Natural Language Understanding on GLUE (7-Task Performance)
76.98CoLADTS-T
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| DTS-Tbackbone=GPT-22025.12 | 76.98 | 81.72 | 79.94 | 87.99 | 89.2 | 65.34 | 91.16 | 81.76 | 0.22 | 5.24 | |
| Individualbackbone=GPT-22025.12 | 76.8 | 81.99 | 80.39 | 88.27 | 89.64 | 65.34 | 91.17 | 81.94 | — | — | |
| DTS-Dbackbone=GPT-22025.12 | 76.69 | 81.58 | 80.63 | 88.24 | 89.46 | 64.98 | 91.05 | 81.8 | 0.17 | 5.24 | |
| DTS-T*backbone=GPT-2, low-memory-config=true2025.12 | 76.6 | 81.86 | 79.94 | 87.95 | 88.18 | 64.62 | 90.13 | 81.33 | 0.75 | 0.93 | |
| DTS-D*backbone=GPT-2, low-memory-config=true2025.12 | 76.4 | 81.68 | 80.41 | 88.44 | 88.81 | 66.06 | 90.94 | 81.39 | 0.67 | 0.93 | |
| T-Switchbackbone=GPT-22025.12 | 76.27 | 81.87 | 79.9 | 88.57 | 88.54 | 63.17 | 90.82 | 81.31 | 0.78 | 6.23 | |
| Twin-Mergingbackbone=GPT-22025.12 | 76.02 | 78.75 | 78.93 | 87.17 | 87.58 | 62.82 | 90.16 | 80.2 | 2.12 | 100 | |
| TALL-Maskbackbone=GPT-22025.12 | 74.78 | 78.59 | 78.43 | 88.15 | 85.61 | 67.5 | 90.82 | 80.55 | 1.82 | 17.41 | |
| SVDbackbone=GPT-22025.12 | 74.49 | 80.91 | 79.41 | 88.12 | 88.21 | 64.14 | 91.16 | 80.92 | 1.24 | 3.8 | |
| EMR-Mergingbackbone=GPT-22025.12 | 72.77 | 81.08 | 79.16 | 84.84 | 88.11 | 66.43 | 90.25 | 80.38 | 1.91 | 17.41 | |
| Task-Arithmeticbackbone=GPT-22025.12 | 68.71 | 68.64 | 69.56 | 70.42 | 81.83 | 47.21 | 83.62 | 70 | 14.58 | 0 | |
| Ties-Mergingbackbone=GPT-22025.12 | 68.48 | 71.46 | 68.49 | 69.56 | 82.57 | 47.68 | 81.82 | 70.01 | 14.56 | 0 | |
| DAREbackbone=GPT-22025.12 | 67.59 | 65.41 | 72.54 | 62.07 | 79.74 | 44.76 | 72.46 | 66.37 | 19.01 | 0 | |
| RegMeanbackbone=GPT-22025.12 | 61.69 | 70.44 | 65.37 | 69.71 | 78.83 | 56.1 | 79.74 | 68.84 | 15.99 | 0 | |
| Weight-Averagingbackbone=GPT-22025.12 | 55.03 | 55.16 | 51.21 | 57.65 | 76.71 | 44.76 | 52.53 | 56.15 | 31.48 | 0 | |
| Fisher-Mergingbackbone=GPT-22025.12 | 54.81 | 58.12 | 39.53 | 63.28 | 81.46 | 49.12 | 64.74 | 58.72 | 28.34 | 0 |