Software Engineering Courses (SECourses)•3y ago

in A1111 i have this setup

PelolisuOP•12/11/23, 5:55 PM

what changes in xformers 0.23?

mtnmecca•12/11/23, 5:57 PM

didn't research changes - just that it was a newly released version and @Dr. Furkan Gözükara recently tested it in latest patreon

PelolisuOP•12/11/23, 5:59 PM

Well, I'm going to test Kohya with its default configuration, after I get good results I'll do the experiment and try to update the dependencies to the most recent ones.

Elisoli•12/11/23, 6:08 PM

Dr, you think we can train this new realistic SDXL model (https://civitai.com/models/152525) with your config files as base? the model contains VAE baked inside already

Realism Engine SDXL - v2.0 VAE | Stable Diffusion Checkpoint | Civitai

Introducing Version 2.0 of the Realism Engine We are excited to announce that Version 2.0 of the Realism Engine is now available, featuring signifi...

PelolisuOP•12/11/23, 6:23 PM

@Dr. Furkan Gözükara when you say that not using xformers with 24gb is better than using xformers with 16gb, does it mean that it only changes the quality of image rendering or affects the quality of training the face correctly?

PPelolisu in A1111 i have this setup

Torcello•12/11/23, 6:47 PM

To use Kandinsky, I had to downgrade my Transformers

TTorcello To use Kandinsky, I had to downgrade my Transformers

PelolisuOP•12/11/23, 7:10 PM

I don't know Kandisky but if it's some dependency restriction I might not be able to update it

Torcello•12/11/23, 7:11 PM

Pytorch 2.1.1 is bad, use 2.1.0+cu118, as cu123 also bad

Torcello•12/11/23, 7:12 PM

I think better off using Kandinsky @fusionbrain

PelolisuOP•12/11/23, 8:45 PM

BBenjamin @FiveSAI sorry if I missed it, but did you already share your training parameter...

FiveSAI•12/11/23, 9:31 PM

I shared sdxl lora ones earlier, still testing all the possible settings, i got about 22 trainings left then will share my best one

PPelolisu @Dr. Furkan Gözükara Do you know if there is any known issue with Torch 2.1.0? ...

Furkan Gözükara SECourses•12/11/23, 9:41 PM

i think depends on each project

Furkan Gözükara SECourses•12/11/23, 9:41 PM

but on automatic1111 working so far

Mmtnmecca also interested in this topic since I tried an update from torch 2.0.1 to 2.1.1 ...

Furkan Gözükara SECourses•12/11/23, 9:42 PM

i replied you from patreon too

Furkan Gözükara SECourses•12/11/23, 9:42 PM

don't forget to activate venv of Automatic1111 Web UI

pip uninstall torch torchvision torchaudio xformers --yes

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

pip install xformers==0.0.23

FFurkan Gözükara SECourses i replied you from patreon too

mtnmecca•12/11/23, 9:43 PM

thanks got it working with torch 2.1.1 and xformers 0.23

EElisoli Dr, you think we can train this new realistic SDXL model (https://civitai.com/mo...

Furkan Gözükara SECourses•12/11/23, 9:43 PM

y4es you can

Furkan Gözükara SECourses•12/11/23, 9:43 PM

i should compare

Mmtnmecca thanks got it working with torch 2.1.1 and xformers 0.23

Furkan Gözükara SECourses•12/11/23, 9:43 PM

great

PelolisuOP•12/11/23, 9:48 PM

when i started training

PelolisuOP•12/11/23, 9:48 PM

PelolisuOP•12/11/23, 9:48 PM

i received this warning

PelolisuOP•12/11/23, 9:49 PM

tensorRT are used to train in kohya?

PelolisuOP•12/11/23, 9:51 PM

however, he continued with the training and is ongoing.

PPelolisu tensorRT are used to train in kohya?

Furkan Gözükara SECourses•12/11/23, 9:54 PM

nope

FFurkan Gözükara SECourses don't forget to activate venv of Automatic1111 Web UI pip uninstall torch torch...

FiveSAI•12/11/23, 9:59 PM

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

need to add the https

FFiveSAI pip3 install torch torchvision torchaudio --index-url https://download.pytorch.o...

Furkan Gözükara SECourses•12/11/23, 10:01 PM

you are right. while copy pasting it was gone

KGTachi•12/11/23, 10:03 PM

Hi I’m still getting the older version of torch and xformers when doing it

KKGTachi Hi I’m still getting the older version of torch and xformers when doing it

PelolisuOP•12/11/23, 10:07 PM

you can try specify version like this pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --extra-index-url https://download.pytorch.org/whl/cu118

PelolisuOP•12/11/23, 10:08 PM

If you have another version in the requirements it will replace the installed version.

KKGTachi Hi I’m still getting the older version of torch and xformers when doing it

Furkan Gözükara SECourses•12/11/23, 10:11 PM

you need to activate venv

KGTachi•12/11/23, 10:11 PM

I did still same issue

PPelolisu you can try specify version like this pip install torch==2.1.0 torchvision==0.16...

KGTachi•12/11/23, 10:11 PM

Tried this too still same

PelolisuOP•12/11/23, 10:12 PM

is on A1111 ?

KGTachi•12/11/23, 10:13 PM

Yep it shows it’s installing etc, but when I load a1111 still on old version

PelolisuOP•12/11/23, 10:14 PM

Are you running on Windows or Linux?

KGTachi•12/11/23, 10:14 PM

Windows

PelolisuOP•12/11/23, 10:15 PM

on webui-user.bat do you have this line TORCH_COMMAND="pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --extra-index-url https://download.pytorch.org/whl/cu118" ?

PelolisuOP•12/11/23, 10:15 PM

or like this

KGTachi•12/11/23, 10:17 PM

Yep tried and still the same

PelolisuOP•12/11/23, 10:19 PM

in modules folder

PelolisuOP•12/11/23, 10:19 PM

file launch_utils.py

PelolisuOP•12/11/23, 10:19 PM

PelolisuOP•12/11/23, 10:20 PM

whenever you run a1111 it will revert to this version. You can change it directly to the version you want, be careful, make a backup of the file first.

KGTachi•12/11/23, 10:23 PM

Even with that still the same

Kallamamran•12/11/23, 10:24 PM

Shit... I've started to get NaN's again

PelolisuOP•12/11/23, 10:25 PM

if you type pip show torch which version do you get?

KGTachi•12/11/23, 10:28 PM

Hmm it shows the old version

PelolisuOP•12/11/23, 10:29 PM

your requirements_versions.txt file is still the original, you didn't make any changes to it, right?

KGTachi•12/11/23, 10:30 PM

I didn’t make any changes to that

in A1111 i have this setup

Similar Threads