Run your first GPU job
Gipper needs a Linux client, Python 3.12+, OpenSSH, and tar with gzip support. Provider accounts supply the GPUs; a missing API key disables that provider for rentals. Your application and dependencies remain part of your job.
Install
uv tool install gipper
gipper --help
Alternatively, use python -m pip install gipper in a virtual environment.
Install uv separately if using the uv commands. The Python runtime has no
third-party dependencies.
This walkthrough uses the example files from the repository:
git clone https://github.com/lavallee/gipper.git
cd gipper
For an installation from this source checkout, use uv tool install . instead
of the package-index command above. See contributing
for development setup.
Configure a provider
mkdir -p ~/.config/gipper
(umask 077; cp -n .env.example ~/.config/gipper/.env)
chmod 600 ~/.config/gipper/.env
Edit that file to set VAST_API_KEY or RUNPOD_API_KEY. Environment variables
override dotenv values. An explicit GIPPER_ENV selects a different file.
Never put API keys in a job, shell command argument, or issue report.
Shadeform is experimental and has not had live qualification.
Inspect before renting
gipper pod estimate examples/smoke.toml --provider vast
gipper pod run examples/smoke.toml --provider vast --dry-run
These commands contact provider APIs but do not rent. Read the estimated phases, fees, and assumptions. The smoke job requests at least 16 GB of GPU memory and limits estimated spend to $0.10 and runtime to 15 minutes. Those are client-side safeguards; provider rounding or failed cleanup can make the bill larger.
Run and verify closure
The next command spends money. Run it only when you intend to rent:
gipper pod run examples/smoke.toml --provider vast
gipper pod ls
The example uploads a text file, runs nvidia-smi, and collects results in
examples/smoke-out/. It checks the transfer and lifecycle, not GPU performance.
Check the provider console if a machine remains listed or closure is uncertain.
Keep cleanup available
mkdir -p ~/.config/systemd/user
cp deploy/gipper-reap.* ~/.config/systemd/user/
systemctl --user daemon-reload
systemctl --user enable --now gipper-reap.timer
The service expects ~/.local/bin/gipper; adjust it if your installation lives
elsewhere. Keep provider credentials in the user-config dotenv so the service
can read them. Timers need a running user manager; user lingering or another
independent scheduler is needed to cover logout. A remote command timeout does
not terminate the rental.
Bring your own workload
Copy the smoke job and change [run] command, [run] setup, [[inputs]], and
[[outputs]]. Set GPU requirements, budgets, and placement hints for your
training, inference, renderer, or other shell-driven application. Use relative
input paths so jobs remain portable. Repeated comparable jobs should share a
workload name; use a separate name or explicit estimates when work changes.
See placement, the Python API, and the security model before running sensitive or long-lived work.