Seems that it helps you quantize models: https://quark.docs.amd.com/latest/index.html
And it looks like they post train models in mxfp4 giving it better quality: https://huggingface.co/amd/MiniMax-M2.5-MXFP4
They only have a couple hundred downloads per model update so maybe its gone unnoticed?
[link] [comments]

