it matters since you are also training unet
it matters since you are also training unet
Fthe more you train unet = bad
NI see so in multiple concepts, it doesnt train the unet so theres some advantage
Fso you train that resolution? SD 1.5 model?
Frealistic vision yes
Fcan you share some parameters like LR, steps, etc? so i give it a try too
Fsame as in my video
Fhow many training pictures, how many reg images, that kind of stuff
F1e-7 lion
Fyou like lion
Ftried 90, 85 class images
F30 training images
Fthat's a lot of images
Fand how many vram does this need?
Fit is very slow
Fyou need 24 gb card
Fbecause i dont use xformers
For gradient checkpointing
Fso the a5000
Fis what i normally use anyway
Fno any 24 gb can work
Flike rtx 3090
Fbut a5000 is cheapest i think
Fi'm gonna try but my training images are already shitty at 512x512 xD
F
ARTX 3090 doesn't work for me on RunPod, always can't find the GPU
Fye looks like prefer started to pefer
AOn RunPod, 3090 is cheaper
Fye i use community
F0.29$
Fit is probably cheaper than electricity it would use in some countries :d
AYes, thats why they say they are going to get rid of the current spot pricing structure because it costs more in energy than what they are charging for spot. They said they might keep it but make it a fixed rate and not have the bidding system anymore.
AIn some countries like mine, we are lucky to even have electricity. They cut out our power in rotation for several hours a day.
AThe joys of living in Africa 
Fwow damn sad
Fby the way spot is not dependable to use for training
AYes they can take it back at any time but I have been using it the whole time, been lucky I guess 
Fi never use spot because i don't want to get disconnected
Fi tried once i think :d
Fnot sure if it's happening or not
Fi love raid 1
Fone of my hdd in my game server had become like this
AMy current spot pod has been up for almost 5 hours 
F
Fthey replaced it, raid 1 auto did rebuild 
Fthe other disk is also failing
Fthey will replace it too
F
Fdue to 8% hdd it was reading and writing like with 5mb per second
Fnow back to 500-1000
PIt might have been training incorrectly but still giving better results, I think current runpod kohya actually trains correctly but something with runpod changed and fixed the issue with training and made it train as it should