TileWeave paper proposes heterogeneous tiling for sparse LLM prefill on mobile SoCs
A new arXiv paper, TileWeave, introduces a heterogeneous tiling method to speed up sparse prefill for large language models on mobile SoCs, targeting long-context use in smartphone apps where prefill latency and memory demands are limiting factors.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-03 05:00 · arXiv cs.LG
TileWeave: Heterogeneous Tiling for Sparse Prefill on Mobile SoCs