Run your training command with evidence saved.
Runs a selected training command and saves structured evidence of its progress, checkpoints, metrics, and outcome.
When to use it
Use after selecting a training command, for startup verification, short-run verification, full training kickoff, or resume handling.
Give it a selected training command and repository; it runs the command conservatively and writes training evidence into train_outputs/.
What you provide
The bundled training runner uses Python 3.
Uses its shared operating principles and bundled shared runtime; the runner instructs installation of ai-research-reproduction when the shared runtime is missing.
This skill
NVIDIA devices
Samples NVIDIA device telemetry