Update docker_run.sh #8
Conversation
| docker run --gpus all -it --rm -v `pwd`:/scratch --user $(id -u):$(id -g) \ | ||
| docker run --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 --gpus all \ | ||
| --mount type=bind,source="$WORKDIR",target=/workspace \ | ||
| -it --rm -v `pwd`:/scratch --user $(id -u):$(id -g) \ |
There was a problem hiding this comment.
"-v pwd:/scratch" mounts your current working directory under /scratch. It looks like your change adds another bindmount for workdir under /workspace. So both /scratch and /workspace should be pointing to your host's current dir.
Perhaps you were using paths like /workspace/something.pkl when /scratch/something.pkl would've worked without the above change?
I will comment on the bug. The docker_run.sh is kind of a sample thing that I expect everyone to adapt to their needs. These kinds of scripts never do exactly what multiple people want from docker run, so the simpler they are in a repo like this, the better.. this way it's easier to understand it and fork it to one's own needs.
There was a problem hiding this comment.
yes thanks Janne - /scratch/something.pkl seems fine. Just need some help for newbies in readme
I like the shell wrapper - it helps hide the guff and also get the user credentials correct. I've dived into docker and end up in the bash - then spitting out images which can't be seen by the host / this is solved by the user id parameter - so far - so good - works elegantly.
I want to load a local pickle file into docker so crafted this extra mapping -
I put pickle file in git repo directory / is it best place? maybe not due to git tracking.
I nest it into a workspace folder on docker container to map back to host.
this works and loads fine. If there's a simpler way - I'm all ears.
P.s Do we we need the gpu guff to get 3090 card working at scale?
docker run --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 --gpus all \