Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon

Ddsienra is the new cuda libraries?

Furkan Gözükara SECourses•8/22/24, 10:42 PM

yes

Furkan Gözükara SECourses•8/22/24, 10:51 PM

Huge updates and improvements for FLUX LoRA training : https://www.patreon.com/posts/kohya-flux-lora-110293257

10 GB, 16 GB, 24 GB and 48 GB GPU configs added - 10 GB config is like 3x to 5x slower sadly

Massed Compute, RunPod and Windows Kohya SS GUI LoRA installers added to the zip file

Patreon

Kohya FLUX LoRA and Fine Tuning Training Full Tutorial For Local Wi...

Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon

FFurkan Gözükara SECourses 3.16 second / it for rtx 4090

nicearoni•8/22/24, 10:54 PM

what speed do you see\expect on a 3090?

Nnicearoni what speed do you see\expect on a 3090?

Furkan Gözükara SECourses•8/22/24, 11:07 PM

4-4.5 second / it

nicearoni•8/22/24, 11:10 PM

okay I'm currently getting 8.8s/it - but I think I am using v2 of your scripts - I'll try v4. already tried some of your earlier configs and am getting great results.

Nnicearoni okay I'm currently getting 8.8s/it - but I think I am using v2 of your scripts -...

Furkan Gözükara SECourses•8/22/24, 11:15 PM

great

Furkan Gözükara SECourses•8/22/24, 11:15 PM

ye use latest zip file

Furkan Gözükara SECourses•8/22/24, 11:15 PM

and run Fix_For_FLUX_Step_2.bat

Furkan Gözükara SECourses•8/22/24, 11:16 PM

dont forget to have Fix_GUI_Bat.py

nicearoni•8/22/24, 11:16 PM

I did step 2, so my libraries are updated and i'm only using 17.5gb vram

Furkan Gözükara SECourses•8/22/24, 11:16 PM

it prevents gui updating back to older libraries

Nnicearoni I did step 2, so my libraries are updated and i'm only using 17.5gb vram

Furkan Gözükara SECourses•8/22/24, 11:16 PM

oh nice

Furkan Gözükara SECourses•8/22/24, 11:16 PM

8.8 second is slow though let me test on my pc

nicearoni•8/22/24, 11:20 PM

I noticed on your earlier configs that used 23.5gb that sampling had to be turned off, otherwise after the first sample, GPU mem usage went up over 30gb, so it has to use shared vram, essentially killing the training. did this happen with you as well which is why you don't have samples included in your configs? now that vram usage is down in the 17gb range, might sampling during training be possible? just curious

Nnicearoni I noticed on your earlier configs that used 23.5gb that sampling had to be turne...

Furkan Gözükara SECourses•8/22/24, 11:25 PM

you can use sampling but i dont prefer it because with my experience

Furkan Gözükara SECourses•8/22/24, 11:25 PM

sampling usually broken

Furkan Gözükara SECourses•8/22/24, 11:25 PM

hurts training