Data, Data Everywhere: A Guide for Pretraining Dataset Construction
This provides actionable guidance for practitioners in AI and NLP to develop high-quality pretraining sets, addressing a critical bottleneck in model development, though it is incremental as it builds on existing techniques.
The paper tackles the lack of open information on constructing effective pretraining datasets for language models by conducting a systematic study across the entire pipeline, identifying methods that yield the largest gains in model accuracy and categorizing web crawl data by attributes like toxicity and quality to refine datasets.
The impressive capabilities of recent language models can be largely attributed to the multi-trillion token pretraining datasets that they are trained on. However, model developers fail to disclose their construction methodology which has lead to a lack of open information on how to develop effective pretraining sets. To address this issue, we perform the first systematic study across the entire pipeline of pretraining set construction. First, we run ablations on existing techniques for pretraining set development to identify which methods translate to the largest gains in model accuracy on downstream evaluations. Then, we categorize the most widely used data source, web crawl snapshots, across the attributes of toxicity, quality, type of speech, and domain. Finally, we show how such attribute information can be used to further refine and improve the quality of a pretraining set. These findings constitute an actionable set of steps that practitioners can use to develop high quality pretraining sets.