# How Linear Attention Remembers

arXiv: https://arxiv.org/abs/2609.33093 · HTML: https://mlsanity.com/paper/2609.33093/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 27 of its 28 references that are in the map.
- First posted on arXiv 27 Sep 2026.

## Builds on (most cited first, 27 of 27)

- [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)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [Pointer Sentinel Mixture Models](https://mlsanity.com/paper/1609.07843/) (2016)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Efficient Streaming Language Models with Attention Sinks](https://mlsanity.com/paper/2309.17453/) (2023)
- [Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention](https://mlsanity.com/paper/2006.16236/) (2020)
- [SnapKV: LLM Knows What You are Looking for Before Generation](https://mlsanity.com/paper/2404.14469/) (2024)
- [H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models](https://mlsanity.com/paper/2306.14048/) (2023)
- [Fast Transformer Decoding: One Write-Head is All You Need](https://mlsanity.com/paper/1911.02150/) (2019)
- [KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache](https://mlsanity.com/paper/2402.02750/) (2024)
- [Efficiently Scaling Transformer Inference](https://mlsanity.com/paper/2211.05102/) (2022)
- [Gated Delta Networks: Improving Mamba2 with Delta Rule](https://mlsanity.com/paper/2412.06464/) (2024)
- [Gated Linear Attention Transformers with Hardware-Efficient Training](https://mlsanity.com/paper/2312.06635/) (2023)
- [Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference](https://mlsanity.com/paper/2406.10774/) (2024)
- [Linear Transformers Are Secretly Fast Weight Programmers](https://mlsanity.com/paper/2102.11174/) (2021)
- [Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Time](https://mlsanity.com/paper/2305.17118/) (2023)
- [Causal Abstractions of Neural Networks](https://mlsanity.com/paper/2106.02997/) (2021)
- [Parallelizing Linear Transformers with the Delta Rule over Sequence Length](https://mlsanity.com/paper/2406.06484/) (2024)
- [Towards Best Practices of Activation Patching in Language Models: Metrics and Methods](https://mlsanity.com/paper/2309.16042/) (2023)
- [MobileLLM: Optimizing Sub-billion Parameter Language Models for On-Device Use Cases](https://mlsanity.com/paper/2402.14905/) (2024)
- [Simple linear attention language models balance the recall-throughput tradeoff](https://mlsanity.com/paper/2402.18668/) (2024)
- [LLM in a flash: Efficient Large Language Model Inference with Limited Memory](https://mlsanity.com/paper/2312.11514/) (2023)
- [MiniCache: KV Cache Compression in Depth Dimension for Large Language Models](https://mlsanity.com/paper/2405.14366/) (2024)
- [What needs to go right for an induction head? A mechanistic study of in-context learning circuits and their formation](https://mlsanity.com/paper/2404.07129/) (2024)
- [Going Beyond Linear Transformers with Recurrent Fast Weight Programmers](https://mlsanity.com/paper/2106.06295/) (2021)
- [Distilling On-device Language Models for Robot Planning with Minimal Human Intervention](https://mlsanity.com/paper/2506.17486/) (2025)

## Abstract (arXiv)

Linear attention replaces the growing key--value (KV) cache of standard attention with a fixed-size recurrent state, substantially reducing memory growth with context length. This efficiency, however, changes how past information is stored: many tokens must share and repeatedly update the same memory. We study how this recurrent state functions as a memory system. Using an analytical decomposition together with controlled causal interventions in pretrained GLA and GDN models, we trace how recalled information is written, retained, and later accessed. We find that fact-specific information enters recurrent memory through concentrated, content-dependent writes and is later accessed through concentrated query-time read pathways. Multiple facts can remain selectively accessible within the same state, yet their internal representations exhibit cross-fact causal coupling rather than independent KV-like storage. As memory load increases, both recall and targeted editability degrade, whereas elapsed context alone has a substantially smaller effect within the tested regime. Causal interventions on subsequent writes further show that interference is shaped by their overlap with existing memory. Finally, in hybrid architectures that combine recurrent layers with full attention, the runtime memory directly supporting recall shifts predominantly to the full-attention KV state. Together, these results reveal how fixed-size recurrent memory supports selective recall despite shared storage, while exposing the interference and capacity limits that distinguish it from token-addressable KV memory.