I have some sysadmin and DevOps experience, but pretty much a newbie in ML and Python, so please bear with me:)
I have set up a docker-compose based JupyterHub deployment for our machine learning team, and we can access the notebooks normally. But since the machine does have a A5000 GPU, of course we want to use it.
Since the default notebook images don’t have GPU support, I tried to use the tensorflow/tensorflow:latest-gpu-jupyter image, but the spawner failed with some error about “start-singleuser.sh” not existing.
After some digging I realized that I probably need to change the DOCKER_SPAWN_CMD variable, but have no idea about what to change it to in the case of the above image.
Could someone help me out here?
If using other images isn’t an option, would it be possible to change the jupyter/tensorflow-notebook image to have GPU support? It’d be great if someone could point me to an example, since I’m pretty new to all of this.