Kohya finetuning error

[etha302]

Hello,

i am getting this error when using "finetuning" method in kohya_ss: fine_tune.py: error: unrecognized arguments: --no_half_vae. It worked for me in the past just fine, but after the update i simply cannot figure out why i am getting this error, this only applies for "finetune". Dreambooth and loras are working just fine.Hope someone can help me out.
i cannot figure out where --no_half_vae, args are in fine_tune.py and if removing it would even help.
edit: also i have no extra arguments specified in the gui itself

[etha302]

i havent been able to find a solution so far, so if anyone has a clue please help me out here. this for finetuning 1.5 but exactly the same happens for sdxl.

[bmaltais]

Hummm... did I broke something in the GUI with the recent SDXL update? You are training SD1.5, right? I will add it to the list of things to check when I am back from vacation.

[etha302]

Hummm... did I broke something in the GUI with the recent SDXL update? You are training SD1.5, right? I will add it to the list of things to check when I am back from vacation.

hi, thanks for your reply. OR im doing something really wrong, but i dont know. i would really appreciate if you take a look when you are back from vacation, here are some screenshots. ignore parameters, but it doesnt even matter what i set them to, same error always.
edit: seems it is not happy with args: --no_half_vae but i looked everywhere for this in .py files but havent found anything that would cause this.

[maifree123]

open SDXL model，--no_half_vae in Parameters

[dneilsterino]

modify your config .json with:

"sdxl_no_half_vae": false

then open the config and train. You might also have to use:

"bucket_no_upscale": false

[asgeorges]

Just to add on to @maifree123 ....

You'll want to open up SDXL model option, even though you might not be using it, uncheck the half vae option, then unselect the SDXL option if you are using 1.5, etc...

Thanks @maifree123 !

[farizy4n]

modify your config .json with:

"sdxl_no_half_vae": false

then open the config and train. You might also have to use:

"bucket_no_upscale": false

Thanks, thats worked to me.