Software Engineering Courses (SECourses)•2y ago

@Dr. Furkan Gözükara what is the current best dreambooth training for kohya? You've done a lot of te

@Dr. Furkan Gözükara what is the current best dreambooth training for kohya? You've done a lot of tests. Is it the 48gb_textencoder.json in your https://www.patreon.com/posts/very-best-for-of-89213064?

Rrafraf @Dr. Furkan Gözükara what is the current best dreambooth training for kohya? You...

Furkan Gözükara SECourses•5/19/24, 5:51 PM

not updated yet

Furkan Gözükara SECourses•5/19/24, 5:51 PM

currently working to update

rafrafOP•5/19/24, 5:52 PM

cool. You're updating the TE training version of db adafactor ?

rafrafOP•5/19/24, 5:53 PM

you have it at TE lr = 1/3 Unet lr

rafrafOP•5/19/24, 5:53 PM

Youre testing for betterTE training?

rafrafOP•5/19/24, 5:55 PM

"learning_rate": 1e-05,
  "learning_rate_te1": 3e-06,
 "learning_rate_te2": 0.0,

"learning_rate": 1e-05,
  "learning_rate_te1": 3e-06,
 "learning_rate_te2": 0.0,

Rrafraf ``` "learning_rate": 1e-05, "learning_rate_te1": 3e-06, "learning_rate_te2": ...

Furkan Gözükara SECourses•5/19/24, 7:45 PM

i am testing both

Furkan Gözükara SECourses•5/19/24, 7:45 PM

TE and Unet right now

Furkan Gözükara SECourses•5/19/24, 7:45 PM

will share on patreon

Furkan Gözükara SECourses•5/19/24, 8:04 PM

started new 9 training too

Furkan Gözükara SECourses•5/19/24, 8:04 PM

so total will be 4+9+9

Furkan Gözükara SECourses•5/19/24, 9:30 PM

RRMK @NicB Could you guys create a way to access JupyterLab inside MasseCompute from ...

NicB•5/19/24, 9:49 PM

Yes. I'll work on getting that fixed tomorrow with the team

NNicB Yes. I'll work on getting that fixed tomorrow with the team

Furkan Gözükara SECourses•5/19/24, 9:52 PM

i think he is asking any way to access jupyterlab interface from his computer

Furkan Gözükara SECourses•5/19/24, 9:52 PM

like gradio share

Furkan Gözükara SECourses•5/19/24, 9:52 PM

or like runpod proxy

Furkan Gözükara SECourses•5/19/24, 10:39 PM

adding adetailer and reactor to kaggle notebook

Furkan Gözükara SECourses•5/19/24, 10:44 PM

Kaggle notebook updated

Furkan Gözükara SECourses•5/19/24, 10:44 PM

Furkan Gözükara SECourses•5/19/24, 10:44 PM

https://www.patreon.com/posts/88714330

FFurkan Gözükara SECourses i think he is asking any way to access jupyterlab interface from his computer

NicB•5/19/24, 11:04 PM

Yes just need to double check that labs and notebook are opening the port and using host 0.0.0.0

I think notebook does automatically but need to check labs

Furkan Gözükara SECourses•5/19/24, 11:14 PM

new kohya trainings just finished

Furkan Gözükara SECourses•5/19/24, 11:14 PM

wow took a lot

Furkan Gözükara SECourses•5/19/24, 11:14 PM

runpod is really really slow

Furkan Gözükara SECourses•5/19/24, 11:14 PM

way slower than my PC RTX 3090 for some reason

Furkan Gözükara SECourses•5/19/24, 11:20 PM

https://www.linkedin.com/posts/furkangozukara_18-different-hyper-parameters-kohya-ss-stable-activity-7198099909412024321-rcxw?utm_source=share&utm_medium=member_desktop

Furkan Gözükara on LinkedIn: 18 different hyper parameters Kohya SS...

