Mmastodon TechnologyAI first seen 18 h ago, last 18 h ago, peak #3

EmbeddingGemma 2: an open-source model indexing text, code, images and audio locally

Original: EmbeddingGemma 2 : un modèle d'embeddings open source qui indexe vos textes, code, images, vidéos et audios dans un même

A new open-source embedding model called EmbeddingGemma 2 is drawing attention for its ability to index text, code, images, video and audio within a single shared space, all locally. It has 740 million parameters, is released under the Apache 2.0 licence, and can run with 270 million parameters for text only. A quantised version reportedly fits in roughly 191 MB of RAM on a Pixel 11 Pro smartphone, making on-device search practical.

Why now: People are discussing a lightweight open-source multimodal embedding model that runs entirely on-device, which is notable for privacy and local AI applications.

EmbeddingGemma 2GooglePixel 11 Pro

Open on mastodon →

Evidence

• EmbeddingGemma 2 : un modèle d'embeddings open source qui indexe vos textes, code, images, vidéos et audios dans un même espace, en local. 740M de paramètres, licence Apache 2.0, dès 270M en texte seul. Il tient dans ~191 Mo de RAM sur un Pixel 11 Pro en version quantifiée. ⬇️… · camilleroux · 1

API: https://socialmediatrends-api.osmike.com/v1/trends/2104936