[Models](/models)/sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2

Apache-2.0

# paraphrase-multilingual-MiniLM-L12-v2

by sentence-transformers

Lightweight multilingual sentence embeddings across 50+ languages

118Mparams

[HuggingFace](https://huggingface.co/sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2)[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)

Identifiers

Model ID

`sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2`

## Deploy paraphrase-multilingual-MiniLM-L12-v2

Single-tenant

Mixpeek has no managed extractor for this model. On a single-tenant deployment you upload the weights and a custom plugin serves them next to the rest of your pipeline.

[Read the customer-hosted guide](https://mixpeek.com/docs/customer-hosted/overview) [Talk to us about single-tenant](/demo)

## Overview

This Sentence-Transformers model maps sentences and short passages from 50+ languages into a shared 384-dimensional space, so semantically similar text clusters together regardless of language. It is small, fast, and CPU-friendly, which makes it a popular default for multilingual semantic search and clustering when latency and cost matter more than top-of-leaderboard accuracy.

On Mixpeek, it is a text embedding extractor for metadata, transcripts, captions, and document chunks: useful when content spans many languages and you need cheap, fast recall.

## Architecture

12-layer multilingual MiniLM encoder fine-tuned with a paraphrase/contrastive objective and mean-pooled to a 384-dim sentence embedding. Distilled for speed, so it runs well on CPU.

## Mixpeek SDK Integration

IngestRetrieve

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so paraphrase-multilingual-MiniLM-L12-v2 runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace\_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns\_your\_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API\_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection\_id: "col\_your\_collection",
      documents: \[
        {
          document\_id: "asset-00412",
          // The vector name has to match a vector index on the collection.
          vectors: { "text-embedding": yourVector },
          payload: { source\_key: "archive/2026/asset-00412" },
        },
      \],
    }),
  },
);

// Managed alternative, if this exact model is not the requirement:
// text\_extractor@v1 runs intfloat/multilingual-e5-large-instruct
// (1024-d) over a bucket, with no inference of your own.

## Capabilities

*   384-dim multilingual sentence embeddings (50+ languages)
*   Cross-lingual semantic similarity and clustering
*   Fast, small, CPU-friendly
*   Drop-in Sentence-Transformers API

## Use Cases on Mixpeek

Multilingual semantic search over metadata and transcripts

Clustering and dedup of multilingual text at low cost

Cheap first-stage recall before a reranker

Tagging and routing of international support content

## Performance

Input SizeUp to 128 tokens

Embedding Dim384

GPU Latency~3ms / sentence (A100); CPU-viable

GPU Throughput~3000 sentences/sec (A100, batch 128)

GPU Memory~0.5 GB

Pick a larger multilingual model (bge-m3, multilingual-e5-large) when accuracy outranks latency

## Common Pipeline Companions

[

BAAI/bge-m3

Higher-accuracy multilingual upgrade

](/model/BAAI/bge-m3)[

Qwen/Qwen3-Reranker-8B

Rerank recall for precision

](/model/Qwen/Qwen3-Reranker-8B)

## Explore on Mixpeek

[

Browse all models

Search the full model catalog

](/models)[

Feature Extractors

See how Mixpeek runs models as extractors

](/extractors)[

Mixpeek Vector Store

Store & search embeddings at scale

](/mvs)[

Pricing

Usage-based pricing for pipelines

](/pricing)[

All Curated Lists

Compare models, APIs & infrastructure

](/curated-lists)

### Specification

Organizationsentence-transformers

Retriever\-

Parameters118M

LicenseApache-2.0

Downloads/moN/A

[

### Research Paper

Sentence-BERT / Making Monolingual Sentence Embeddings Multilingual

arxiv.org](https://huggingface.co/sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2)

### Build a pipeline with paraphrase-multilingual-MiniLM-L12-v2

Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.

[Run it on your own data, free](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=model_page&utm_campaign=mvs_onramp)