Volumes

Volumes are persistent, workspace-scoped storage. They are mounted at /mnt/<name> in any sandbox or run that requests them, and survive across runs.

Commands

ts volume create datasets
ts volume put datasets ./imagenet.tar /imagenet.tar
ts volume ls datasets /
ts volume get datasets /outputs/model.pt ./model.pt
ts volume rm datasets /tmp
ts volume delete datasets

Mounting

ts run train.py --gpu h100 --volume datasets --volume checkpoints
ts sandbox create --gpu a100-80 --volume datasets

Files written inside the container are persisted when the process finishes or the sandbox stops. The console lets you browse directories, upload files and download files up to 256 MB; use the CLI for larger transfers.

Environments

Volumes belong to an environment. A volume created in staging is separate from one with the same name in main.