Mitigating Hallucinations in Large Vision-Language Models via DPO: On-Policy Data Hold the Key Paper • 2501.09695 • Published Jan 16 • 1