How Can We Synthesize High-Quality Pretraining Data? A Systematic Study of Prompt Design, Generator Model, and Source Data
Joel Niklaus · Atsuki Yamaguchi · Michal Štefánik · Guilherme Penedo · Hynek Kydlíček · Elie Bakouch · Lewis Tunstall · Edward Emanuel Beeching · Thibaud Frere · Colin Raffel · Leandro von Werra · Thomas Wolf
Controlled trillion-token experiments identify structured formats and source selection as key factors in synthetic pretraining data, yielding FinePhrase, a 486B-token dataset outperforming baselines at 30x lower cost.
摘要
Synthetic data is a standard component in training large language models, yet systematic comparisons across design dimensions, including rephrasing strategy, generator model, and source data, remain absent. We conduct extensive controlled experiments, generating over one trillion tokens, to identify critical factors in rephrasing web text into synthetic pretraining data. Our results reveal that structured output formats, such as tables, math problems, FAQs, and tutorials, consistently outperform both curated web baselines and prior synthetic methods. Notably, increasing the size of the generator model beyond 1B parameters provides no additional benefit. Our analysis also demonstrates that the selection of the original data used for mixing substantially influences performance. By applying our findings, we develop FinePhrase, a 486-billion-token open dataset of rephrased web text. We show that FinePhrase outperforms all existing synthetic data baselines while reducing generation costs by up to 30 times. We provide the dataset, all prompts, and the generation framework to the research community.