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

• URL: http://arxiv.org/abs/2601.20520v1
• Date: Wed, 28 Jan 2026 11:54:42 GMT
• Title: Context Tokens are Anchors: Understanding the Repetition Curse in dMLLMs from an Information Flow Perspective
• Authors: Qiyan Zhao, Xiaofeng Zhang, Shuochen Chang, Qianyu Chen, Xiaosong Yuan, Xuhang Chen, Luoqi Liu, Jiajun Zhang, Xu-Yao Zhang, Da-Han Wang,
• Abstract summary: cache mechanisms often introduce undesirable repetitive text generation.<n>We analyze repetition generation through the lens of information flow.<n>We present textbfCoTA, a plug-and-play method for mitigating repetition.
• Score: 40.28551750991027
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Recent diffusion-based Multimodal Large Language Models (dMLLMs) suffer from high inference latency and therefore rely on caching techniques to accelerate decoding. However, the application of cache mechanisms often introduces undesirable repetitive text generation, a phenomenon we term the \textbf{Repeat Curse}. To better investigate underlying mechanism behind this issue, we analyze repetition generation through the lens of information flow. Our work reveals three key findings: (1) context tokens aggregate semantic information as anchors and guide the final predictions; (2) as information propagates across layers, the entropy of context tokens converges in deeper layers, reflecting the model's growing prediction certainty; (3) Repetition is typically linked to disruptions in the information flow of context tokens and to the inability of their entropy to converge in deeper layers. Based on these insights, we present \textbf{CoTA}, a plug-and-play method for mitigating repetition. CoTA enhances the attention of context tokens to preserve intrinsic information flow patterns, while introducing a penalty term to the confidence score during decoding to avoid outputs driven by uncertain context tokens. With extensive experiments, CoTA demonstrates significant effectiveness in alleviating repetition and achieves consistent performance improvements on general tasks. Code is available at https://github.com/ErikZ719/CoTA

Related papers

• Deep Learning for Contextualized NetFlow-Based Network Intrusion Detection: Methods, Data, Evaluation and Deployment [5.402853794565817]
  This paper synthesizes recent research on context-aware deep learning for flow-based intrusion detection.<n>We organize existing methods into a four-dimensional taxonomy covering temporal context, graph or relational context, multimodal context, and multi-resolution context.<n>We review common failure modes that can inflate reported results, including temporal leakage, data splitting, dataset design flaws, limited dataset diversity, and weak cross-dataset generalization.
  arXiv  Detail & Related papers  (2026-02-05T12:25:18Z)
• Streaming-dLLM: Accelerating Diffusion LLMs via Suffix Pruning and Dynamic Decoding [36.74241893088594]
  Diffusion Large Language Models (dLLMs) offer a compelling paradigm for natural language generation.<n>Recent works have accelerated inference via KV cache reuse or decoding, but overlook the intrinsic inefficiencies within the block-wise diffusion process.<n>We propose Streaming-dLLM, a training-free framework that streamlines inference across both spatial and temporal dimensions.
  arXiv  Detail & Related papers  (2026-01-25T17:36:04Z)
• DeLeaker: Dynamic Inference-Time Reweighting For Semantic Leakage Mitigation in Text-to-Image Models [55.30555646945055]
  Text-to-Image (T2I) models are vulnerable to semantic leakage.<n>We introduce DeLeaker, a lightweight approach that mitigates leakage by directly intervening on the model's attention maps.<n>SLIM is the first dataset dedicated to semantic leakage.
  arXiv  Detail & Related papers  (2025-10-16T17:39:21Z)
• Explaining multimodal LLMs via intra-modal token interactions [55.27436637894534]
  Multimodal Large Language Models (MLLMs) have achieved remarkable success across diverse vision-language tasks, yet their internal decision-making mechanisms remain insufficiently understood.<n>We propose enhancing interpretability by leveraging intra-modal interaction.
  arXiv  Detail & Related papers  (2025-09-26T14:39:13Z)
• Decoding Memories: An Efficient Pipeline for Self-Consistency Hallucination Detection [17.792828844969033]
  We propose a novel Decoding Memory Pipeline (DMP) that accelerates generation through selective inference and decoding.<n>Our method achieves up to a 3x speedup without sacrificing AUROC performance.
  arXiv  Detail & Related papers  (2025-08-28T21:39:53Z)
• LayerCake: Token-Aware Contrastive Decoding within Large Language Model Layers [53.43862310647276]
  Large language models (LLMs) excel at natural language understanding and generation but remain vulnerable to factual errors.<n>We introduce a token-aware, layer-localized contrastive decoding method that aligns specific token types with their most influential transformer layers to improve factual generation.<n>Our method requires no additional training or model modification, and experiments demonstrate that our method consistently improves factuality across multiple LLMs and various benchmarks.
  arXiv  Detail & Related papers  (2025-07-06T14:35:43Z)
• FIRP: Faster LLM inference via future intermediate representation prediction [54.897493351694195]
  FIRP generates multiple tokens instead of one at each decoding step. We conduct extensive experiments, showing a speedup ratio of 1.9x-3x in several models and datasets.
  arXiv  Detail & Related papers  (2024-10-27T15:53:49Z)
• Maintaining Informative Coherence: Migrating Hallucinations in Large Language Models via Absorbing Markov Chains [6.920249042435973]
  Large Language Models (LLMs) are powerful tools for text generation, translation, and summarization. LLMs often suffer from hallucinations-instances where they fail to maintain the fidelity and coherence of contextual information. We propose a novel decoding strategy that leverages absorbing Markov chains to quantify the significance of contextual information.
  arXiv  Detail & Related papers  (2024-10-27T04:51:18Z)
• Parallel Decoding via Hidden Transfer for Lossless Large Language Model Acceleration [54.897493351694195]
  We propose a novel parallel decoding approach, namely textithidden transfer, which decodes multiple successive tokens simultaneously in a single forward pass. In terms of acceleration metrics, we outperform all the single-model acceleration techniques, including Medusa and Self-Speculative decoding.
  arXiv  Detail & Related papers  (2024-04-18T09:17:06Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.