Natural Language Inference on MNLI (Accuracy, ECE, Brier)
89.5AccuracyDeep-Ens-LoRA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Deep-Ens-LoRABase=LLaMA-22026.02 | 89.5 | 0.03 | 0.12 | |
| SBABase=LLaMA-22026.02 | 89.4 | 0.027 | 0.118 | |
| LoRABase=LLaMA-22026.02 | 89.2 | 0.061 | 0.152 | |
| Laplace-LoRABase=LLaMA-22026.02 | 89.1 | 0.039 | 0.131 | |
| Gauss+ProjBase=LLaMA-22026.02 | 89 | 0.036 | 0.128 | |
| Deep-Ens-LoRABase=RoBERTa2026.02 | 88.1 | 0.029 | 0.125 | |
| SBABase=RoBERTa2026.02 | 87.9 | 0.025 | 0.121 | |
| LoRABase=RoBERTa2026.02 | 87.8 | 0.058 | 0.168 | |
| Laplace-LoRABase=RoBERTa2026.02 | 87.6 | 0.037 | 0.139 | |
| Gauss+ProjBase=RoBERTa2026.02 | 87.5 | 0.034 | 0.135 |