Environment variables

The repository ships .env.example. Copy it to .env, edit the paths for your allocation, and keep that file private. The example uses snicXXXX-XX-X and USER as placeholders; neither is a real project or account. A model file, SIF, native runtime, and results directory are artifacts outside Git.

For the first Arrhenius notebook run, usually edit MLIP_PROJECT_ROOT in the private .env and check that its derived model, SIF, native runtime, and results paths actually exist. Change an individual derived path only when you placed that artifact elsewhere. The remaining variables below belong to optional builds, scaling jobs, or the other site; do not fill them all in before the first exercise.

Part B artifacts, builds and site runs:

Variable	Example or source	Used for
MLIP_PROJECT_ROOT	/nobackup/proj/disk/snicXXXX-XX-X/personal/USER/mlip-md	Convenience base path in .env.example only.
MLIP_MODEL	$MLIP_PROJECT_ROOT/artifacts/models/mace-mp-0a-small.model	Original pinned MACE checkpoint mounted into the SIF.
MLIP_MLIAP_MODEL	$MLIP_PROJECT_ROOT/artifacts/models/mace-mp-0a-small-mliap.pt	Validated LAMMPS ML-IAP export of the checkpoint.
MLIP_ALCHEMI_SIF	$MLIP_PROJECT_ROOT/artifacts/alchemi-aarch64.sif	ALCHEMI runtime image selected for MD.
MLIP_NATIVE_PREFIX	$MLIP_PROJECT_ROOT/artifacts/lammps-mpi	Extracted native Python and MPI-LAMMPS runtime root.
MLIP_RESULTS_DIR	$MLIP_PROJECT_ROOT/results	Existing private directory for benchmark output.
MLIP_JUPYTER_PORT	8888	Compute-node loopback port; Jupyter token remains private.
MLIP_SIF_OUTPUT	A new .sif path outside Git	Destination when building a new image.
MLIP_NATIVE_ARCHIVE	Pinned native Python archive	Build input containing CPython, headers, shared library, and packages.
MLIP_LAMMPS_SOURCE_ARCHIVE	Pinned LAMMPS source archive	MPI-LAMMPS build input.
MLIP_MPI_BUILD_ROOT	Existing private artifact directory	Destination for a distinct MPI build candidate.
MLIP_MPI_ARCHIVE	Verified MPI runtime archive	Input for the 1/2/4-GPU scaling job.
MLIP_MPI_ARCHIVE_SHA256	64-character digest	Verifies that archive before extraction.
MLIP_JUPITER_ROOT	/e/project1/<PROJECT>/<USER>/mlip-hands-on	Private JUPITER inputs, build candidates, and results, outside Git.
MLIP_JUPITER_BUILD_ID	Successful build job ID	Selects the exact native JUPITER build for Python packaging and runtime checks.
MLIP_WHEELHOUSE_OUTPUT	Fresh private .tar path	Login-node destination for pinned aarch64 Python packages.
MLIP_JUPITER_WHEELHOUSE	Reviewed wheelhouse tar path	Offline Python package input on a JUPITER compute node.
MLIP_JUPITER_WHEELHOUSE_SHA256	64-character digest	Verifies the wheelhouse before installation.
MLIP_JUPITER_ALCHEMI_ENV_ID	Successful environment job ID	Selects one fresh native ALCHEMI environment.
MLIP_JUPITER_REPLICAS	1 or 8	Selects the short one-GPU ALCHEMI functional smoke.
MLIP_ALCHEMI_RUNNER	native on JUPITER, container on Arrhenius	Selects how the same notebook cell starts ALCHEMI.
MLIP_NATIVE_ALCHEMI_PYTHON	Reviewed private Python executable	Runs the pinned native ALCHEMI example on JUPITER.
MLIP_JUPITER_MYST_WHEEL / MLIP_JUPITER_MYST_WHEEL_SHA256	Reviewed private wheel and digest	Input for the private JUPITER MyST renderer; the selected build is hash-checked.
MLIP_JUPITER_JUPYTER_ENV	Fresh private environment path	JupyterLab/Jupytext/Matplotlib environment for the JUPITER notebook server.
MLIP_JOB_LOG_DIR	Owner-only directory outside Git	Token-bearing Jupyter job logs and TLS fingerprint for either site.
MLIP_METATOMIC_SOURCE	Patched private LAMMPS checkout	Pinned Metatomic fork with the reviewed Kokkos compatibility patch.
MLIP_METATOMIC_OVERLAY	Private package directory	Metatomic build/runtime and export dependencies.
MLIP_MACE_EXPORT_OVERLAY	Private package directory	MACE 0.3.14 needed by the experimental Metatomic exporter.
MLIP_METATOMIC_EXPORT_DIR	Fresh private output directory	Receives model.pt, export options, and diagnostic log.
MLIP_JUPITER_METATOMIC_BUILD_ID	Successful build job ID	Selects the Metatomic/Kokkos LAMMPS executable.
MLIP_METATOMIC_MODEL / MLIP_METATOMIC_MODEL_SHA256	Exported file and reviewed digest	Select and verify the Metatomic model before a run.
MLIP_METATOMIC_INPUT	Staged silicon input	Optional 512-atom multi-GPU input; default is 64 atoms.
MLIP_LESSON_ROOT	Private source stage on a site	Selects the exact lesson example scripts inside a site job.
MLIP_LOCAL_RANK	Set by a launcher when needed	Optional rank annotation in the ALCHEMI result. Do not set for a one-GPU notebook.
MLIP_ALLOCATED_CUDA_DEVICES	Set inside the MPI job	Preserves the allocation’s peer-visible GPU list for each rank. Do not set manually.

