Fact Verification on HOVER (test)
56.6AUROCDiffuTruth
Evaluation Results
| Method | Links | |
|---|---|---|
| DiffuTruth2026.02 | 56.6 | |
| Hybrid2026.02 | 56.6 | |
| Direct NLI2026.02 | 52.5 | |
| GEPABackbone=Qwen3 8B, Optimization Budget (# rollouts)=70512025.07 | 52.33 | |
| GEPA+MergeBackbone=Qwen3 8B, Optimization Budget (# rollouts)=70512025.07 | 51.67 | |
| MIPROv2Backbone=Qwen3 8B2025.07 | 47.33 | |
| GRPOBackbone=Qwen3 8B, Optimization Budget (# rollouts)=240002025.07 | 38.67 | |
| BaselineBackbone=Qwen3 8B2025.07 | 35.33 |