I attempted using the 24gig config file for finetune on my 4090 but it estimated around 700 hours..
I attempted using the 24gig config file for finetune on my 4090 but it estimated around 700 hours.. is this normal?
Fnope you must be using shared vram.
Fincrease number of blocks to swap or reduce your vram usage
Falso make a fresh kohya installation if still fails
CI basically force quit everything I could except kohya. It is a fresh install
Fi tested and even on my RTX 3060 works great :d
Fso how much VRAM used before starting did you check?
Caround 0.6
CI do suspect something else on my system or hardware though. Even running base flux1-dev in forge is super slow.
Fi think should work
Ftry 16 gb config
Fand report me back speed
Cwill do, thank you
Falso check how much shared vram used
Falways a possibility
Fyou should get over 2 it second with 4090 on swarmui with --fast
AHmm, I have installed Visual Studio (2017 and 2022 CE) , but it seems that I get a problem with XFormers (not installed) after installing Windows_Install_Torch_2_5_Dev_Huge_Speed_Up.bat... any clue?
A(2017 is needed for other projects 
)
)
Fxformers not available yet
Ffor torch 2.5
Fi am waiting
Fso on windows we use SDPA
Fon linux you can use torch 2.4.1 and xformers 0.0.28.post1
AAh thanks
AIt drove me crazy^^
AI have win 11 here and any linux on runpod...
ADo you already have a quick tut for Dreambooth? I used kohya some times ago however only for sd 1.5 lora tests...
AIf not, it will happen through learning by doing, where “doing” is the sound that is made when I hit my face with my hand much too hard. 
Fdid you watch lora?
Fit is exactly same
Fjust load config into dreaambooth tab
FYouTubeSECourses
Ultimate Kohya GUI FLUX LoRA training tutorial. This tutorial is product of non-stop 9 days research and training. I have trained over 73 FLUX LoRA models and analyzed all to prepare this tutorial video. The research still going on and hopefully the results will be significantly improved and latest configs and findings will be shared. Please wat...
FYouTubeSECourses
Unlock the power of FLUX LoRA training, even if you're short on GPUs or looking to boost speed and scale! This comprehensive guide takes you from novice to expert, showing you how to use Kohya GUI for creating top-notch FLUX LoRAs in the cloud. We'll cover everything: maximizing quality, optimizing speed, and finding the best deals. With our exc...
AAh!
ANow I see the beauty of kohya
AYour first vid above was an eye opener
ABut what happens if the training images have different resolutions, let's say I can't crop them down to 1024x1024, but I can resize them to 1024x677? Should I fill the rest with a spare color, e-g. black?
Fi would get all images 1024x1024
Fif you cant get you can enable bucketing
Fit will auto crop
AThanks! Is there any whitepaper available about this theme?
Fnot that i know
Fbut flux works best with 1024x1024 i find that
Fbest is comparing
F1024 vs bucketing
V@Furkan Gözükara SECourses I want to train a flux lora based on oil paintings. I just want the lora to capture the textures, brushstrokes and art style, not necessarily the colors. Would fully captioning the dataset be preferred over using an activation word?
T@Furkan Gözükara SECourses I'm a bit confused, the youtube video says to load the configs into the lora tab, but with the latest best configv5 they load into the dreambooth tab. Which tab do you do the config in, and which tab do you hit start training? (UC Arcade on patreon)
SSame thing is happening on my GTX4090. You in Windows 10? I'm about to nuke the computer and throw Ubuntu on it because I think Windows is the problem
VCheck to see if its using shared memory, if so that might be the issue, which means the config needs tweaking for your machine.
SYeah, I've tried everything everything google will tell me to kill the shared memory. Gonna try the 16gb config file and see how that goes before I blow up the OS
Fi think you can still train without captions
