Hallucination Evaluation on AMBER-G (full discriminative split)
7.2CHAIR ScoreLLaDA-V
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LLaDA-VModel=LLaDA-V, Setting=Baseline2026.05 | 7.2 | 61.6 | 41.1 | |
| LLaDA-V + OursModel=LLaDA-V, Inference-time Techniques=Mask Prior Suppression & Monotonic RoPE Scaling2026.05 | 6.9 | 61.6 | 40.6 |