symbiontconjurer/sdwebui repository overview

⁠Lambda Labs Launch Info

To start, launch an instance using your cloud instance dashboard.

SSH into machine with port forwarding:

ssh -L:7860:localhost:7860 ubuntu@<machine IP>

Start a tmux

tmux

Configure nvidia docker runtime per https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/user-guide.html#daemon-configuration-file⁠

sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

Pull and run image

sudo docker run --rm -it --init --runtime=nvidia --gpus=all --ipc=host --name sdwebui -p 7860:7860 --volume $PWD:/data symbiontconjurer/sdwebui:latest

⁠Command Line Args

Extra command line args are passed through from docker run to the Automatic1111 execution script.

⁠Example: Full Precision

If you want to run in full precision, you could run this instead:

sudo docker run --rm -it --init --runtime=nvidia --gpus=all --ipc=host --name sdwebui -p 7860:7860 --volume $PWD:/data symbiontconjurer/sdwebui:latest --no-half --no-half-vae --precision=full

⁠Example: Gradio

If you want to expose your Web UI using a Gradio tunnel:

sudo docker run --rm -it --init --runtime=nvidia --gpus=all --ipc=host --name sdwebui -p 7860:7860 --volume $PWD:/data symbiontconjurer/sdwebui:latest --share

⁠Notes

This image downloads models the first time you run it (i.e., the models are not baked into the docker image). As such you pretty much always want to run with --volume <dir> (e.g., --volume $PWD:/data as in the commands above) to persist the models and any outputs outside of the running container.