Skip to main content
Start with dreamscale doctor (or dreamscale doctor so101, dreamscale doctor yam). Every error names the problem and the next step.

Stuck on Starting

Starting means your robot’s server is getting a GPU and loading the model, usually in under a minute. The CLI, the SDK and dreamscale deployments show <deployment> say what it waits for:
  • Waiting for a GPU: the cloud hasn’t given the server one yet. After about a minute, a second server starts in the next-closest region, and the first one ready serves your robot. To keep a deployment in its own region: dreamscale deployments fallback <deployment> off.
  • Queue: every server is busy. The CLI and SDK show your place.
  • New serving image: after a model update, a deployment built on the previous version passes a short check before it starts, usually a minute or two. The CLI and SDK wait.
dreamscale.connect() gives up after 15 minutes (startup_timeout changes that). Still starting? Email team@dreamscalelabs.com what dreamscale deployments show <deployment> says.

Still stuck?

Email team@dreamscalelabs.com with the model, dreamscale --version and the exact error. Never send your API key.