# Breaking the KV Cache Bottleneck: Fan Duality Model Achieves O(1) Decode Memory with Superior Associative Recall

arXiv: https://arxiv.org/abs/2604.07716 · HTML: https://mlsanity.com/paper/2604.07716/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving.
- Builds on 6 of its 8 references that are in the map.
- First posted on arXiv 9 Apr 2026.

## Placed in

- Efficient LLM inference and serving / [KV cache optimization](https://mlsanity.com/inference/kv-cache-optimization/)

## Builds on (most cited first, 6 of 6)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://mlsanity.com/paper/2312.00752/) (2023)
- [Longformer: The Long-Document Transformer](https://mlsanity.com/paper/2004.05150/) (2020)
- [Efficiently Modeling Long Sequences with Structured State Spaces](https://mlsanity.com/paper/2111.00396/) (2021)
- [Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality](https://mlsanity.com/paper/2405.21060/) (2024)
- [Simple linear attention language models balance the recall-throughput tradeoff](https://mlsanity.com/paper/2402.18668/) (2024)

## Abstract (arXiv)

We present FDM (Fan Duality Model), a linear sequence architecture that resolves the fundamental tension between memory efficiency and associative recall in sequence modeling. FDM separates sequence processing into two components: a wave component (recurrent scan via phase-preserving Givens rotations) that compresses long-range patterns into a fixed-size complex hidden state, and a particle component (local-global cache) that retrieves specific tokens via learned associative addressing with W+K=272 slots independent of sequence length N. This yields strictly O(1) decode memory: 867 MB fixed across all prompt lengths 128-8,192 tokens, versus Transformer's 853-4,247 MB (4.9x reduction at N=8,192). Beyond the architecture, we discover that jointly training the wave and particle components leads to suboptimal convergence. We propose Freeze-Scan, a two-phase training strategy that freezes the recurrent scan and optimizes the cache jointly with embeddings, achieving PPL=64.9 on WikiText-103 in 44K steps -- a 7.5x improvement over full fine-tuning (PPL=487). On Multi-Query Associative Recall (MQAR), FDM achieves 0.966 accuracy, surpassing Transformer (0.606) by 59.5%, while pure scan without cache scores only 0.011, confirming the necessity of the particle component. Finally, we introduce Holographic Reference Beam Decoding, interpreting the complex hidden state h_t as a holographic plate encoding the entire temporal history. Using the current input x_t as a reference beam to modulate h_t reduces PPL by up to 2.13 points (PPL=62.79) with a 4-head orthogonal reference beam using only 1.3M additional parameters, providing empirical support for the holographic interpretation. Code and pretrained weights: https://github.com/YasongFan/FDM