Reproduce “Highlight-Then-Summarize: Learning to Compress Evidence for Long-Context Understanding” 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
Zhaoyuan Xia, Qinghongbing Xie, Yung Xiang Hue, Jianguang Jiang, Gaofeng Lu, Zhenyu Jiao, Xing Yuan, Dai Dai and 2 more
The paper introduces Highlight-Then-Summarize, a method for answering long-document questions by selecting source-linked evidence, organizing its claims and relationships into a question-focused summary, then answering with the original document available. On H2S-Bench, its 14B model scores 32.60 on the unweighted average across seven tasks, versus 22.43 for Qwen3.8-27B under shared token limits; this is not a win on every task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.