Multi-hop Fact Verification on HoVer few-shot
56RecallLLaDA (Public)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LLaDA (Public)Model configuration=Public checkpoint, Evaluation protocol=Few-shot, Training=No additional training2026.04 | 56 | 26.8 | 43.5 | |
| LLaDA (FS)Model configuration=Full-sequence masking (FS), Evaluation protocol=Few-shot, SFT=Essential for unlocking prompt infilling2026.04 | 55.6 | 27.9 | 55 | |
| LLaDA (FS+RO)Model configuration=FS + response-only refinement (RO), Evaluation protocol=Few-shot2026.04 | 54.7 | 21.1 | 60.4 | |
| LLaDA (RO)Model configuration=Response-only refinement (RO), Evaluation protocol=Few-shot2026.04 | 53.3 | 16.3 | 50.3 |