# multilingual-e5-large

> Microsoft's state-of-the-art multilingual text embedding model supporting 100 languages with 1024-dimensional embeddings, trained on 1 billion multilingual text pairs for robust cross-lingual retrieval.

- **URL:** https://huggingface.co/intfloat/multilingual-e5-large
- **Category:** Machine Learning Models
- **Tags:** multilingual, embedding, microsoft
- **Updated:** 2026-03-14 12:39
- **Canonical page:** https://vectordb.ever.works/items/multilingual-e5-large

## Details

## Overview

The multilingual-e5-large model is a sophisticated embedding model developed at Microsoft, supporting 100 languages from xlm-roberta. It's designed for robust text representation across diverse languages and tasks.

## Model Specifications

- **Architecture**: 24 layers based on XLM-RoBERTa-large
- **Embedding Size**: 1024 dimensions
- **560M Parameters**: Optimal balance of performance and efficiency
- **Multilingual Support**: 100 languages

## Training Methodology

The training procedure adheres to the English E5 model recipe:
1. Contrastive pre-training on 1 billion multilingual text pairs
2. Fine-tuning on a combination of labeled datasets

## Performance

- Achieves 51.4 on BEIR benchmark
- Strong cross-lingual retrieval capabilities
- Robust performance across various text representation tasks

## Model Family

The E5 family includes:
- **multilingual-e5-small**: 12 layers, 384 dimensions
- **multilingual-e5-base**: 12 layers, 768 dimensions
- **multilingual-e5-large**: 24 layers, 1024 dimensions (this model)
- **multilingual-e5-large-instruct**: Instruction-tuned version with 52.5 BEIR score

## Use Cases

- Information retrieval
- Semantic textual similarity
- Text reranking
- Cross-lingual search
- Document classification
- Clustering

## Resources

- Hugging Face: intfloat/multilingual-e5-large
- GitHub: microsoft/unilm/e5
- Technical Report: arXiv:2402.05672

## Pricing

Free and open-source model available on Hugging Face.

---

_This Markdown mirror is generated by Ever Works for AI agents. The canonical HTML page is at https://vectordb.ever.works/items/multilingual-e5-large._