i would reinstall python
i would reinstall python
Fif it cant find pipo
Fyes
Fhello
Fyes i am checking
RHi I'm trying to train an SD1.5 model with my face on Kohya SS Dreambooth (checkpoint training, not lora) and wanted to see what settings you would recommend. Appreciate it! Here is my current setup:
CPU: AMD 5900x (12core/24processors)
GPU: RTX 3090 (24gb vram)
RAM: 16GB
Training images: 16 images / 40 repeat
Regularization images: 5200 images from your patreon post / 1 repeat
Image size: 768 x 1024
Kohya setting:
accelerate launch
--num_cpu_threads_per_process=24 "./train_db.py"
--pretrained_model_name_or_path="C:/Users/foodmass/Stable-diffusion/t3_Ver09.safetensors"
--train_data_dir="C:/Users/foodmass/Documents/asdf-t3-v2\img"
--reg_data_dir="C:/Users/foodmass/Documents/asdf-t3-v2\reg"
--resolution="768,1024"
--output_dir="C:/Users/foodmass/Documents/asdf-t3-v2\model"
--logging_dir="C:/Users/foodmass/Documents/asdf-t3-v2\log"
--save_model_as=safetensors
--full_bf16
--output_name="asdf"
--lr_scheduler_num_cycles="8"
--max_data_loader_n_workers="1"
--learning_rate_te="1e-06"
--learning_rate="0.0001"
--lr_scheduler="constant"
--train_batch_size="4"
--max_train_steps="2560"
--save_every_n_epochs="1"
--mixed_precision="bf16"
--save_precision="bf16"
--cache_latents
--optimizer_type="Adafactor"
--max_data_loader_n_workers="1"
--clip_skip=2
--bucket_reso_steps=64
--gradient_checkpointing
--xformers
--bucket_no_upscale
--noise_offset=0.0
CPU: AMD 5900x (12core/24processors)
GPU: RTX 3090 (24gb vram)
RAM: 16GB
Training images: 16 images / 40 repeat
Regularization images: 5200 images from your patreon post / 1 repeat
Image size: 768 x 1024
Kohya setting:
accelerate launch
--num_cpu_threads_per_process=24 "./train_db.py"
--pretrained_model_name_or_path="C:/Users/foodmass/Stable-diffusion/t3_Ver09.safetensors"
--train_data_dir="C:/Users/foodmass/Documents/asdf-t3-v2\img"
--reg_data_dir="C:/Users/foodmass/Documents/asdf-t3-v2\reg"
--resolution="768,1024"
--output_dir="C:/Users/foodmass/Documents/asdf-t3-v2\model"
--logging_dir="C:/Users/foodmass/Documents/asdf-t3-v2\log"
--save_model_as=safetensors
--full_bf16
--output_name="asdf"
--lr_scheduler_num_cycles="8"
--max_data_loader_n_workers="1"
--learning_rate_te="1e-06"
--learning_rate="0.0001"
--lr_scheduler="constant"
--train_batch_size="4"
--max_train_steps="2560"
--save_every_n_epochs="1"
--mixed_precision="bf16"
--save_precision="bf16"
--cache_latents
--optimizer_type="Adafactor"
--max_data_loader_n_workers="1"
--clip_skip=2
--bucket_reso_steps=64
--gradient_checkpointing
--xformers
--bucket_no_upscale
--noise_offset=0.0
KIrritating when it suddenly starts doing this ona model it has not done it with previously
NThank you Furkan, I did exactly that and the problem is solved.
Fsadly i dont have best parameters for sd 1.5
Fi only have for sdxl
Bi have the problem where i cant generate images, is my problem not having the correct checkpoint?
Fshow error here
Bthis is the cmd prompt when i click generate
Fsadly dont show error here
For at least nothing obvious
Bthis is what is below where image is supposed to be
Bdo i need sdxl to generate images?
Fno you can generate with sd 1.5 too
Fadd your
Fcommand line arguments
F--no-half
Bdo i put that in the webui-user batch file?
Bi gotta somehow get sd 1.5 now
Fread here please
F@Bophedes
Bwhere did you want me to read?
Fhere
FGitHub
I'm using RX6600 and my parameteres are: --medvram --precision full --no-half --opt-sub-quad-attention --sub-quad-q-chunk-size 512 --sub-quad-kv-chunk-size 512 --sub-quad-chunk-threshold 80 Is ...
Fdoing R&D for a client 
Bim assuming this goes in the params folder of the stable-diffusion-webui. it doesnt say where they put the parameters
Fparameter are put into
Fwebui . bat file
Bahhh, its taking a while but an image is generating
Bwe are getting somewhere!!!
Bi think i need cuda, what that may be
SWhat are your opinions on identical training parameters but different GPUs and how that relates to quality? i.e. It seems to me that when I train on a 4090, the results are superior than if I train on a A5000 - this is with identical training parameters. I would expect some speed differences, but not quality. Am I crazy? Has anybody else noticed similar things?
Chello can i talk to you on DM's pls
SI don't think anything else is using my VRAM, unless it's starting simultaneously.
This is right before:
This is right before:
S
SThis is after:
S
CWhat is your favorite SDXL model?
KJuggernaught
CAlright thx, I checked around civit and gonna test out Realism Engine SDXL and New Reality XL
CDo you know difference of Juggarnaut V7+RunDiffusion and V7 FP16 VAE Fix ?
KNope
KPrecission maybe
KDon't think there is a big difference
OHey guys, I've been trying to train lora SDXL but I seem to always run out of CUDA in runpod, despite not running anything.
My parameters are just like in SEcourse videos, i'm running in 3090 runpod with 24gb vram
My parameters are just like in SEcourse videos, i'm running in 3090 runpod with 24gb vram
O
OI also got Untyped Storage error like this
