Software Engineering Courses (SECourses)•17mo ago

Native windows but with Apply T5 Attention Mask off, allmost the same speed as linux

nicearoni•8/28/24, 7:00 PM

very nice. i'm trying first style lora. 696 images lol. T5 attention mask is on because i joy captioned all the images. took the images down to 512 since it's such a huge dataset. getting 3.04s/it

dsienraOP•8/28/24, 7:02 PM

nice! Lets se how 512 resolution affects the quality

dsienraOP•8/28/24, 7:04 PM

T5 attention mask affects speed big time

dsienraOP•8/28/24, 7:06 PM

may be it was why I was having 4.1 it/s yesterday on wsl, maybe T5 attention mask was disabled

nicearoni•8/28/24, 7:30 PM

i really want to stay up till 1am and play with it when it's done but i'll probably fall asleep lol

Poppy•8/28/24, 7:33 PM

@Dr. Furkan Gözükara hey

on runpod when I deploy comfy ui template, Jupiter lab doesn’t allow me to open the checkpoint folder to add safetensors. Is this normal? I’m able to open the the other folders such as Lora etc

Nnicearoni very nice. i'm trying first style lora. 696 images lol. T5 attention mask is on ...

Sipriyani•8/28/24, 7:37 PM

when you do , lora training, how do you, implement captions to the training?

nicearoni•8/28/24, 7:40 PM

just a text file with the same base name as the picture. if extension is .txt, kohya should detect and use by default.

nicearoni•8/28/24, 7:41 PM

Joy Caption started every caption with "This photograph..." so I used Doc's caption editor app to batch replace with my token

mikemenders•8/28/24, 7:50 PM

I'm a bit of a Pony. I managed to find the right parameters under Fooocus for my personal Loras, which I trained to PonyRealism 2.1. Unfortunately none of the realistic Pony models are enough on their own, so I'm happy with the presets below:

pony_realistic_person_lora.json1.1KB

dsienraOP•8/28/24, 7:57 PM

Native windows 1024x1024 same speed as wsl, T5 attention mask disabled

dsienraOP•8/28/24, 7:58 PM

the new torch worked

dsienraOP•8/28/24, 8:00 PM

I eliminated the class from the captions, Lets see if the bleeding is gone.

dsienraOP•8/28/24, 8:17 PM

@Dr. Furkan Gözükara what lr do you recommend for TE and UNET or you are still testing?

dsienraOP•8/28/24, 8:20 PM

Is for training multiple subjects I want to increase resemblance and decrease bleeding, I removed the class from the captions

Djudge•8/28/24, 8:21 PM

Could someone tell me why I've been stuck at 0% for 2 hours without an error message with an RTX 4090

DDjudge Could someone tell me why I've been stuck at 0% for 2 hours without an error mes...

dsienraOP•8/28/24, 8:25 PM

you are doing dreambooth or lora?

Djudge•8/28/24, 8:38 PM

Hi, thx you. I try to creat lora

Djudge•8/28/24, 8:38 PM

with Kohya

dsienraOP•8/28/24, 8:39 PM

you are in the lora tab?

Djudge•8/28/24, 8:40 PM

yes

Djudge•8/28/24, 8:40 PM

Djudge•8/28/24, 8:41 PM

i folllow the tuto on patreon

dsienraOP•8/28/24, 8:41 PM

it should work then

dsienraOP•8/28/24, 8:42 PM

I saw that in your log, I cant find it in mine, thats why I asked

Djudge•8/28/24, 8:44 PM

Thanks, that's interesting, I need to understand why now.

dsienraOP•8/28/24, 8:46 PM

you are runing on masses compute or in you own computer?

dsienraOP•8/28/24, 8:46 PM

is in your computer

dsienraOP•8/28/24, 8:46 PM

I can see your path

Djudge•8/28/24, 8:46 PM

my computer

dsienraOP•8/28/24, 8:46 PM

that preset consumes to muach vram

dsienraOP•8/28/24, 8:47 PM

you can't run it in your computer

dsienraOP•8/28/24, 8:48 PM

pick one for your graphics card

Djudge•8/28/24, 8:48 PM

However, I have a very large configuration

dsienraOP•8/28/24, 8:48 PM

what grapic card do you have?

Djudge•8/28/24, 8:49 PM

rtx 4090

dsienraOP•8/28/24, 8:49 PM

not enough

Djudge•8/28/24, 8:49 PM

-_-' really !

dsienraOP•8/28/24, 8:49 PM

yes

dsienraOP•8/28/24, 8:49 PM

you have 24gb

dsienraOP•8/28/24, 8:49 PM

that preset consumes 28gb

Djudge•8/28/24, 8:50 PM

hooooooo you are so stong

Djudge•8/28/24, 8:50 PM

thx yo so much, i gona test other preset

dsienraOP•8/28/24, 8:50 PM

28700mb

Djudge•8/28/24, 8:50 PM

ok i undestand

Djudge•8/28/24, 8:51 PM

thx you so much

dsienraOP•8/28/24, 8:51 PM

you welcome

Native windows but with Apply T5 Attention Mask off, allmost the same speed as linux

Similar Threads

Similar Threads

Similar Threads