaLMQE Head=Softmax Head, Evaluation Mode=Eval Others2026.01 | 0.081 | |
| TransformerBaseQE Head=Softmax Head, Evaluation Mode=Eval Others2026.01 | 0.076 |