monitor-experiment

You can monitor running experiments across SSH servers, Vast.ai instances, and Modal apps by checking live statuses and capturing logs to verify progress without manual intervention. The skill aggregates these checks into a structured summary.

Collecting experiment results from Vast.ai involves fetching latest logs, discovering JSON outputs, and enforcing deterministic checks like exit codes and epoch logging. This process aggregates final results for review without manual log scraping.

Yes, you can pull Weights and Biases metrics when checking experiment status if WandB is enabled. The skill supports optional WandB data retrieval to fetch evaluation metrics alongside standard log capture and JSON result discovery.

Yes, this experiment monitoring approach works with Modal apps. It covers live status checks for running, completed, or failed apps, enforcing deterministic checks like file presence and exit codes to ensure accurate progress verification.

To verify if an experiment failed or completed on an SSH server, the skill enforces deterministic checks using exit codes, file presence, and epoch logging. It outputs a structured summary detailing the live status of each experiment.

Deterministic checks are important for experiment monitoring because they rely on exit codes, file presence, and epoch logging to accurately verify progress. This prevents ambiguous status reporting and ensures reliable result aggregation across platforms.