you are supposed to divide your epoch to / number of GPUs
you are supposed to divide your epoch to / number of GPUs
Fi told this in video
Fit is experimental
Fonce i analyzed results we will see
Fwhich steps yielding good :d
Flet me upload them to hugging face
Fyou can also test
NI'm just asking, I inquired with the SDXL models coach lye a long time ago. they did their experiments and gave their metrics which I adhered to. Characters are in the 1200-1800-2000 range.
Objects 1500-2500.
Styles 1200-3000 depending on elements and creativity.
Architectural objects up to 4000 with increased dim and alpfa.
Now I wonder what the figures are for Flux
Objects 1500-2500.
Styles 1200-3000 depending on elements and creativity.
Architectural objects up to 4000 with increased dim and alpfa.
Now I wonder what the figures are for Flux
Fi am uploading models
Ffrom 25 to 450 epoch
F114 images
Fi will also test hopefully
Nthe base step is 1?
Fsince 4x gpu, 1 step = 4 images
Fso 1 epoch = 114 / 4 = steps
N114*25/n gpu
Fyep
Fuploading models
Ffiles uploaded
Ffeel free to use 
Fdataset and captions are there too
G8 is divisible by 4
G2 epoch each
Fno not like that
GEpoch can be any number regardless right?
Fhow many images you train?
G10000
Fok
Fso for 10000 images
Ftrain 1 epoch
Fthat will make 2500 steps on each gpu
Fhowever if you want to be sure
Fthat you wont undertrain
GBut i want to train 8 epochs and compare each one
Fyou can train like 5 epoch
Fok for 8 epoch training
Fyes make value 2
Fand save every 1 epoch
G2 Epoch save every 1
Fyep
GThen it will work?
Fyou know default config is 50 epoch
Fmake it 2
Fyes
Fand save every 1 epoch
GDid u see my errors?
Fon the screen you should see 20000 steps on each gpu
Fthey dont looks like error
Fi just saw
Fat the end of the video