Software Engineering Courses (SECourses)•3y ago

Stable Diffusion XL (SDXL) DreamBooth Training eror (((

John CeedOP•11/27/23, 6:08 PM

File "/opt/conda/bin/accelerate", line 8, in <module>
sys.exit(main())
File "/opt/conda/lib/python3.10/site-packages/accelerate/commands/accelerate_cli.py", line 47, in main
args.func(args)
File "/opt/conda/lib/python3.10/site-packages/accelerate/commands/launch.py", line 977, in launch_command
multi_gpu_launcher(args)
File "/opt/conda/lib/python3.10/site-packages/accelerate/commands/launch.py", line 646, in multi_gpu_launcher
distrib_run.run(args)
File "/opt/conda/lib/python3.10/site-packages/torch/distributed/run.py", line 785, in run
elastic_launch(
File "/opt/conda/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 134, in call
return launch_agent(self._config, self._entrypoint, list(args))
File "/opt/conda/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 250, in launch_agent
raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
=====================================================
./sdxl_train.py FAILED
-----------------------------------------------------
Failures:
<NO_OTHER_FAILURES>
-----------------------------------------------------
Root Cause (first observed failure):
[0]:
time : 2023-11-27_18:01:49
host : c4565fefbe4d
rank : 1 (local_rank: 1)
exitcode : -6 (pid: 1053)
error_file: <N/A>
traceback : Signal 6 (SIGABRT) received by PID 1053
=====================================================

mikemenders•11/27/23, 8:01 PM

I discover the hidden things of Prodigy. For example, if I don't have 25 pictures, but 21 or 26, I get more exciting results. For example, with 26 images, I had to lower the d_coef from 1.02 to 0.85, and even then Lora was a bit strong.
Even at 21 images it might be too strong if I set it to 4 repetitions instead of 5. So there are still some small things to explore, but the fact is that whether Lora is stronger or weaker, it still gives a nice result, as long as I fine tune its flexibility.

mikemenders•11/27/23, 8:18 PM

Here is a sample from SD 1.5 with 768x768 and without Hires Fix

01427-2719919611-RAW_photo_a_cvillaloboss_woman_in_beautifil_dress_sitting_in_a_chair_in_restaurant_portrait_looking_at_viewer__lora_cvillalo.png

mikemenders•11/27/23, 8:22 PM

I trained model with 512x512 images

Troubleman•11/27/23, 8:25 PM

Does anyone know the formula for number or regularization images? Something like repeats * train image count / batch size == reg images?

John CeedOP•11/27/23, 8:25 PM

I can't use more than 13 training photos. error

JJohn Ceed I can't use more than 13 training photos. error

Troubleman•11/27/23, 8:27 PM

Including the error in your message might help

SStefanoid9343 no training on this new model yet right?

Furkan Gözükara SECourses•11/27/23, 10:03 PM

they have script but 0 info

DDR.Siva Thanks to @Dr. Furkan Gözükara , Kaggle (for this wonderful free platform) , Hug...

Furkan Gözükara SECourses•11/27/23, 10:03 PM

good job

AAIGambino Will you have a tutorial on using esrgan models @Dr. Furkan Gözükara ?

Furkan Gözükara SECourses•11/27/23, 10:04 PM

you just put them into folder test

Furkan Gözükara SECourses•11/27/23, 10:04 PM

but i plan to do

NNed Shoaei I am using RunPod's Template

Furkan Gözükara SECourses•11/27/23, 10:04 PM

can you tell me templat ename

JJohn Ceed File "/opt/conda/bin/accelerate", line 8, in <module> sys.exit(main()) Fil...

Furkan Gözükara SECourses•11/27/23, 10:04 PM

this doesnt tell the error

Furkan Gözükara SECourses•11/27/23, 10:04 PM

can you record video and send me from dm

TTroubleman Does anyone know the formula for number or regularization images? Something like...

Furkan Gözükara SECourses•11/27/23, 10:05 PM

repeat * train count

JJohn Ceed I can't use more than 13 training photos. error

Furkan Gözükara SECourses•11/27/23, 10:05 PM

you can use

FFurkan Gözükara SECourses can you tell me templat ename

Ned Shoaei•11/27/23, 11:22 PM

RunPod has two SD templates, one is fast, I am using the other one. it's webui 10.2.1

NNed Shoaei RunPod has two SD templates, one is fast, I am using the other one. it's webui 1...

Furkan Gözükara SECourses•11/27/23, 11:45 PM

ok i have web ui controlnet installer

Furkan Gözükara SECourses•11/27/23, 11:46 PM

https://www.patreon.com/posts/84896373

DR.Siva•11/28/23, 12:33 AM

Just throwing an idea, how will be the outcome of Lora trained with base model used from the output of dreambooth..

DR.Siva•11/28/23, 1:42 AM

using extra weight to lora and ohwx woman only brings out better results for me(my trained lora) , giving atleast 1.1 to lora and ((ohwx woman)/ ohwx woman:2 gives better results

Le•11/28/23, 2:14 AM

Or of this is not even the top option we have now, and there is a better way to make it with even more quality

AIGambino•11/28/23, 2:31 AM

I have two custom models I trained in SDXL, I want to have them both in the same photo, how can I do this?

