Multilingual E5 Text Embeddings: A Technical Report

TL;DR

This report details the development and evaluation of open-source multilingual E5 text embedding models, highlighting their training process, different sizes, and competitive performance with state-of-the-art models.

Contribution

It introduces a new multilingual E5 embedding model family with multiple sizes and an instruction-tuned variant, based on contrastive pre-training and fine-tuning.

Findings

Models achieve competitive performance with state-of-the-art English models

Three sizes offer a trade-off between efficiency and quality

Open-source release facilitates broader adoption and research

Abstract

This technical report presents the training methodology and evaluation results of the open-source multilingual E5 text embedding models, released in mid-2023. Three embedding models of different sizes (small / base / large) are provided, offering a balance between the inference efficiency and embedding quality. The training procedure adheres to the English E5 model recipe, involving contrastive pre-training on 1 billion multilingual text pairs, followed by fine-tuning on a combination of labeled datasets. Additionally, we introduce a new instruction-tuned embedding model, whose performance is on par with state-of-the-art, English-only models of similar sizes. Information regarding the model release can be found at https://github.com/microsoft/unilm/tree/master/e5 .