Hey, just wanted to thank you for your posts. This is complex stuff, a lot of it only cursorily explained in docs, and these videos are incredibly helpful. Big thumbs up👏
I know this might be outdated not but for some reason When I train using BF16 I can only use a batch size of 1 or I get out of memory. even at 512,512. for a test. How are you able to do 8 I have a 3090 by the way. followed your settings im using BF16 and 23 images.
Thank you for the tutorial Bernard. I was wondering if it was possible to generate a smaller file (I am thinking LoRA size with dreambooth results). Do you know if that can be accomplished?
Warning: Python version 3.10.9 is recommended. venv folder already exists, skipping creation... El sistema no puede encontrar la ruta especificada. Do you want to uninstall previous versions of torch and associated files before installing? Usefull if you are upgrading from torch 1.12.1 to torch 2.0.0 or if you are downgrading from torch 2.0.0 to torch 1.12.1.
Make sure that both "py --version" and "python --version" both work and return the same value in CMD. If they don't, try swapping around the root python folder and the python/Scripts folder in your environment variables. If they aren't there at all... Add them
i cant use it , when i want to train always shouts GPU needed and i have 24GB Vram cuda dll installed .... Traceback (most recent call last): File "F:\kohya_ss\train_db.py", line 1229, in train(args) File "F:\kohya_ss\train_db.py", line 832, in train accelerator = Accelerator(gradient_accumulation_steps=1, mixed_precision=args.mixed_precision, File "C:\Users\xxxx\anaconda3\envs\kohya\lib\site-packages\accelerate\accelerator.py", line 355, in __init__ raise ValueError(err.format(mode="fp16", requirement="a GPU")) ValueError: fp16 mixed precision requires a GPU Traceback (most recent call last): File "C:\Users\xxx\anaconda3\envs\kohya\lib unpy.py", line 196, in _run_module_as_main return _run_code(code, main_globals, None, File "C:\Users\xxx\anaconda3\envs\kohya\lib unpy.py", line 86, in _run_code exec(code, run_globals) File "C:\Users\xxx\anaconda3\envs\kohya\Scripts\accelerate.exe\__main__.py", line 7, in File "C:\Users\xxx\anaconda3\envs\kohya\lib\site-packages\accelerate\commands\accelerate_cli.py", line 45, in main args.func(args) File "C:\Users\xxxx\anaconda3\envs\kohya\lib\site-packages\accelerate\commands\launch.py", line 1104, in launch_command simple_launcher(args) File "C:\Users\xxxxxx\anaconda3\envs\kohya\lib\site-packages\accelerate\commands\launch.py", line 567, in simple_launcher
@@lehikot7600 not yet but will have another go tomorrow and in case I’ll make it work will let you know. Just comment tomorrow to remind in case I forget …
@@adriangpuiu I got it working too once I reinstalled python. Now I need to figure out how to install deepspeed on windows so I can offload part of the work to my CPU from my gpu lol....
@@lehikot7600 oh man, freaking accelerate was driving me nuts at the beginning now its fine, but frankly i dont know if youll see much of improvement from that deep speed module.. ill dig into it now that you told me :)))
It depend on your subject... but typically the more images the fewer repeats you need. It really is a trial and error for each model. I wish there was a single easy config... but alas there is not.
Hey, just wanted to thank you for your posts. This is complex stuff, a lot of it only cursorily explained in docs, and these videos are incredibly helpful. Big thumbs up👏
This is incredible thank you so much for making this available I used your old repo with kohya before re the GUI it looks much easier now.
How to train SDXL model with large number of images? like example 30k images
I know this might be outdated not but for some reason When I train using BF16 I can only use a batch size of 1 or I get out of memory. even at 512,512. for a test. How are you able to do 8 I have a 3090 by the way. followed your settings im using BF16 and 23 images.
Thank you for the tutorial Bernard. I was wondering if it was possible to generate a smaller file (I am thinking LoRA size with dreambooth results). Do you know if that can be accomplished?
I have a 4090, if i don't care about speed, should i do float precision for better accuracy? whats a good basic config for highest accuracy?
hello does kohya_ss upload my training images to a server or something? i have alot output traffic if i try to train with it
"None of the inputs have requires_grad=True. Gradients will be None" i constantly get this Error. Do you know what this could be ?
