monitor-experiment

To monitor machine learning training jobs in remote SSH screen sessions, you can locate running processes, tail recent log outputs, and read JSON result artifacts to verify training progress and retrieve metrics.

Yes, you can pull Weights and Biases history and run summaries to confirm convergence, detect divergence or OOM risk, and link to the dashboard for deeper review of remote training instances.

You can check cloud or Modal app status by verifying running sessions and determining whether work is complete or stalled, helping you decide next steps without guesswork.

The best way to aggregate experiment results is by hard-copying recent log output from each session and inspecting result directories for JSON artifacts to fetch the latest metrics across environments.

No, you do not need to manually inspect logs to detect divergence or OOM risk; the skill optionally pulls W&B curves and run summaries to interpret failures and convergence automatically.

If your monitoring lacks actionable next steps, ensure it produces a comparison-style results summary and an interpretation with suggested next steps based on the gathered logs and metrics.