view-progress

You monitor training run progress by checking live job state, step progress, and latest activity. The Skill surfaces reward scalars and logs so you can track whether your training run is progressing, stalled, or failed.

You inspect logs to debug a stalled or failed training run by retrieving environment or runtime error outputs for a specified run ID. This lets you identify why training behavior is underperforming or not progressing as expected.

Yes, you can compare evaluation performance and reward curves across multiple training runs. You retrieve scalar metrics and status outputs for different run IDs to evaluate which run is performing better.

Yes, monitoring training runs requires castform runs status, scalars, list, logs, and stop commands with JSON output. This deterministic JSON format is necessary for proper inspection and control of active or completed runs.

The best way to stop an underperforming training run you own is to issue a stop command for the specified run ID after reviewing its status and evaluation metrics. This halts runs that are no longer improving.