Software Engineering Courses (SECourses)•14mo ago

so.. I've been training on massed_compute recently (A6000) but thought I'd give my 4090 a test. I'm

so.. I've been training on massed_compute recently (A6000) but thought I'd give my 4090 a test. I'm actually quite suprised at the performance using the 24GB_GPU_23150MB_10.2_second_it_Tier_1.json

1nfern0OP•11/6/24, 12:22 AM

fresh install of Koyha.

11nfern0 so.. I've been training on massed_compute recently (A6000) but thought I'd give ...

Furkan Gözükara SECourses•11/6/24, 12:33 AM

rtx 4090 will be faster than a6000 with newest update - not merged into main yet :d

Furkan Gözükara SECourses•11/6/24, 12:33 AM

for fine tuning

abtx•11/6/24, 10:29 AM

What is the fastest kohya config to use if you have 80GB vram? (H100)

Aabtx What is the fastest kohya config to use if you have 80GB vram? (H100)

Furkan Gözükara SECourses•11/6/24, 11:17 AM

batch size 7 flux fine tuning

yassinemenace•11/6/24, 2:06 PM

doesnt batch size decrese quality ?

Yyassinemenace doesnt batch size decrese quality ?

Furkan Gözükara SECourses•11/6/24, 8:23 PM

it does

Furkan Gözükara SECourses•11/6/24, 8:23 PM

but for speed mandatory

Saxon•11/6/24, 10:05 PM

when I try to train with dreambooth, the safetensor files are massive (23gb each). Is that to be expected?

SSaxon when I try to train with dreambooth, the safetensor files are massive (23gb each...

Furkan Gözükara SECourses•11/6/24, 10:09 PM

yep

Furkan Gözükara SECourses•11/6/24, 10:10 PM

you can later convert into fp8 - half size

Saxon•11/6/24, 10:11 PM

I tried training one on a particular person, and the sample image during training looked perfect, but when I actually use it with flux, it just makes generic people. Is there something else I might've forgotten to do?

SSaxon I tried training one on a particular person, and the sample image during trainin...

Furkan Gözükara SECourses•11/6/24, 10:12 PM

yep obviously :d

Furkan Gözükara SECourses•11/6/24, 10:12 PM

i dont know what though

Saxon•11/6/24, 10:15 PM

Upon finishing the training, can I simply move the safetensor files to the lora folder, and generate an image with them using the trigger word? or are there steps I'm unwittingly skipping?

SSaxon Upon finishing the training, can I simply move the safetensor files to the lora ...

Furkan Gözükara SECourses•11/6/24, 10:15 PM

that is not lora

Furkan Gözükara SECourses•11/6/24, 10:15 PM

please read this article : https://www.patreon.com/posts/112099700

Patreon

Kohya FLUX Fine Tuning (Full Checkpoints) Training Full Tutorial Fo...

Get more from SECourses: FLUX, Tutorials, Guides, Resources, Training, Scripts on Patreon

Furkan Gözükara SECourses•11/6/24, 10:15 PM

i assume you are telling 23gb files

Furkan Gözükara SECourses•11/6/24, 10:16 PM

for lora please watch again : https://youtu.be/nySGu12Y05k

YouTubeSECourses

FLUX LoRA Training Simplified: From Zero to Hero with Kohya SS GUI ...

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...

Saxon•11/6/24, 10:16 PM

lol, I understand. Thanks.

Zet•11/6/24, 10:50 PM

@Furkan Gözükara SECourses Anything we need to do about new branch in bmaltais?

Zet•11/6/24, 10:50 PM

and are the faster speeds in the current branch (since it's sd-scripts that are faster)?

Vigilence•11/6/24, 10:59 PM

@Furkan Gözükara SECourses Any latest studies on stable diffusion 3.5 training?

ZZet <@205854764540362752> Anything we need to do about new branch in bmaltais?

Furkan Gözükara SECourses•11/6/24, 11:03 PM

i didnt test i copy pasted sd scripts to test yesterday

VVigilence <@205854764540362752> Any latest studies on stable diffusion 3.5 training?

