GGUF conversion of https://civarchive.com/models/2731187/moody-krea-2-mix-uncensored by @catlover1937 - all credits belong to them!
Get the GGUF loader nodes from here: https://github.com/molbal/ComfyUI-GGUF
Description
FAQ
Comments (6)
Lots of time, Q8 is better than FP8 in quality and/or speed. Have you tested if its the case?
Correct, that is because Q8 GGUFs often leave some layers in FP16 or FP32. In this case however, the original weighs I converted from were in FP8 not FP16, so those layers, while technically are in a higher precision, will not actually benefit from it.
I did not actually make comparisons, maybe when I have my rtx 3090 server set up I will begin with that practice :)
There have been a number of new versions of this model- is there any chance you'll quantize v4?
Sure thing! Just started to download v4
Hey there, https://civitai.red/models/2739784/moody-krea-2-mix-gguf?modelVersionId=3136876 v4 gguf finishes uploading in ~15 minutes
@molbal Wow, excellent! Thank you, I'll try it out!