Announcing the full release of Gemma 3n, bringing powerful multimodal capabilities to edge devices for developers š ā https://developers.googleblog....
Gemma 3n delivers quality improvements across multilinguality (supporting 140 languages for text and 35 for multimodal understanding), math, coding, and reasoning. It is the first model under 10B parameters with an @lmarena_ai score above 1300.
Gemma 3n models are available in both effective 2B and 4B sizes, with Per-Layer Embeddings (PLEs) improving quality while reducing the VRAM footprint compared to similar models.
Dive into the new foundation of Gemma 3nās unique mobile-first architecture and learn how it all kicks off with MatFormer.
Download model weights on @kaggle and @huggingface: https://huggingface.co/collect... And use your preferred tools and frameworks like @ollama, llama.cpp, @unslothai, @lmstudio, and others.
Start experimenting with Gemma 3n on Google AI Studio today: https://aistudio.google.com/pr...