EmbeddingGemma 2 Puts Multimodal AI Search in Your Pocket
Google DeepMind has just released a new AI model that could make it much easier to search through photos, videos, audio, and text without sending any of that data to the cloud. Called EmbeddingGemma 2, the open-weight model was designed specifically for edge devices, including smartphones and computers with relatively limited memory.
EmbeddingGemma 2 is a 740-million-parameter multimodal embedding model. Rather than generating text like a conventional large language model, it converts information into numerical vectors that represent its meaning. Those vectors can then be compared to determine how closely different pieces of information are related.
It is possible for the model to do this across several different types of media. Text, images, video frames, and audio are all mapped into the same vector space, allowing a text description to be compared directly against a photograph or moment in a video.
Built to run locally
Google designed EmbeddingGemma 2 with...
Copyright of this story solely belongs to www.hackster.io. To see the full text click HERE
