Natural Language Inference on MNLI (Accuracy, Average Accuracy)
91.7AccuracyFinetuned
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FinetunedBackbone=DeBERTa-V2-XXL, Number of Parameters=1.5B2025.12 | 91.7 | 94.4 | |
| Task ArithmeticBackbone=DeBERTa-V2-XXL, Number of Parameters=1.5B, lambda=12025.12 | 91.2 | 78.6 | |
| SAMergingBackbone=DeBERTa-V2-XXL, Number of Parameters=1.5B, k=16002025.12 | 90.5 | 93.5 | |
| TIES-MergingBackbone=DeBERTa-V2-XXL, Number of Parameters=1.5B2025.12 | 88.7 | 92.3 | |
| Task ArithmeticBackbone=DeBERTa-V2-XXL, Number of Parameters=1.5B, lambda=0.52025.12 | 72.5 | 63.5 | |
| AdaMergingBackbone=DeBERTa-V2-XXL, Number of Parameters=1.5B, k=16002025.12 | 72.5 | 63.5 | |
| Pre-trainedBackbone=DeBERTa-V2-XXL, Number of Parameters=1.5B2025.12 | 33 | 41.5 |