Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Estève, Louis, Servan, Christophe, Lavergne, Thomas, Savary, Agata
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2602.22014
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911468337758208
author Estève, Louis
Servan, Christophe
Lavergne, Thomas
Savary, Agata
author_facet Estève, Louis
Servan, Christophe
Lavergne, Thomas
Savary, Agata
contents Diversity has been gaining interest in the NLP community in recent years. At the same time, state-of-the-art transformer models such as ModernBERT use very large pre-training datasets, which are driven by size rather than by diversity. This summons for an investigation of the impact of diversity on the ModernBERT pre-training. We do so in this study, with the express intent of reducing pre-training dataset size, while retaining at least comparable performance. We compare diversity-driven sampling algorithms, so as to pick the best one. We find that diversity-driven sampling allows in some tasks to gain 10 points relative to randomly-sampled pre-training data of commensurate size. We also see that a model pre-trained for 483h on a diversity-driven dataset of 150M tokens can yield a commensurate performance to a model pre-trained for 1,775h on a randomly-driven dataset of 2.4B tokens.
format Preprint
id arxiv_https___arxiv_org_abs_2602_22014
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle A Diversity Diet for a Healthier Model: A Case Study of French ModernBERT
Estève, Louis
Servan, Christophe
Lavergne, Thomas
Savary, Agata
Computation and Language
Diversity has been gaining interest in the NLP community in recent years. At the same time, state-of-the-art transformer models such as ModernBERT use very large pre-training datasets, which are driven by size rather than by diversity. This summons for an investigation of the impact of diversity on the ModernBERT pre-training. We do so in this study, with the express intent of reducing pre-training dataset size, while retaining at least comparable performance. We compare diversity-driven sampling algorithms, so as to pick the best one. We find that diversity-driven sampling allows in some tasks to gain 10 points relative to randomly-sampled pre-training data of commensurate size. We also see that a model pre-trained for 483h on a diversity-driven dataset of 150M tokens can yield a commensurate performance to a model pre-trained for 1,775h on a randomly-driven dataset of 2.4B tokens.
title A Diversity Diet for a Healthier Model: A Case Study of French ModernBERT
topic Computation and Language
url https://arxiv.org/abs/2602.22014