Discover how to deploy and run Gemma 4 12B and DiffusionGemma directly on standard 16GB VRAM hardware. We will demonstrate how to leverage a unified, encoder-free architecture to natively process text, image, and audio inputs locally while using multi-token prediction to maximize generation speeds. Finally, you'll learn how to utilize the new Gemma Skills Repository to build secure, offline tool-calling agents that analyze massive datasets within a 256K context window.

RESOURCES FROM THE SESSION:

• Codelab: https://goo.gle/wbh-gemma4-cloudrun-labs

• huggingface.co/spaces/lgusm/gemma4vramestimator

• github.com/google-gemma/cookbook

• newsletter.maartengrootendorst.com/p/a-visual-guide-to-gemma-4

• newsletter.maartengrootendorst.com/p/a-visual-guide-to-gemma-4-12b