Announcement
EmbeddingGemma 2 maps text, images, video, audio and code on-device
Google Gemma says the 740-million-parameter model can support multimodal apps running entirely in browsers via WebGPU and Transformers.js.
TLDR
Google Gemma says EmbeddingGemma 2 maps text, images, video, audio and code while running locally on a device. It says developers can use it for multimodal search, classification and clustering in browser apps. A separate post says Google released a paper on the model under an Apache 2.0 license and reports that the full model takes about 567MB on a Pixel 11 Pro with 4-bit weights.
Combined views
24.8K
4 Sources, first seen ago
