Software Engineering Courses (SECourses)•3y ago

lion was best with that extension correct

Furkan Gözükara SECoursesOP•10/9/23, 1:00 AM

but didnt work well for me with kohya

Furkan Gözükara SECoursesOP•10/9/23, 1:00 AM

i shared all my testings in this post : https://www.patreon.com/posts/89213064

Deleted User•10/9/23, 1:01 AM

i've noticed more flexibility when uses multiple batch sizes like 3-4-5

Deleted User•10/9/23, 1:01 AM

though some quality tradeoff

FFurkan Gözükara SECourses lion was best with that extension correct

Gerdes of Earth•10/9/23, 1:03 AM

You provide in this best settings based on different GB sizes. Did you use runpod to do these tests?

Gerdes of Earth•10/9/23, 1:06 AM

Also is this for Loras or Dreambooth? I noticed the # of epochs is 4, that seems small.

Furkan Gözükara SECoursesOP•10/9/23, 1:06 AM

https://www.youtube.com/watch?v=EEV8RPohsbw

YouTubeSECourses

How To Do Stable Diffusion XL (SDXL) DreamBooth Training (Full Fine...

GGerdes of Earth You provide in this best settings based on different GB sizes. Did you use runpo...

Furkan Gözükara SECoursesOP•10/9/23, 1:06 AM

yes i use them on runpod too

Gerdes of Earth•10/9/23, 1:07 AM

Got it. Thank you for sharing

GGerdes of Earth Also is this for Loras or Dreambooth? I noticed the # of epochs is 4, that seems...

Furkan Gözükara SECoursesOP•10/9/23, 1:07 AM

4 epoch is for 40 repeat

Furkan Gözükara SECoursesOP•10/9/23, 1:07 AM

so 160 epochs

Furkan Gözükara SECoursesOP•10/9/23, 1:07 AM

here dreambooth quick tutorial for patreon supporters : https://www.youtube.com/watch?v=EEV8RPohsbw

YouTubeSECourses

How To Do Stable Diffusion XL (SDXL) DreamBooth Training (Full Fine...

Praveen Miriyala•10/9/23, 2:28 AM

I'm training LoRa SDXL on my 3060 with 12GB VRAM, 768*768 resolution and it's showing 2 hours and 30 minutes of training time. I've used 15 images with 20 repeats. Is this configuration fine?

--num_cpu_threads_per_process=2 "./sdxl_train_network.py"
--network_alpha="1"
--save_model_as=safetensors
--network_module=networks.lora
--text_encoder_lr=0.0004
--unet_lr=0.0004
--network_dim=32
--output_name="local_test_1"
--lr_scheduler_num_cycles="8"
--no_half_vae
--learning_rate="0.0004"
--lr_scheduler="constant"
--train_batch_size="1"
--max_train_steps="4800"
--save_every_n_epochs="1"
--mixed_precision="fp16"
--save_precision="fp16"
--caption_extension=".txt"
--cache_latents
--cache_latents_to_disk
--optimizer_type="Adafactor"
--optimizer_args scale_parameter=False relative_step=False warmup_init=False
--max_data_loader_n_workers="0"
--bucket_reso_steps=64
--gradient_checkpointing
--full_fp16
--xformers
--bucket_no_upscale
--noise_offset=0.0
--lowram

SSuperTurboHero Guys. My RTX3060 12gb is very limited. Can’t really train SDXL models on it and ...

mikemenders•10/9/23, 4:52 AM

It's interesting, because I also have an RTX 3060 12 GB card and I can train a person Lora in 20 minutes on SD 1.5 with 512x512

PPraveen Miriyala I'm training LoRa SDXL on my 3060 with 12GB VRAM, 768*768 resolution and it's sh...

mikemenders•10/9/23, 4:56 AM

I would use 7 repeat and classification images with 10 epochs, no captions, 128/64 weights. For Adafactor use classic parameters for LR and Unet LR: 0.0001, TE: 5e-05, token length: 225, model: RealisticVision 5.1, optimizer args: scale_parameter=False relative_step=False warmup_init=False weight_decay=0.01

Praveen Miriyala•10/9/23, 5:00 AM

How were the results obtained?

mikemenders•10/9/23, 5:02 AM

I have also looked at the parameters from Furkan and other loras. I experimented with weights and trained on RealisticVision with Adafactor.

mikemenders•10/9/23, 5:08 AM

Here is a few sample from yesterday model, and no used hires fix for generation now.

