Natural Language Inference on SNLI (test) (Accuracy & Efficiency)
81.06AccuracyBudgetFormer
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| BudgetFormerModel architecture=BudgetFormer, Inference selection=top-k adaptive head selection2026.04 | 81.06 | 6.05 | 0.0202 | 0.364 | |
| TransformerModel architecture=Standard Transformer encoder2026.04 | 78.35 | 6.18 | 0.0206 | — |