BGE Multilingual Gemma2
bge-multilingual-gemma2
About
BGE Multilingual Gemma2 is BAAI's large multilingual embedding model built on Google's Gemma 2 9B decoder architecture (42 layers, 3584 hidden dim). It supports instruction-based encoding for retrieval tasks and achieves state-of-the-art results on MIRACL, MTEB-pl, and MTEB-fr benchmarks, with strong performance across MTEB, C-MTEB, and AIR-Bench. It covers diverse languages including English, Chinese, Japanese, Korean, and French, and supports a 4,096-token context window.
BGE Multilingual Gemma2 has a 4K-token context window.
Capabilities
VisionMultimodalReasoningFunction CallingTool UseStructured OutputsCode Execution
