convert_to_quant settings
#2
by dubsta - opened
Hi Trying to create my own qwen image 2512 int8 convrot from a different finetune but all I am getting is noise after converting it.
Can you share your convert_to_quant settings please?
I made a YouTube tutorial on it because you need to feed Claude the results and it will calculate the math. Convrot requires that you calculate the rounded number like 64, for example and that's how you create the "convrot group". The process requires investigating each model, whether it's a diffuser model or text encoder and using the ctq --help-filters flag to see if the model is supported or not.