Reproduce “Subquadratic Attention with a Hierarchical Tree-Structured Indexer” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Leon Lufkin, Cutter Dawes, Alexander Speicher, Kamesh Krishnamurthy, Tomás Figliolia, Beren Millidge
ZHA uses hierarchical summaries to let each query search a KV cache without scoring every raw key. It compresses key chunks into a tree, routes queries from coarse to fine, and applies ordinary attention to selected keys while retaining the full cache. On Qwen3.6-35B-A3B, it scores 63.4% on LongBench v2, versus MSA’s 50.7%, but this is a benchmark comparison, not measured speedup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.