This is the easiest way to convert Z-Image Turbo fp16 models(12GB) into Fp8 models(6GB). On the first node on the left pick the big fp16 file you want to convert and on the last node on the right, write the path and the name for your new Fp8 model and where you want it to be saved.
Description
FAQ
Comments (6)
Thanks for sharing, this is so useful for graphics cards with limited video memory! 😍 How are those all-in-one graphics cards (like RED-AIO) implemented? Are they made using similar toolnodes? Or, it would be even better if they could be converted to GGUF format with selectable sizes (Q5, Q8, etc.).
Are there any readily available .EXE tools on GitHub or Huggingface? (For direct model conversion without using COMFYUI) Thank you again for your help.
"How are those all-in-one graphics cards (like RED-AIO) implemented?" er, just a checkpoint
"f they could be converted to GGUF format with selectable sizes" llama.cpp with a patch to do that or something else
someone else has to answer this. but the node pack you had to download for the fp8 files conversion has also nodes for gguf conversion. if you want to try it, but the install can be a bummer
@qek THX~
@Piamertens if doesn't support fp8 scaled or something else, it isn't useful because it can be done without the custom node(s)
@qek it is useful to those that don't know how to do all this. how do you convert your fp8 scaled models and which node have you created yourself to make it work for zit? https://civitai.com/models/2242985?modelVersionId=2530631