40436-2240229857-bbl_woman_red_dress_sitting_in_a_chair_restaurant_lipstick_some_man_in_background_upper_body_shot_top_view_look_at_vie.png

40440-1223372462-bbl_woman_red_dress_sitting_in_a_chair_restaurant_lipstick_some_man_in_background_upper_body_shot_top_view_look_at_vie.png

40438-367875223-bbl_woman_red_dress_sitting_in_a_chair_restaurant_lipstick_some_man_in_background_upper_body_shot_top_view_look_at_vie.png

40437-4019934250-bbl_woman_red_dress_sitting_in_a_chair_restaurant_lipstick_some_man_in_background_upper_body_shot_top_view_look_at_vie.png

mikemenders•10/9/23, 5:09 AM

Look her hands, it's not fixed, just pure prompt

40444-2754387671-bbl_woman_green_dress_sitting_in_a_chair_restaurant_some_man_in_background_upper_body_shot_top_view_look_at_viewer_and.png

40443-3481190311-bbl_woman_green_dress_sitting_in_a_chair_restaurant_some_man_in_background_upper_body_shot_top_view_look_at_viewer_and.png

40442-4031526407-bbl_woman_green_dress_sitting_in_a_chair_restaurant_some_man_in_background_upper_body_shot_top_view_look_at_viewer_and.png

40441-2800161235-bbl_woman_green_dress_sitting_in_a_chair_restaurant_some_man_in_background_upper_body_shot_top_view_look_at_viewer_and.png

JM•10/9/23, 5:10 AM

Is 6700 steps enough for sdxl?

mikemenders•10/9/23, 5:14 AM

and here is a few portrait

40450-2492411505-bbl_woman_blue_dress_short_hair_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many.png

40449-1065977894-bbl_woman_blue_dress_short_hair_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many.png

40447-1578977074-bbl_woman_green_dress_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_people_sit.png

40446-1230343363-bbl_woman_green_dress_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_people_sit.png

40445-1387945743-bbl_woman_green_dress_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_people_sit.png

mikemenders•10/9/23, 5:19 AM

generates stable images even under aux10 model

40464-2163930282-bbl_woman_dress_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_people_sitting_i.png

40463-1992744207-bbl_woman_dress_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_people_sitting_i.png

40462-1257673000-bbl_woman_dress_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_people_sitting_i.png

40461-1801482298-bbl_woman_dress_short_hair_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_peop.png

40460-1897398850-bbl_woman_dress_short_hair_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_peop.png

40459-1473175472-bbl_woman_dress_short_hair_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_peop.png

40458-3196843830-bbl_woman_dress_short_hair_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_peop.png

mikemenders•10/9/23, 5:26 AM

With artUniverse model:

40489-3077305307-bbl_woman_dress_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_people_sitting_i.png

40488-3117019385-bbl_woman_dress_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_people_sitting_i.png

40487-2920139718-bbl_woman_dress_sitting_in_a_chair_restaurant_some_man_in_background_portrait_look_at_viewer_and_many_people_sitting_i.png

FFurkan Gözükara SECourses i think memory efficient attention doing nothing 😄

Metaphysix•10/9/23, 5:39 AM

Agree I made some test for speed/memory. No gradient checkpointing:
xformers + Mem Efficient Attention = xformers without Mem. E. attention
Same speed, same memory usage.

mikemenders•10/9/23, 6:37 AM

I noticed a very interesting thing. I'm training for style and I haven't changed anything else, just the number of epochs from 10 to 30, and I get completely different sample images and the training values are absolutely different. I'm using Prodigy now.

