Dots Studio's DoSA speeds sparse attention indexer up to 2.74x over SGLang, open-sourced
dots-studio · hf · 2026-10-14
Dots Studio (Xiaohongshu) introduces DoSA, a lossless sparse attention method that makes indexer scoring conditional while preserving exact Top-K token selection. Its certificate-based approach uses a seed-set threshold lower bound, nonnegative head projections, and activation-informed bounds to skip full scoring of candidates that cannot win. The GPU implementation accelerates the full indexer up to 2.74x over SGLang and 1.69x over LiteTopK across English, Chinese, and code workloads, delivering up to 1.94x and 1.50x time-to-first-token speedups on GLM-5.2 and DeepSeek-V3.2. Code is open-sourced.
More from Infra
- Only 7% of frontier lab compute to pretraining by Q4 2026, but the viral chart is being misread — repligate · 2026-10-14
- AI agents hit monorepo scale: you need Google-style copy-on-write primitives, not Git — rickasaurus · 2026-10-14
- Modal trains RL models on mixed-accelerator clusters, matching homogeneous setups at lower cost — charles_irl · 2026-10-14
- A backyard "distributed data center" ADU startup for local LLM enthusiasts — dee_hw · 2026-10-14
- ShellElastic trains one supernet yielding all sizes, with provable hyperparameter transfer — Can Jin · 2026-10-14
- Report: Microsoft strains under data center spending and AI talent war, mulls cost cuts — SumitGup · 2026-10-14