EmbeddingGemma 2: One On-Device Model for Multimodal Embeddings
Melvin Vivas · X video post · 2026-10-07 · 0:53 · 1,806 views · Open on X
Topics: Embeddings & Vector Databases, Retrieval-Augmented Generation (RAG), Industry Trends & Job Market · Level: intermediate
Summary
Melvin Vivas shares Google's announcement of EmbeddingGemma 2. It is a compact (740M-parameter) open embedding model built for on-device and edge apps. It maps text, images, video and audio into one shared embedding space, so a single model can handle cross-modal search, retrieval, classification and clustering. Before this, each data type needed its own model.
Key points
- EmbeddingGemma 2 is a lightweight open model with 740M parameters, built for on-device and edge apps.
- The first-generation EmbeddingGemma embedded text only. Version 2 natively supports multiple types of data (multimodal).
- Cross-modal retrieval used to need separate models for text, vision and audio. EmbeddingGemma 2 puts text, images, video, audio, or any mix of them, into one shared embedding space.
- Example: a picture of a cat, the word 'cat' and the sound of a cat all land close together in the shared embedding space.
- One model can power search, retrieval, classification and clustering across different data types.
- The model weights are on Hugging Face, and starter notebooks are in the Gemma Cookbook.
Resources mentioned
- EmbeddingGemma 2 · tool · huggingface.co · free
Google's open 740M-parameter multimodal embedding model that puts text, images, video and audio into one shared vector space for on-device use.
Also in: Google AI Edge Foresight: On-Device Meeting Notes with EmbeddingGemma 2 (Melvin Vivas on X · notes), Try EmbeddingGemma 2 multimodal embeddings in Colab (Melvin Vivas on X · notes), A day of work done from a phone with Grok Bot agents (Melvin Vivas on X · notes), EmbeddingGemma 2 Colab Notebook for FAQ Semantic Search (Melvin Vivas on X · notes) and 2 more - Google Gemma (@googlegemma) on X · person · x.com · free
The official Google Gemma account on X, which posts news about Gemma model releases.
Also in: Running Gemma 4 Models Offline on an iPhone (Melvin Vivas on X · notes), Running Gemma4-E2B tool calling locally on an iPhone (Melvin Vivas on X · notes) - Hugging Face · website · huggingface.co · free · recommended by both Bashiri Smith & Melvin Vivas
Named in the prompt as a beginner-friendly but powerful AI platform for models used in AI projects.
Also in: llama.app: Run Local Models on a Mac Without the llama.cpp CLI (Melvin Vivas on X · notes), Grok Bot workflow: turn model-release posts into Colab notebooks (Melvin Vivas on X · notes), Generate Portfolio AI Project Ideas with Claude and Hugging Face Tasks (Bashiri Smith on Facebook · notes), EmbeddingGemma 2 released on Hugging Face (Melvin Vivas on X · notes) and 48 more - Gemma Cookbook · repo · github.com · free
Google's collection of example notebooks for building with Gemma models, including EmbeddingGemma.
Try this
- Download the EmbeddingGemma 2 model weights from Hugging Face.
- Work through the EmbeddingGemma notebooks in the Gemma Cookbook to get started.
- Build cross-modal search where a text query finds matching images, video or audio, all using one embedding model.
- Build on-device classification or clustering of mixed media (text, images, audio) with EmbeddingGemma 2.
More in Embeddings & Vector Databases
- Try EmbeddingGemma 2 multimodal embeddings in Colab
- EmbeddingGemma 2 Colab Notebook for FAQ Semantic Search
- EmbeddingGemma 2 released on Hugging Face
- EmbeddingGemma 2: open multimodal embeddings model for on-device use
- Liquid AI LFM 2.5 Encoder: a CPU-friendly encoder model
- How RAG Finds the Right Document Fast: Graph-Based Vector Search