Natural Language Processing on BERT NLP Task Suite (ANLI, Rotten Tomatoes, CoLA, SMS) (test)
51.5ANLI AccuracyTA
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TAMerging method=Task Arithmetic2025.12 | 51.5 | 51.9 | 68 | 14.5 | 46.5 | |
| Iso-C w/ DSCMerging method=Iso-C, Calibration=DSC2025.12 | 50.4 | 59.6 | 59.6 | 96.4 | 66.5 | |
| Iso-CTS w/ DSCMerging method=Iso-CTS, Calibration=DSC2025.12 | 49.1 | 55.4 | 62.5 | 88.2 | 63.8 | |
| TSV-MMerging method=TSV-M2025.12 | 48.6 | 61.6 | 72.7 | 88.4 | 67.8 | |
| TIES w/ DSCMerging method=Ties-Merging, Calibration=DSC2025.12 | 47.6 | 75 | 53.5 | 92.7 | 67.2 | |
| DAREMerging method=DARE2025.12 | 47.5 | 55.4 | 69.2 | 44.1 | 54 | |
| DARE w/ DSCMerging method=DARE, Calibration=DSC2025.12 | 47.3 | 70.2 | 53.3 | 97.8 | 67.2 | |
| TA w/ DSCMerging method=Task Arithmetic, Calibration=DSC2025.12 | 46.9 | 67.3 | 59.8 | 97.7 | 67.9 | |
| TSV-M w/ DSCMerging method=TSV-M, Calibration=DSC2025.12 | 46.7 | 65.4 | 67.5 | 97.1 | 69.2 | |
| TIESMerging method=Ties-Merging2025.12 | 46.5 | 74.6 | 55.4 | 87.6 | 66 | |
| Iso-CTSMerging method=Iso-CTS2025.12 | 44.6 | 49 | 65.1 | 46.6 | 51.3 | |
| Iso-CMerging method=Iso-C2025.12 | 43.5 | 48 | 62.9 | 54.7 | 52.3 |