I started a lora training, like an hour back.
I started a lora training, like an hour back.
Fweird
Fi wonder if only runpod broken
Fdue to accleerate
Fwill check now
Fbut it was working previously on runpod too
FGitHub
Urgent fix needed for SD3 branch @kohya-ss this error is after latest @bmaltais update this config worked yesterday INFO prepare accelerator flux_train.py:177 accelerator device: cuda Saving traini...
DIn the logs, like an hour back. Seems to say it is already up to date?
Fweird how yours works 
Fupdate was like 5 hours ago
DThen, not one, but two of mine are going right now lol
Fkohya fixing the error
Fman this guy awesome
Fhe replied my from twitter
Fye he fixed
Fnow we have to wait bmaltais
Fkohya fixed
DFux-De-Distilled I'm training right now great speed T5 and T5 att mask
Dwas training slower this morning.. same dataset, same settings
Didk
Adr did you have images available of you to download for onetrainer testing
A
PWhat error?
DHi, can i know the training time for flux dev lora on a 3060 12gb card for a total of 15 images?
DI tried that, it was something like 80 hours or so
Dif i remember correctly
Nwhat is gold membership
Nok thanks
Shello, did anyone try multi object lora finetuning? can someone provide guide or tips for it
Fi have reg images
Fwhich config you used
Fon which gpu?
Fit is fixed
Fwas a parsin error
F40 second / it
Fyou need 48 gb ram
F
FFyour objects has to be in same picture
Fotherwise may bleed
SI have multiple product images, if I train lora for each product and merge loras will work?
did you try merging loras earlier?
did you try merging loras earlier?
Fit will bleed
Fthat is what de distilled models may fix
Fhopefully it is next research
Sgot it, even sdxl models wont produce good results with lora merge? or checkpoint train + merge will work?
FSDXL should work fairly well
Ffor flux still bleeds with base 23.8 gb dev model
Sunderstood, thank you!
Ftriton 3 published i think testing