PPraveen Miriyala I'm training LoRa SDXL on my 3060 with 12GB VRAM, 768*768 resolution and it's sh...

Furkan Gözükara SECoursesOP•10/9/23, 9:27 AM

yep looking decent

Mmikemenders It's interesting, because I also have an RTX 3060 12 GB card and I can train a p...

Furkan Gözükara SECoursesOP•10/9/23, 9:27 AM

512x512 much lower resolution

JJM Is 6700 steps enough for sdxl?

Furkan Gözükara SECoursesOP•10/9/23, 9:28 AM

it is decent steps yes

MMetaphysix Agree I made some test for speed/memory. No gradient checkpointing: xformers + M...

Furkan Gözükara SECoursesOP•10/9/23, 9:28 AM

yep i noticed same

FFurkan Gözükara SECourses yep looking decent

Praveen Miriyala•10/9/23, 9:28 AM

but result was not good

Mmikemenders I noticed a very interesting thing. I'm training for style and I haven't changed...

Furkan Gözükara SECoursesOP•10/9/23, 9:28 AM

i never got good results with Prodigy

PPraveen Miriyala but result was not good ☹️

Furkan Gözükara SECoursesOP•10/9/23, 9:29 AM

try bf16 too

Furkan Gözükara SECoursesOP•10/9/23, 9:29 AM

but reason could be 768x768

Furkan Gözükara SECoursesOP•10/9/23, 9:29 AM

also add this parameter to optimizer : --max_grad_norm=0.0

Furkan Gözükara SECoursesOP•10/9/23, 9:29 AM

i will do another research for LoRA

Praveen Miriyala•10/9/23, 9:30 AM

ok i will follow these tips for next training

JM•10/9/23, 9:38 AM

: A tensor with all NaNs was produced in VAE. This could
be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check. How to fix?

SuperTurboHero•10/9/23, 10:05 AM

I wonder what is quality difference if I don't use classification images? training would be much faster, but would I lose a lot of quality?

mikemenders•10/9/23, 10:40 AM

I used this parameters for Prodigy, and trained SD 1.5 base model: "use_bias_correction=True" "weight_decay=0.5" "decouple=True" "betas=(0.9, 0.99)" "d_coef=2" "safeguard_warmup=False"

JJM : A tensor with all NaNs was produced in VAE. This could be because there's not ...

Furkan Gözükara SECoursesOP•10/9/23, 10:41 AM

this happens when model totally overtrained

FFurkan Gözükara SECourses this happens when model totally overtrained

JM•10/9/23, 10:57 AM

I don't use with model lora I use model checkpoint Every model doesn't work.Basicly sd 1.5 I found this problem.

Sandstormleader•10/9/23, 10:57 AM

so is SDXL training essentially the same as 1.5 training just using an SDXL model? @Dr. Furkan Gözükara

SSandstormleader so is SDXL training essentially the same as 1.5 training just using an SDXL mode...

Furkan Gözükara SECoursesOP•10/9/23, 11:15 AM

the parameters changes too

Furkan Gözükara SECoursesOP•10/9/23, 11:15 AM

learning rate

Furkan Gözükara SECoursesOP•10/9/23, 11:15 AM

Sandstormleader•10/9/23, 11:18 AM

so not 0.000001

SSandstormleader so not 0.000001

Furkan Gözükara SECoursesOP•10/9/23, 11:19 AM

yes

Furkan Gözükara SECoursesOP•10/9/23, 11:19 AM

I am searching best parameters for SDXL DreamBooth for weeks now

Furkan Gözükara SECoursesOP•10/9/23, 11:38 AM

Huge Stable Diffusion XL (SDXL) Text Encoder DreamBooth training comparison

U-NET is always trained

PNG info (prompts) are in alts

All images are 1024x1024 so download full sizes

https://twitter.com/GozukaraFurkan/status/1711345323900100643

Benjamin•10/9/23, 11:40 AM

v23.10.0 adetailer got released. Just noticed because now in the preview of the image a red box appears around the detected face

lion was best with that extension correct

Similar Threads