Section 3 — Launch Your First ML Job

SkyPortal simplifies training job launches across any connected infrastructure.

Steps to Launch

1. Select a Host
   Choose from your connected hosts or clusters.

2. Choose Environment
   Pick the appropriate Python environment and container if needed.

3. Define the Job

   • Training script
   • Dataset location
   • Hyperparameters
   • Resource requirements (GPUs, CPUs)

4. Run
   Click Start Job or run the equivalent CLI/agent command.

What Happens Next

• The job is scheduled on your chosen host
• Logs begin streaming live
• Metrics such as loss, accuracy, and resource usage begin populating dashboards
• Budget tracking starts automatically

Best Practices

• Attach datasets via supported storage backends (e.g., S3, MinIO)
• Track experiments for reproducibility
• Use tagging to organize runs