Episode
Hierarchical Sparse Attention Done Right: Toward Infinite Context Modeling
- Podcast
- Daily Paper Cast
- Published
- Jul 9, 2026
- Duration seconds
- 1257
- Processing state
not_requested- Canonical source
- https://share.transistor.fm/s/2a4aad18
Actions
POST https://stenobird.com/v1/public/podcasts/daily-paper-cast-7079649/episodes/hierarchical-sparse-attention-done-right-toward-infinite-context-modeling/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/daily-paper-cast-7079649/hierarchical-sparse-attention-done-right-toward-infinite-context-modeling.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
🤗 Upvotes: 42 | cs.CL, cs.AI Authors: Xiang Hu, Xinyu Wei, Hao Gu, Minshen Zhang, Tian Liang, Huayang Li, Lei Zhu, Yan Wang, Sirui Han, Yushi Bai, Kewei Tu, Haitao Mi, Leo Liang Title: Hierarchical Sparse Attention Done Right: Toward Infinite Context Modeling Arxiv: http://arxiv.org/abs/2607.02980v1 Abstract: Scaling modern large language models (LLMs) to long contexts is limited by the quadratic computation cost, and poor length extrapolation of dense attention. Chunk-wise sparse attention offers a promising alternative, but all existing methods fall short of full attention because of their inaccurate chunk selection. We propose Hierarchical Landmark Sparse (HiLS) Attention, a chunk-wise sparse attention mechanism that learns chunk selection end-to-end under the language-modeling (LM) loss. HiLS factorizes attention hierarchically: each query performs attention independently with each retrieved chunk to extract chunk-specific information, and the resulting outputs are fused according to chunk retrieval scores. By incorporating retrieval scores into the forward attention computation, HiLS optimizes them directly with the LM loss, enabling end-to-end retrieval learning and native sparse training. Experimental results show that HiLS-Attention achieves performance comparable to, and in some cases better than, full attention at in-domain context lengths. Meanwhile, HiLS-Attention extrapolates more than $64\times$ the training context length with 90% retrieval accuracy, far beyond full attention. Moreover, existing full-attention models can be converted to HiLS-Attention with lightweight continued pretraining, preserving in-domain performance while acquiring ultra-long-context extrapolation. Together with its sparse KV access and computation, HiLS-Attention breaks the usual…