UEmbed
Alibaba Tongyi Lab🇨🇳Research
Embeddingopen
UEmbed is a decoder-only multimodal embedding model that produces both dense embeddings and SPLADE-style sparse lexical embeddings from a single causal forward pass. It supports text, image, video, and mixed-modal inputs for retrieval, multimodal search, and visual-document retrieval.
3 variants9B4B2B