How much RAM does your graphics card have? :)
I have a folder with 200+ images in it, and Kohya_SS says that there are not images in that folder.
You need to put in a folder containing the folder with the images, aka the parent folder of the image folder.
anything specific for reg images? i made a bunch images with the class promp i want to use , but when i try training says reg images: 0
Warning: Python version 3.10.9 is recommended.
venv folder already exists, skipping creation...
El sistema no puede encontrar la ruta especificada.
Do you want to uninstall previous versions of torch and associated files before installing? Usefull if you are upgrading from torch 1.12.1 to torch 2.0.0 or if you are downgrading from torch 2.0.0 to torch 1.12.1.
Can you give us New Tutorial they is change everything now :(
My problem is installing Kohya_ss. I get the error that Python can't be found, even though I have Python installed for Automatic1111
Make sure that both "py --version" and "python --version" both work and return the same value in CMD. If they don't, try swapping around the root python folder and the python/Scripts folder in your environment variables. If they aren't there at all... Add them
@@pm146 Thank you for the advice - which I will look into - although I am not very hand with commands etc:-)
i cant use it , when i want to train always shouts GPU needed and i have 24GB Vram cuda dll installed .... Traceback (most recent call last):
File "F:\kohya_ss\train_db.py", line 1229, in
train(args)
File "F:\kohya_ss\train_db.py", line 832, in train
accelerator = Accelerator(gradient_accumulation_steps=1, mixed_precision=args.mixed_precision,
File "C:\Users\xxxx\anaconda3\envs\kohya\lib\site-packages\accelerate\accelerator.py", line 355, in __init__
raise ValueError(err.format(mode="fp16", requirement="a GPU"))
ValueError: fp16 mixed precision requires a GPU
Traceback (most recent call last):
File "C:\Users\xxx\anaconda3\envs\kohya\lib
unpy.py", line 196, in _run_module_as_main
return _run_code(code, main_globals, None,
File "C:\Users\xxx\anaconda3\envs\kohya\lib
unpy.py", line 86, in _run_code
exec(code, run_globals)
File "C:\Users\xxx\anaconda3\envs\kohya\Scripts\accelerate.exe\__main__.py", line 7, in
File "C:\Users\xxx\anaconda3\envs\kohya\lib\site-packages\accelerate\commands\accelerate_cli.py", line 45, in main
args.func(args)
File "C:\Users\xxxx\anaconda3\envs\kohya\lib\site-packages\accelerate\commands\launch.py", line 1104, in launch_command
simple_launcher(args)
File "C:\Users\xxxxxx\anaconda3\envs\kohya\lib\site-packages\accelerate\commands\launch.py", line 567, in simple_launcher
Did you find a solution for this? I was having the same issue where fp16 wasn't seeing my GPU with a different training repo.
@@lehikot7600 not yet but will have another go tomorrow and in case I’ll make it work will let you know. Just comment tomorrow to remind in case I forget …
@@lehikot7600 i installed automatic1111 in the end and made it work with xformers and all shebangs ....
@@adriangpuiu I got it working too once I reinstalled python. Now I need to figure out how to install deepspeed on windows so I can offload part of the work to my CPU from my gpu lol....
@@lehikot7600 oh man, freaking accelerate was driving me nuts at the beginning now its fine, but frankly i dont know if youll see much of improvement from that deep speed module.. ill dig into it now that you told me :)))
what would be the "Repeats" for 16, 30 and 100 Images? any logic we can use to choose "Repeats" based on images?
It depend on your subject... but typically the more images the fewer repeats you need. It really is a trial and error for each model. I wish there was a single easy config... but alas there is not.
Kohya's interface changed so much since this tutorial that I can't follow it.
so i did 3 epochs and i got 3 output files, why ?
By default the trainer save every epoch. You can configure how many epoch to save instead of 1. Look in the training tab.
grateful for video but damn that audio makes me wanna die
I came pretty far
many error!!!!!
"quickstart"
horrible tutorial
Your audio is fucking awful with an echo in it. We can't understand you. I just had to mute it and use CC instead.
ANNNNNDDDD he's eating
"None of the inputs have requires_grad=True. Gradients will be None" i constantly get this Error. Do you know what this could be ?
ignore that. it always says that