Question Answering on NQ (EM, F1, Trigger rate (T))
52.4EMVerbal-Calibrate
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Verbal-CalibrateScope=global2026.04 | 52.4 | 54.4 | 25 | |
| Uncertain-CalibrateScope=local2026.04 | 41.4 | 51 | 52 | |
| SR-13BBackbone=13B2026.04 | 30.8 | 36.9 | 5.4 | |
| No-Ret2026.04 | 29.2 | 42 | — | |
| ADARAGUE2026.04 | 28.6 | 40 | 21.8 | |
| Ret-All2026.04 | 25.4 | 35.8 | 100 | |
| DRAGIN2026.04 | 23.8 | 35.9 | 70.4 | |
| FLARE2026.04 | 19.8 | 31.5 | 99 | |
| Base-UncTok2026.04 | 18.8 | 28.1 | 3.7 | |
| SR-7BBackbone=7B2026.04 | 17.8 | 22 | 18 | |
| Base-Verbal2026.04 | 17.8 | 29.7 | 22.8 |