Fhello. did you make a new installation ?
▲3 times 
no luck
no luck▲i always chose torch 2 cause i use it for auto1111 (not sure if is shared or whatever)
should i redo install choosing torch 1?
should i redo install choosing torch 1?
Fdont do that
▲ok
Fkohya makes its own venv
Ffollow as i did
Fit wont affect your main install
▲ok so i reinstall and choose torch 1
Fbut on runpod dont i wonder the difference
Fi suggest you to do manual install. explained in the end of this video . it is so easy. also watching this video help you tremendously https://youtu.be/B5U7LJOvH6g
YouTubeSECourses
How to install Python, have multiple Python installations, set system wide default Python version. How to compose venv for any Python installation, change Python default path, and install SD web UI properly. Discord: https://bit.ly/SECoursesDiscord.
If I have been of assistance to you and you would like to show your support for my work, please...
If I have been of assistance to you and you would like to show your support for my work, please...
▲thnaks
Fyes it is amazing and controlnet is getting this feature . they already made inpainting. working on outpainting
Fi dont think has spyware. otherwise the author wouldnt remove it :d
Fnot yet what does it do?
Fvery nice pictures. which model is this?
Fi made those images with dreamshaper 6 using the terran lora i trained. haven't had great success on the first 3 versions of it, but dadaptation 3 helped this one quite a lot
Ftrained the lora using the base 1.5
Fi see thanks
Fwhat is dadaptation 3
Fthey added support in the latest kohya update. optimizer from facebook research
Fcan you show screenshot
Fhttps://github.com/facebookresearch/dadaptation https://github.com/kohya-ss/sd-scripts#25-may-2023-20230525
what do you want the screenshot of?
what do you want the screenshot of?
GitHub
D-Adaptation for SGD, Adam and AdaGrad. Contribute to facebookresearch/dadaptation development by creating an account on GitHub.
GitHub
Contribute to kohya-ss/sd-scripts development by creating an account on GitHub.
Fwith d adaptation you dont chose learning rate right?
Fi had heard this
Fdefinitely should test
F
Fyeah you set the learn rate to 1 and it adjusts it all on the fly
Fif i understand it right, the lr acts more of a weight towards the on the fly estimates
Fthat lion is the best one?
Flion is what i've been using but it's all i've tried. using an automatic algorithm to tame it seems suiting
Factually i am plannin to compare all learning optimizers haha
FThat's good! i've stuck only to adam8bit so far until jumping to lion. i've not enough exepreince to say what's better than others yet. Don't even know how to effectively test these models other than throwing spaghetti at the wall
Fi've only recently started using the kohya setup. i think itsa much better workflow than what the webui-extension provides
Fnow lates roop is supported . updated github file : https://github.com/FurkanGozukara/Stable-Diffusion/blob/main/Tutorials/1-Click-DeepFake-Tutorial.md
GitHub
Best Stable Diffusion and AI Tutorials, Guides, News, Tips and Tricks - Stable-Diffusion/1-Click-DeepFake-Tutorial.md at main · FurkanGozukara/Stable-Diffusion
Fwhat network rank you are using? 128?
Fthat terran lora was rank 8 alpha 1. wanted to see how low i could take it
Faccelerate launch --num_cpu_threads_per_process=2 "train_network.py" --enable_bucket --pretrained_model_name_or_path="runwayml/stable-diffusion-v1-5" --train_data_dir="J:\ImageSets\terran\set3" --resolution=640,640 --output_dir="J:\loras\terrantechv4" --logging_dir="J:\loras\terrantechv4\logs" --network_alpha="1" --save_model_as=safetensors --network_module=networks.lora --text_encoder_lr=1.0 --unet_lr=1.0 --network_dim=8 --output_name="terrantechv4" --lr_scheduler_num_cycles="3" --learning_rate="1.0" --lr_scheduler="cosine_with_restarts" --lr_warmup_steps="280" --train_batch_size="4" --max_train_steps="2800" --save_every_n_epochs="5" --mixed_precision="bf16" --save_precision="bf16" --caption_extension=".txt" --optimizer_type="DAdaptLion" --max_data_loader_n_workers="0" --bucket_reso_steps=64 --mem_eff_attn --color_aug --flip_aug --shuffle_caption --gradient_checkpointing --xformers --bucket_no_upscale --random_crop --multires_noise_iterations="7" --multires_noise_discount="0.1" --sample_sampler=k_dpm_2_a --sample_prompts="J:\loras\terrantechv4\sample\prompt.txt" --sample_every_n_epochs="1"
RThanks! It worked out. But now I don't now how to solve this one:
ERROR: triton-2.0.0-cp310-cp310-win_amd64.whl is not a supported wheel on this platform.
I tried looking for a differnt version but the one I've tried apparently doesn't satisdies the riquirement triton
ERROR: triton-2.0.0-cp310-cp310-win_amd64.whl is not a supported wheel on this platform.
I tried looking for a differnt version but the one I've tried apparently doesn't satisdies the riquirement triton
Fokay so i had simeilar problems. i think it's cause preexisting stuff exists on the system outside of venvs. like when i was installing things for other ui's but forgot to activate a venv
Fterminal > make sure no venv is active >
cd \kohya_ss\ [whereever it is]pip freeze > uninstall.txtpip uninstall -r uninstall.txtthen i deleted the whole kohya folder and started over. worked
Fi dont know if that's what i did to fix it, but it worked when i reinstalled and that's the only step i tried before
Fgot the info here https://github.com/bmaltais/kohya_ss#filenotfounderror
Flooks like latest web ui commit breaking dreambooth
Ftesting your template @ashleyk right now
Fat which step did you get this error
RAt the beggining. I got the 3 error at the same time
RWhen I run .\setup.bat