BGE-M3: multilingual embeddings for retrieval and RAG

One embedding model that does three kinds of retrieval: dense vectors, sparse lexical weights and multi-vector (ColBERT-style) matching. That makes it a good fit for hybrid search and RAG pipelines.

• Handles inputs up to 8,192 tokens
• Covers 100+ languages, including cross-lingual search

Summary written by SolxLab. Weights and code belong to their authors and are distributed under the license shown; always check the official page before use.