kermt-monitor

Track detached KERMT workflow progress by parsing log files and querying Docker container status to extract epoch, step, and validation loss metrics automatically, eliminating manual log inspection.

Track epoch, step, and validation loss metrics during pretraining and finetuning by parsing log files and querying Docker container status, providing live progress updates and automated summary reporting upon completion.

Yes, Docker and jq are required dependencies. Docker queries container status, uptime, and exit codes, while jq parses run manifests and extracts structured data from log files to enable real-time workflow monitoring.

Point the monitoring skill to your run directory (e.g., runs/continue-pretrain-2026-05-17) to parse associated log files and query Docker container status, reporting real-time progress, container health, and ETA calculations.

Use automated log parsing and Docker container querying to generate summary metrics, ETA calculations, test results, and TensorBoard launch commands upon workflow completion, replacing manual log inspection for detached training runs.

Monitoring fails when Docker or jq is missing from the environment. Docker is required to inspect container environments and query status, while jq extracts structured data from run manifests; absence of either prevents metric extraction.