experiment-run

Automated GPU training runs require pre-flight checks to verify environment readiness and GPU availability before dispatching execution locally or remotely. This process ensures metadata creation, launch validation, and persistent logging for each experiment.

Remote experiment dispatch sends training runs to the best available machine rather than executing locally. It integrates GPU selection, environment setup, and automated launch sequences to streamline orchestration across different hardware resources.

Yes, experiment execution supports both remote and local launch modes. The system dispatches to the best available machine but can fall back to running locally, provided the pre-flight checks confirm local GPU availability and environment readiness.

Persistent experiment logs are generated automatically in logs/experiments/{exp_id}.yaml format. This standardized schema captures the experiment state, metadata, and traceability details for each run, ensuring consistent logging across automated workflows.

No, automated experiment workflows handle GPU selection automatically during the pre-flight checks. The system verifies GPU availability and environment readiness autonomously, selecting the best available GPU for the training run without manual configuration.

Automated experiment orchestration depends on successful pre-flight validation; if environment readiness or GPU availability checks fail, the launch is blocked. Additionally, persistent logging requires write access to the logs/experiments directory for metadata storage.