Natural Language Understanding on GLUE (SST2, MRPC, QQP, RTE, WNLI, QNLI Subset)
57.11SST-2 AccuracySHE-LoRA
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SHE-LoRADirichlet distribution parameter (rho)=0.32025.05 | 57.11 | 70.88 | 72.52 | 50.18 | 57.75 | 59.63 | |
| HeterLoRADirichlet distribution parameter (rho)=0.32025.05 | 55.73 | 68.38 | 72.17 | 44.72 | 48.86 | 49.14 | |
| Flex-LoRADirichlet distribution parameter (rho)=0.32025.05 | 52.29 | 74.81 | 75.31 | 46.93 | 49.66 | 49.51 | |
| FedSADirichlet distribution parameter (rho)=0.32025.05 | 48.23 | 33.71 | 66.32 | 43.56 | 48.27 | 48.26 | |
| FedITDirichlet distribution parameter (rho)=0.32025.05 | 47.41 | 31.62 | 64.71 | 43.07 | 46.34 | 48.87 |