Software Engineering Courses (SECourses)•3y ago

best hyper parametrs , its about these you posted on patreon 24GB_TextEncoder kohya ? , and how man

best hyper parametrs , its about these you posted on patreon 24GB_TextEncoder kohya ? , and how many images did you select for training, and repeats while using class images

wantedrbkOP•11/30/23, 2:05 PM

if im gonna use 3 4 gpus , i also need consider about for choosing right number of repeats , as u said , if i have two gpus need to devide it on number of gpus

Mmikemenders Now tell me why you use so many reps and spend hours training when it was done i...

aligg73•11/30/23, 2:14 PM

Very inspiring, have you considered checkpoint training with EveryDream2 instead of LoRa...and watch these same parameters? Just wondering whether this could lead to even better results. Although I agree with you, the quick turnaround of LoRa training is great from a feedback cycle point of view.

Wwantedrbk best hyper parametrs , its about these you posted on patreon 24GB_TextEncoder k...

Furkan Gözükara SECourses•11/30/23, 2:31 PM

i used 14 images

Wwantedrbk if im gonna use 3 4 gpus , i also need consider about for choosing right number ...

Furkan Gözükara SECourses•11/30/23, 2:32 PM

multiple gpu means increased batch size

Furkan Gözükara SECourses•11/30/23, 2:32 PM

so 2 gpu will make it batch size 2. so you should halve the number of steps

Furkan Gözükara SECourses•11/30/23, 2:32 PM

yes that training done with 24gb config on patreon

Furkan Gözükara SECourses•11/30/23, 2:32 PM

https://www.patreon.com/posts/very-best-for-of-89213064

Furkan Gözükara SECourses•11/30/23, 2:33 PM

24GB_TextEncoder.json

FFurkan Gözükara SECourses multiple gpu means increased batch size

wantedrbkOP•11/30/23, 3:02 PM

Thanks , Repeats should be around 160 for sdxl , I remember for 1.5 it’s easy to overtrain and 25-40 was okay

DR.Siva•11/30/23, 3:23 PM

this time , the image creation needed very little effort to bring the face output. Using correct , HQ face images , makes lora training much easier.

AIGambino•11/30/23, 3:54 PM

Do we use extract faces or resizer?

AIGambino•11/30/23, 4:24 PM

is this correct? I just want to do 1024x1024

Aaligg73 Very inspiring, have you considered checkpoint training with EveryDream2 instead...

mikemenders•11/30/23, 5:43 PM

I find the big model training unnecessary. I'd only do it if I really wanted to make a new model, but that would require millions of photos, and I only do DreamBooth very rarely (for example for a pose that no model can do by default, like a full undereye). For everything else, I have Lora, which is fast, perfect for me, and the quality is absolutely comparable to a DB model, plus I can put it on top of any model, mix it with other Lora.

AAIGambino is this correct? I just want to do 1024x1024

mikemenders•11/30/23, 5:48 PM

Need only one DATA array only for all size what you need, example:

DATA = [
    ((512, 512), r"G:\man dataset\new_cropped\512x512"),
    ((768, 768), r"G:\man dataset\new_cropped\768x768"),
    ((1024, 1024), r"G:\man dataset\new_cropped\1024x1024")
]

DATA = [
    ((512, 512), r"G:\man dataset\new_cropped\512x512"),
    ((768, 768), r"G:\man dataset\new_cropped\768x768"),
    ((1024, 1024), r"G:\man dataset\new_cropped\1024x1024")
]

AIGambino•11/30/23, 5:49 PM

Can I just delete the second data = [] one?

mikemenders•11/30/23, 5:49 PM

yes

mikemenders•11/30/23, 5:50 PM

last comma no need

AIGambino•11/30/23, 5:51 PM

its being strange

AIGambino•11/30/23, 5:51 PM

cropping and rotating

AIGambino•11/30/23, 5:52 PM

and also its 512?

mikemenders•11/30/23, 5:53 PM

Yes, it's 512. Have you used the Yolo script before?

AIGambino•11/30/23, 5:54 PM

nope

mikemenders•11/30/23, 5:54 PM

First is the Yolo script, it cuts the person. Then select the best images by hand or with CompreFace and then use this cropping tool.

AIGambino•11/30/23, 5:55 PM

whats extract_faces for?

mikemenders•11/30/23, 5:56 PM

There is a Yolo script under 30th october update:
https://www.patreon.com/posts/sota-subject-and-88391247

mikemenders•11/30/23, 5:56 PM

"Yolo v7 based cropper script updated and now works much better."

mikemenders•11/30/23, 5:57 PM

or Auto cropper (that's same)

mikemenders•11/30/23, 5:57 PM

Under this section found newer version on same patreon link:
How To Use Subject Cropper

mikemenders•11/30/23, 5:58 PM

I use (1536, 1536) aspect ratio only, because need square crop later and this size is largest. No need other ratios from there

mikemenders•11/30/23, 6:00 PM

cropper_v7.py, line 13

mikemenders•11/30/23, 6:00 PM

Once you're done cutting and sorting, then use the resizer and it will cut it out really well.

Mishroom•11/30/23, 6:01 PM

Hi! I have some issues with Dreambooth Kohya, I tried to start learning but it stopped because of ‘ Doesn’t exist img folder’. I upload train dataset in workspace and pointed in dreambooth. It has happened on Runpod

Mmikemenders There is a Yolo script under 30th october update: https://www.patreon.com/posts/...

AIGambino•11/30/23, 6:02 PM

So I have the yolo v7 in its completely own folder?

mikemenders•11/30/23, 6:03 PM

Yes, create a different folder for yolo script

mikemenders•11/30/23, 6:06 PM

and after install change cropper_v7.py constants, then go run_cropper.bat

mikemenders•11/30/23, 6:09 PM

in cropper_v7.py comment second ASPECT_RATIOS line, need first one only.

AIGambino•11/30/23, 6:09 PM

this look good?