Natural Language Understanding on GLUE (CoLA, MRPC, QNLI, RTE, SST-2, STS-B Subset)
100Average GLUE ScoreIndividual
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| IndividualBackbone=RoBERTa-Base2026.05 | 100 | — | — | — | — | — | — | |
| IndividualBackbone=RoBERTa-Large2026.05 | 100 | — | — | — | — | — | — | |
| SA-MergingBackbone=RoBERTa-Large2026.05 | 90.2 | — | — | — | — | — | — | |
| WUDI-MergingBackbone=RoBERTa-Large2026.05 | 88.8 | — | — | — | — | — | — | |
| PERAModel=RoBERTa-large, #Param=0.8M2026.04 | 88.13 | 96.24 | 90.78 | 68.4 | 94.89 | 86.47 | 91.98 | |
| FFTModel=RoBERTa-large, #Param=355M2026.04 | 88 | 96 | 91.7 | 68.2 | 93.8 | 85.8 | 92.6 | |
| REDModel=RoBERTa-large, #Param=0.05M2026.04 | 87.6 | 96 | 90.3 | 68.1 | 93.5 | 86.2 | 91.3 | |
| LoRAModel=RoBERTa-large, #Param=0.8M2026.04 | 87.3 | 96 | 89.8 | 65.5 | 94.7 | 86.3 | 91.7 | |
| SA-MergingBackbone=RoBERTa-Base2026.05 | 87.1 | — | — | — | — | — | — | |
| WUDI-MergingBackbone=RoBERTa-Base2026.05 | 85.3 | — | — | — | — | — | — | |
| PERAModel=RoBERTa-base, #Param=0.3M2026.04 | 85.1 | 94.13 | 89.71 | 64.8 | 92.94 | 77.84 | 91.13 | |
| DeLoRAModel=RoBERTa-base, #Param=0.3M2026.04 | 84.6 | 94.1 | 89 | 63.6 | 92.8 | 77.1 | 90.9 | |
| FFTModel=RoBERTa-base, #Param=125M2026.04 | 84.4 | 94.4 | 87.9 | 62.4 | 92.5 | 78.3 | 90.6 | |
| REDModel=RoBERTa-base, #Param=0.02M2026.04 | 83.9 | 93.9 | 89.2 | 61 | 90.7 | 78 | 90.4 | |
| HiRAModel=RoBERTa-base, #Param=0.3M2026.04 | 83.45 | 94.22 | 89.53 | 60.3 | 92.39 | 74.68 | 89.58 | |
| LoRAModel=RoBERTa-base, #Param=0.3M2026.04 | 83.4 | 93.9 | 88.7 | 59.7 | 92.6 | 75.3 | 90.3 | |
| PCB MergingBackbone=RoBERTa-Large2026.05 | 79 | — | — | — | — | — | — | |
| PCB MergingBackbone=RoBERTa-Base2026.05 | 76.5 | — | — | — | — | — | — | |
| TIES + DAREBackbone=RoBERTa-Large2026.05 | 72.8 | — | — | — | — | — | — | |
| TIES-MergingBackbone=RoBERTa-Large2026.05 | 72.4 | — | — | — | — | — | — | |
| Task ArithmeticBackbone=RoBERTa-Large2026.05 | 70.9 | — | — | — | — | — | — | |
| TA + DAREBackbone=RoBERTa-Large2026.05 | 70.9 | — | — | — | — | — | — | |
| Task ArithmeticBackbone=RoBERTa-Base2026.05 | 67.8 | — | — | — | — | — | — | |
| TIES + DAREBackbone=RoBERTa-Base2026.05 | 65.6 | — | — | — | — | — | — | |
| TIES-MergingBackbone=RoBERTa-Base2026.05 | 64.7 | — | — | — | — | — | — | |
| TA + DAREBackbone=RoBERTa-Base2026.05 | 63.7 | — | — | — | — | — | — | |
| Weight Avg.Backbone=RoBERTa-Large2026.05 | 53.3 | — | — | — | — | — | — | |
| Weight Avg.Backbone=RoBERTa-Base2026.05 | 52.6 | — | — | — | — | — | — | |
| PretrainedBackbone=RoBERTa-Base2026.05 | 41.7 | — | — | — | — | — | — | |
| PretrainedBackbone=RoBERTa-Large2026.05 | 38.2 | — | — | — | — | — | — |