It's Not RoPE that Creates Sinks: The Role of Self-Concentration and Value-Non-Mixing in Attention

• URL: http://arxiv.org/abs/2609.09085v1
• Date: Tue, 08 Sep 2026 17:32:31 GMT
• Title: It's Not RoPE that Creates Sinks: The Role of Self-Concentration and Value-Non-Mixing in Attention
• Abstract summary: Large Language Models (LLMs) often exhibit "Attention Sink" (AS) and the accompanying "Massive Activations" (MAs) at the initial position of a sequence.<n>We analyze the factors underlying AS and MAs that emerge at the initial position regardless of the token occupying it.<n>Our experiments suggest that self-concentration of attention, resulting from the causal mask, and the subsequent Value-non-mixing in attention outputs contribute to AS and MAs.
• Score: 13.912622084285493
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models (LLMs) often exhibit "Attention Sink" (AS) and the accompanying "Massive Activations" (MAs) at the initial position of a sequence. These phenomena frequently co-occur, and MAs can pose challenges for low-bit quantization. In this study, we analyze the factors underlying AS and MAs that emerge at the initial position regardless of the token occupying it. Our experiments suggest that self-concentration of attention, resulting from the causal mask, and the subsequent Value-non-mixing in attention outputs contribute to AS and MAs. These findings provide new empirical evidence on the internal dynamics of LLMs, offering insights that may inform future quantization strategies and advance our understanding of the internal mechanisms of attention layers.

Related papers

• CoSToM:Causal-oriented Steering for Intrinsic Theory-of-Mind Alignment in Large Language Models [23.229692182223157]
  Theory of Mind (ToM) is the ability to attribute mental states to others.<n>LLMs demonstrate promising performance on standard ToM benchmarks.<n>LLMs often fail to generalize to complex task-specific scenarios.
  arXiv  Detail & Related papers  (2026-04-11T05:04:37Z)
• Revealing the Attention Floating Mechanism in Masked Diffusion Models [52.74142815156738]
  Masked diffusion models (MDMs) leverage bidirectional attention and a denoising process.<n>This paper investigates the attention behaviors in MDMs, revealing the phenomenon of Attention Floating.
  arXiv  Detail & Related papers  (2026-01-12T09:10:05Z)
• Beyond Benchmarks: Understanding Mixture-of-Experts Models through Internal Mechanisms [55.1784306456972]
  Mixture-of-Experts (MoE) architectures have emerged as a promising direction, offering efficiency and scalability by activating only a subset of parameters during inference.<n>We use an internal metric to investigate the mechanisms of MoE architecture by explicitly incorporating routing mechanisms and analyzing expert-level behaviors.<n>We uncover several findings: (1) neuron utilization decreases as models evolve, reflecting stronger generalization; (2) training exhibits a dynamic trajectory, where benchmark performance alone provides limited signal; (3) task completion emerges from collaborative contributions of multiple experts, with shared experts driving concentration; and (4) activation patterns at the neuron level provide a fine-grained proxy for data diversity.
  arXiv  Detail & Related papers  (2025-09-28T15:13:38Z)
• 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)
• Quantifying Self-Awareness of Knowledge in Large Language Models [19.886013407039517]
  Hallucination prediction in large language models is often interpreted as a sign of self-awareness.<n>We argue that such performance can arise from question-side shortcuts rather than true model-side introspection.<n>We introduce SCAO, a method that enhances the use of model-side signals.
  arXiv  Detail & Related papers  (2025-09-18T18:29:14Z)
• Unveiling Super Experts in Mixture-of-Experts Large Language Models [5.36587814108884]
  Sparsely activated Mixture-of-Experts (MoE) models have shown promise in enhancing the learning capacity of large language models (LLMs)<n>We present the first discovery and investigation of a distinct subset of experts that play a crucial role in the underlying mechanisms during the model's forward inference.<n>Our findings confirm that MoE LLMs rely on SEs to induce attention sinks, which are crucial for the distribution of attention scores but are significantly disrupted by SE pruning.
  arXiv  Detail & Related papers  (2025-07-31T06:35:33Z)
• Hallucination Detection in LLMs with Topological Divergence on Attention Graphs [60.83579255387347]
  Hallucination, i.e., generating factually incorrect content, remains a critical challenge for large language models.<n>We introduce TOHA, a TOpology-based HAllucination detector in the RAG setting.
  arXiv  Detail & Related papers  (2025-04-14T10:06:27Z)
• How do Large Language Models Understand Relevance? A Mechanistic Interpretability Perspective [64.00022624183781]
  Large language models (LLMs) can assess relevance and support information retrieval (IR) tasks.<n>We investigate how different LLM modules contribute to relevance judgment through the lens of mechanistic interpretability.
  arXiv  Detail & Related papers  (2025-04-10T16:14:55Z)
• Don't Take Things Out of Context: Attention Intervention for Enhancing Chain-of-Thought Reasoning in Large Language Models [32.71672086718058]
  Few-shot Chain-of-Thought (CoT) significantly enhances the reasoning capabilities of large language models (LLMs)<n>We observe that isolated segments, words, or tokens within CoT demonstrations can unexpectedly disrupt the generation process of LLMs.<n>We propose a Few-shot Attention Intervention method (FAI) that dynamically analyzes the attention patterns of demonstrations to accurately identify these tokens.
  arXiv  Detail & Related papers  (2025-03-14T07:46:33Z)
• Active-Dormant Attention Heads: Mechanistically Demystifying Extreme-Token Phenomena in LLMs [77.66717051042032]
  Practitioners have consistently observed three puzzling phenomena in transformer-based large language models. These phenomena are characterized by certain so-called "sink tokens" receiving disproportionately high attention weights. We elucidate the mechanisms behind extreme-token phenomena.
  arXiv  Detail & Related papers  (2024-10-17T17:54:06Z)
• Look Before You Decide: Prompting Active Deduction of MLLMs for Assumptive Reasoning [77.72128397088409]
  We show that most prevalent MLLMs can be easily fooled by the introduction of a presupposition into the question.<n>We also propose a novel reinforcement learning paradigm to encourage the model to actively perform composite deduction.
  arXiv  Detail & Related papers  (2024-04-19T15:53:27Z)
• Multilingual Multi-Aspect Explainability Analyses on Machine Reading Comprehension Models [76.48370548802464]
  This paper focuses on conducting a series of analytical experiments to examine the relations between the multi-head self-attention and the final MRC system performance. We discover that passage-to-question and passage understanding attentions are the most important ones in the question answering process. Through comprehensive visualizations and case studies, we also observe several general findings on the attention maps, which can be helpful to understand how these models solve the questions.
  arXiv  Detail & Related papers  (2021-08-26T04:23:57Z)

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.