Key Info
Google DeepMind introduced EmbeddingGemma 2, described as its first natively multimodal open model built for on-device embeddings. It extends embeddings beyond text to unify code, images, audio, and video within a single shared space.
Highlights
- First natively multimodal open model from Google DeepMind focused on on-device embeddings
- Unifies code, images, audio, and video embeddings in one shared space
- Extends beyond text-only embedding approaches
- Distributed as an open model, enabling on-device use cases