bge-multilingual-gemma2

Name: bge-multilingual-gemma2
Author: BAAI

by BAAI

LLM-based multilingual text embedding: strong cross-lingual retrieval across 100+ languages

9B (Gemma 2 backbone)params

HuggingFace Run it on your own data, free

Identifiers

Model ID

BAAI/bge-multilingual-gemma2

Feature URI

Overview

BGE-Multilingual-Gemma2 is an embedding model built on the Gemma 2 decoder, trained with contrastive learning over a broad multilingual and cross-lingual corpus. Unlike encoder-only embedders, it inherits the wide language coverage and instruction-following of an LLM backbone, which makes it a common default when a corpus spans many languages or when queries and documents are in different languages.

On Mixpeek, BGE-Multilingual-Gemma2 is a text embedding extractor for multilingual document and metadata search. It pairs well with a reranker for precision and with a multimodal encoder when text is only one of several modalities an agent must search.

Architecture

Decoder-based (Gemma 2 9B) embedding model. The final hidden state is pooled into a dense sentence embedding and trained with contrastive (InfoNCE) loss over multilingual query-document pairs, including cross-lingual positives so a query in one language retrieves documents in another.

Mixpeek SDK Integration

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so bge-multilingual-gemma2 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

Dense text embeddings across 100+ languages
Strong cross-lingual retrieval (query and document in different languages)
Instruction-aware query encoding inherited from the LLM backbone
Long-context passages for document-level retrieval

Use Cases on Mixpeek

Multilingual document and metadata search over a global content library

Cross-lingual retrieval where an English query must surface non-English documents

Agent perception over international support tickets, contracts, or transcripts

First-stage recall feeding a reranker for precision

Performance

Input SizeUp to 8192 tokens

GPU Latency~28ms / passage (A100, 9B backbone)

GPU Throughput~250 passages/sec (A100, batch 32)

GPU Memory~18 GB (bf16)

Heavier than encoder embedders: reserve for genuinely multilingual/cross-lingual corpora; use BGE-M3 or gte-modernbert when English-dominant and latency-sensitive