🗂️ AI & Agents · View mindmap

EmbeddingGemma 2

EmbeddingGemma 2 is a compact, open-source multimodal embedding model developed by Google, designed to facilitate efficient Retrieval Augmented Generation (RAG) workflows, particularly in on-device scenarios.

Key Specifications

• Parameters: 740 million
• License: Apache 2.0
• Capabilities: Unified cross-modal embedding space mapping diverse data types:
  • Text
  • Code
  • Images
  • Video
  • Audio

Core Features

• Unified Cross-Modal Embeddings: Maps heterogeneous data types into a single vector space, enabling semantic search across modalities.
• On-Device Optimization: Designed for efficiency, allowing deployment on edge devices with limited computational resources.
• Multimodal RAG: Enhances RAG pipelines by allowing retrieval from mixed-media corpora using a single embedding model.

References

• EmbeddingGemma 2: On-Device Multimodal RAG with Unified Cross-Modal Embeddings
• Embedding Gemma 2: On-Device Multimodal RAG Made Easy by Prompt Engineering