sidebar_position | title | sidebar_label |
---|---|---|
1.0 |
Overview |
Overview |
For prototyping/debugging/analyzing, sometimes you need a LIVE machine. We call these Sessions.
Session features:
- Change Machine Instance Type
- Use Spot Instance
- Mount multiple GPUs
- Auto-mount datastores
- Pre-installed JupyterLab
- SSH access
- Visual Studio Code access
- Pause the session to save work and pause expenses
- Resume where you left off
Click here for the 1-minute tour of Sessions
The equivalent via the CLI command:
# session with 2 M60 GPUs
grid session create --instance_type 2_m60_8gb
Listing sessions in the CLI:
grid session
The equivalent CLI command:
grid session delete GRID_SESSION_NAME
As we will not be working 24/7 on our sessions it is important to pause them to not consume our credits. Resuming will bring back your session with all packages you have installed previously.
grid session pause GRID_SESSION_NAME
grid session resume GRID_SESSION_NAME
# Note: at the moment it is required that you first pause the session
grid session pause GRID_SESSION_NAME
grid session change-instance-type GRID_SESSION_NAME g4dn.xlarge
More details on how to change Session instance type can be found here