experiment-execution

You can execute distributed training experiments locally or remotely by specifying explicit inputs for execution targets, project roots, and node configuration. This ensures experiments run safely and reproducibly across different environments.

ML experiment execution handles failure recovery through evidence-backed error handling and checkpoint-based replanning. It automatically captures all logs and manages reruns, ensuring jobs recover mid-way if they fail.

Mode-aware interaction adapts confirmation prompts based on your selected execution mode, such as full-auto, moderate, or detailed. This controls the level of manual confirmation required during experiment launches and monitoring.

Yes, remote execution for MLOps workflows maintains clear traceability and reproducibility. It requires explicit inputs for proxy and tracking configuration to ensure all experiment runs are logged and verifiable.

You must provide explicit inputs for the execution target, project roots, node configuration, proxy, and tracking. These inputs enable controlled execution and manage experiment launches with specific environment and runtime settings.

Failed reruns during remote execution often result from missing explicit environment controls or tracking inputs. The Skill uses smoke validation and evidence-backed error triage to enforce reproducibility and diagnose unknown errors.