| Startup takes minutes | Normal for a cold start while the model loads. Keep the command running; don’t retry in parallel |
| Capacity unavailable | No GPU is free in that region. Retry, or use --region available (region="available" in Python) |
| Out of credits | Add credits on the dashboard |
| A session is still open after a crash | dreamscale sessions list, then dreamscale sessions stop <id> |
| Billing continues after a run | --keep-warm keeps the GPU reserved after each task. Use --keep-warm 0, or dreamscale sessions stop --all |
This session ended because its GPU server reached its 24-hour limit. | Reconnect. A deployment kept on Always has a replacement ready |