Software Engineering Courses (SECourses)•2y ago

Getting quite good results with sd1.5 lora training with new settings

�🕯Fox Plaid🕯Ahh okay

FiveSAIOP•12/31/23, 2:03 AM

you can also try training the sdxl model on 768x768

�

🕯Fox Plaid🕯•12/31/23, 2:04 AM

I did that on 768x768 but still out of memory

�

🕯Fox Plaid🕯•12/31/23, 2:04 AM

I just think the program is too overkill for my pc

AIGambino•12/31/23, 2:09 AM

Can someone tell me the setting theyre using ot train a 1.5 model I cannot get it working to save my life and nobody seems to have any insight on specific settings

�🕯Fox Plaid🕯I just think the program is too overkill for my pc

FiveSAIOP•12/31/23, 2:09 AM

are you using xformers and gradient checkpointing ?

�

🕯Fox Plaid🕯•12/31/23, 2:09 AM

Yes

FiveSAIOP•12/31/23, 2:10 AM

memory efficient attention?

AAIGambino Can someone tell me the setting theyre using ot train a 1.5 model I cannot get i...

FiveSAIOP•12/31/23, 2:12 AM

depends what are you aiming for ? realism?

AIGambino•12/31/23, 2:14 AM

yeah

FFiveSAI Getting quite good results with sd1.5 lora training with new settings

mikemenders•12/31/23, 7:23 AM

Can you share your training script?

Deleted User•12/31/23, 9:51 AM

--ddp_gradient_as_bucket_view

DDeleted User --ddp_gradient_as_bucket_view

Deleted User•12/31/23, 9:51 AM

Is not working in kaggle

Deleted User•12/31/23, 9:51 AM

Need too much memory

tom•12/31/23, 10:32 AM

I'm having trouble with kaggle. I tried the original Lora training video and got into the Kohya gui without any issues. But, the json file was a no go.

tom•12/31/23, 10:33 AM

Tried the updated tut and now the notebook keeps timing out? Any ideas?

tom•12/31/23, 10:41 AM

Ok, so I got back into the gui. But loading the json file gives me this : Error
Unexpected token '<', " <!DOCTYPE "... is not valid JSON

tom•12/31/23, 10:41 AM

Am I missing a step somewhere?

Ttom Ok, so I got back into the gui. But loading the json file gives me this : Error ...

Benjamin•12/31/23, 10:43 AM

looks like the json file is not correct

Benjamin•12/31/23, 10:43 AM

some error when you downloaded the file

tom•12/31/23, 10:46 AM

Hmm. Tried to redownload the fille. Then noticed kaggle stopped the session again: ^C
Keyboard interruption in main thread... closing server.
Killing tunnel 127.0.0.1:7860

tom•12/31/23, 10:47 AM

This is the file I'm using Kaggle_SDXL_DreamBooth_Best.json. I shows up fine. Copy and paste folder link in gui. But the gui won't even let me create a file like "test1.json

Benjamin•12/31/23, 10:51 AM

by Unexpected token '<', " <!DOCTYPE "... is not valid JSONUnexpected token '<', " <!DOCTYPE "... is not valid JSON it's saying that the file starts with <!DOCTYPE<!DOCTYPE and this isn't the case for a valid json

Ttom Hmm. Tried to redownload the fille. Then noticed kaggle stopped the session agai...

Benjamin•12/31/23, 10:51 AM

are you using the latest kaggle template? normally it is only killing the session with the old kaggle templates that don't use ngrok

Deleted User•12/31/23, 10:55 AM

I have this problem in kaggle

Deleted User•12/31/23, 11:15 AM

update to kaggle dreambooth best .json is not working in kaggle for me

DDeleted User update to kaggle dreambooth best .json is not working in kaggle for me

tom•12/31/23, 11:27 AM

Same. I think maybe kaggle updated something somewhere and errors are popping up. Maybe give it a few days. or ... sigh ... might have to find a gpu. I'm going to muddle around on colab.

tom•12/31/23, 11:27 AM

getting frustrated with kaggle at the moment

Sathish H•12/31/23, 12:27 PM

@Dr. Furkan Gözükara I have been trying to do SDXL dreambooth training on Kaggle. I'm getting below issue. Pls assist

Sathish H•12/31/23, 12:31 PM

My bad.. i figured out localserver was not running. Its resolved. Thanks

Ttom Same. I think maybe kaggle updated something somewhere and errors are popping up...

Deleted User•12/31/23, 1:45 PM

Yeah they fixing the problem first there was catche issue

Original message was deleted

Deleted User•12/31/23, 1:46 PM

You can go with it

AAIGambino Any idea why sometimes itll only export the same image over and over despite it ...

Furkan Gözükara SECourses•12/31/23, 2:17 PM

sometimes bugged

VVlad Hello @Dr. Furkan Gözükara can I please ask you 2 questions? 1. is it better to ...

Furkan Gözükara SECourses•12/31/23, 2:18 PM

i would suggest this. do dreambooth on SDXL . then extract lora. are you really need sd 1.5?

JJMPA May someone answer this?

Furkan Gözükara SECourses•12/31/23, 2:19 PM

please do screen record hard to understand from there

KKallamamran @Dr. Furkan Gözükara I have 1962 reg-images i need to filter out which ones I ca...

Furkan Gözükara SECourses•12/31/23, 2:19 PM

resizer will resize

Furkan Gözükara SECourses•12/31/23, 2:19 PM

crop will not

�🕯Fox Plaid🕯hello im getting this issue when trying to make a sdxl lora with my pc using the...

Furkan Gözükara SECourses•12/31/23, 2:20 PM

hello. you cant do lora of sdxl with 8 gb vram gpu

Furkan Gözükara SECourses•12/31/23, 2:20 PM

you can only do sd 1.5

AAIGambino Can someone tell me the setting theyre using ot train a 1.5 model I cannot get i...

Furkan Gözükara SECourses•12/31/23, 2:21 PM

i am still yet to explore this. sadly couldnt find time yet

DDeleted User Is not working in kaggle

Furkan Gözükara SECourses•12/31/23, 2:22 PM

i updated the patreon post

Ttom Same. I think maybe kaggle updated something somewhere and errors are popping up...

Furkan Gözükara SECourses•12/31/23, 2:22 PM

updated the atreon post

Furkan Gözükara SECourses•12/31/23, 2:23 PM

https://www.patreon.com/posts/88397937

DJRich•12/31/23, 2:29 PM

Hi @Dr. Furkan Gözükara Is there a good formula on setting repeats/epochs when you're using a larger number of images for a DreamBooth training? For example, I have a subject that has 90 images. If I use the default DreamBooth SDXL preset you created for 16GB textencoding, it defaults to doing 18000 steps which takes 7 hours on my mobile 4090. I can reduce the image count, but even if I cut the number down to 30 images, it still takes between 3.5-4 hours. I am using all of the default settings in your preset as well as 20 repeats for my images and 1 repeat for the regularization images. The default epochs in the preset is 8.