OBELICS: An Open Web-Scale Filtered Dataset of Interleaved Image-Text Documents

TL;DR

OBELICS is a large, openly available web-scale dataset of interleaved image-text documents, enabling training of multimodal models that outperform prior models on benchmarks.

Contribution

The paper introduces OBELICS, a comprehensive, filtered dataset of 141 million web pages with interleaved images and text, and demonstrates its effectiveness by training large multimodal models.

Findings

Models trained on OBELICS achieve competitive benchmark performance.

The dataset enables training of models with up to 80 billion parameters.

Open release of dataset, models, and code facilitates further research.

Abstract

Large multimodal models trained on natural documents, which interleave images and text, outperform models trained on image-text pairs on various multimodal benchmarks. However, the datasets used to train these models have not been released, and the collection process has not been fully specified. We introduce the OBELICS dataset, an open web-scale filtered dataset of interleaved image-text documents comprising 141 million web pages extracted from Common Crawl, 353 million associated images, and 115 billion text tokens. We describe the dataset creation process, present comprehensive filtering rules, and provide an analysis of the dataset's content. To show the viability of OBELICS, we train vision and language models of 9 and 80 billion parameters named IDEFICS, and obtain competitive performance on different multimodal benchmarks. We release our dataset, models and code.