Natural Language Understanding on GLUE (val) - Subset (MNLI, SST-2, CoLA, QNLI, MRPC)
86.33MNLI AccuracyFull
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| FullBackbone=T5-Base2026.05 | 86.33 | 94.75 | 80.7 | 93.19 | 84.56 | 87.91 | |
| FullBackbone=T5-base, Evaluation Protocol=Finetuned2026.03 | 86.07 | 94.11 | 80.5 | 93.34 | 84.8 | — | |
| LoRA+Backbone=T5-Base2026.05 | 85.81 | 93.85 | 77.53 | 93.14 | 74.43 | 84.95 | |
| LoRA-Over-MPORBackbone=T5-Base2026.05 | 85.79 | 94.5 | 81.21 | 93.26 | 85.38 | 88.03 | |
| PiSSABackbone=T5-Base2026.05 | 85.75 | 94.07 | 74.27 | 93.15 | 76.31 | 84.71 | |
| rsLoRABackbone=T5-Base2026.05 | 85.73 | 94.19 | 72.32 | 93.12 | 52.86 | 79.64 | |
| LoRA-GABackbone=T5-Base2026.05 | 85.7 | 94.11 | 80.57 | 93.18 | 85.29 | 87.77 | |
| DoRABackbone=T5-Base2026.05 | 85.67 | 94.04 | 72.04 | 93.04 | 68.08 | 82.57 | |
| LoRA-Over-MPOPBackbone=T5-Base2026.05 | 85.59 | 94.41 | 81.02 | 93.19 | 84.07 | 87.66 | |
| LoRA-OneBackbone=T5-base, Evaluation Protocol=Finetuned2026.03 | 85.49 | 93.99 | 80.15 | 92.83 | 84.47 | — | |
| AdaLoRABackbone=T5-Base2026.05 | 85.45 | 93.69 | 69.16 | 91.66 | 68.14 | 81.62 | |
| LoRA-Over-MPOBackbone=T5-Base2026.05 | 85.42 | 94.35 | 80.79 | 93.14 | 82.84 | 87.31 | |
| LoRA-GABackbone=T5-base, Evaluation Protocol=Finetuned2026.03 | 85.36 | 94.07 | 78.74 | 92.95 | 83.98 | — | |
| CG-LoRA (shift)Backbone=T5-base, Evaluation Protocol=Finetuned2026.03 | 85.36 | 94.34 | 80.34 | 93.03 | 84.31 | — | |
| LoRA-Over-SVDBackbone=T5-Base2026.05 | 85.33 | 94.13 | 74.52 | 93.05 | 75.76 | 84.56 | |
| CG-LoRA (no-shift)Backbone=T5-base, Evaluation Protocol=Finetuned2026.03 | 85.32 | 94.38 | 79.86 | 93.14 | 84.39 | — | |
| LoRABackbone=T5-Base2026.05 | 85.3 | 94.04 | 69.35 | 92.96 | 68.38 | 82.08 | |
| rsLoRABackbone=T5-base, Evaluation Protocol=Finetuned2026.03 | 85.18 | 93.78 | 79.51 | 93.07 | 83.49 | — |