Context Tokens are Anchors: Understanding the Repetition Curse in dMLLMs from an Information Flow Perspective

Documenti analoghi

• Diffusion-CAM: Faithful Visual Explanations for dMLLMs
  di: Zuo, Haomin, et al.
  Pubblicazione: (2026)
• D$^{3}$ToM: Decider-Guided Dynamic Token Merging for Accelerating Diffusion MLLMs
  di: Chang, Shuochen, et al.
  Pubblicazione: (2025)
• MCA-LLaVA: Manhattan Causal Attention for Reducing Hallucination in Large Vision-Language Models
  di: Zhao, Qiyan, et al.
  Pubblicazione: (2025)
• Understanding In-Context Learning from Repetitions
  di: Yan, Jianhao, et al.
  Pubblicazione: (2023)
• Image Generation Based on Image Style Extraction
  di: Chang, Shuochen
  Pubblicazione: (2025)