create-slurm-job

Files (2)

• SKILL.md
• job.template.sh

SKILL.md (first lines)

# Create Slurm Job Script

You are helping the user generate a `job.sh` file for the UNIGE HPC clusters
(Baobab, Yggdrasil, or Bamboo). Your job is to ask a short series of focused
questions, then produce a clean, ready-to-submit script.

**Template location:** `skills/hpc-job/job.template.sh`
Read the template first so you understand all the placeholders before you
start asking questions.

---

## Step 0 — Read the template

Before asking anything, read `skills/hpc-job/job.template.sh`.
This is the file you will copy and fill in. Every `__PLACEHOLDER__` in that
file corresponds to one or more questions below.

---

## Step 1 — Ask the user these questions

Work through the questions in order. Ask **all of them in a single message**
so the user can answer in one go. Do not ask one question at a time.

Format your questions clearly, like a numbered form. Provide sensible defaults
in square brackets wherever possible — the user can just press Enter or type
a value to override.

```
I'll create a job.sh for you. Answer these questions (press Enter to use the
default shown in [brackets]):

1. Job name (no spaces, used for output file names)  [myjob]

2. What command will you run?
   e.g. "python3 train.py", "Rscript analysis.R", "matlab -r run_sim"

3. HPC project path — the absolute path to your project on the cluster.
   UNIGE HPC path structure: /home/users/a/USERNAME/PROJECT_NAME
   e.g. "/home/users/a/amiet/tst-llm"
   (This is added as a `cd` before your command so all relative paths work)

4. Which software modules do you need to load?
   e.g. "Python/3.11.3-GCCcore-12.3.0", "R/4.3.2-foss-2023a"
   (Leave blank if you will use conda or a venv — add activation line manually)

5. Partition — choose one:
   [debug-cpu]   15 min — use while testing your script
   shared-cpu    12 h   — most production CPU jobs
   public-cpu    4 days — longer CPU jobs
   public-bigmem 4 days — high-RAM CPU jobs
   shared-gpu    12 h   — GPU jobs under 12 h
   public-gpu    4 days — longer GPU jobs
   → Your choice:

6. Time limit (HH:MM:SS)  [00:15:00]
   Tip: be accurate — over-requesting blocks backfill scheduling.

7. Job type:
   [1] Single-threaded  (one core — Python, R, Stata SE)
    2  Multi-threaded   (one node, N cores — Matlab, Stata-MP, OpenMP, multiprocessing)
    3  MPI              (N tasks across nodes — OpenMPI, OpenFOAM; only if your program uses MPI)
   → Your choice:

8. Number of CPUs / tasks:
   (If single-threaded: how many cores? Usually 1)
   (If multi-threaded:  how many cores does your program use?)
   (If MPI:            how many MPI tasks?)
   → [1]

9. Memory per CPU in MB  [3000]
   Default is 3 GB/core. Increase if your job runs out of memory.
   (For a total memory cap instead, type "total:XXXX")

10. GPU?  [no]
    If yes, specify type and count, e.g. "1" (any), "a100:1", "titan:2"

11. Email notifications when job ends or fails?  [no]
    If yes, enter your email address:
```

---

## Step 2 — Confirm before writing

After the user responds, echo back a brief summary of the settings you
understood, like this:

```
Here's what I'll put in job.sh:

  Job name:   myjob
  Command:    python3 train.py
  Modules:    Python/3.11.3-GCCcore-12.3.0
  Partition:  debug-cpu
  Time:       00:15:00
  CPUs:       1 task × 1 core
  Memory:     3000 MB/core
  GPU:        no
  Email:      no

Shall I write this to job.sh in the current directory? [yes]
```

If the user says yes (or just presses Enter), proceed. If they want to
correct something, apply the correction and confirm again.

---

## Step 3 — Write job.sh

Read the template at `skills/hpc-job/job.template.sh`, then write
`job.sh` in the current working directory with all placeholders replaced.

### Placeholder substitution rules

| Placeholder              | Value                                                              |
|--------------------------|--------------------------------------------------------------------|

From pascalamiet/ai-setup, MIT.

More from pascalamiet/ai-setup