Best practices
Contents
Best practices#
How to clone a repo#
One of the billions of how to guide: https://docs.gitlab.com/ee/gitlab-basics/start-using-git.html \n
How to run a script on server#
Before running a script, check if nobody else is using a GPU with nvidia-smi or even better with nvtop. With this command, you will see the active jobs running in the server. If a particular GPU is free, you can run your script reserving the specific card with these commands:
export CUDA_VISIBLE_DEVICES=<gpu-id>
YOUR SCRIPT TO BE EXECUTED
or if you want to export the environment variable just for the command you are running with
CUDA_VISIBLE_DEVICES=<gpu-id> YOUR SCRIPT TO BE EXECUTED
For example:
From this image, it is evident that the GPU with id 0 has a running job (python) of almost 11 GB of memory, while the GPU 1 has no running jobs. So you can run your script python main.py, with
CUDA_VISIBLE_DEVICES=1 python main.py
or
export CUDA_VISIBLE_DEVICES=1
python main.py