gotcha okay, i have a 4090 and 3090, was hoping 48GB would work between the two
gotcha okay, i have a 4090 and 3090, was hoping 48GB would work between the two
Fhave consistent dataset thats all
Fno it cant merge
Fbut multi gpu lora may work
Mokay
Mso in your different configurations are you saying those are the required memory for each in MB?
DAt all? Not even "photo of name"?
Fuse ohwx class
Fnot matters much but not hurting either 
Mso shouldnt the 23100MB work for me then? i was confused since you said it required 80GB but the config file says 23.1GB
F23100 mb
Fmulti gpu 80 gb
Fyou can do seperate training on each gpu
Mahhh okay that makes sense now, okay cool, thank you so much and looking forward to your dreambooth tutorial
Fawesome. i am experimenting batch size impact right now
Moh yah another quick question, i saw you have buckets disabled, is that better?
Funtil you have a base model
Fdisable
Fafter that you can enable and test
�@Furkan Gözükara SECourses please see my message
MFor Flux DreamBooth are you using the sd3-flux.1 branch of Kohya or the main branch?
Fsd3 flux branch
Hhello bro! I want ask, I want train sdxl , so use onetrainer better or kohya_ss ?
Fonetrainer
M@Furkan Gözükara SECourses , any news on Flux Dreambooth finetuning recommended number of steps? The 17500 steps (175 training images) training I ran the other night is disappointing… I trained on one person and the ohwx person in generated images only bears a passing resemblance…
Fit shound resonable steps
Fcurrently i am training myself with 256 images
Falso working on batch size 7 config
Falmost ready
F40 epoch is amazing
MCan you detail how many repeats, epochs, total steps and batch size please?
F256 images, batch size 7, epoch 40 = 256 * 40 / 7 steps = 1462
Ffine tuned on RTX A6000
F
MThis is Dreambooth finetuning on Kohya, right?
Fdreambooth in kohya for flux
F256 images
F256 images is a total overkill can be reduced a lot
MYou didn't mention repeats, so your folder was 1_ohwx man, right?
Falways 1 repeat
Ffor flux
MWell ,maybe that's where I went wrong, I did use repeats
F
Fwhy?
AHas anyone had issues with the vae when trying to finetune in kohya? It looks like it's loading the full flux dev checkpoint as the vae rather than the vae I'm pointing it to in the config...and so it fails to generate latents. I think I can work around this by just not caching to disk and by turning off "use latent files".
Just wondering if anyone has seen this. kohya_ss is all up to date.
Just wondering if anyone has seen this. kohya_ss is all up to date.
MI thought I was using your settings, but maybe I increased the repeats in addition to the epochs to get more steps? I can't recall 100% but I guess it was a mistake on my part