Question Answering on QASC and WikiQA
82.4AccuracyT5-Base Experts
Evaluation Results
| Method | Links | |
|---|---|---|
| T5-Base ExpertsExpert training steps=10242025.06 | 82.4 | |
| T5-Base ExpertsSelection strategy=Early stop v0, Steps=485±3502025.06 | 82.3 | |
| T5-Base ExpertsExpert training steps=20482025.06 | 82 | |
| T5-Base ExpertsSelection strategy=Early stop v1, Steps=269±1992025.06 | 81.3 | |
| TIES MergingSelection strategy=Early stop v1, Steps=269±1992025.06 | 77 | |
| TIES MergingSelection strategy=Early stop v0, Steps=485±3502025.06 | 76.9 | |
| TIES MergingSelection strategy=Best (# steps)2025.06 | 75.7 | |
| Task ArithmeticSelection strategy=Early stop v0, Steps=485±3502025.06 | 75 | |
| Task ArithmeticSelection strategy=Early stop v1, Steps=269±1992025.06 | 74 | |
| Task ArithmeticSelection strategy=Best (# steps)2025.06 | 73.7 | |
| TIES MergingExpert training steps=10242025.06 | 72.2 | |
| TIES MergingExpert training steps=20482025.06 | 72.2 | |
| Task ArithmeticExpert training steps=10242025.06 | 70.9 | |
| Task ArithmeticExpert training steps=20482025.06 | 70.1 | |
| Average MergingExpert training steps=10242025.06 | 65.2 | |
| Average MergingSelection strategy=Best (# steps)2025.06 | 65.2 | |
| Average MergingExpert training steps=20482025.06 | 65 | |
| Average MergingSelection strategy=Early stop v0, Steps=485±3502025.06 | 63.4 | |
| Average MergingSelection strategy=Early stop v1, Steps=269±1992025.06 | 62.8 |