Reproduce “ReScraper: Unified Scraping and Cleaning of Web Data for Effective LLM Pretraining” 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

Zichun Yu, Jiarui Yan, Shlok Sanghvi, Nihar Atri, Chenyan Xiong

ReScraper uses a small model to read rendered webpages, extract main content, and keep, delete, edit, or rewrite each page before pretraining. It replaces whole-page filtering with page-specific treatment. In the reported 1.4B-model setting, its corpus reaches 0.27348 DCLM Core versus 0.26152 for UltraX, a 4.6% relative improvement, within one English Common Crawl and DCLM setup.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.