- First you need to Get out of following runtime env dependency hell... docker nvidia drivers (sorry if you are on AMD RX GPUs as I'm not sure AMD's rocm has gone mainstream or not..) nvidia-docker nvidia-container-toolkit
you can test this out by running dpkg -l | grep -i nvidia (on ubuntu) docker run --gpus all dvcorg/cml-py3 nvidia-smi
- Start your self-hosted runner docker run --name myrunner -d --gpus all iterativeai/cml:0-dvc2-base1-gpu runner --repo=$my_repo_url --token=$my_repo_token --labels="local,runner" --idle-timeout=1800