Organize the Web: Constructing Domains Enhances Pre-Training Data Curation
This work addresses the problem of pre-training data curation for language model developers, providing a valuable complement to quality-based methods.
The authors tackled the problem of pre-training data curation for language models by organizing web pages into domains, resulting in improved model performance on downstream tasks. By combining insights about effective topics and formats, they further boosted performance, with domain mixing also improving existing quality-based methods.
Modern language models are trained on large, unstructured datasets consisting of trillions of tokens and obtained by crawling the web. The unstructured nature makes it difficult to reason about their contents and develop systematic approaches to data curation. In this paper, we unpack monolithic web corpora by developing taxonomies of their contents and organizing them into domains. We introduce WebOrganizer, a framework for organizing web pages in terms of both their topic and format. Using these two complementary notions of domains, we automatically annotate pre-training data by distilling annotations from a large language model into efficient classifiers. This allows us to study how data from different domains should be mixed to improve models on downstream tasks, and we show that we can combine insights about effective topics and formats to further boost performance. We demonstrate that our domain mixing also improves existing methods that select data based on quality. Furthermore, we study and compare how quality-based methods will implicitly change the domain mixture. Overall, our work demonstrates that constructing and mixing domains provides a valuable complement to quality-based data curation methods, opening new avenues for effective and insightful pre-training data curation.