Furkan Gözükara SECourses•11/6/24, 11:03 PM

coıuldnt start yet planning very soon hopefully

FFurkan Gözükara SECourses i didnt test i copy pasted sd scripts to test yesterday

Zet•11/6/24, 11:52 PM

what did you copy, just sd-scripts flux branch?

Zet•11/7/24, 3:46 AM

@Furkan Gözükara SECourses Did you ever do a rank test on Lora Extraction? We know 64 is max, but 63 is only 1.2GB, is there really a difference? 64=12GB 63=1.2GB

Zet•11/7/24, 3:47 AM

Personally, I have only done a small test, but with the wrong model, so it was hard to tell. I am training a new one at the moment, and will test again

ZZet <@205854764540362752> Did you ever do a rank test on Lora Extraction? We know 64...

wardensc2•11/7/24, 5:22 AM

The max rank that can be extracted is 640 not 64 my friend. Even training with a6000 we already achieve rank 128

Wwardensc2 The max rank that can be extracted is 640 not 64 my friend. Even training with a...

Zet•11/7/24, 12:04 PM

Thanks

Wwardensc2 The max rank that can be extracted is 640 not 64 my friend. Even training with a...

Zet•11/7/24, 12:04 PM

I totally read that as 64, bummer. More tests

ZZet <@205854764540362752> Did you ever do a rank test on Lora Extraction? We know 64...

Furkan Gözükara SECourses•11/7/24, 1:17 PM

i did tes

Furkan Gözükara SECourses•11/7/24, 1:18 PM

https://www.reddit.com/r/SECourses/comments/1f0le85/here_most_wondered_comparison_different_lora/

From the SECourses community on Reddit: Here most wondered comparis...

Explore this post and more from the SECourses community

ZZet what did you copy, just sd-scripts flux branch?

Furkan Gözükara SECourses•11/7/24, 1:18 PM

just copy paste sd scripts and overwrite gui sd scripts

wardensc2•11/7/24, 2:13 PM

Hi @Furkan Gözükara SECourses I wonder if I choose the same epoch to save and use. Are the quality of the same epoch checkpoint when training with 200 epoch and 100 epoch with same learning rate equal or different ? For example I use checkpoint saved after 80 epoch trained one from 200 epoch training trained other from 100 epoch trained

Wwardensc2 Hi <@205854764540362752> I wonder if I choose the same epoch to save and use. Ar...

Furkan Gözükara SECourses•11/7/24, 2:17 PM

with my config should be same

Furkan Gözükara SECourses•11/7/24, 2:17 PM

because i use constant learning rate

Furkan Gözükara SECourses•11/7/24, 2:18 PM

kohya made important fixes to SD 3.5

wardensc2•11/7/24, 2:19 PM

Thank, because after training with your 200 epoch those checkpoint with more than 120 epoch become overtrained so I just want to reduce my training time ?

FFurkan Gözükara SECourses kohya made important fixes to SD 3.5

wardensc2•11/7/24, 2:20 PM

Can't wait to download your config with SD 3.5 and test and pay 5 usd more

Wwardensc2 Can't wait to download your config with SD 3.5 and test and pay 5 usd more

Furkan Gözükara SECourses•11/7/24, 2:22 PM

Wwardensc2 Thank, because after training with your 200 epoch those checkpoint with more tha...

Furkan Gözükara SECourses•11/7/24, 2:22 PM

sure you can

Furkan Gözükara SECourses•11/7/24, 2:23 PM

so let me clarify

Furkan Gözükara SECourses•11/7/24, 2:23 PM

with my config

Furkan Gözükara SECourses•11/7/24, 2:23 PM

100 epoch trained

Furkan Gözükara SECourses•11/7/24, 2:23 PM

saved

Furkan Gözükara SECourses•11/7/24, 2:23 PM

then continued another 100

Furkan Gözükara SECourses•11/7/24, 2:23 PM

should be equal to directly training 200

wardensc2•11/7/24, 2:26 PM

thank Doc

wardensc2•11/7/24, 2:29 PM

to continue training I need to tick on these option right ? Maybe you can make a guide about how to continue train a model which require very long training time by separate it into many training section

so.. I've been training on massed_compute recently (A6000) but thought I'd give my 4090 a test. I'm

Similar Threads

Similar Threads

Similar Threads