NeurIPS 2024PosterAccept (Poster)

DataComp-LM: In search of the next generation of training sets for language models

Jeffrey Li, Alex Fang, Georgios Smyrnis, Maor Ivgi, Matt Jordan, Samir Yitzhak Gadre, Hritik Bansal, Etash Guha, Sedrick Scott Keh, Kushal Arora, Saurabh Garg, Rui Xin, Niklas Muennighoff, Reinhard Heckel, Jean Mercat, Mayee Chen, Suchin Gururangan, Mitchell Wortsman, Alon Albalak, Yonatan Bitton, Marianna Nezhurina, Amro Abbas, Cheng-Yu Hsieh, Dhruba Ghosh, Josh Gardner, Maciej Kilian, Hanlin Zhang, Rulin Shao, Sarah Pratt, Sunny Sanyal, Gabriel Ilharco, Giannis Daras, Kalyani Marathe, Aaron Gokaslan, Jieyu Zhang, Khyathi Chandu, Thao Nguyen, Igor Vasiljevic, Sham Kakade, Shuran Song, Sujay Sanghavi, Fartash Faghri, Sewoong Oh, Luke Zettlemoyer, Kyle Lo, Alaaeldin El-Nouby, Hadi Pouransari, Alexander Toshev, Stephanie Wang, Dirk Groeneveld, Luca Soldaini, Pang Wei Koh, Jenia Jitsev, Thomas Kollar, Alex Dimakis, Yair Carmon, Achal Dave, Ludwig Schmidt, Vaishaal Shankar

Department of Computer Science, University of Washington · Stanford University · University of Texas at Austin · Tel Aviv University · Ai2 · Columbia University · University of California, Los Angeles (UCLA) · University of Washington · Carnegie Mellon University · Toyota Research Institute · TUM / Rice University · Meta · SynthLabs · Google · LAION, JSC · African Institute For Mathematical Secience · Apple · Stability AI · Harvard University · The University of Texas at Austin · MIT · University of Washington, Seattle · Cornell University · Allen Institute for Artificial Intelligence · University of Washington, Meta FAIR · Harvard University & Amazon · UT-Austin · University of Toronto · University of Washington; Meta · Allen Institute for AI · Meta AI / INRIA · Apple Inc · Apple ML Research · LAION, Juelich Supercomputing Center (JSC) · Wayve · Electrical Engineering & Computer Science Department, University of California, Berkeley · Anthropic

论文页面 ↗ 登录后记笔记

PDF 由论文原始站点提供，PaperCompass 不保存论文文件。

摘要

We introduce DataComp for Language Models, a testbed for controlled dataset experiments with the goal of improving language models.As part of DCLM, we provide a standardized corpus of 240T tokens extracted from Common Crawl, effective pretraining recipes based on the OpenLM framework, and a broad suite of 53 downstream evaluations.Participants in the DCLM benchmark can experiment with data curation strategies such as deduplication, filtering, and data mixing atmodel scales ranging from 412M to 7B parameters.As a baseline for DCLM, we conduct extensive experiments and find that model-based filtering is key to assembling a high-quality training set.The resulting dataset, DCLM-Baseline, enables training a 7B parameter language model from scratch to 63% 5-shot accuracy on MMLU with 2T training tokens.Compared to MAP-Neo, the previous state-of-the-art in open-data language models, DCLM-Baseline represents a 6 percentage point improvement on MMLU while being trained with half the compute.Our results highlight the importance of dataset design for training language models and offer a starting point for further research on data curation. We release the \dclm benchmark, framework, models, and datasets at https://www.datacomp.ai/dclm/

论文信息

会议: NeurIPS 2024
年份: 2024
主题: Datasets and Benchmarks