Skip to content

Files & storage

Understanding where your files live is important: some locations persist across workspaces, some persist only within one workspace, and some are temporary.

Every workspace mounts ~/work from the host as /work inside the container.

  • Files written to /work persist across all your workspaces and survive stops, restarts, and even workspace deletion.
  • A different workspace you create later sees the same /work.
  • Put datasets, checkpoints, code, and anything you can’t afford to lose here.

Your home directory inside the container is backed by a per-workspace directory on the host (~/workspaces/<id>/workspace-private). This means:

  • conda environments, pip user-installs, dotfiles, and shell history persist across stop and start of that workspace.
  • When you create a new workspace, the home directory starts fresh - the environments and installs from an old workspace don’t carry over.
  • When you delete a workspace, its per-workspace home is archived under your account. Ask your admin if you need to recover something from it. Your /work is not removed.

If you want to preserve a conda environment across workspaces, keep it under /work or export and reinstall it.

Your admin may have configured shared directories - datasets, pre-downloaded model weights - that appear at fixed paths like /data inside every workspace. These are read-only by default. Ask your admin what paths are available.

Workspaces run as your Linux user with --userns=keep-id. Files you create inside the container appear as your Linux user on the host - no chown required, and no UID mismatch when you access them outside LabPod (e.g. over SSH).

The workspace detail page has a Files button (folder icon) that opens the server-side file manager. From there you can:

  • Browse your /work and per-workspace home directories.
  • Upload files from your local machine.
  • Download individual files or entire directories as a .tar.gz archive - useful for backing up a workspace home before deleting, or for extracting results.
  • Create directories and rename or delete files.

The file manager works even when the workspace is stopped.

LabPod sets HF_HOME=/work/.hf-cache inside every workspace. This means model weights and datasets you download with transformers, datasets, or huggingface-hub are stored once per user under your /work and are automatically shared across all your workspaces - instead of downloading 16+ GB again every time you create a new workspace.

If you want a different cache location, override HF_HOME in workspace Settings → Environment. If your lab shares common weights, ask your admin to configure a shared read-only mount and point HF_HUB_CACHE at it.