Software Engineering Courses (SECourses)•17mo ago

I will not finish this training, so I can't tell you result. But maybe I will try to start training

I will not finish this training, so I can't tell you result. But maybe I will try to start training at night. So could be some results tomorrow...or not)

Nnovsky I will not finish this training, so I can't tell you result. But maybe I will tr...

Furkan Gözükara SECourses•8/28/24, 3:38 PM

i will do same :d

macba•8/28/24, 7:43 PM

Hello everyone, quick question: what does this script do?

https://huggingface.co/MonsterMMORPG/Generative-AI/resolve/main/kohua_gui_pass_update.sh

As far as I understand, it replaces kohya_ss's gui.shgui.shwhen you use Furkan's kohya installer script (at least on Linux). What's the advantage? Thanks…

daroth•8/28/24, 7:45 PM

okay, lets see the local flux kohya training

Mmacba Hello everyone, quick question: what does this script do? https://huggingface.c...

nicearoni•8/28/24, 7:46 PM

I think updates a few libraries to newer versions, then patches original gui to not roll them back? not sure though.

daroth•8/28/24, 7:46 PM

do we still need to use the kohya training folder structure? even with the folder name #_name_class?

Ddaroth do we still need to use the kohya training folder structure? even with the folde...

nicearoni•8/28/24, 7:47 PM

that's what kohya is still expecting as of now (space between name and class though, not _ )

FFurkan Gözükara SECourses nicer

sergio_gonzo•8/28/24, 9:06 PM

Dr Furkan i hope everyting is ok, wich is your estimate tha could be a video tutorial of FLUX Lora step by steps? wonderfull work

Mmacba Hello everyone, quick question: what does this script do? https://huggingface.c...

Furkan Gözükara SECourses•8/28/24, 10:57 PM

it prevents kohya to revert back requirements

Furkan Gözükara SECourses•8/28/24, 10:58 PM

so our custom torch installation remains

Ssergio_gonzo Dr Furkan i hope everyting is ok, wich is your estimate tha could be a video tut...

Furkan Gözükara SECourses•8/28/24, 10:58 PM

thank you. still editing video

Evivrus•8/29/24, 3:02 AM

Has anyone gotten a slider Lora training working for Flux yet? I have seen a few sliders on Civit but I don't think they are trained the same way SD sliders were trained.

Anyone seen anything? I tried modifying the ai-toolkit code to get it to work but it didn't work out ;P

Furkan Gözükara SECourses•8/29/24, 3:49 AM

nope i didnt see yet

daroth•8/29/24, 9:02 AM

when the trained person should be blonde but her last name is "Blue"

novskyOP•8/29/24, 3:33 PM

what is better to have more epochs or more repeats?

Nnovsky what is better to have more epochs or more repeats?

Furkan Gözükara SECourses•8/29/24, 4:34 PM

they have different purposes

Furkan Gözükara SECourses•8/29/24, 4:34 PM

explained in video :d

Furkan Gözükara SECourses•8/29/24, 4:34 PM

normally you dont need repeat

novskyOP•8/29/24, 4:34 PM

I'm in a middle of watching))

Sandstormleader•8/29/24, 9:27 PM

@Furkan Gözükara SECourses how long is the new flux training take?

SSandstormleader <@205854764540362752> how long is the new flux training take?

Furkan Gözükara SECourses•8/29/24, 9:29 PM

totally depends on your total steps and gpu

Furkan Gözükara SECourses•8/29/24, 9:29 PM

but rtx 3090 can get amazing model within 5 hours

Furkan Gözükara SECourses•8/29/24, 9:29 PM

rtx 4090 likely to with like 2 hours 2.5 hours

daroth•8/29/24, 9:31 PM

2-2.5 hours for the 4090 is correct

Sandstormleader•8/29/24, 9:34 PM

Nice I'll have to check it out. I have a 4090

Ddaroth 2-2.5 hours for the 4090 is correct

Furkan Gözükara SECourses•8/29/24, 9:50 PM

yep

SSandstormleader Nice I'll have to check it out. I have a 4090

Furkan Gözükara SECourses•8/29/24, 9:50 PM

great

Furkan Gözükara SECourses•8/29/24, 9:50 PM

use torch 2.5 to have best speed on windows

Sandstormleader•8/29/24, 10:57 PM

i have like 60 more models to merge but i want to check it out this weekend for sure

