Double-P: Hierarchical Top-P Sparse Attention for Long-Context LLMs
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866915775630016512 |
|---|---|
| author | Ni, Wentao Zhang, Kangqi Yu, Zhongming Nelson, Oren Lee, Mingu Cai, Hong Porikli, Fatih Kim, Jongryool Liu, Zhijian Zhao, Jishen |
| author_facet | Ni, Wentao Zhang, Kangqi Yu, Zhongming Nelson, Oren Lee, Mingu Cai, Hong Porikli, Fatih Kim, Jongryool Liu, Zhijian Zhao, Jishen |
| contents | As long-context inference becomes central to large language models (LLMs), attention over growing key-value caches emerges as a dominant decoding bottleneck, motivating sparse attention for scalable inference. Fixed-budget top-k sparse attention cannot adapt to heterogeneous attention distributions across heads and layers, whereas top-p sparse attention directly preserves attention mass and provides stronger accuracy guarantees. Existing top-p methods, however, fail to jointly optimize top-p accuracy, selection overhead, and sparse attention cost, which limits their overall efficiency. We present Double-P, a hierarchical sparse attention framework that optimizes all three stages. Double-P first performs coarse-grained top-p estimation at the cluster level using size-weighted centroids, then adaptively refines computation through a second top-p stage that allocates token-level attention only when needed. Across long-context benchmarks, Double-P consistently achieves near-zero accuracy drop, reducing attention computation overhead by up to 1.8x and delivers up to 1.3x end-to-end decoding speedup over state-of-the-art fixed-budget sparse attention methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_05191 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Double-P: Hierarchical Top-P Sparse Attention for Long-Context LLMs Ni, Wentao Zhang, Kangqi Yu, Zhongming Nelson, Oren Lee, Mingu Cai, Hong Porikli, Fatih Kim, Jongryool Liu, Zhijian Zhao, Jishen Machine Learning Artificial Intelligence As long-context inference becomes central to large language models (LLMs), attention over growing key-value caches emerges as a dominant decoding bottleneck, motivating sparse attention for scalable inference. Fixed-budget top-k sparse attention cannot adapt to heterogeneous attention distributions across heads and layers, whereas top-p sparse attention directly preserves attention mass and provides stronger accuracy guarantees. Existing top-p methods, however, fail to jointly optimize top-p accuracy, selection overhead, and sparse attention cost, which limits their overall efficiency. We present Double-P, a hierarchical sparse attention framework that optimizes all three stages. Double-P first performs coarse-grained top-p estimation at the cluster level using size-weighted centroids, then adaptively refines computation through a second top-p stage that allocates token-level attention only when needed. Across long-context benchmarks, Double-P consistently achieves near-zero accuracy drop, reducing attention computation overhead by up to 1.8x and delivers up to 1.3x end-to-end decoding speedup over state-of-the-art fixed-budget sparse attention methods. |
| title | Double-P: Hierarchical Top-P Sparse Attention for Long-Context LLMs |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2602.05191 |