Part 2 — Submit a Job Array
Part 1 got one job onto the scheduler. This part is the jump from one to many: a single submission that fans out into hundreds of tasks, made safe to rerun, then pushed up against the ceiling the scheduler actually enforces.
10:30 to noon. A short lecture, then the work block.
Hands-On Lab
| What you’ll learn | |
|---|---|
| 1. Hello World Array | One submission, four tasks, four logs — the shortest proof an array works |
| 2. Scale the Loop to an Array | Work out how a loop becomes an array, write both files yourself, run 100 filings |
| 3. Make Your Tasks Rerun-Safe | Make a task skip work it has already done, so a resubmit costs nothing |
| 4. Yen-Slurm Array Limits | Where the array ceiling is, and how to get ~992 filings through a scheduler that caps you at 512 |
| 5. Bonus — GPUs & Local LLMs | Ask Slurm for a GPU, serve an open-weight model on it, and query it without your data leaving the Yens |