AIGambino•11/28/23, 2:47 AM

wow what causes this? these colors arent in my training images at all :\

Rae•11/28/23, 3:31 AM

any tool to mass rename images, or have an easy tool to easily adjust filenames with drag / drop filename crumbs?

RRae any tool to mass rename images, or have an easy tool to easily adjust filenames ...

GeekDadKevin•11/28/23, 3:59 AM

I use https://www.advancedrenamer.com/

Advanced Renamer - Free and fast batch rename utility for files and...

Batch rename files and folders in a snap. Perfect for any kind of file renaming including music and photo files. Preview your files before renaming and undo erroneous renaming jobs.

Advanced Renamer - Free and fast batch rename utility for files and...

Batch rename files and folders in a snap. Perfect for any kind of file renaming including music and photo files. Preview your files before renaming and undo erroneous renaming jobs.

Nora•11/28/23, 4:36 AM

What is the most recent video about using dreambooth in google colab without a laptop with gpu

Neuromasters•11/28/23, 9:18 AM

@Dr. Furkan Gözükara You've released a lot of new stuff! Now I just have to find the time to study it all)
I don't understand how classes work, I've been trying to figure it out for how long, I can't get it. Physically, when we train a large model on characters, we roughly divide into a man and a woman. And if on subjects? Then every class is a subject? Car, Chair, Door, bag or sheep, Dragon, whale

And then I haven't seen anyone use the indicators, and here is the Reg_1 folder, how the learning structure should look in it. How to write files in it and what should they be?

Original message was deleted

Furkan Gözükara SECourses•11/28/23, 10:48 AM

looks like that service D-ID and eleven labs

AAIGambino I have two custom models I trained in SDXL, I want to have them both in the same...

Furkan Gözükara SECourses•11/28/23, 10:49 AM

extract lora and do regional prompting

RRae any tool to mass rename images, or have an easy tool to easily adjust filenames ...

Furkan Gözükara SECourses•11/28/23, 10:50 AM

i would use chatgpt to get written easy python script

Pan_Pierdziadek•11/28/23, 10:51 AM

guys, i got 770 images of dataset. how many steps should i set to train it on kaggle? it takes 2 minutes per step.....

NNora What is the most recent video about using dreambooth in google colab without a l...

Furkan Gözükara SECourses•11/28/23, 10:53 AM

i have kaggle sdxl dreambooth

Furkan Gözükara SECourses•11/28/23, 10:53 AM

https://youtu.be/16-b1AjvyBE

YouTubeSECourses

How To Do Stable Diffusion XL (SDXL) DreamBooth Training For Free -...

Master Stable Diffusion XL Training on Kaggle for Free!

Welcome to this comprehensive tutorial where I'll be guiding you through the exciting world of setting up and training Stable Diffusion XL (SDXL) with Kohya on a free Kaggle account. This video is your one-stop resource for learning everything from initiating a Kaggle session with dual ...

PPan_Pierdziadek guys, i got 770 images of dataset. how many steps should i set to train it on ka...

Furkan Gözükara SECourses•11/28/23, 10:53 AM

what is the purpose of training

Pan_Pierdziadek•11/28/23, 10:54 AM

just put some realistic images to train on

PPan_Pierdziadek just put some realistic images to train on

Furkan Gözükara SECourses•11/28/23, 10:57 AM

ye but training a person or what else?

Pan_Pierdziadek•11/28/23, 10:57 AM

instagram photos of girls

Furkan Gözükara SECourses•11/28/23, 10:57 AM

so you are looking for fine tuning

Furkan Gözükara SECourses•11/28/23, 10:58 AM

well cant tell certain number of epochs

Furkan Gözükara SECourses•11/28/23, 10:58 AM

for example even 100 epoch wasnt overtraining for 177 images when training a style

Pan_Pierdziadek•11/28/23, 10:58 AM

btw should i use dreambooth tab or finetuning tab in kohya?

FFurkan Gözükara SECourses looks like that service D-ID and eleven labs

Le•11/28/23, 2:14 PM

Thanks Professor

FFurkan Gözükara SECourses looks like that service D-ID and eleven labs

Le•11/28/23, 2:19 PM

Is there another option to D-ID that is not paid? D-ID is so expensive.

FFurkan Gözükara SECourses extract lora and do regional prompting

AIGambino•11/28/23, 2:28 PM

do you have a tutorial on regional prompting and lora extractions?

Praveen•11/28/23, 2:45 PM

Im trying to find captioning techniques in your guides but couldnt get a clear idea. The SOTA you mentioned captions the image. Is that the one we need to use for Dreambooth/ Lora? Is Ohwx still best way to go to name someone? I saw some YT videos saying name to celebrity with high similarity is better way.

PPan_Pierdziadek btw should i use dreambooth tab or finetuning tab in kohya?

Furkan Gözükara SECourses•11/28/23, 3:12 PM

for dreambooth yes

Furkan Gözükara SECourses•11/28/23, 3:12 PM

dreambooth and fine tune same actually

Furkan Gözükara SECourses•11/28/23, 3:13 PM

only difference is using reg images

LLe Is there another option to D-ID that is not paid? D-ID is so expensive.

Furkan Gözükara SECourses•11/28/23, 3:13 PM

yes paid and expensive

Stable Diffusion XL (SDXL) DreamBooth Training eror (((

Similar Threads

Similar Threads

Similar Threads