on runpod A40 vs RTX4090 which better for training lora?
on runpod A40 vs RTX4090 which better for training lora?
Frtx 4090 faster
Falmost same quality
Fa40 may get just a little bit quality boost from fp16 training
D@Furkan Gözükara SECourses someone seems to have achieved FLUX training on 8GB vram
DDsettings in video description
Didk if you also have, haven't followed FLUX closely yet. But when I last looked at your settings I think it was still > 16 GB, unless you use very slow kohya model splitting technique
SDr Furkan when OneTrainer have full support on flux do you think regularitation images could be usefull again?
Fwe already have config for that 
Fuses 7.5 GB
Fi havent tested yet plan to test
Fonetrainer as well
J@Furkan Gözükara SECourses after finetune, is there a script to compare results with all different settings ( different samplers, different refiners etc, it almost not fun anymore to do it manually for every saved epoch). Some saved Epoch looks overtrained, but when use another sampler or weight, suddenly it becomes much better as the next saved epoch?
FYes grid feature of swarm ui
FShown in both latest 2 tutorials
Jyes, i did found them just now (sooo many tutorials) thankz
FYou are welcome. This grid is amazing
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
DDoc do u have shares in Swarmui and Massed compute lol
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fthey give 10% if you use my link , 10% of what you spend
Dusing OT, without kohya model split?
Fi started ot trainings
Fand its interface sucks
F
UYouTubeuisato![Neo Ballet / Experiments [New SDXL & FLUX LORA | Available]](https://cdn.answeroverflow.com/embeds/1283844624787439761/0/thumbnail.jpeg)
You can access this newly trained LORA [for both SDXL and FLUX] + 5 GB of images and videos with metadata [prompts and configs], plus many more project files and experiments, through my Patreon profile: https://linktr.ee/uisato
PS: Excited to see some of you next month during my travels. I’m still sorting out the details of where I’ll be, so fe...
PS: Excited to see some of you next month during my travels. I’m still sorting out the details of where I’ll be, so fe...
XI got caption files with 300 tokens, so it wont train?
Fit will
Fbut only 225
Xis there any way to make it 300?
Xsad
Fwhat do you train sdxl?
P@Furkan Gözükara SECourses for flux what is the best trainer currently? Konya, AI toolki, onetrainer, something - something else? Do you see OneTrainer catching up?
Fonetrainer not good unless you have 40 gb gpu :d
Fit doesnt have fp8 yet
Fkohya works great
Fi didnt compare ai toolkit
Phow about training time on a A6000 - what should one expect for a 15 person picture Lora with your configs?
Feasy to calculate
Flets say you used Rank_2_27700MB_4_85_Second_IT.json
Fit is 4.85 second / it
Fyou need 3000 steps - assuming 200 epoch
F3000 x 4.85 second
PI see - thank you. By what factor does speed increase on mass compute multi GPU training ?
Falmost linear
Fif you rent 8 gpu you get like 7x speed up maybe more
M@Furkan Gözükara SECourses congrats on the latest training article. Can you add something about captioning (if you used it, how you made it…)?
MI mean, how did the model learn what each expression was? What expression does it inherently know, as far as your dataset goes?
Fi added it actually :d
FI didn't caption them only ohwx man since captioning reduces likeliness and doesn't bring improvements.
Fi updated article added more details
MThe Patreon one?
