# Attention Sinks: A 'Catch, Tag, Release' Mechanism for Embeddings

arXiv: https://arxiv.org/abs/2502.00919 · HTML: https://mlsanity.com/paper/2502.00919/

## Key facts

- Cited by 16 papers in the map (data through 14 Oct 2026).
- Builds on 35 of its 60 references that are in the map.
- First posted on arXiv 2 Feb 2025.

## Builds on (most cited first, 30 of 35)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [DINOv2: Learning Robust Visual Features without Supervision](https://mlsanity.com/paper/2304.07193/) (2023)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [Efficient Streaming Language Models with Attention Sinks](https://mlsanity.com/paper/2309.17453/) (2023)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [Vision Transformers Need Registers](https://mlsanity.com/paper/2309.16588/) (2023)
- [The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets](https://mlsanity.com/paper/2310.06824/) (2023)
- [KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache](https://mlsanity.com/paper/2402.02750/) (2024)
- [Massive Activations in Large Language Models](https://mlsanity.com/paper/2402.17762/) (2024)
- [Query-Key Normalization for Transformers](https://mlsanity.com/paper/2010.04245/) (2020)
- [Extracting Latent Steering Vectors from Pretrained Language Models](https://mlsanity.com/paper/2205.05124/) (2022)
- [QuALITY: Question Answering with Long Input Texts, Yes!](https://mlsanity.com/paper/2112.08608/) (2021)
- [When Attention Sink Emerges in Language Models: An Empirical View](https://mlsanity.com/paper/2410.10781/) (2024)
- [A mathematical perspective on Transformers](https://mlsanity.com/paper/2312.10794/) (2023)
- [Quantizable Transformers: Removing Outliers by Helping Attention Heads Do Nothing](https://mlsanity.com/paper/2306.12929/) (2023)
- [Signal Propagation in Transformers: Theoretical Perspectives and the Role of Rank Collapse](https://mlsanity.com/paper/2206.03126/) (2022)
- [The emergence of clusters in self-attention dynamics](https://mlsanity.com/paper/2305.05465/) (2023)
- [Round and Round We Go! What makes Rotary Positional Encodings useful?](https://mlsanity.com/paper/2410.06205/) (2024)
- [BERT Busters: Outlier Dimensions that Disrupt Transformers](https://mlsanity.com/paper/2105.06990/) (2021)
- [Transformers need glasses! Information over-squashing in language tasks](https://mlsanity.com/paper/2406.04267/) (2024)
- [Unveiling and Harnessing Hidden Attention Sinks: Enhancing Large Language Models without Training through Attention Calibration](https://mlsanity.com/paper/2406.15765/) (2024)
- [Attention Score is not All You Need for Token Importance Indicator in KV Cache Reduction: Value Also Matters](https://mlsanity.com/paper/2406.12335/) (2024)
- [Active-Dormant Attention Heads: Mechanistically Demystifying Extreme-Token Phenomena in LLMs](https://mlsanity.com/paper/2410.13835/) (2024)
- [Outliers Dimensions that Disrupt Transformers Are Driven by Frequency](https://mlsanity.com/paper/2205.11380/) (2022)
- [Outlier-Efficient Hopfield Layers for Large Transformer-Based Models](https://mlsanity.com/paper/2404.03828/) (2024)
- [Prefixing Attention Sinks can Mitigate Activation Outliers for Large Language Model Quantization](https://mlsanity.com/paper/2406.12016/) (2024)
- [Mitigating the Impact of Outlier Channels for Language Model Quantization with Activation Regularization](https://mlsanity.com/paper/2404.03605/) (2024)

## Built on by (most cited first, 16 of 16)

- [The Spike, the Sparse and the Sink: Anatomy of Massive Activations and Attention Sinks](https://mlsanity.com/paper/2603.05498/) (2026)
- [Attention Sinks Are Provably Necessary in Softmax Transformers: Evidence from Trigger-Conditional Tasks](https://mlsanity.com/paper/2603.11487/) (2026)
- [FreeText: Training-Free Text Rendering in Diffusion Transformers via Attention Localization and Spectral Glyph Injection](https://mlsanity.com/paper/2601.00535/) (2026)
- [Surgery: Mitigating Harmful Fine-Tuning for Large Language Models via Attention Sink](https://mlsanity.com/paper/2602.05228/) (2026)
- [What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs](https://mlsanity.com/paper/2603.06591/) (2026)
- [Gradient Flow Polarizes Softmax Outputs towards Low-Entropy Solutions](https://mlsanity.com/paper/2603.06248/) (2026)
- [A Single Layer to Explain Them All:Understanding Massive Activations in Large Language Models](https://mlsanity.com/paper/2605.08504/) (2026)
- [Practical Online KV Cache Compaction for LLM Agents: An Empirical Study](https://mlsanity.com/paper/2608.00902/) (2026)
- [Learning the ARTS of Search for Automated Discovery](https://mlsanity.com/paper/2606.21891/) (2026)
- [Text Capability Loss in Vision-Language Adaptation: An Attention-Sink Diagnosis](https://mlsanity.com/paper/2609.00746/) (2026)
- [Metaphor Is Not All Attention Needs](https://mlsanity.com/paper/2605.12128/) (2026)
- [When Graph Tokens Sink: A Mechanistic Analysis of Graph Language Models](https://mlsanity.com/paper/2606.03712/) (2026)
- [A Unifying View of Attention Sinks: From Mechanisms to Architectural Interventions](https://mlsanity.com/paper/2606.08105/) (2026)
- [Lung-SRAD: Spectral-Aware Regularized Audio DASS with Dual-Axis Patch-Mix Contrastive Learning for Respiratory Sound Classification](https://mlsanity.com/paper/2606.11922/) (2026)
- [Do New Attention Mechanisms Actually Fix Attention Sinks at Million-Token Context?](https://mlsanity.com/paper/2609.08574/) (2026)
- [OmniTide: Co-Designing Algorithms and Systems for Efficient On-Device Omni-LLM Streaming](https://mlsanity.com/paper/2609.34653/) (2026)

## Abstract (arXiv)

Large language models (LLMs) often concentrate their attention on a few specific tokens referred to as attention sinks. Common examples include the first token, a prompt-independent sink, and punctuation tokens, which are prompt-dependent. While the tokens causing the sinks often lack direct semantic meaning, the presence of the sinks is critical for model performance, particularly under model compression and KV-caching. Despite their ubiquity, the function, semantic role, and origin of attention sinks -- especially those beyond the first token -- remain poorly understood. In this work, we conduct a comprehensive investigation demonstrating that attention sinks: catch a sequence of tokens, tag them using a common direction in embedding space, and release them back into the residual stream, where tokens are later retrieved based on the tags they have acquired. Probing experiments reveal these tags carry semantically meaningful information, such as the truth of a statement. These findings extend to reasoning models, where the mechanism spans more heads and explains greater variance in embeddings, or recent models with query-key normalization, where sinks remain just as prevalent. To encourage future theoretical analysis, we introduce a minimal problem which can be solved through the 'catch, tag, release' mechanism, and where it emerges through training.