NVIDIA-搭建Kohya's GUI

Koyha_ss 项目使用了gradio对lora的训练任务进行了封装，提供了可视化交互界面，通过交互界面，可以方面修改训练参数的配置，对小白更加友好。

Github仓库：https://github.com/bmaltais/kohya_ss

本篇的主要目的是带你掌握如何在 Nvidia Jetson AGX Orin 中搭建运行环境

先决条件

• Nvidia Jetson AGX Orin 基本环境安装
• Python 3.10
• CUDA 12.6

安装步骤

# export https_proxy=http://proxy.to.host
sudo apt-get install python3-tk tk-dev libopenblas-dev libatlas-base-dev -y

git clone --recursive https://github.com/bmaltais/kohya_ss.git
cd kohya_ss/

cat > requirements_linux.txt <<EOF
# Custom index URL for specific packages
--index-url https://pypi.jetson-ai-lab.dev/jp6/cu126/+simple/

torch==2.7.0
torchvision==0.22.0
xformers==0.0.30+c5c0720.d20250414

bitsandbytes==0.44.2.dev0
tensorboard<2.19,>=2.18
tensorflow==2.18.0
onnxruntime-gpu==1.20.2

-r requirements.txt
EOF

cat > requirements.txt <<EOF
accelerate==0.33.0
aiofiles==23.2.1
altair==4.2.2
dadaptation==3.2
diffusers[torch]==0.32.2
easygui==0.98.3
einops==0.7.0
fairscale==0.4.13
ftfy==6.1.1
gradio>=5.23.1
huggingface-hub==0.29.3
imagesize==1.4.1
invisible-watermark==0.2.0
lion-pytorch==0.0.6
lycoris_lora==3.1.0
omegaconf==2.3.0
onnx==1.16.1
prodigyopt==1.1.2
protobuf==3.20.3
open-clip-torch==2.20.0
opencv-python==4.10.0
prodigy-plus-schedule-free==1.8.0
pytorch-lightning==1.9.0
pytorch-optimizer==3.5.0
rich>=13.7.1
safetensors==0.4.4
schedulefree==1.4
scipy==1.11.4
# for T5XXL tokenizer (SD3/FLUX)
sentencepiece==0.2.0
timm==0.6.12
tk==0.1.0
toml==0.10.2
transformers==4.44.2
voluptuous==0.13.1
wandb==0.18.0
# for kohya_ss sd-scripts library
-e ./sd-scripts
EOF

sudo chmod +x steup.sh
bash setup.sh