UniME is a series of multimodal large language models trained for learning universal multimodal embedding.