wemm-embedding

Definition

A name for a multimodal embedding model in the WeChat line of work. Embedding models exist to map images and text into fixed-length vectors for retrieval, recommendation, and deduplication, and they are judged on representation quality and retrieval accuracy rather than on generation. Their output is not human-readable text but coordinates in a vector space, which is what makes nearest-neighbor search over hundreds of millions of items practical.

Explainers using this term

• Paper Walkthrough: WeMM-Embedding — Putting Text, Images and Video on One RulerWeMM-Embedding: WeChat Multi-Modal Embedding Technical Report