Vision-Language Hallucination Assessment on MM-HallBench
3.64Average ScoreGreedy
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GreedyBackbone=Qwen2.5-VL (3B)2026.04 | 3.64 | 34 | |
| Top-p (nucleus)Backbone=Qwen2.5-VL (3B)2026.04 | 3.58 | 36 | |
| η-SamplingBackbone=Qwen2.5-VL (3B)2026.04 | 3.57 | 38 | |
| Beam SearchBackbone=Qwen2.5-VL (3B)2026.04 | 3.54 | 34 | |
| Temp’ OnlyBackbone=Qwen2.5-VL (3B)2026.04 | 3.48 | 36 | |
| Top-kBackbone=Qwen2.5-VL (3B)2026.04 | 3.38 | 42 | |
| ε-SamplingBackbone=Qwen2.5-VL (3B)2026.04 | 3.36 | 35 | |
| Min-pBackbone=Qwen2.5-VL (3B)2026.04 | 3.32 | 43 |