Question Answering on QA domain average
85.2Best AccuracyTask-level fine-tuning
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Task-level fine-tuning2025.06 | 85.2 | — | |
| Many-shot (ours)Masking Strategy=mask all targets2025.06 | 84.5 | 84.5 | |
| Many-shot (ours)Masking Strategy=mask last target2025.06 | 84 | 84 | |
| Few-shotMasking Strategy=mask all targets2025.06 | 82.2 | 82.2 | |
| Zero shot2025.06 | 81.8 | 81.3 | |
| Few-shotMasking Strategy=mask last target2025.06 | 81.8 | 81.8 | |
| Many-shotStrategy=autoregressive2025.06 | 80.8 | 80.8 | |
| No fine-tuning2025.06 | 80.1 | 79.8 |