not even sure how i can cleanly remove them
not even sure how i can cleanly remove them
Fno install onetrainer
Fand installation cmd logs
Zsure
Fmake a fresh one delete old one
ZI did that once
Zoriginally i installed on e drive
Zthen didnt work so I installed on c
Zwill try again and send u the logs
Zthanks
Fyou are welcome
Zhow can i uninstall it? by just deleting the folder?
Dis there a mixmum grid size A1111/forge can do? ETA is more than one hour...
DI hope it doesn't fail at the end
Fyep
Fyes there is
Fbut it checks before you startr
Dthanks
Dthe kohya model splitting / offloading feature that you are using, is it with bf16? or does kohya even use a lower precision for full fine tune
Fnope it cant train fp8 yet
Ffine tuning
Fthat is how works
Falso on lora, it still train bf16 but model is converted into fp8 weights
Dfor lora it's easy to use a model in low precision, because the values you are changing are still higher precision. but for full finetune he uses less than 16bit?
Ffor fine tune only 16 bit
Fnot lesser
Dfor full finetune, those 8 bit are the weights that you are training
Dokay that makes sense
Zcurrently installing @Dr. Furkan Gözükara
Dso that 2 bytes * # of parameters gotta go somewhere. does it need a lot of ram if he offloads?
DiPNDM does look nice, thanks for the tip. forge has it too
Zand its gone
Zits working! the issue was python
Zcant find this link tho
Zokay ill get your patreon
EI can see it at the bottom here: https://www.patreon.com/posts/96028218?utm_campaign=postshare_fan&utm_content=android_share&utm_source=copyLink
Patreon
Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon
OAny best practices on inpainting in Flux?
Zhow can i just start training on images instead of text or doing complicated stuff?
Fawesome
Zworking on training now
Fah you did already
Fgreat
Zyes
Zwatching your one trainer video
Zone concept is done
Fawesome
Falso use reg images
Fhelps huge
Ffor sdxl
Fwhen training a person