this-cluster-general

name

this-cluster-general

description

Reference card for a single-node SLURM GPU server. General template — fill in all {{PLACEHOLDER}} values in FILL_IN.md before use. Use when the user invokes /this-cluster-general, or automatically before: running local commands on this box, submitting/cancelling SLURM jobs, writing sbatch scripts, choosing a Python interpreter, deploying code locally, or reading job logs. Also consult when other skills need local machine facts. Triggers on: /this-cluster-general, sbatch, squeue, scancel, sinfo, SLURM, 本机, 这台机子, 这台服务器, 提交任务, 跑实验, gpu partition, 本地部署, 本地提交.

This Cluster ({{HOSTNAME}}) — Reference Card

TEMPLATE: Replace every {{PLACEHOLDER}} with real values before using. See FILL_IN.md for the full list of values to fill in.

Route to the right sub-skill based on what you are about to do. Read ONLY the relevant reference file — do not load all of them.

Last verified: {{LAST_VERIFIED_DATE}}

---

Sub-skill Router

When you are about to...	Read this	Why
Access the machine / check identity / network	references/connection.md	Host facts, user/group membership, SSH access, whether we're already ON the box. Getting this wrong = running local commands as if remote, or vice versa.
Run Python or pick an interpreter	references/python-envs.md	The primary venv path, which packages are installed, PyTorch+CUDA version. Wrong interpreter = missing packages / wrong torch / broken GPU.
Submit, cancel, or monitor a SLURM job	references/job-submission.md	SLURM commands, gold-standard sbatch template, partition choices, common pitfalls. Wrong template = wasted GPU hours.
Move code or data into/out of this machine	references/deploy.md	Storage layout, where to put code/data, internet connectivity status.
Check hardware specs or node/GPU status	references/hardware.md	CPU/RAM/GPU specs, disk mounts, SLURM resource view. Required before sizing --mem or --gres.

---

Quick Cheat Sheet

# We're ALREADY on this machine — no SSH needed for normal work.
hostname              # {{HOSTNAME}}

# Submit a job
sbatch /path/to/script.sbatch

# Check queue
squeue
squeue -u {{USERNAME}}

# Cancel a job
scancel <job_id>

# Node / partition info
sinfo
scontrol show node {{HOSTNAME}}

# Run Python (hardcoded venv path — no activate needed)
{{PYTHON_BIN}} script.py

Canonical Paths

Thing	Path
User home	/home/{{USERNAME}}
Python venv	{{VENV_PATH}} (interpreter: {{PYTHON_BIN}})
Main project	{{PROJECT_ROOT}}
SLURM templates	/home/{{USERNAME}}/slurm-templates/
Fast NVMe scratch	{{SCRATCH_PATH}} ({{SCRATCH_SIZE}}, mostly free)
Large bulk storage	{{BULK_STORAGE_PATH}} ({{BULK_SIZE}}, check df -h {{BULK_STORAGE_PATH}} before writing)