Natural Language Understanding on GLUE (MNLI, QNLI, QQP, SST2 Subset)
85.59MNLI AccuracyGLoRA
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GLoRADistribution=Heavy-tail2026.05 | 85.59 | 90.89 | 94.5 | 84.7 | 88.3 | 85.84 | |
| GLoRADistribution=Normal2026.05 | 85.57 | 91.4 | 94.04 | 84.45 | 88.34 | 86.25 | |
| HetLoRADistribution=Heavy-tail2026.05 | 85.37 | 90.81 | 94.18 | 83.96 | 87.92 | 85.29 | |
| HetLoRADistribution=Normal2026.05 | 83.92 | 91.07 | 95.07 | 83.52 | 87.56 | 84.22 | |
| GLoRADistribution=Uniform2026.05 | 83.02 | 90.12 | 95.07 | 83.39 | 87.05 | 83.66 | |
| FedTreeLoRAnumber of clients=16, clients per task=42026.02 | 82.94 | 89.31 | 94.19 | 84.75 | 87.8 | — | |
| HetLoRADistribution=Uniform2026.05 | 82.74 | 89.57 | 94.61 | 82.99 | 86.57 | 82.96 | |
| FlexLoRADistribution=Heavy-tail2026.05 | 82.71 | 88.54 | 90.09 | 83.17 | 85.45 | 82.72 | |
| FlexLoRADistribution=Uniform2026.05 | 81.89 | 89.54 | 93.67 | 78.57 | 85.28 | 82.74 | |
| FlexLoRADistribution=Normal2026.05 | 81.63 | 90.54 | 93.32 | 80.71 | 85.79 | 82.75 | |
| FedLEASEnumber of clients=16, clients per task=42026.02 | 79.91 | 87.54 | 94.17 | 83.46 | 86.27 | — | |
| FedDPAnumber of clients=16, clients per task=42026.02 | 75 | 86.29 | 93.58 | 81.92 | 84.2 | — | |
| FedSAnumber of clients=16, clients per task=42026.02 | 74.29 | 86.33 | 93.87 | 82.12 | 84.15 | — | |
| FedALTnumber of clients=16, clients per task=42026.02 | 73.25 | 83 | 93.69 | 84.31 | 83.56 | — | |
| FFA-LoRAnumber of clients=16, clients per task=42026.02 | 72.17 | 82.75 | 93.79 | 81.25 | 82.49 | — | |
| FedITnumber of clients=16, clients per task=42026.02 | 71.46 | 83.71 | 93.25 | 78.54 | 81.74 | — |