With the introduction of its embedding, Google provides a multilingual text inlaid model designed to run directly on mobile phones, laptops and other edge devices for mobile generative AI.
Presented September 4IngredingGemma presents a design of 308 million parameters that allows developers to build applications using techniques such as RAG (aquatic recovery generation) and semantic search that will be executed directly in the directed hardware, Google explained. Based on the architecture of light models of Gemma 3, IncedDinggemma is trained in more than 100 languages and is small enough to function with less than 200 MB of RAM with quantization. Customizable exit dimensions are presented, ranging from 768 dimensions to 128 dimensions through Representation of Matryoshka and a token 2k context window.
IngredingGemma empowers developers to build applications in disposition, flexible and privacy, according to Google. You can download model weights to increase Hugged face, Kaggyand VERTEX AI. Working with him Gemma 3n Model, incredpinggemma can unlock new use cases for mobile rag pipes, semantic search and more, Google said. IncreddingGemma works with tools such as prayers-transformators, flame.cpp, mlx, ollama, litert, transformers.js, lmstudio, weaviate, cloud, flame index and langchain. Documentation for increddingMma can be found in ai.google.dev.
#Google #IntroductionGemma #device