18 different hyper parameters Kohya SS Stable Diffusion XL (SDXL) DreamBooth / FineTuning trainings just finished

After recent updates of Kohya how the…

Furkan Gözükara SECourses•5/20/24, 1:07 AM

all the test files

now need to manually check and compare all

FFurkan Gözükara SECourses you cant

ForumlaLover•5/20/24, 1:40 AM

ForumlaLover•5/20/24, 1:41 AM

Question @Dr. Furkan Gözükara what is reccomended steps to upscale photo faces with default setting? still 50?

ForumlaLover•5/20/24, 1:41 AM

SUPIR

AIGambino•5/20/24, 1:51 AM

Does anyone know how to fix this?

FForumlaLover Question @Dr. Furkan Gözükara what is reccomended steps to upscale photo faces ...

Furkan Gözükara SECourses•5/20/24, 2:03 AM

i use 40 steps to generate org image

Furkan Gözükara SECourses•5/20/24, 2:03 AM

and 70 steps 50% denoise to restore face with adetailer

AAIGambino Does anyone know how to fix this?

Furkan Gözükara SECourses•5/20/24, 2:03 AM

you are missing models

Furkan Gözükara SECourses•5/20/24, 2:03 AM

my auto installers i am about to update

Furkan Gözükara SECourses•5/20/24, 2:03 AM

Furkan Gözükara SECourses•5/20/24, 2:03 AM

doing final test

FFurkan Gözükara SECourses you are missing models

AIGambino•5/20/24, 2:56 AM

What models?

Furkan Gözükara SECourses•5/20/24, 3:43 AM

Furkan Gözükara SECourses•5/20/24, 3:43 AM

https://www.patreon.com/posts/86307255

AAIGambino What models?

Furkan Gözükara SECourses•5/20/24, 3:43 AM

ip adapter

dsienra•5/20/24, 5:03 AM

@Dr. Furkan Gözükara I'm doing my first dreamboth training on kohya_ss previously I only used lora, I want to train many people at the same time on an xl model, friends and family from many ages and different genders, so what is the best to use dreamboth or fine tuning, one problem with dreamboth is regularization images because I'm training many classes at the same time, the other thing is the learning rate, I downloaded your preset for 24gb and the learning rate is too high compared to what I was used to use in 1.5 models that were 1e-6, I've read your YouTube post in the community that you are trying new learning rates for the new vesion of kohya, any advice?

FFurkan Gözükara SECourses updated and fixed

neo•5/20/24, 6:13 AM

finally worked!

thanks

FFurkan Gözükara SECourses all the test files 😄 now need to manually check and compare all

gohar•5/20/24, 7:06 AM

waiting for the best final config

Tuanny•5/20/24, 7:28 AM

trying to use supir on confyui with controlnet so i can up the hallucination while keeping similar composition not sure how to connect it. i did a LatentAdd node of the canny + the supir latent and connected to the supir sampler but getting this error.
"!!! Exception during processing!!! Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!
Traceback (most recent call last):
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\execution.py", line 151, in recursive_execute
output_data, output_ui = get_output_data(obj, input_data_all)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\execution.py", line 81, in get_output_data
return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\execution.py", line 74, in map_node_over_list
results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Tuan Do\Desktop\ComfyUI - TEST\ComfyUI\comfy_extras\nodes_latent.py", line 27, in op
samples_out["samples"] = s1 + s2

Tuanny•5/20/24, 7:30 AM

im able to do it on a regular img2img workflow with controlnet but not sure how to connect it with supir.

Ddsienra @Dr. Furkan Gözükara I'm doing my first dreamboth training on kohya_ss previousl...

Furkan Gözükara SECourses•5/20/24, 12:05 PM

kohya need new learning rate

Furkan Gözükara SECourses•5/20/24, 12:05 PM

i am today gonna announce

Furkan Gözükara SECourses•5/20/24, 12:05 PM

with good examples

Furkan Gözükara SECourses•5/20/24, 12:05 PM

did 22 trainings