For RunPod - Automatic Kohya SS LoRA Installer (Patreon)

Patreon exclusive posts index

Join discord and tell me your discord username to get a special rank : SECourses Discord

RunPod Automatic1111 updater : https://www.patreon.com/posts/1-click-runpod-86438018

Updated 12 November 2023:

• downloadSDXL.sh will now download SDXL FP16 VAE so you don't need --no-half-vae anymore
• It will also download SDXL FP32 VAE which you can use during training
• chmod +x downloadSDXL.sh
• ./downloadSDXL.sh

Updated 18 October 2023:

• Updated for missing scipy library

Updated 10 September 2023:

• Execute below code 1 time after each restart of the Pod to get super insane speed with A5000 GPU. Same price of RTX 3090. 1.64 IT/s for 1024x1024 full SDLX DreamBooth training
• apt update
• apt install libcudnn8=8.9.7.29-1+cuda11.8 libcudnn8-dev=8.9.7.29-1+cuda11.8

Updated 11 August 2023

• To use certain GPU if you have multiple GPUs : --gpu_ids=1
• Changes : It will auto download Realistic_Vision_V5.1.safetensors and vae-ft-mse-840000-ema-pruned.ckpt (best VAE for SD 1.5 models) and sdxl_vae.safetensors (best VAE for SDXL) into the relative folders in both templates.
• Best SDXL LoRA Training Tutorial published 8 August 2023 > Become A Master Of SDXL Training With Kohya SS LoRAs - Combine Power Of Automatic1111 & SDXL LoRAs

Instructions

• downloadSDXL. sh file will automatically download latest models mentioned above
• RunPod io link > https://bit.ly/RunPodIO
• * Select stable-diffusion:web-ui or stable-diffusion:fast-stable-diffusion templates
• * Make container disk size at least 15 GB
• Download attached kohya_installer.sh , downloadSDXL.sh and then upload it into your Workspace folder
• Screenshot of upload
• Click here to open install commands on GitHub
• Or here quick command
• chmod +x kohya_installer.sh
• ./kohya_installer.sh
• If you pick stable-diffusion:web-ui template download attached relauncher.py, enter inside stable-diffusion-webui folder, upload relauncher. py. It will ask you to Overwrite so overwrite. Then restart your pod 1 time. This is only necessary 1 time to prevent permanent relaunching of Automatic1111 Web UI.

You can also use this runpodctl auto installer from this Patreon post > https://www.patreon.com/posts/1-click-bat-file-87505171

Check section Auto Installer For RunPod

Screenshot of install commands

Use a custom model otherwise not working.

• .e.g : /workspace/stable-diffusion-webui/models/Stable-diffusion/v1-5-pruned.ckpt
• Screenshot of custom model select

Use below command to kill Web UI instance to save more VRAM before doing Kohya Training

• fuser -k 3000/tcp

Use below command to start Kohya on RunPod after installation

• apt update 
• yes | apt-get install python3.10-tk
• fuser -k 7860/tcp
• cd /workspace/kohya_ss
• source venv/bin/activate
• export CUDA_VISIBLE_DEVICES=0
• bash gui.sh --share --headless

By changing CUDA_VISIBLE_DEVICES  to different CUDA you can start different training on each GPU

SDXL training presets : https://www.patreon.com/posts/very-best-for-of-89213064