Text-only versions of gemma-4 without the vision encoders for a smaller memory and storage footprint.