Software Engineering Courses (SECourses)•3y ago

is this where to ask for help?

DJ_Body manOP•7/12/23, 4:25 AM

w sdzl

DJ_Body manOP•7/12/23, 4:25 AM

sdxl

FFurkan Gözükara SECourses how many images you have and what is your per instance class count?

Narva•7/12/23, 6:28 AM

I have 24 Training dataset and 200 per instance class, I followed your tutorial, it only generated 500 class image

Narva•7/12/23, 6:32 AM

also dreambooth found the rest of the class images and after that the training started, maybe some class images were corrupted

FFurkan Gözükara SECourses Click to see attachment

Khang•7/12/23, 7:33 AM

Those are nice pics Dr. Furkan. Did you create them with ComfyUI? I hope you will make a video comparing the pros and cons of each UI.

DDJ_Body man is this where to ask for help?

Furkan Gözükara SECourses•7/12/23, 10:22 AM

yeyes

Vveldonlives i am legitimately curious - the config settings must be somewhat "best practice"...

Furkan Gözükara SECourses•7/12/23, 10:22 AM

i agree

KKhang Those are nice pics Dr. Furkan. Did you create them with ComfyUI? I hope you wil...

Furkan Gözükara SECourses•7/12/23, 10:22 AM

yes

Furkan Gözükara SECourses•7/12/23, 10:22 AM

and i am working on it

NNarva I have 24 Training dataset and 200 per instance class, I followed your tutorial,...

Furkan Gözükara SECourses•7/12/23, 10:23 AM

generate class images from txt2img tab

Furkan Gözükara SECourses•7/12/23, 10:23 AM

i did a lora training on sdxl yesterday on my computer

Furkan Gözükara SECourses•7/12/23, 10:23 AM

looks like my LR was too low

Furkan Gözükara SECourses•7/12/23, 10:24 AM

dmn this thing taking time

Furkan Gözükara SECourses•7/12/23, 10:24 AM

to experiment

Furkan Gözükara SECourses•7/12/23, 11:23 AM

started another lora training

Furkan Gözükara SECourses•7/12/23, 11:24 AM

90 min on rtx 3090 for 13 images 200 epochs + class images

FFurkan Gözükara SECourses 90 min on rtx 3090 for 13 images 200 epochs + class images

Jehex•7/12/23, 11:43 AM

hi there, I have one question about the " steps " of the images training ( in kohya ss we have to rename our training folder like 30_namemodel ) my question is how many steps do you recommend for get a decent result ? I see that divise 1000 by the number of training images can give good results ( for exemple we have 60 training images so 1000/60 = 16-17 steps per images ) I did it today and the result was honestly not too bad. Thanks

JJehex hi there, I have one question about the " steps " of the images training ( in ...

Furkan Gözükara SECourses•7/12/23, 11:48 AM

i calculate based on epoch

Furkan Gözükara SECourses•7/12/23, 11:48 AM

so 10 epoch 20 repeat means 200 epochs

Furkan Gözükara SECourses•7/12/23, 11:48 AM

usually 200 epochs is sufficient

Furkan Gözükara SECourses•7/12/23, 11:48 AM

if overtrains or dont learns change learning rate

Jehex•7/12/23, 11:48 AM

ok thanks, I will test that

Brofessor•7/12/23, 12:45 PM

hello

Brofessor•7/12/23, 12:45 PM

trying to set up Lora Kohya on runpod according to guide

Brofessor•7/12/23, 12:45 PM

anyone seen this error?: train_network.py: error: unrecognized arguments: --cache_text_encoder_outputs

Brofessor•7/12/23, 12:45 PM

TF-TRT Warning: Could not find TensorRT
usage: train_network.py [-h] [--v2] [--v_parameterization]

Brofessor•7/12/23, 12:45 PM

tensorrt should be installed

weilu•7/12/23, 12:46 PM

@Dr. Furkan Gözükara Do you know how to add LoRA safetensors files trained by kohya script to sdxl pipeline?

revoerom•7/12/23, 12:56 PM

hello, i've tried follwing this guide(How To Install DreamBooth & Automatic1111 On RunPod & Latest Libraries - 2x Speed Up - cudDNN - CUDA https://www.youtube.com/watch?v=c_S2kFAefTQ) to install SD A1111 dreambooth on runpod, but run into this error at the python realuncher.py stage:

