Saved in:
Bibliographic Details
Main Authors: Warner, Benjamin, Chaffin, Antoine, Clavié, Benjamin, Weller, Orion, Hallström, Oskar, Taghadouini, Said, Gallagher, Alexis, Biswas, Raja, Ladhak, Faisal, Aarsen, Tom, Cooper, Nathan, Adams, Griffin, Howard, Jeremy, Poli, Iacopo
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2412.13663
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913618061164544
author Warner, Benjamin
Chaffin, Antoine
Clavié, Benjamin
Weller, Orion
Hallström, Oskar
Taghadouini, Said
Gallagher, Alexis
Biswas, Raja
Ladhak, Faisal
Aarsen, Tom
Cooper, Nathan
Adams, Griffin
Howard, Jeremy
Poli, Iacopo
author_facet Warner, Benjamin
Chaffin, Antoine
Clavié, Benjamin
Weller, Orion
Hallström, Oskar
Taghadouini, Said
Gallagher, Alexis
Biswas, Raja
Ladhak, Faisal
Aarsen, Tom
Cooper, Nathan
Adams, Griffin
Howard, Jeremy
Poli, Iacopo
contents Encoder-only transformer models such as BERT offer a great performance-size tradeoff for retrieval and classification tasks with respect to larger decoder-only models. Despite being the workhorse of numerous production pipelines, there have been limited Pareto improvements to BERT since its release. In this paper, we introduce ModernBERT, bringing modern model optimizations to encoder-only models and representing a major Pareto improvement over older encoders. Trained on 2 trillion tokens with a native 8192 sequence length, ModernBERT models exhibit state-of-the-art results on a large pool of evaluations encompassing diverse classification tasks and both single and multi-vector retrieval on different domains (including code). In addition to strong downstream performance, ModernBERT is also the most speed and memory efficient encoder and is designed for inference on common GPUs.
format Preprint
id arxiv_https___arxiv_org_abs_2412_13663
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference
Warner, Benjamin
Chaffin, Antoine
Clavié, Benjamin
Weller, Orion
Hallström, Oskar
Taghadouini, Said
Gallagher, Alexis
Biswas, Raja
Ladhak, Faisal
Aarsen, Tom
Cooper, Nathan
Adams, Griffin
Howard, Jeremy
Poli, Iacopo
Computation and Language
Artificial Intelligence
Encoder-only transformer models such as BERT offer a great performance-size tradeoff for retrieval and classification tasks with respect to larger decoder-only models. Despite being the workhorse of numerous production pipelines, there have been limited Pareto improvements to BERT since its release. In this paper, we introduce ModernBERT, bringing modern model optimizations to encoder-only models and representing a major Pareto improvement over older encoders. Trained on 2 trillion tokens with a native 8192 sequence length, ModernBERT models exhibit state-of-the-art results on a large pool of evaluations encompassing diverse classification tasks and both single and multi-vector retrieval on different domains (including code). In addition to strong downstream performance, ModernBERT is also the most speed and memory efficient encoder and is designed for inference on common GPUs.
title Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2412.13663