# PAT: Accelerating LLM Decoding via Prefix-Aware Attention with Resource Efficient Multi-Tile Kernel

arXiv: https://arxiv.org/abs/2511.22333 · HTML: https://mlsanity.com/paper/2511.22333/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving.
- Builds on 8 of its 10 references that are in the map.
- First posted on arXiv 27 Nov 2025.

## Placed in

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

## Builds on (most cited first, 8 of 8)

- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration](https://mlsanity.com/paper/2306.00978/) (2023)
- [Mixture-of-Experts with Expert Choice Routing](https://mlsanity.com/paper/2202.09368/) (2022)
- [FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision](https://mlsanity.com/paper/2407.08608/) (2024)
- [QuIP: 2-Bit Quantization of Large Language Models With Guarantees](https://mlsanity.com/paper/2307.13304/) (2023)
- [CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion](https://mlsanity.com/paper/2405.16444/) (2024)
- [ChunkAttention: Efficient Self-Attention with Prefix-Aware KV Cache and Two-Phase Partition](https://mlsanity.com/paper/2402.15220/) (2024)
- [RelayAttention for Efficient Large Language Model Serving with Long System Prompts](https://mlsanity.com/paper/2402.14808/) (2024)

## Built on by (most cited first, 5 of 5)

- [Characterization of Multi-Model Agentic AI Systems on General Tasks via Trace-Driven Simulation](https://mlsanity.com/paper/2606.01725/) (2026)
- [Bole: Efficient Tree Speculation for Hybrid-Attention Language Models](https://mlsanity.com/paper/2608.01651/) (2026)
- [Mosaic: Unlocking Long-Context Inference for Diffusion LLMs via Global Memory Planning and Dynamic Peak Taming](https://mlsanity.com/paper/2601.06562/) (2026)
- [Requests of a Feather Must Flock Together: Batch Size vs. Prefix Homogeneity in LLM Inference](https://mlsanity.com/paper/2605.06046/) (2026)
- [FoldAttention: Declared-Reference Softmax for Fast Decode and Deterministic Backward](https://mlsanity.com/paper/2609.33410/) (2026)

## Abstract (arXiv)

LLM serving is increasingly dominated by decode attention, which is a memory-bound operation due to massive KV cache loading from global memory. Meanwhile, real-world workloads exhibit substantial, hierarchical shared prefixes across requests (e.g., system prompts, tools/templates, RAG). Existing attention implementations fail to fully exploit prefix sharing: one-query-per-CTA execution repeatedly loads shared prefix KV cache, while one-size-fits-all tiling leaves on-chip resources idle and exacerbates bubbles for uneven KV lengths. These choices amplify memory bandwidth pressure and stall memory-bound decode attention. This paper introduces PAT, a prefix-aware attention kernel implementation for LLM decoding that organizes execution with a pack-forward-merge paradigm. PAT packs queries by shared prefix to reduce repeated memory accesses, runs a customized multi-tile kernel to achieve high resource efficiency. It further applies practical multi-stream forwarding and KV splitting to reduce resource bubbles. The final merge performs online softmax with negligible overhead. We implement PAT as an off-the-shelf plugin for vLLM. Evaluation on both real-world and synthetic workloads shows that PAT reduces attention latency by 53.5% on average and TPOT by 17.0-93.1% under the same configurations against state-of-the-art attention kernels. PAT's source code is publicly available at https://github.com/flashserve/PAT.