Homelab

Google's Gemma 4 12B is a fully open-weights, encoder-free multimodal model that handles text, images, and audio — and fits inside 16GB of VRAM. Here's what makes the architecture interesting and whether it changes anything for local AI workflows.