• Docker installed
• (optionally) Docker GPU support set up, see https://docs.docker.com/config/containers/resource_constraints/#gpu

Run start-jupyter.sh to start Jupyter lab server on localhost:

sudo ./start-jupyter.sh

or by calling the Docker manually:

sudo docker run --gpus all --rm -it -p 8888:8888 -p 8797:8787 -p 8796:8786 --ipc=host --cap-add SYS_NICE -v $(pwd):/nwsi166-merlin nvcr.io/nvidia/merlin/merlin-tensorflow:22.11 /bin/bash -c "cd / ; jupyter-lab --allow-root --ip='0.0.0.0' --NotebookApp.token=''"

In case the --gpus all is not supported (e.g. because no NVIDIA GPU is present or the docker is not set up to expose it), the command in start-jupyter.sh can be adapted:

sudo docker run --rm -it -p 8888:8888 -p 8797:8787 -p 8796:8786 --ipc=host --cap-add SYS_NICE -v $(pwd):/nwsi166-merlin nvcr.io/nvidia/merlin/merlin-tensorflow:22.11 /bin/bash -c "cd / ; jupyter-lab --allow-root --ip='0.0.0.0' --NotebookApp.token=''"

If successful, the running instance should be found on http://localhost:8888/lab by default. There, you can find examples provided by NVIDIA as a part of the container image. Our experiment can be found in /nswi166-merlin directory: http://127.0.0.1:8888/lab/tree/nwsi166-merlin