Quantizing Gemma 4 on Mac with llama.cpp

A guide shows how to set up llama.cpp on a Mac to run Gemma 4 locally.

It covers downloading the model, converting Hugging Face weights to BF16 GGUF, and quantizing to Q4_K_M.

The workflow uses llama-quantize and llama-cli to test the model on macOS.

The article highlights a practical way to reduce memory use and keep inference on-device.