Maintenance¶
Routine¶
| Frequency | Action |
|---|---|
| Each session | restart_ui.py if UI is unstable |
| After testing | clean_output.py for a clean history |
| Monthly | Check disk space in data/output/ and models/ |
| Updates | git pull + pip install -e ".[local]" |
Disk space¶
| Folder | Typical size |
|---|---|
models/parakeet-*.nemo |
~2.5 GB |
models/mt5-small/ |
~1.1 GB |
| Single job | Source audio + a few MB of text |
queue.db |
KB–MB |
Backup¶
Save:
data/output/jobs/— user resultsmodels/— if you do not want to re-download (optional, re-downloadable)
No need to back up .venv/.
Model updates¶
Remove files/folders in models/ and re-run the download scripts.
Logs¶
- Worker: stdout of the
sbobina workerprocess - FastAPI: terminal where
restart_ui.pyruns - Failed jobs:
job.jsonerrorfield orsbobina jobs show ID
Recovery after crash¶
The worker automatically recovers orphaned running jobs.
Git cleanup¶
Output and models are in .gitignore — do not commit queue.db or audio files.