Keep your data

Whether your files survive depends on the Workspace storage you pick at launch. Decide before you launch: the setting can’t be changed on a running instance.

Temporary or persistent

Option	What happens to `/workspace`
Temporary workspace (local storage)	Fast local disk on the node. Everything is deleted when the instance is destroyed. This is the default.
Persistent /workspace	Everything under `/workspace` is kept and appears again in your next instance that also uses persistent storage.

Only /workspace is kept. Packages you install with pip, files in your home directory and anything else outside /workspace are gone when the instance is destroyed. New instances start in /workspace.

To pick persistent storage, turn on Customize in the launch form and set Workspace storage to Persistent /workspace.

Capacity

Persistent storage is 25 GB. Some accounts can choose 100 GB large storage under Persistent workspace capacity.

The 25 GB and 100 GB options are two separate volumes. Files saved with one don’t appear when you launch with the other, so stick to one capacity.

What to put under /workspace

Code, notebooks, configuration and results.
Training checkpoints, saved at regular intervals.
A requirements.txt listing the extra packages you need. Packages don’t carry over between instances, so reinstall them after each launch:
Terminal window
```
pip install -r /workspace/requirements.txt
```
Leave torch out of the file, so the image’s ROCm build stays in place.

Large datasets and model weights fill 25 GB quickly. Check usage with du -sh /workspace/* and df -h /workspace.

Mounted models

Mount a model in the launch form mounts a prepared model read-only. Read the weights from there instead of downloading them. To change them, copy them into /workspace first.

Most models are mounted at /models. ComfyUI is mounted at /comfyui_workspace/ComfyUI/models, where ComfyUI looks for its models.

Before you destroy an instance

The destroy confirmation reminds you that temporary workspace data is deleted and persistent workspace data is kept. Before you confirm, move anything you need:

Into /workspace, if the instance uses persistent storage.
To your computer, from the JupyterLab file browser (right-click a file and choose Download) or with scp over SSH.
To a Git repository or your own object storage.