- Go to
Compute Engine -> VM Instances
- Choose resources (CPU, GPU etc.)
- Choose a disk image, you can either:
- Start from a standard image such as Ubuntu
- Start from a custom image we've made, such as
kansal-image-2
which has cuda and PyTorch installed. You can also make your own image and use that.
- Select 'Allow HTTP traffic' and 'Allow HTTPS traffic'.
- Create the image.
Follow Nhan's instructions here https://fastml.slack.com/files/T6RPRPG92/FL217F4RG?origin_team=T6RPRPG92. You can find your instance's IP address on the VM Instances page.
Some notes for using them.
- Can be used just like a standard server you
ssh
into. - Remember to shut them down if you are not using them
Compute Engine -> VM Instances -> <your instance name>
and clickShutdown
.- The instance state will be saved so you can continue from where you left off if you want to start it again.
- You can leave a process running and exit the instance by using, for example, nohup or screen.
- You can start an http server to access your server files from your browser using
python -m http.server
, which can be run in the background with screen or nohup. (e.g.screen -d -m python3 -m http.server 80
)
- Switch to user with
su user
and the home directory/app/
. - You may need to add miniconda to your path:
export PATH="/home/user/miniconda3/bin:$PATH"
- Go nuts.