HuggingFace used an NVFP4 quant of GLM-5.2 to investigate their latest hack, so that might also be worth a try:

https://huggingface.co/nvidia/GLM-5.2-NVFP4