Job submission and notebook servers:

Variable	Example or source	Used for
MLIP_ACCOUNT	Your current project account	Slurm account for the Arrhenius and JUPITER Jupyter submit scripts.
MLIP_RESERVATION	Empty unless given one	Optional Slurm reservation for the Arrhenius Jupyter job.
MLIP_TIME_LIMIT	02:00:00	Optional Jupyter job time limit, HH:MM:SS.
MLIP_ENV_FILE	Private .env path	Artifact paths sourced by the Arrhenius Jupyter job.
MLIP_NOTEBOOK_VENV	/path/outside/git/mlip-notebook-venv	Private GPU-node notebook environment on Arrhenius.
MLIP_JUPYTER_TLS	Set by the job scripts	Starts Jupyter with TLS. Do not set manually.
MLIP_SSH_CONFIG	$HOME/.ssh/config	Optional SSH config for connect-from-laptop.sh.
MLIP_MACE_MODEL	Original reviewed MACE checkpoint	Input for the JUPITER Metatomic export.

Part A (A1, A2) and Leonardo smokes:

Variable	Example or source	Used for
MLIP_TORCHSIM_CHECKPOINT	<SCRATCH>/models/ MACE checkpoint	MACE-MP-0b small for the A1 and A2 TorchSim jobs.
MLIP_ORB_CHECKPOINT	<SCRATCH>/models/orb-v3-conservative-inf-omat.ckpt	Orb-v3 checkpoint for A2; check its SHA-256.
MATGL_CACHE	MatGL download directory	Where MatGL models are cached (A2).
MLIP_REPLICAS	1 or 8	Leonardo ALCHEMI smoke.
MLIP_LEONARDO_PYTHON	Private native Python 3.12 executable	Leonardo LAMMPS ML-IAP smoke.
MLIP_LEONARDO_LAMMPS_PREFIX	Verified CUDA-Kokkos build	Leonardo LAMMPS ML-IAP smoke.
MLIP_LEONARDO_WRAPPER_ROOT	Matching LAMMPS Python wrapper root	Leonardo LAMMPS ML-IAP smoke.
MLIP_LEONARDO_OVERLAY_ROOT	Matching CuPy overlay root	Leonardo LAMMPS ML-IAP smoke.

LUMI (A3 and A4):

Variable	Example or source	Used for
MLIP_TRAINING_DIR	<SCRATCH>/mlip-training	Venv, data and results for A4, prepared by lumi-training-setup.sh.
MLIP_TASKS	eform finetune (default)	Which A4 examples the job runs.
MLIP_FLOAT_BITS	64 in the LUMI job	A4 float width; default 32.
MATGL_FLOAT_BITS	64 in the LUMI job	A3 float width; default 32.

LUMI (A6, Orb and UMA):

Variable	Example or source	Used for
HF_TOKEN_PATH	$HOME/.cache/huggingface/token	Hugging Face token for the gated UMA weights (licence accepted on the model page).
HF_HOME	<SCRATCH>/mlip-uma/hf-home	Hugging Face cache, kept off home.
FAIRCHEM_CACHE_DIR	<SCRATCH>/mlip-uma/fairchem-cache	UMA checkpoint cache.
CACHED_PATH_CACHE_ROOT	<SCRATCH>/mlip-uma/cached-path	Orb-v3 and OrbMol weights cache.

The scripts check additional standard site variables such as SLURM_JOB_ID, SLURM_NTASKS, and CUDA_VISIBLE_DEVICES. Their values come from the allocation, not .env. Recheck the actual site policy and artifact hashes before a class; this page does not authorize a scheduler submission.