SparseGDN：线性注意力中的分块稀疏框架
SparseGDN: Chunk Sparse in Linear Attention
AI 导读
SparseGDN 是一种硬件感知框架，通过分块内结构化近似与依赖输入的循环事件选择，降低 GDN 长上下文 prefill 的计算与显存开销。在 Qwen3.5-4B 和 Qwen3.5-9B 上的 RULER 任务（上下文 4K 到 128K）评测显示，其平均仅保留约 29% 循环事件，相对原生 FLA 内核实现 1.50-1.94x 的 GDN-core prefill 加速。
正文 · 原文
Abstract:Gated Delta Networks (GDNs) offer linear-time sequence modeling, yet long-context prefill remains costly due to dense intra-chunk computation, recurrent-state updates, and intermediate state storage. We introduce SparseGDN, a hardware-aware framework that combines structured intra-chunk approximation with input-dependent recurrent-event selection while producing outputs for every token. A block-adjacent solver reduces local Delta-solve computation, while a dynamic selector allocates recurrent-update budgets across head pairs. The GPU implementation materializes updates only for selected events and reconstructs token outputs using compact state snapshots and group-local interactions, reducing computation and memory traffic. We evaluate SparseGDN on Qwen3.5-4B and Qwen3.5-9B across RULER tasks and context lengths from 4K to 128K. Retaining approximately 29% of recurrent events on average, SparseGDN achieves 1.50-1.94x GDN-core prefill speedups over native FLA kernels, including online selection overhead.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.41420 [cs.LG] |
| (or arXiv:2610.41420v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.41420 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jiedong Zhuang [view email]
[v1]
Fri, 30 Oct 2026 08:27:36 UTC (681 KB)
来源：arXiv cs.AI · arxiv.org