granite-embedding 278m for RAG

A high-efficiency 278-million parameter model optimized for enterprise RAG pipelines and multilingual document retrieval.

IBM's Granite 278M embedding model delivers a precise balance of performance and speed for Retrieval-Augmented Generation (RAG). It supports a 512-token context window and handles 15 languages (including English, German, and Japanese) to ensure global utility. By mapping text to 768-dimensional vectors, it enables sub-millisecond similarity searches across massive document stores. This model is specifically tuned for business use cases where low latency and high retrieval accuracy are non-negotiable requirements for LLM grounding.

https://huggingface.co/ibm-granite/granite-embedding-278m-multilingual