Fthat is why
Fdelete logs
Fit happened me in past
Fi told kohya this
Fhe didnt fix 
Fhere my machine
GDelete log and it will work?
Fyep
Fdont set logs folder
Fcheck my machine
F
Fbeen runnnign trainng for 37 hours
GAnd 2 Epoch. Save every Epoch thats it
Fyep
GIt will do 80000 since images is 10000
G10k * 8
Fthe math is easy
Fif 1 gpu batch size 1 = 80000 steps
GI think 1 Epoch will be enough but want to do more just in case
F4 gpu = 20000 steps on each gpu
Fmakes sense
GAnd each file has a .txt desciption
GThanks gn, ill be on later and will try again
GBut it doesnt use all 4 gpus to work on 1 at a time?
GI thought all 4 worked on 1 at a time to do 1 epoch faster
Fall 4 must work
Fbut all 4 clones
Fso each one does a seperate training and then results averaged
Flike batch size 4
Moh wow t5 lora training available
M@Dr. Furkan Gözükara time to update all your configs again
Fi saw it
Fbut i am not sure if worthy
Fwill test later hopefully
Mtesting it with sd scripts to see
Mprobably only if you decide to caption
Fclip l certainly improved results
Fif i can go lower unet T5 can help too
Mthere are some issues though
Mtried some custom trained flux checkpoints
Myou need to increase strength to 1.5 or 2 for it to work good
Fit is really hard to make a better custom checkpoint for flux
Mmost of them are only using the unet version extracted which is only 12gb 
Mtested dreambooth but its like 75gb vram for decent quality
Fthere are so many optimizations 75gb dreambooth sounds too extreme
Fi expect under 24 gb
Shi guys, just subscribed to patreon page to train my lora. Downloaded the given zip file, start Windows_Install_Step_1 and getting bunch of errors
S'pip' is not recognized as an internal or external command, same for python and git. Do I need to install something before I run this Windows_Install_Step_1.bat?
Fyes you are missing requirements
SI found out that I need python and in the video you say to use version 3.10, but it is no longer available on the python official webiste