Sign inSign up

moophlo/stable-diffusion

By moophlo

•Updated over 1 year ago

Hassle free AUTOMATIC1111 stable-diffusion-webui docker imagefor AMD RX5700XT Navi10

Image
Machine learning & AI
0

5.2K

moophlo/stable-diffusion repository overview

⁠Scope

I struggled a lot to have AUTOMATIC1111 WebUI implementation for Stable Diffusion locally on my AMD RX5700XT. I decide to build this image to be able to run it hassle free just with a single command. Everything is already cofigured, you've just to place your models in the mapped folder.

⁠Reference

This image is based on:

rocm/pytorch:rocm5.2_ubuntu20.04_py3.7_pytorch_1.11.0_navi21

and on the github repository:

https://github.com/AUTOMATIC1111/stable-diffusion-webui⁠

⁠Usage

To use your own authentication username and password adjust accordingly the environment variables WEBUI_USER and WEBUI_PASSWORD in the docker-compose.yml file and issue this commands:

sudo docker compose up -d

P.S.: if you have more than one GPU you can user the environment variable HIP_VISIBLE_DEVICES

Otherwise just run:

sudo docker compose up -d

⁠NOTE:

  1. default user: admin

    default password: ThisIsMySuperSecurePassword.123

  2. running the compose command without sudo does not allow docker daemon to correctly manage the GPU

Dockerfile:

FROM rocm/pytorch:rocm5.2_ubuntu20.04_py3.7_pytorch_1.11.0_navi21

WORKDIR /dockerx/stable-diffusion-webui

RUN apt update && apt full-upgrade -y && apt install -y bc google-perftools && apt autoclean -y
RUN conda update conda
RUN conda create -y --name sd python=3.10.6 
RUN conda install -y -n sd -c conda-forge libstdcxx-ng=12.1.0 --no-deps 
RUN conda install -y -n sd -c conda-forge insightface=0.7.3 --no-deps
RUN conda init bash && . ~/.bashrc && echo "conda activate sd" >> ~/.bashrc
RUN rm -rf /dockerx/stable-diffusion-webui
RUN cd /dockerx && git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui && cd stable-diffusion-webui && git checkout tags/v1.10.0
RUN conda run --no-capture-output -n sd pip cache purge
ADD run.sh .
RUN chmod +x run.sh

ENTRYPOINT ["./run.sh"]

Docker compose file docker-compose.yml:

name: stable
services:
    diffusion:
        build: .
        stdin_open: true
        restart: always
        tty: true
        privileged: true
	environment:
	  WEBUI_USER: "admin"
	  WEBUI_PASSWORD: "ThisIsMySuperSecurePassword.123"
	  HIP_VISIBLE_DEVICES: "0"
        devices:
            - /dev/kfd
            - /dev/dri
        group_add:
            - video
        ipc: host
        cap_add:
            - SYS_PTRACE
        security_opt:
            - seccomp=unconfined
        volumes:
            - ./models:/dockerx/stable-diffusion-webui/models
            - ./extensions:/dockerx/stable-diffusion-webui/extensions
        ports:
            - 7860:7860
        healthcheck:
          test: curl --fail http://localhost:7860/sdapi/v1/options || exit 1
          interval: 30s
          timeout: 30s
          retries: 3
          start_period: 240s

run.sh file:

#!/bin/bash -x



if [ -n "$COMMANDLINE_ARGS" ]; then
        echo "export COMMANDLINE_ARGS=\"$COMMANDLINE_ARGS\""  >> /dockerx/stable-diffusion-webui/webui-user.sh
else
        echo 'export COMMANDLINE_ARGS="--listen --device-id=0 --precision full --no-half --medvram --skip-version-check --api --gradio-auth-path auth --enable-insecure-extension-access --opt-sub-quad-attention"' >> /dockerx/stable-diffusion-webui/webui-user.sh
fi

if [ -n "$WEBUI_USER" ] && [ -n "$WEBUI_PASSWORD" ]; then
        echo "$WEBUI_USER:$WEBUI_PASSWORD" > /dockerx/stable-diffusion-webui/auth
else
        echo 'admin:ThisIsMySuperSecurePassword.123' > /dockerx/stable-diffusion-webui/auth
fi

echo 'export TORCH_COMMAND="pip install https://repo.radeon.com/rocm/manylinux/.private-9bd8754a636553e1d1ce3e107bf18c00/20231018/torch-2.1.0%2Brocm5.2-cp310-cp310-linux_x86_64.whl https://github.com/kmsedu/pytorch_rocm52_wheels/raw/main/torchvision-0.17.0a0+rocm5.2-cp310-cp310-linux_x86_64.whl"' >> /dockerx/stable-diffusion-webui/webui-user.sh

mv /dockerx/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/libMIOpen.so /dockerx/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/libMIOpen.so_ORIG
cp /opt/rocm/lib/libMIOpen.so.1.0.50200.nvv5.ms1 /dockerx/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/lib/libMIOpen.so

sed -i 's|^can_run_as_root.*|can_run_as_root=1|g' /dockerx/stable-diffusion-webui/webui.sh

conda run --no-capture-output -n sd ./webui.sh

This docker compose file will build the image for you if placed in the same path of the Dockerfile, otherwise you can comment the build line and uncomment the image line like this:

        #build: .
        image: moophlo/stable-diffusion:latest

Tag summary

Content type

Image

Digest

sha256:5beb6d65a…

Size

22 GB

Last updated

over 1 year ago

docker pull moophlo/stable-diffusion