Software Engineering Courses (SECourses)•15mo ago

<@205854764540362752> I'm looking at the 48gb one. It says the repeat is set at 1 but the config i

@Furkan Gözükara SECourses I'm looking at the 48gb one. It says the repeat is set at 1 but the config is at 40. Should it be instead set at 1?

VVigilence <@205854764540362752> I'm looking at the 48gb one. It says the repeat is set a...

Furkan Gözükara SECourses•10/15/24, 8:39 PM

rename your folder to 1

VigilenceOP•10/15/24, 8:40 PM

@Furkan Gözükara SECourses I appreciate the confirmation. I would recommend double checking the epoch and repeat for your config, it might have mistakes in this area.

Furkan Gözükara SECourses•10/15/24, 8:42 PM

ye i think e4m3fn better
https://imgsli.com/MzA3OTY0/0/2

VVigilence <@205854764540362752> I appreciate the confirmation. I would recommend double ...

Furkan Gözükara SECourses•10/15/24, 8:43 PM

how many images you have to train?

VigilenceOP•10/15/24, 8:43 PM

VVigilence 89

Furkan Gözükara SECourses•10/15/24, 8:43 PM

ok train up to 100 epoch at least

Furkan Gözükara SECourses•10/15/24, 8:43 PM

preferably 150

Furkan Gözükara SECourses•10/15/24, 8:43 PM

and compare

VigilenceOP•10/15/24, 8:44 PM

So 150 epoch at 1 repeat or is higher recommended?

VVigilence So 150 epoch at 1 repeat or is higher recommended?

Furkan Gözükara SECourses•10/15/24, 8:47 PM

always 1 repeat with flux

VigilenceOP•10/15/24, 8:48 PM

@Furkan Gözükara SECourses I see. That makes the results even more impressive! But the images are all of your face. Do you know if its recommended for styles like oil paintings?

VVigilence <@205854764540362752> I see. That makes the results even more impressive! But th...

Furkan Gözükara SECourses•10/15/24, 8:49 PM

it is totally same

Furkan Gözükara SECourses•10/15/24, 8:49 PM

i have style training

Furkan Gözükara SECourses•10/15/24, 8:50 PM

https://huggingface.co/MonsterMMORPG/3D-Cartoon-Style-FLUX

MonsterMMORPG/3D-Cartoon-Style-FLUX · Hugging Face

VigilenceOP•10/15/24, 8:54 PM

That's actually impressive. I always see in your videos pictures of yourself.

Have you tried mixing it up with styles to see if you get more views/subscriptions?

VVigilence That's actually impressive. I always see in your videos pictures of yourself. ...

Furkan Gözükara SECourses•10/15/24, 8:56 PM

i may

VigilenceOP•10/15/24, 8:56 PM

I would recommend it. Training the images of the face is good for people who make celebrity loras etc, but most of us want to train on styles, concepts etc.

VigilenceOP•10/15/24, 8:57 PM

You can even do some on characters etc.

VigilenceOP•10/15/24, 9:07 PM

@Furkan Gözükara SECourses Have you looked into flux rapid lora training? Both fal and civit.ai offer it.

https://education.civitai.com/quickstart-guide-to-flux-1/#rapid-flux-training

VVigilence <@205854764540362752> Have you looked into flux rapid lora training? Both fal ...

Furkan Gözükara SECourses•10/15/24, 9:08 PM

i checked civitiai

Furkan Gözükara SECourses•10/15/24, 9:08 PM

as expected it was inferior

Furkan Gözükara SECourses•10/15/24, 9:08 PM

i didnt pay for fal

Furkan Gözükara SECourses•10/15/24, 9:08 PM

i know it will be too

VigilenceOP•10/15/24, 9:10 PM

I think if you make your own version you would get lots of begginers to payfor the config. I know I would have signed up for it back then.

wardensc2•10/16/24, 4:32 AM

Hi @Furkan Gözükara SECourses I use your Batch_Size_7_48GB_GPU_46250MB_28.9_second_it_Tier_1. In the config I only see 1 GPU not multi GPU, so you only increase the batch file to 7 right. What about the time and quality if I use your batch 7, does it save me 7 times less than your batch 1 config and the quality is the same ?

Constipated•10/16/24, 6:01 AM

@Furkan Gözükara SECourses saw your Flux 15 images vs 256 analysis and it was very insightful. Obviously 256 images may not be practical in many cases and would take a long time on Batch 1. Let's say I have the more realistic option of using 30 or 40 images. Do you think that is better than 15?

Constipated•10/16/24, 6:02 AM

Assuming all are high quality images

Wwardensc2 Hi <@205854764540362752> I use your Batch_Size_7_48GB_GPU_46250MB_28.9_second_it...

Furkan Gözükara SECourses•10/16/24, 9:50 AM

if you get 7 GPU SXM machine with A100

Furkan Gözükara SECourses•10/16/24, 9:50 AM

it will be like 7 times faster

Furkan Gözükara SECourses•10/16/24, 9:50 AM

and you can use same learning rate

CConstipated <@205854764540362752> saw your Flux 15 images vs 256 analysis and it was very in...

Furkan Gözükara SECourses•10/16/24, 9:50 AM

definitely better

Furkan Gözükara SECourses•10/16/24, 9:50 AM

15 images underperform

FFurkan Gözükara SECourses if you get 7 GPU SXM machine with A100

wardensc2•10/16/24, 10:36 AM

So I can't use A6000 right ? because based your named 48gb I though 7xA6000 are enough

Carlos•10/16/24, 10:42 AM

Is Torch 2.5 OK now or not? I can't figure it out from the Video Tutorial, this is for Dreambooth Flux

Wwardensc2 So I can't use A6000 right ? because based your named 48gb I though 7xA6000 are ...

Furkan Gözükara SECourses•10/16/24, 12:28 PM

you can use

Furkan Gözükara SECourses•10/16/24, 12:28 PM

but not for fine tuning

Furkan Gözükara SECourses•10/16/24, 12:28 PM

but for lora multi gpu

CCarlos Is Torch 2.5 OK now or not? I can't figure it out from the Video Tutorial, this ...

Furkan Gözükara SECourses•10/16/24, 12:28 PM

it is ok to use for windows

Will Scarlett Ohara•10/16/24, 1:44 PM

https://1drv.ms/u/s!AoU1QUjSr6pJrbdRgnQNUMMyMe2qTg?e=xCOQuv
Hi, could you guide me through what's wrong with this training. My latest samples at 4400 steps are still far from imitating Justin Sweet's style. Thanks