Google Gemma releases EmbeddingGemma 2, a 740M-parameter multimodal embedding model

AIGoogle Gemma announces EmbeddingGemma 2, which maps text, images, video, audio, and code into embeddings. The model has 740M parameters and runs 100% locally on the device, including in the browser via WebGPU and Transformers.js.

AIWhy it matters.

The post gives the model's size and browser-based runtime path, which show how multimodal embeddings can run locally without a server.

Google GemmaOfficial

@googlegemma

View on X

EmbeddingGemma 2 maps text, images, video, audio, and code, running 100% locally on your device.

At just 740M parameters, you can build powerful multimodal apps running entirely in your browser via WebGPU & Transformers.js. From multimodal search to cross-modal classification and clustering, local embeddings unlock endless possibilities.

Open source

Source details

Google Gemma · x.comPublished Oct 29, 12:54 PM ET · added here 1:13 PM ET

Original title: EmbeddingGemma 2 maps text, images, video, audio, and code, running 100% locally on your device.

Topics

Google / Gemini Multimodal AI On-device AI Open-source ecosystem Model releases