Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion Makefile
Original file line number Diff line number Diff line change
Expand Up @@ -82,7 +82,7 @@ py-test: py-dev

py-test-benchmarks: py-dev
@command -v $(UV) >/dev/null 2>&1 || (echo "uv not found; install from https://docs.astral.sh/uv/" && exit 1)
cd atompack-py && UV_CACHE_DIR=$(UV_CACHE_DIR) $(UV) run --extra dev --locked pytest tests/benchmarks
cd atompack-py && UV_CACHE_DIR=$(UV_CACHE_DIR) $(UV) run --extra dev --extra benchmarks --locked pytest tests/benchmarks

docs-sync:
@command -v $(UV) >/dev/null 2>&1 || (echo "uv not found; install from https://docs.astral.sh/uv/" && exit 1)
Expand Down
8 changes: 5 additions & 3 deletions atompack-py/benchmarks/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -64,7 +64,8 @@ Notes:
- `get_molecules_flat(...)` remains the fastest tensor-batch path when the
consumer can work with concatenated arrays directly.
- `hdf5_soa` reflects the usual chunked-dataset HDF5 layout used for
fixed-shape materials data and is part of the default benchmark environment.
fixed-shape materials data; install with `pip install atompack-db[benchmarks]`
to pull in the `h5py` dependency.
- `lmdb_packed` now participates in the synthetic custom-property variants in
this suite too; `lmdb_soa` remains the builtins-only helper baseline.
- In this suite, `hdf5_soa` is intentionally read one molecule at a time so it
Expand Down Expand Up @@ -330,8 +331,9 @@ tuning that matters most for dataloader-like access:
- reader-side raw chunk cache sizing through `h5py` open options
- batch reads that regroup requested indices by chunk before rebuilding payloads

This backend is included in the default benchmark environment through the base
project dependencies.
This backend depends on `h5py`, which is shipped as the `[benchmarks]` extra:
install with `pip install atompack-db[benchmarks]` (or `make py-test-benchmarks`
which now uses both the `dev` and `benchmarks` extras).

### `atom_lmdb_soa.py`

Expand Down
7 changes: 6 additions & 1 deletion atompack-py/pyproject.toml
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,6 @@ classifiers = [
"Programming Language :: Python :: 3.14",
]
dependencies = [
"h5py>=3.10",
"huggingface_hub>=0.24",
"numpy>=1.20",
]
Expand All @@ -51,6 +50,12 @@ dev = [
"mypy",
"ruff",
]
# Used only by the comparison scripts under atompack-py/benchmarks/. Not
# imported anywhere under python/atompack/, so it has no business in the
# base wheel — pulls ~50 MB of HDF5 binaries onto every install.
benchmarks = [
"h5py>=3.10",
]

[tool.maturin]
python-source = "python"
Expand Down
12 changes: 7 additions & 5 deletions atompack-py/uv.lock

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

Loading