Research
Omni-Woodpecker reduces cross-modal hallucinations in omni LLMs without training
The training-free method rebalances fused hidden representations to separate grounding evidence from priors. On Qwen2.5-Omni-7B it gains 12.9 points on CMM and 8.6 on AVHBench, and 0.67 to 2.25 points across four general reasoning benchmarks.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
Omni-Woodpecker: Hallucination Correction via Evidence-Prior Rebalancing for Omni LLMs