QuRating: Selecting High-Quality Data for Training Language Models
This addresses the challenge of data quality selection for language model training, offering a novel approach that improves efficiency and performance, though it is incremental in advancing existing data selection methods.
The paper tackles the problem of selecting high-quality pre-training data for language models by introducing QuRating, a method that uses LLMs to rate data based on four qualities, and shows that models trained on data selected with these ratings achieve lower perplexity and stronger in-context learning, with the best model matching performance of a baseline trained for 50% more steps.
Selecting high-quality pre-training data is important for creating capable language models, but existing methods rely on simple heuristics. We introduce QuRating, a method for selecting pre-training data that can capture human intuitions about data quality. In this paper, we investigate four qualities - writing style, required expertise, facts & trivia, and educational value - and find that LLMs are able to discern these qualities, especially when making pairwise judgments of texts. We train a QuRater model to learn scalar ratings from pairwise judgments, and use it to annotate a 260B training corpus with quality ratings for each of the four criteria. In our experiments, we select 30B tokens according to the different quality ratings and train 1.3B-parameter language models on the selected data. We find that it is important to balance quality and diversity. When we sample using quality ratings as logits over documents, our models obtain lower perplexity and stronger in-context learning performance than baselines. Our best model is based on educational value and performs similarly to a model trained with uniform sampling for 50% more steps. Beyond data selection, we use the quality ratings to construct a training curriculum which improves performance without changing the training dataset. We extensively analyze the quality ratings and discuss their characteristics, biases, and wider implications.