Software Engineering Courses (SECourses)•3y ago

i need the sdxl.tensor file i assume. but the tesnor download gives me error

Pelolisu•12/13/23, 2:40 PM

@Dr. Furkan Gözükara kohya dont train v1.5 models ? i tried realistic_vision_5.1 and got errors in load unet model.

BBophedes i need the sdxl.tensor file i assume. but the tesnor download gives me error

Furkan Gözükara SECourses•12/13/23, 2:46 PM

TensorRT wont work on AMD. it is NVIDIA feature

PPelolisu @Dr. Furkan Gözükara kohya dont train v1.5 models ? i tried realistic_vision_5....

Furkan Gözükara SECourses•12/13/23, 2:46 PM

it supports

Furkan Gözükara SECourses•12/13/23, 2:46 PM

you need to uncheck sdxl option

Furkan Gözükara SECourses•12/13/23, 2:46 PM

in first page

Pelolisu•12/13/23, 2:47 PM

i uncheck this and checked v_prediction then booth unchecked?

Pelolisu•12/13/23, 2:52 PM

I'll try, another thing I noticed is that in sdxl with 200 repeat and 1 epoch I got 1280 steps. Without sdxl I got 6500 steps. My dataset is 16.

PPelolisu I'll try, another thing I noticed is that in sdxl with 200 repeat and 1 epoch I ...

Furkan Gözükara SECourses•12/13/23, 2:52 PM

it makes 3200

Furkan Gözükara SECourses•12/13/23, 2:52 PM

without reg images

Pelolisu•12/13/23, 2:54 PM

Even if I provide regularization images?

PPelolisu Even if I provide regularization images?

Furkan Gözükara SECourses•12/13/23, 2:55 PM

if you provide

Furkan Gözükara SECourses•12/13/23, 2:55 PM

it becomes 6400

Pelolisu•12/13/23, 2:57 PM

yes that's what happened, sdxl then generates fewer steps. So it must be right, I'll try to run it with SDXL unchecked.

Pelolisu•12/13/23, 2:58 PM

i can specify vae-ft-mse-840000 for training ?

Pelolisu•12/13/23, 3:07 PM

there must be a bug or unsupported parameter.

Charkel•12/13/23, 3:30 PM

Does Kaggle allow Automatic or Comfy? I tried Furkans notebook for Fooocus which was horrible UI to use

CCharkel Does Kaggle allow Automatic or Comfy? I tried Furkans notebook for Fooocus which...

୧

୧ʕ•̀ᴥ•́ʔ୨•12/13/23, 3:49 PM

Yes it allows automatic

୧

୧ʕ•̀ᴥ•́ʔ୨•12/13/23, 3:49 PM

There's a notebook for it too

୧୧ʕ•̀ᴥ•́ʔ୨ Yes it allows automatic

Charkel•12/13/23, 3:53 PM

Alright I'll search for that notebook thx then I know I won't be wasting my time

Zono50•12/13/23, 4:18 PM

anyone heard of a program called unstable diffusion which is uncensored?

CCharkel Alright I'll search for that notebook thx then I know I won't be wasting my time...

୧

୧ʕ•̀ᴥ•́ʔ୨•12/13/23, 4:19 PM

It's here https://www.patreon.com/posts/run-sd-web-ui-on-88714330

ZZono50 anyone heard of a program called unstable diffusion which is uncensored?

Pelolisu•12/13/23, 4:34 PM

i think this should be the right name for stable diffusion

Pelolisu•12/13/23, 4:34 PM

haha

Zono50•12/13/23, 4:35 PM

lol

Pflanzen die Tanzen•12/13/23, 5:07 PM

FFurkan Gözükara SECourses if you cant afford runpod we have kaggle SDXL DreamBooth

JeremytheJedi•12/13/23, 6:10 PM

Ahh okay. Yeah the 6GB was less than expected. I restarted and changed the batch sizes to 1, and then it said I had 10GB. I’ll fiddle around a little more. Thanks so much for making the video. It’s been really helpful.

FFurkan Gözükara SECourses https://youtu.be/sBFGitIvD2A

nissanle•12/13/23, 6:17 PM

Hello Furkan,
I did all the installation again from scratch however, when I double click “webui-user.bat” in order to launch SD I get the enclosed pip error.
What did I do wrong and how do I rectify it?
Thank you for your help.

Digital [Starburst]•12/13/23, 7:09 PM

There's a live summit right now about Google Cloud and how to use it with something called Vertex. https://cloudonair.withgoogle.com/events/summit-applied-ml-summit-23/watch?talk=kn_demo

Google Cloud Applied AI Summit, Americas

Discover how you and your fellow developers, data scientists, and ML engineers can build a new generation of applications and boost productivity with the latest generative AI research and technology from Google Cloud.

Baran•12/13/23, 8:27 PM

does fine-tuning on a pruned fp16 model work?

Kallamamran•12/13/23, 9:28 PM

Fu**ing hate this:

Kallamamran•12/13/23, 9:28 PM

Only some models suddenly

KKallamamran Fu**ing hate this:

Furkan Gözükara SECourses•12/13/23, 10:01 PM

this happens when model cooked as well

Tadeas•12/13/23, 10:15 PM

Hello everyone.

Nnissanle Hello Furkan, I did all the installation again from scratch however, when I dou...

nissanle•12/13/23, 10:15 PM

Dear Furkan, I solved the problem. TY

Tadeas•12/13/23, 10:15 PM

@Dr. Furkan Gözükara are you available for paid consultation?

Tadeas•12/13/23, 10:16 PM

Please check your DMs.

PPelolisu i can specify vae-ft-mse-840000 for training ?

Furkan Gözükara SECourses•12/13/23, 10:34 PM

you can

CCharkel Does Kaggle allow Automatic or Comfy? I tried Furkans notebook for Fooocus which...

Furkan Gözükara SECourses•12/13/23, 10:34 PM

it allows automatic1111

Furkan Gözükara SECourses•12/13/23, 10:35 PM

we have notebook for that

Furkan Gözükara SECourses•12/13/23, 10:35 PM

https://www.patreon.com/posts/run-on-free-like-88714330

Nnissanle Hello Furkan, I did all the installation again from scratch however, when I dou...

Furkan Gözükara SECourses•12/13/23, 10:36 PM

i would reinstall python

Furkan Gözükara SECourses•12/13/23, 10:36 PM

if it cant find pipo

BBaran does fine-tuning on a pruned fp16 model work?

Furkan Gözükara SECourses•12/13/23, 10:36 PM

yes

TTadeas Hello everyone.

Furkan Gözükara SECourses•12/13/23, 10:37 PM

hello

TTadeas @Dr. Furkan Gözükara are you available for paid consultation?

Furkan Gözükara SECourses•12/13/23, 10:37 PM

yes i am checking

revoerom•12/13/23, 11:19 PM

Hi 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

FFurkan Gözükara SECourses this happens when model cooked as well

Kallamamran•12/13/23, 11:24 PM

Irritating when it suddenly starts doing this ona model it has not done it with previously

FFurkan Gözükara SECourses i would reinstall python

nissanle•12/13/23, 11:32 PM

Thank you Furkan, I did exactly that and the problem is solved.

Rrevoerom Hi I'm trying to train an SD1.5 model with my face on Kohya SS Dreambooth (check...

Furkan Gözükara SECourses•12/14/23, 12:23 AM

sadly i dont have best parameters for sd 1.5

Furkan Gözükara SECourses•12/14/23, 12:23 AM

i only have for sdxl

i need the sdxl.tensor file i assume. but the tesnor download gives me error

Similar Threads

Similar Threads

Similar Threads