🤖EmbeddingGemma 2: Compact Multimodal Model for On-Device Search

Efficient On-Device Search for Mobile and Laptops

EmbeddingGemma 2 is a compact, multimodal model for on-device search, designed to run efficiently on consumer hardware like mobile devices and laptops. This model delivers low-latency semantic representations for on-device applications, making semantic search and retrieval-augmented generation more efficient. Developers can now build retrieval systems that search across different media types without maintaining separate embedding spaces. The model supports a wide range of retrieval tasks, including finding relevant documents, screenshots, and recordings in response to a query. It also supports searching source code using natural language and retrieving video content based on textual descriptions. EmbeddingGemma 2 significantly outperforms its predecessor on code search and technical retrieval, with 270 million parameters for text and code and 740 million for full multimodal support.