Predictive Data Selection: The Data That Predicts Is the Data That Teaches

Kashun Shum, Yuzhen Huang, Hongjian Zou, Qi Ding, Yixuan Liao, Xiaoxin Chen, Qian Liu, Junxian He
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:55427-55450, 2025.

Abstract

Language model pretraining involves training on extensive corpora, where data quality plays a pivotal role. In this work, we aim to directly estimate the contribution of data during pretraining and select pretraining data in an efficient manner. Specifically, we draw inspiration from recent findings showing that compression efficiency (i.e., normalized loss) of diverse models on certain text correlates strongly with their downstream performance, when the text domain aligns with the downstream benchmarks (Huang et al., 2024). Building on this observation, we hypothesize that data on which model losses are predictive of downstream abilities also contribute effectively to learning, which shares similar intuition with Thrush et al. (2024). To leverage this insight, we introduce predictive data selection (PreSelect), a lightweight and efficient data selection method that requires training and deploying only a fastText-based scorer. Through comprehensive experiments with 1B and 3B parameter models, we demonstrate that models trained on 30B tokens selected with PreSelect surpass the performance of the vanilla baseline trained on 300B tokens, achieving a 10x reduction in compute requirements. Furthermore, PreSelect significantly outperforms other competitive data selection baselines, such as DCLM and FineWeb-Edu on a scale of 3B models trained on 100B tokens. We open-source our trained data selection scorer along with the curated datasets at https://github.com/hkust-nlp/PreSelect.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-shum25a, title = {Predictive Data Selection: The Data That Predicts Is the Data That Teaches}, author = {Shum, Kashun and Huang, Yuzhen and Zou, Hongjian and Ding, Qi and Liao, Yixuan and Chen, Xiaoxin and Liu, Qian and He, Junxian}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {55427--55450}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/shum25a/shum25a.pdf}, url = {https://proceedings.mlr.press/v267/shum25a.html}, abstract = {Language model pretraining involves training on extensive corpora, where data quality plays a pivotal role. In this work, we aim to directly estimate the contribution of data during pretraining and select pretraining data in an efficient manner. Specifically, we draw inspiration from recent findings showing that compression efficiency (i.e., normalized loss) of diverse models on certain text correlates strongly with their downstream performance, when the text domain aligns with the downstream benchmarks (Huang et al., 2024). Building on this observation, we hypothesize that data on which model losses are predictive of downstream abilities also contribute effectively to learning, which shares similar intuition with Thrush et al. (2024). To leverage this insight, we introduce predictive data selection (PreSelect), a lightweight and efficient data selection method that requires training and deploying only a fastText-based scorer. Through comprehensive experiments with 1B and 3B parameter models, we demonstrate that models trained on 30B tokens selected with PreSelect surpass the performance of the vanilla baseline trained on 300B tokens, achieving a 10x reduction in compute requirements. Furthermore, PreSelect significantly outperforms other competitive data selection baselines, such as DCLM and FineWeb-Edu on a scale of 3B models trained on 100B tokens. We open-source our trained data selection scorer along with the curated datasets at https://github.com/hkust-nlp/PreSelect.} }
Endnote
%0 Conference Paper %T Predictive Data Selection: The Data That Predicts Is the Data That Teaches %A Kashun Shum %A Yuzhen Huang %A Hongjian Zou %A Qi Ding %A Yixuan Liao %A Xiaoxin Chen %A Qian Liu %A Junxian He %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-shum25a %I PMLR %P 55427--55450 %U https://proceedings.mlr.press/v267/shum25a.html %V 267 %X Language model pretraining involves training on extensive corpora, where data quality plays a pivotal role. In this work, we aim to directly estimate the contribution of data during pretraining and select pretraining data in an efficient manner. Specifically, we draw inspiration from recent findings showing that compression efficiency (i.e., normalized loss) of diverse models on certain text correlates strongly with their downstream performance, when the text domain aligns with the downstream benchmarks (Huang et al., 2024). Building on this observation, we hypothesize that data on which model losses are predictive of downstream abilities also contribute effectively to learning, which shares similar intuition with Thrush et al. (2024). To leverage this insight, we introduce predictive data selection (PreSelect), a lightweight and efficient data selection method that requires training and deploying only a fastText-based scorer. Through comprehensive experiments with 1B and 3B parameter models, we demonstrate that models trained on 30B tokens selected with PreSelect surpass the performance of the vanilla baseline trained on 300B tokens, achieving a 10x reduction in compute requirements. Furthermore, PreSelect significantly outperforms other competitive data selection baselines, such as DCLM and FineWeb-Edu on a scale of 3B models trained on 100B tokens. We open-source our trained data selection scorer along with the curated datasets at https://github.com/hkust-nlp/PreSelect.
APA
Shum, K., Huang, Y., Zou, H., Ding, Q., Liao, Y., Chen, X., Liu, Q. & He, J.. (2025). Predictive Data Selection: The Data That Predicts Is the Data That Teaches. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:55427-55450 Available from https://proceedings.mlr.press/v267/shum25a.html.

Related Material