EmbeddingGemma 2: 740 million parameters, open multimodal embedding model
Open-source EmbeddingGemma 2 (740 million parameters, Apache 2.0) maps text, code, images, audio, and video into a shared embedding space for on-device semantic search and retrieval.
What changed
Google DeepMind released EmbeddingGemma 2, a 740 million-parameter Apache 2.0 multimodal embedding model for on-device inference.
Why it matters
Developers can run cross-modal semantic search and local RAG to improve data privacy and reduce pipeline latency by generating embeddings on-device.
Who should care
DevelopersResearchersEnterprises
Worth trying
Yes — Available on Hugging Face and Kaggle; optimized for on-device use.