Docs / For hosts
Troubleshooting
The messages the installer and the client print, and what to do about each.
| You see | What to do |
|---|---|
nvidia-smi was not found | Install the NVIDIA driver, check that nvidia-smi prints a table, then run the install command again. |
your largest GPU has N GB of VRAM | Ballast needs 8 GB or more. A smaller GPU cannot join. |
Python's venv module is missing | On Debian or Ubuntu, run sudo apt install python3-venv, then install again. |
This machine is not linked yet | Run ballast login and approve the code. |
That code did not match | Codes last 10 minutes and work once. Run ballast login again and use the new code. |
This account already has 4 devices | Remove one from the dashboard first. An account can link 4. |
This device was removed or its token was revoked | Run ballast login to link it again. |
| The dashboard says offline | A device is offline after 30 seconds without a heartbeat. Run ballast status, then journalctl --user -u ballast -f. If the service stops when you log out, run loginctl enable-linger $USER. |
Ollama is not installed | Run the install command again, or install Ollama from ollama.com/download. |
A model never shows loaded | The test completion may be failing, or the pull is still running. ballast status shows loading models and any problem. Check free disk space. |
| The device shows quarantined | Ballast stops sending jobs to a device that fails spot checks. Keep the client running. It recovers after passing three checks in a row. |
| A model is unavailable for a while | The device is over 88 C, or its free VRAM is below the model size. It comes back when that clears. |
Still stuck? Send the output of ballast status through Support. Do not send your token.
Something here wrong or missing? Tell us on Support.