Research map / Architectures and efficiency / Attention and transformers
DisSparse: Pipelined Top-$p$ Sparse Attention for Long-Context LLM Serving
Nurlan Nazaraliyev, Elaheh Sadredini, Nael Abu-Ghazaleh
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes DisSparse, a pipelined top-p sparse attention system for long-context LLM serving that addresses the cost of dynamic top-p selection relative to fixed-budget top-k.
Problem
Long-context LLM inference is bottlenecked by KV cache memory; dynamic sparse attention usually relies on fixed-budget top-k selection, while top-p (adaptive budget) selection is harder to serve efficiently.
Approach
Pipelines top-p sparse attention so that the adaptive, variable-size token selection is overlapped with computation and KV fetching in serving, rather than running it serially as in prior top-p approaches. (Details beyond the abstract not available.)
Closest papers by meaning
- SparDA: Sparse Decoupled Attention for Efficient Long-Context LLM Inference (NeurIPS 2026 · similarity 0.90)
- MInference 1.0: Accelerating Pre-filling for Long-Context LLMs via Dynamic Sparse Attention (NeurIPS 2024 · similarity 0.89)
- Loki: Low-rank Keys for Efficient Sparse Attention (NeurIPS 2024 · similarity 0.89)
- QUEST: Query-Aware Sparsity for Efficient Long-Context LLM Inference (ICML 2024 · similarity 0.89)
- Full Attention Strikes Back: Transferring Full Attention into Sparse within Hundred Training Steps (NeurIPS 2026 · similarity 0.88)
- ShadowKV: KV Cache in Shadows for High-Throughput Long-Context LLM Inference (ICML 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
