sue-run-preparation

To prevent failed ML scale-up experiment runs, you can use a preflight workflow that validates runtime environments, checks path configurations, and prepares datasets before execution. This ensures all dependencies and backend constraints are met prior to launching.

Preparing datasets for HPC scale-up experiments requires enforcing inode-safe dataset formatting rules, such as converting files to LMDB format. This prevents inode quota violations and ensures the dataset is correctly structured for the target sandbox backend.

Yes, preflight checks can be run on both LUMI and RunPod sandboxes. The process applies backend-specific runtime constraints, such as enforcing container-only policies on LUMI, while coordinating environment installation across these diverse HPC and cloud platforms.

The best way to coordinate environment installation and dataset preparation simultaneously is by dispatching parallel subagent tasks. This approach cuts down preflight time for large experiments by handling both processes concurrently rather than sequentially.

No, preflight environment setup does not automatically edit configuration files directly. It returns proposed runtime.yaml configuration patches, ensuring all changes are safely reconciled and applied consistently by the parent process after preflight tasks complete.

You need to enforce strict workspace root paths before a dryrun execution to prevent path mismatches, inode waste, and quota violations. Validating these contracts ensures the runtime environment correctly locates datasets and output directories.