t1-software

T1-LArTPC public data-challenge container (T1-FM foundation-models thrust).

One shared base image (Torch x JAX, plus a ROOT x GEANT4 conda-forge layer used by event-generation subgroup overlays) plus one overlay per detector technology subgroup. Only the LArTPC subgroup exists so far.

t1-software/
+-- base/
|   +-- base.def          # Bootstrap: docker (nvidia/cuda) -> installs Torch + JAX + ROOT/GEANT4 (conda-forge)
|   +-- versions.env       # CUDA_VARIANT, TORCH_VERSION, TORCH_INDEX_URL, JAX_VERSION, ROOT_VERSION, GEANT4_VERSION, ...
+-- subgroups/
|   +-- lartpc/
|       +-- lartpc.def      # Bootstrap: localimage (t1-base) -> clones + pins JAXTPC, pimm-data, goop, DLPGenerator, edep-sim
|       +-- versions.env    # JAXTPC_COMMIT, PIMM_DATA_COMMIT, GOOP_COMMIT, DLPGENERATOR_COMMIT, EDEPSIM_COMMIT
|       +-- tests/
|           +-- test_integration.py         # JAXTPC -> pimm-data HDF5 seam smoke test
|           +-- test_generator_edepsim.sh   # DLPGenerator -> edep-sim seam smoke test (bomb-macro -> H5DLP)
+-- scripts/
|   +-- nightly-build.sh   # builds both images, runs %test + integration test
+-- images/                 # gitignored default output dir (override via $T1_SOFTWARE_IMAGES_DIR)

Building locally

scripts/nightly-build.sh          # cuda12, the only variant built today

Requires apptainer with unprivileged --fakeroot builds (confirmed working without a /etc/subuid//etc/subgid entry on the nodes this has been built on so far). No docker/podman daemon needed or used. Apptainer's build scratch/cache, and the finished .sif images themselves, are staged under $LSCRATCH (required) and only moved into the images dir below once both builds and the %test step succeed -- building a multi-GB image under a small default /tmp fails or thrashes, and writing the finished multi-GB .sif directly to a shared/NFS images dir mid-build is much slower than one local build plus one final move.

Final images land in images/ next to this repo by default (gitignored -- see .gitignore). Set $T1_SOFTWARE_IMAGES_DIR before invoking the script to redirect them elsewhere -- e.g. a workspace embedding this repo under its own src/ typically points it at that workspace's own image store rather than accepting the default. Re-running is idempotent per tag (see below): the same commit on the same day overwrites its own dated file, and always re-points the -latest.sif symlinks.

Runtime environment

The image has one Python: conda-forge CPython 3.11 at /opt/conda, carrying Torch, JAX, PyROOT, and every subgroup package. It is active in every apptainer exec/run/shell with no setup step -- python3, root, dlpgen, edep-sim, and the Geant4 data variables all work directly.

To compile against ROOT/Geant4 (e.g. rebuilding DLPGenerator or edep-sim yourself), first run source /opt/setup-g4root.sh. It adds only conda's compiler settings (CC/CXX/CFLAGS/LDFLAGS, matching the GCC ROOT was built with); these are kept out of the default environment so unrelated builds never pick them up.

Image versioning

Every image nightly-build.sh produces is named:

<layer>-<cuda-variant>-<YYYYMMDD>-<git-sha>[-dirty].sif

• layer: base, or a subgroup name (lartpc; future subgroups add their own).
• cuda-variant: cuda12 today; cuda13 once a driver >=580.65.06 host exists to build and verify it.
• YYYYMMDD: UTC build date.
• git-sha: this repo's short (8-char) commit at build time, suffixed -dirty if the working tree had uncommitted changes. This ties an image back to the exact .def/versions.env recipe that produced it, independent of the upstream pins (JAXTPC/pimm-data/goop commits), which are recorded separately.

Two more files land next to each dated image in the output directory:

• <name>.manifest.json -- build timestamp, cuda variant, this repo's git commit, and the layer's upstream pins (from its versions.env).
• <layer>-<cuda-variant>-latest.sif -- a symlink to the newest dated build for that layer+variant. This is the pointer to follow for "the latest nightly build"; scripts and downstream consumers should target it (e.g. images/t1-lartpc-cuda12-latest.sif) rather than a dated filename, unless they specifically need to pin an exact historical build. The symlink is updated atomically (write-then-rename) on every successful run.

Dated images accumulate across runs (nothing here prunes old builds); that is a deliberate simplification for now, not a limitation.

Adding a new detector-technology subgroup

Copy subgroups/lartpc/ as a template: one <name>.def bootstrapping Bootstrap: localimage / From: {{ BASE_SIF }}, one versions.env for its own third-party pins, and a tests/ directory with at least one integration smoke test proving the subgroup's own data-pipeline seam. Wire it into scripts/nightly-build.sh's subgroup loop.