Embeddings · Microsoft (intfloat)
Run multilingual-e5-large-instruct embeddings in the EU through an OpenAI-compatible API
multilingual-e5-large-instruct is a compact embedding model from Microsoft researchers that turns short texts in about 100 languages into 1024-dimension vectors. It takes a task instruction on the query side, which helps it adapt to search, classification and clustering. On inference.eu you call it through the OpenAI embeddings format, processed in EU data centres and never used for training.
- 01560M
- 02EU data centres
- 03MIT
What multilingual-e5-large-instruct is
multilingual-e5-large-instruct is part of the E5 family of text embedding models published by Microsoft researchers under the intfloat account. It was released in February 2024 under the MIT licence. It is initialised from XLM-RoBERTa large, with 24 layers, about 560 million parameters and an embedding size of 1024.
The model was trained on a large multilingual mix of text pairs and then fine-tuned with task instructions. It supports the 100 languages of XLM-RoBERTa, although quality is lower for languages with little training data. Inputs are limited to 512 tokens, so it is designed for sentences, paragraphs and short passages rather than whole documents.
What it does well, and where it falls short
The model is small, fast and well understood, which makes it a good fit for large indexes of short texts: product descriptions, FAQ entries, support tickets and paragraphs of policies. Because queries carry an instruction, the same model can be steered towards retrieval, semantic similarity, classification or clustering without fine-tuning.
Its main limit is the 512-token input, which forces you to chunk long documents into short passages. Newer models such as bge-m3 accept 8192 tokens and Qwen3 Embedding 8B accepts 32K, and the larger Qwen model usually retrieves better on hard queries. Forgetting the query instruction also reduces quality noticeably, so it must be part of your integration.
Calling multilingual-e5-large-instruct through the API
inference.eu uses the OpenAI request format, so OpenAI SDKs, LangChain, LlamaIndex and vector database integrations that call /v1/embeddings work unchanged. Set the base URL to the inference.eu API, use your inference.eu key and set the model to multilingual-e5-large-instruct.
Each query needs a one-sentence instruction in the form Instruct: {task} followed by a new line and Query: {text}, for example Instruct: Given a web search query, retrieve relevant passages that answer the query. Documents are embedded as plain text, without any instruction. Keep passages under 512 tokens, since longer input is truncated.
Data residency for multilingual-e5-large-instruct
Indexing your content means sending every passage to the embedding model. On inference.eu, multilingual-e5-large-instruct runs on infrastructure in the European Union, so your documents and queries are processed in EU data centres and are never used to train any model.
Every customer receives a data processing agreement under Article 28 of the GDPR and our subprocessor list, which makes it straightforward to document your search or RAG pipeline in your records of processing. Billing is in euros, on the same invoice as the chat models that use the retrieved passages. The MIT licence allows commercial use with few conditions.
import os
from openai import OpenAI
client = OpenAI(base_url="https://api.inference.eu/v1", api_key=os.environ["INFERENCE_EU_API_KEY"])
result = client.embeddings.create(
model="multilingual-e5-large-instruct",
input=["Where is my data processed?"],
)
print(len(result.data[0].embedding))What teams use it for
- 1
FAQ and help-centre search
Match customer questions in many languages to the right help article or FAQ entry.
- 2
Classification of short texts
Sort tickets, reviews or form submissions into categories with an instruction that describes the task.
- 3
Semantic similarity
Find similar products, duplicate questions or related paragraphs across a large catalogue.
multilingual-e5-large-instruct: common questions
The questions buyers, developers and legal teams ask us most. Something missing? Ask us directly.
Related
- BGE-M3Run bge-m3 on inference.eu: a compact multilingual embedding model for search and RAG, 100+ languages and 8192-token inputs, processed in the EU.Embeddings
- Qwen3 Embedding 8BRun Qwen3 Embedding 8B on inference.eu: multilingual text embeddings for search and RAG, up to 4096 dimensions and 32K tokens, processed in the EU.Embeddings
- BGE Reranker v2 M3Run bge-reranker-v2-m3 on inference.eu: a light multilingual reranker that sorts search and RAG results by relevance, processed in the EU.Reranking
Move your AI workloads to EU infrastructure
Tell us which models you use and roughly how much traffic you have. We reply within one working day with pricing and next steps.
- Reply within one working day
- Pricing for your volume
- DPA ready to sign