BGE-M3: Advanced Multilingual Text Retrieval Model

Originally published at: NVIDIA NIM | bge-m3

Experience the versatile embedding model designed for multilingual, multi-functional, and multi-granularity text retrieval tasks, excelling in dense, multi-vector, and sparse retrieval for inputs from short sentences to long documents.

Hi,
does anyone know how to generate “sparse” vectors for which the BGE-M3 model is perfect for? I assume that the api returns only dense vectors.

Thank you in advance.