FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data Processing to Every Language

TL;DR

FineWeb2 introduces an adaptable pre-training data pipeline that efficiently creates high-quality multilingual datasets, significantly improving model performance across diverse languages and scaling to over 1000 languages.

Contribution

The paper presents a novel, adaptable data curation pipeline for multilingual pre-training, along with a large-scale dataset and evaluation methods for diverse languages.

Findings

Pipeline improves multilingual model performance

Dataset scaling to over 1000 languages

Enhanced data balancing and deduplication methods

Abstract

Pre-training state-of-the-art large language models (LLMs) requires vast amounts of clean and diverse text data. While the open development of large high-quality English pre-training datasets has seen substantial recent progress, training performant multilingual LLMs remains a challenge, in large part due to the inherent difficulty of tailoring filtering and deduplication pipelines to a large number of languages. In this work, we introduce a new pre-training dataset curation pipeline based on FineWeb that can be automatically adapted to support any language. We extensively ablate our pipeline design choices on a set of nine diverse languages, guided by a set of meaningful and informative evaluation tasks that were chosen through a novel selection process based on measurable criteria. Ultimately, we show that our pipeline can be used to create non-English corpora that produce more performant models than prior datasets. We additionally introduce a straightforward and principled approach to rebalance datasets that takes into consideration both duplication count and quality, providing an additional performance uplift. Finally, we scale our pipeline to over 1000 languages using almost 100 Common Crawl snapshots to produce FineWeb2, a new 20 terabyte (5 billion document) multilingual dataset which we release along with our pipeline, training, and evaluation codebases.