unsloth/gemma-4-26B-A4B-it-qat-GGUF · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
AI Model Release
Summary
Google's full release of Gemma 3n, a mobile-first multimodal AI model designed for on-device deployment. The developer guide details architectural innovations like MatFormer, Per-Layer Embeddings, and KV Cache Sharing that enable 5B and 8B parameter models to run with memory footprints comparable to 2B and 4B models.
Highlights
- MatFormer architecture provides compute flexibility, while Per-Layer Embeddings (PLE) and KV Cache Sharing dramatically reduce memory footprint
- Natively multimodal: supports image, audio, video, and text inputs with text outputs
- Two model sizes (E2B and E4B) operate with as little as 2GB and 3GB of memory respectively
- Supports 140 languages for text and 35 languages for multimodal understanding
- Broad ecosystem support including Hugging Face Transformers, llama.cpp, Ollama, MLX, and Google AI Edge
auto-generated
Context
Audience
Developers building on-device or edge AI applications, mobile app developers seeking multimodal capabilities, and ML engineers interested in efficient model architectures
DomainOn-Device Machine Learning
Formatdeveloper guide / technical blog post
Accessfree online, open source model weights
Source Overview
The Gemma 3n model has been fully released, building on the success of previous Gemma models and bringing advanced on-device multimodal capabilities to edge devices with unprecedented performance. Explore Gemma 3n’s innovations, including its mobile-first architecture, MatFormer technology, Per-Layer Embeddings, KV Cache Sharing, and new audio and MobileNet-V5 vision encoders, and how developers can start building with it today.
Topics
Related
Gemma 3MatFormerPer-Layer EmbeddingsMobileNet-V5llama.cppOllama MLX backend for on-device inference
