Embedić Released: A Suite of Serbian Text Embedding Models Optimized for Information Retrieval and RAG

We are a community of AI/ ML/Generative AI enthusiasts/researchers/journalists/writers who share interesting news and articles about the applications of AI. 

Machine Learning News

Novak Zivanic has released Embedić, a suite of state-of-the-art Serbian text embedding models designed for Information Retrieval and Retrieval-Augmented Generation tasks. The models, optimized from multilingual-e5, come in three sizes (small, base, large) and exhibit high efficiency, with the smallest model outperforming previous benchmarks while using significantly fewer parameters. They support both Serbian (Cyrillic and Latin scripts) and English, making them versatile for cross-lingual tasks such as clustering and semantic search. The models underwent rigorous evaluation with comprehensive Serbian datasets to ensure their effectiveness.