Study Proposes Step-Aware Decoding to Reduce Hallucinations in Multimodal Diffusion Language Models
A new arXiv paper introduces step-aware decoding to mitigate hallucinations in multimodal diffusion language models, which use iterative refinement for strong global modeling.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-11 05:00 · arXiv cs.CL
Rethinking in Spikes: Mitigating Hallucinations in MDLMs with Step-Aware Decoding