BGE Embeddings
An embedding model family supports semantic search and RAG with multilingual retrieval and reranking through the MIT-licensed FlagEmbedding toolkit.
BGE Embeddings is a family of models for developers who need to find relevant passages for search or retrieval-augmented generation (RAG). It includes embedding models for finding candidate text and rerankers for assessing how closely those results match a query.
- Multilingual search: BAAI/bge-m3 combines dense, sparse and ColBERT-style multi-vector retrieval and accepts long text inputs.
- Instruction-based retrieval: BAAI/bge-en-icl uses task instructions and example pairs for English embeddings; BAAI/bge-multilingual-gemma2 supports instructions across languages and tasks.
- Result reranking: BAAI/bge-reranker-v2-m3 and BAAI/bge-reranker-v2-gemma score retrieved documents against a query.
- Adjustable inference work: BAAI/bge-reranker-v2-minicpm-layerwise allows output layer selection, giving developers a way to reduce the computation used for reranking.
Developed by the Beijing Academy of Artificial Intelligence, BGE works with the open-source Python toolkit FlagEmbedding for inference, evaluation and fine-tuning. The toolkit uses the MIT license.
