Metraipolis › archive › dc_40f20ffff68f
TileWeave: Heterogeneous Tiling for Sparse Prefill on Mobile SoCs — arXiv:2611.00408
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.00408
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T18:35:10.614Z
- SHA-256 of the archived text
e8637f3b17f5da6820f4d74b1aafeb4c78ba859551c13bfcbdca533fa04f2c53- Stable URL of this page
https://metraipolis.com/documento/dc_40f20ffff68f
How to cite
TileWeave: Heterogeneous Tiling for Sparse Prefill on Mobile SoCs — arXiv:2611.00408 — arXiv (en), https://arxiv.org/abs/2611.00408 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_40f20ffff68f (Metraipolis archive dc_40f20ffff68f, sha256 e8637f3b17f5da6820f4d74b1aafeb4c78ba859551c13bfcbdca533fa04f2c53)
BibTeX
@misc{dc40f20ffff68f,
title = {TileWeave: Heterogeneous Tiling for Sparse Prefill on Mobile SoCs — arXiv:2611.00408},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.00408},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_40f20ffff68f; sha256 e8637f3b17f5da6820f4d74b1aafeb4c78ba859551c13bfcbdca533fa04f2c53},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.00408. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Smartphone applications increasingly require large language models to process long contexts, whereas high prefill latency and memory demands constrain local inference. Block-sparse attention reduces computation and memory traffic, but its irregular selections conflict with mobile NPUs' preference for large, regular tiles. Efficient execution must exploit cross-query reuse while avoiding redundant computation and preserving selections needed for accuracy. In this paper, we present TileWeave, which partitions sparse attention by this cross-query reuse: highly shared key blocks run in coarse NPU tiles, while low-reuse blocks run on the GPU only for the query blocks that selected them. In addition, we design an NPU-native pooled-key selector that keeps selection inexpensive, and a runtime that publishes GPU partials in the order the NPU consumes them and merges them in its existing output store. On RULER, HeteroSparse matches XAttention's accuracy with Qwen3-1.7B and improves it with Llama-3.2-3B-Instruct; its mean LongBench scores match XAttention as well. On Snapdragon SM8750 and SM8850 phones, it achieves 1.35--3.23$\times$ faster full-model prefill than XAttention over 1k--16k tokens and up to 1.58$\times$ faster than dense attention. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2611.00408 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-11-03T18:12:54.415Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).