Software Engineering Courses (SECourses)•16mo ago

@Dr. Furkan Gözükara Did you use bucketing enabled with your style training? I remember you talking

@Dr. Furkan Gözükara Did you use bucketing enabled with your style training? I remember you talking about it but I don't remember where you talked about it.

LuckyStrikeOP•9/4/24, 5:49 PM

(I think you said it hurt the quality of the result in the case of getting a person's likeness but I don't think there was a consensus on styles)

LLuckyStrike @Dr. Furkan Gözükara Did you use bucketing enabled with your style training? I r...

Furkan Gözükara SECourses•9/4/24, 5:52 PM

i didnt all

Furkan Gözükara SECourses•9/4/24, 5:52 PM

was 1024x1024

Furkan Gözükara SECourses•9/4/24, 5:52 PM

grid is being generated

LuckyStrikeOP•9/4/24, 5:53 PM

Great I'll keep it off for mine and stick to 1024, thank you!

ihsass•9/4/24, 6:07 PM

How to use realvisxl in sdxl training

ihsass•9/4/24, 6:07 PM

If anyone have link please

ihsass•9/4/24, 6:08 PM

Realvisxlv4.0

ihsass•9/4/24, 6:08 PM

There is only sdxl base model in a1111

Iihsass Realvisxlv4.0

Furkan Gözükara SECourses•9/4/24, 6:12 PM

just download it

Furkan Gözükara SECourses•9/4/24, 6:12 PM

https://youtu.be/X5WVZ0NMaTg

YouTubeSECourses

How to Download (wget) Models from CivitAI & Hugging Face (HF) & up...

Notebook updated to V5 with huge improvements. If you are having trouble to download models onto your cloud platforms such as RunPod, Google Colab, Kaggle, Massed Compute or anywhere, this tutorial is made for you. Moreover, you can upload your files / models into your private Hugging Face repository and save them permanently privately and downl...

Furkan Gözükara SECourses•9/4/24, 6:13 PM

https://huggingface.co/SG161222/RealVisXL_V4.0/resolve/main/RealVisXL_V4.0.safetensors

ihsass•9/4/24, 6:13 PM

I am using kaggle

ihsass•9/4/24, 6:13 PM

I haved used sdxl base before

ihsass•9/4/24, 6:14 PM

I wnat to try with realvisxl

Iihsass I haved used sdxl base before

Furkan Gözükara SECourses•9/4/24, 6:14 PM

i know that is why i linked you

Furkan Gözükara SECourses•9/4/24, 6:14 PM

watch that

Furkan Gözükara SECourses•9/4/24, 6:14 PM

i am afk

ihsass•9/4/24, 6:14 PM

Ok sir thank you

Websteria•9/4/24, 7:10 PM

Good morning! I am doing the KOHYA flux training according to your video. I have an RTX 4090 however I'm only getting 4.39it/s even with the torch 2.5.... Any ideas on what I am doing that might cause it to be so slow?

Websteria•9/4/24, 7:15 PM

running windows 11

WWebsteria Good morning! I am doing the KOHYA flux training according to your video. I have...

Furkan Gözükara SECourses•9/4/24, 7:25 PM

Which config?

Websteria•9/4/24, 7:32 PM

Rank_2_27700MB_4_85_Second_IT

Websteria•9/4/24, 7:33 PM

wait

Websteria•9/4/24, 7:33 PM

Rank_3_18950MB_9_05_Second_IT

Websteria•9/4/24, 7:34 PM

GPU Memory is at 21GB

Websteria•9/4/24, 7:35 PM

@Dr. Furkan Gözükara

Dazzastrous [4090]•9/4/24, 7:37 PM

erm the first prompt wheres the airplane

WWebsteria Rank_2_27700MB_4_85_Second_IT

Furkan Gözükara SECourses•9/4/24, 7:37 PM

Great speed for rank 3

Furkan Gözükara SECourses•9/4/24, 7:37 PM

Use rank 4 for faster

DDazzastrous [4090]erm the first prompt wheres the airplane 🙂

Furkan Gözükara SECourses•9/4/24, 7:38 PM

Prompt was wrong fixed it :))

Websteria•9/4/24, 7:38 PM

gotcha. Thanks! Your video was very detailed and informative.

Furkan Gözükara SECourses•9/4/24, 7:38 PM

Will update grids

Websteria•9/4/24, 7:42 PM

@Dr. Furkan Gözükara Using Rank_4_16960MB_Fast.json and still getting

Websteria•9/4/24, 7:42 PM

12/2800 [00:41<2:42:26, 3.50s/it,

WWebsteria 12/2800 [00:41<2:42:26, 3.50s/it,

Furkan Gözükara SECourses•9/4/24, 7:43 PM

Wait 150 steps and let me know

Dazzastrous [4090]•9/4/24, 7:43 PM

Furkan Gözükara SECourses•9/4/24, 7:43 PM

But it is already like 2x of a6000

Websteria•9/4/24, 7:50 PM

@Dr. Furkan Gözükara After 150 steps these are the stats: 150/2800 [08:46<2:35:08, 3.51s/it, avr_loss=0.393]

slipkn67•9/4/24, 8:21 PM

can I use flux dev with 12GB vram?

WWebsteria @Dr. Furkan Gözükara After 150 steps these are the stats: 150/2800 [08:46<2:35:...

Furkan Gözükara SECourses•9/4/24, 8:22 PM

ok it is a good speed but slower than linux i guess

Furkan Gözükara SECourses•9/4/24, 8:22 PM

A6000 gets like 4.5 second it on linux

Sslipkn67 can I use flux dev with 12GB vram?

Furkan Gözükara SECourses•9/4/24, 8:22 PM

yep

FFurkan Gözükara SECourses ok it is a good speed but slower than linux i guess

GomezBro•9/4/24, 8:39 PM

Final question I just loaded up a 4GPU L40S and going to train, I did 4GPU config, but learning rate is 0.00005? Change it to 0.0001? Your config comes preloaded at 0.00005

GGomezBro Final question I just loaded up a 4GPU L40S and going to train, I did 4GPU confi...

Furkan Gözükara SECourses•9/4/24, 8:40 PM

change 0.0001

Furkan Gözükara SECourses•9/4/24, 8:40 PM

0.00005 i,s fo single gpu / b atch size

FFurkan Gözükara SECourses change 0.0001

GomezBro•9/4/24, 8:41 PM

you have your config saved that way, just FYI

GomezBro•9/4/24, 8:41 PM

I wonder how fast this will do 10000 images 1 Epoch about to find out

Furkan Gözükara SECourses•9/4/24, 8:41 PM

guys i need your careful analysis and your opinions which check point looks best

trained style images as a single grid : https://huggingface.co/MonsterMMORPG/3D-Cartoon-Style-FLUX/resolve/main/Trained_Style_Images.jpg
trained checkpoints grid comparison : https://huggingface.co/MonsterMMORPG/3D-Cartoon-Style-FLUX/resolve/main/Trained_Checkpoints_Massive_Grid.jpg

GGomezBro you have your config saved that way, just FYI

Furkan Gözükara SECourses•9/4/24, 8:41 PM

sure

@Dr. Furkan Gözükara Did you use bucketing enabled with your style training? I remember you talking

Similar Threads