check out a custom 4-bit quant I made today https://huggingface.co/macwhisperer/Gemma4-12B-SuperDense https://huggingface.co/macwhisperer/Gemma4-12B-SuperDense

check out a custom 4-bit quant I made today https://huggingface.co/macwhisperer/Gemma4-12B-SuperDense https://huggingface.co/macwhisperer/Gemma4-12B-SuperDense should run perfect for 12-16gb with maybe 10-20k context seems intelligent enough that I would recommend this as a daily driver for friends who just want a local ai that can do most things relatively quickly (getting 10 tps on my m2 air)