LLM Digest

Story

arxiv_cs_cl · Nov 3, 2026 · paper

Source brief

Coarse Semantic Blocks for Efficient Training-Free Sparse Attention

arxiv.orgNov 3, 2026
original source linked

In brief

We introduce a training-free sparse attention method that builds a prompt-specific key index to accelerate both prefill and decoding. The method clusters the keys into large semantic blocks that group tokens from nonc...

Feed lens

eval

Read the original at arxiv.org →Open in live feed Read that day’s brief

Coarse Semantic Blocks for Efficient Training-Free Sparse Attention

Earlier in this thread 4 items

Vortex: Efficient and Programmable Sparse Attention Serving for AI Agents

Multi-layer Indexer: Overlapping Selection and Attention in DeepSeek Sparse Attention

Latent Sparse Attention: Towards Subquadratic Ultra-Long Context Attention

DoubtLess: Training-Free Contrastive Decoding for Efficient Reasoning