ya, Bartowski and Unsloth quants I've used both. My go to quants.
On my 24GB MacBook Pro I'm using Unsloth MXFP4 quant of Gemma-4 26B it just fits with 16k context window, maybe I could increase a tad more. I can't load the multi-modal encoder. So it's a text only LLM.
Gemma-4 writes concise code.