Sandstormleader•8/29/24, 11:07 PM

Sandstormleader•8/29/24, 11:07 PM

fine tuning and adding flexibility to this 1.5 model

Sandstormleader•8/29/24, 11:09 PM

while keeping the most accuracy

Sandstormleader•8/29/24, 11:18 PM

i wish i knew the science behind how merging works but i just mix them up and hope for the best

SSandstormleader i wish i knew the science behind how merging works but i just mix them up and ho...

Sharinel•8/30/24, 9:14 AM

This man Sciences

SSandstormleader i wish i knew the science behind how merging works but i just mix them up and ho...

Furkan Gözükara SECourses•8/30/24, 9:30 AM

there is no science

it averages neural network weights and you expect to work

Sandstormleader•8/30/24, 11:56 AM

hahah well i just mean in a basic sense, like if i do 50% + 50% of one model and then merge that with another at 50% does it equal out to be 25% +25% +50% or what lol

Sandstormleader•8/30/24, 11:57 AM

or am i actually losing any data or does other data just cover it up to where i can bring other data forward by adjusting weights of tokens and what not

daroth•8/30/24, 3:20 PM

hm, I seem to be using less VRAM than supposed to
odd

Ddaroth hm, I seem to be using less VRAM than supposed to odd

Furkan Gözükara SECourses•8/30/24, 3:32 PM

maybe torch 2.5 diff

Furkan Gözükara SECourses•8/30/24, 3:32 PM

are you using torch 2.5? vrams tested on torch 2.4

Absentis•8/30/24, 3:47 PM

Thanks to your guides and configs I fine tuned an SDXL model with pics of my dog and of me using OneTrainer (will post some results later). The thing I'm struggling with is extracting a Lora from the model. I was able to do this with kohya_ss but the Lora comes out at almost 2Gb. I was wondering am I using wrong settings in kohya_ss or should I extract the Lora and then run the reduce Lora option in kohya_ss?

AAbsentis Thanks to your guides and configs I fine tuned an SDXL model with pics of my dog...

Furkan Gözükara SECourses•8/30/24, 3:52 PM

why you need to reduce its size?

Furkan Gözükara SECourses•8/30/24, 3:52 PM

you can reduce by

Furkan Gözükara SECourses•8/30/24, 3:52 PM

set precision to bf16 or fp16 compare each one quality

Furkan Gözükara SECourses•8/30/24, 3:52 PM

set rank 128 or lower if you need lower size

Furkan Gözükara SECourses•8/30/24, 3:52 PM

may reduce quality

Furkan Gözükara SECourses•8/30/24, 3:52 PM

i use kohya gui extractor

Absentis•8/30/24, 4:25 PM

Yeah, using the Kohya gui. Main reason I'd like to reduce the size is I wanted to share it + my hard drive was getting full (I really should buy a bigger drive). I played a bit with the resize option in the gui. Dropped it to a rank of 64 and the final size was about 380MB without noticeable quality loss. I was mainly curious if there was a recommend way to get smaller Loras.

Absentis•8/30/24, 4:29 PM

But you are right. I should just follow your example of just testing the quality of the Lora at different ranks and precision till I find a good mix.

AAbsentis Yeah, using the Kohya gui. Main reason I'd like to reduce the size is I wanted t...

Furkan Gözükara SECourses•8/30/24, 4:45 PM

awesome

FFurkan Gözükara SECourses maybe torch 2.5 diff

daroth•8/30/24, 4:52 PM

not sure, I just installed from the files in the latest kohya pack

I will not finish this training, so I can't tell you result. But maybe I will try to start training

Similar Threads