Launching Web UI with arguments: -f --port 3000 --xformers --skip-install --listen --enable-insecure-extension-access
Traceback (most recent call last):
File "/workspace/stable-diffusion-webui/launch.py", line 38, in <module>
main()
File "/workspace/stable-diffusion-webui/launch.py", line 34, in main
start()
File "/workspace/stable-diffusion-webui/modules/launch_utils.py", line 340, in start
import webui
File "/workspace/stable-diffusion-webui/webui.py", line 14, in <module>
from fastapi import FastAPI, Response
ModuleNotFoundError: No module named 'fastapi'
Relauncher: Process is ending. Relaunching in 2s...

does anyone know how to fix the fastapi error? thanks in advance!

YouTubeSECourses

How To Install DreamBooth & Automatic1111 On RunPod & Latest Librar...

I have shown how to install latest version of Automatic1111 Web UI for Stable Diffusion and DreamBooth extension of Auto1111 on RunPod in this video. Moreover, I show how to upgrade to latest Cuda, Torch and cuDNN DLL files. With these upgrades the image generation speed literally doubles.

GitHub Readme File

https://github.com/FurkanGozukara...

How To Install DreamBooth & Automatic1111 On RunPod & Latest Librar...

Rrevoerom hello, i've tried follwing this guide(How To Install DreamBooth & Automatic1111 ...

ashley•7/12/23, 1:05 PM

How did you launch WebUI? Looks like you launched it incorrectly and your venv was not activated.

Aashley How did you launch WebUI? Looks like you launched it incorrectly and your venv ...

revoerom•7/12/23, 1:06 PM

well i just followed these steps exactly...

Furkan Gözükara SECourses•7/12/23, 1:06 PM

ok train done with 1e-4 time to download and test

BBrofessor hello

Furkan Gözükara SECourses•7/12/23, 1:07 PM

hello. did you use auto installer?

FFurkan Gözükara SECourses and i am working on it

Khang•7/12/23, 1:07 PM

Awesome I know many people use ComfyUI but I'm not sure what can it do that Auto1111 can't so I haven't spent time to learn it

Furkan Gözükara SECourses•7/12/23, 1:07 PM

also they have some argument errors

Wweilu @Dr. Furkan Gözükara Do you know how to add LoRA safetensors files trained by ko...

Furkan Gözükara SECourses•7/12/23, 1:07 PM

yep i am preparing a tutorial

Khang•7/12/23, 1:08 PM

Also is your grid_add_image_number extension working ok Dr.Furkan? Without that extension it's hard to find images after I generated >100 images

Furkan Gözükara SECourses•7/12/23, 1:08 PM

did you do this part @revoerom

Furkan Gözükara SECourses•7/12/23, 1:08 PM

Furkan Gözükara SECourses•7/12/23, 1:09 PM

so it is a 2 main part install

FFurkan Gözükara SECourses Click to see attachment

revoerom•7/12/23, 1:11 PM

yes i did that part as well! wow thanks for replying!!! the only part that I can see that is different from your tutorial is that when creating the pod, the template i could use was this one (as opposed to the 8.0.3 version in your (excellent:1.2):) tutorial)

FFurkan Gözükara SECourses hello. did you use auto installer?

Brofessor•7/12/23, 1:13 PM

i used manual commands, but they were same as in autoinstaller

Brofessor•7/12/23, 1:13 PM

apt update
yes | apt-get install python3.10-tk

cd /workspace

git clone https://github.com/bmaltais/kohya_ss.git

cd /workspace/kohya_ss

#git checkout 429d2f282fb32e7c63b5e5fe6dbc1ddef92bba58

python3 -m venv venv

source venv/bin/activate

yes | apt-get install python3.10-tk

./setup.sh -n

GitHub

GitHub - bmaltais/kohya_ss

Contribute to bmaltais/kohya_ss development by creating an account on GitHub.

Brofessor•7/12/23, 1:14 PM

i think this tensotrt error has been around before, i checked the chat

Rrevoerom yes i did that part as well! wow thanks for replying!!! the only part that I can...

Furkan Gözükara SECourses•7/12/23, 1:16 PM

Ye they updated

Rrevoerom yes i did that part as well! wow thanks for replying!!! the only part that I can...

Furkan Gözükara SECourses•7/12/23, 1:17 PM

Let me verify

BBrofessor i used manual commands, but they were same as in autoinstaller