# modelscope可以存文件的地方：
"""
目录: / (可用空间: 60 GB)
写入速度: 1556.90 MB/s
读取速度: 5127.25 MB/s

目录: /dev/shm (可用空间: 30 GB)
写入速度: 3954.25 MB/s
读取速度: 4598.10 MB/s

目录: /etc/sgpu/pmem (可用空间: 48 GB)
写入速度: 7371.38 MB/s
读取速度: 6247.02 MB/s

目录: /proc/driver/nvidia (可用空间: 32 GB) (容量超出会导致严重系统错误)
写入速度: 3942.13 MB/s
读取速度: 4594.20 MB/s
"""
# 下列代码将sd webui安装到/etc/sgpu/pmem/，可自行批量替换
apt update && apt install -y aria2
rm -rf /etc/sgpu/pmem/* #避免重复安装，所以全部移除，注意重要文件是否在安装目录
aria2c -x 16 -s 16 "https://www.modelscope.cn/models/ACCC1380/Fulx_dev_Model/resolve/master/code/ssh7865.py" -o ssh7865.py -d /root
aria2c -x 16 -s 16 "https://www.modelscope.cn/models/ACCC1380/Fulx_dev_Model/resolve/master/code/aria2.sh" -o aria2.sh -d /root
echo 安装SD
cd /etc/sgpu/pmem/ && aria2c -x 16 -s 16 -c -k 1M "https://www.modelscope.cn/models/ACCC1380/SD-WebUI-pack/resolve/master/sd-webui.zip" -o sd-webui.zip
cd /etc/sgpu/pmem/ && unzip ./sd-webui.zip && rm ./sd-webui.zip
echo 安装venv和模型
cd /etc/sgpu/pmem/ && aria2c -x 16 -s 16 -c -k 1M "https://www.modelscope.cn/models/ACCC1380/SD-WebUI-pack/resolve/master/sdvenv.tar.safetensors" -o venv.tar
cd /etc/sgpu/pmem/ && tar -xvf venv.tar & aria2c -x 16 -s 16 -c -k 1M "https://www.modelscope.cn/models/ACCC1380/Noob_Final/resolve/master/Noob_v1.0.safetensors" -o NoobXL-v1.0.safetensors -d /etc/sgpu/pmem/sdmodels
cd /etc/sgpu/pmem/ && rm venv.tar
echo 创建模型目录
mkdir -p /etc/sgpu/pmem/sdmodels
mkdir -p /etc/sgpu/pmem/lora
mkdir -p /etc/sgpu/pmem/vae
echo 安装必要文件
cd /etc/sgpu/pmem/tmp/stable-diffusion-webui/models/VAE-approx && wget -O vaeapprox-sdxl.pt "https://www.modelscope.cn/models/ACCC1380/Noobxl/resolve/master/vaeapprox-sdxl.pt" --no-check-certificate
cd /etc/sgpu/pmem/tmp/ && wget -O rep_github.py "https://www.modelscope.cn/models/ACCC1380/Fulx_dev_Model/resolve/master/code/rep_github.py" --no-check-certificate
cd /etc/sgpu/pmem/tmp/ && python rep_github.py
echo 开始启动sd
cd /etc/sgpu/pmem/tmp/stable-diffusion-webui && HF_ENDPOINT=https://hf-mirror.com /etc/sgpu/pmem/venv/bin/python launch.py --api --port=7865 --xformers --ckpt-dir=/etc/sgpu/pmem/sdmodels --lora-dir=/etc/sgpu/pmem/lora --vae-dir=/etc/sgpu/pmem/vae & python /root/ssh7865.py & bash /root/aria2.sh