Repository navigation
This release focuses on production deployment of DeePMD-GNN models: MACE exportable-backend support, MPI communication for GNN message passing, CUDA graph ops, and a more robust build/test pipeline.
Highlights
- Added MPI
border_opcommunication support for both MACE and NequIP, enabling parallel periodic LAMMPS simulations through DeePMD-kit's message-passing path. - Added optional MACE cuEquivariance training via
enable_cueq. - Added MACE support for DeePMD-kit's PyTorch exportable backend (required DeePMD-kit v3.2.0):
dp --pt-expt traindp --pt-expt freeze -o model.pt2
- Enabled DeePMD-kit's
torch.compiletraining path for MACE with thept_exptbackend. - Added CUDA support for the
edge_indexop and CUDA-enabled wheel builds. - Switched graph ops to the LibTorch Stable ABI for improved compatibility with modern PyTorch.
Compatibility Notes
- Python 3.10 or newer is now required.
- PyTorch 2.10 or newer is now required.
What's Changed
- feat(op): switch edge index to LibTorch Stable ABI by @njzjz in #114
- chore(deps): bump codecov/codecov-action from 6 to 7 by @dependabot[bot] in #112
- chore(deps): bump pypa/cibuildwheel from 3.4 to 4.1 by @dependabot[bot] in #113
- fix: derive atomic virial via DeePMD output transform by @njzjz-bot in #108
- fix: fix GPU pytest failures by @njzjz in #115
- fix: fix pip install without nvcc by @njzjz in #116
- test: add LAMMPS integration test by @njzjz in #117
- fix: avoid linking torch cuda for CPU op by @njzjz in #118
- ci: use mdformat for markdown by @njzjz-bot in #120
- feat: enable MACE MPI communication in LAMMPS by @njzjz in #119
- ci: add CUDA test workflow by @njzjz-bot in #122
- feat: add CUDA edge_index op by @njzjz-bot in #123
- CI: build CUDA ops for wheels by @njzjz in #127
- Support NequIP border_op communication by @njzjz in #126
- feat: support pt_expt graph model export by @njzjz in #124
- feat: export MACE pt2 with comm artifact by @njzjz in #128
- docs: document pt_expt MACE export by @njzjz-bot in #130
- fix: avoid forcing torch CUDA link for CUDA ops by @njzjz in #131
- feat: dynamically load CUDA runtime for ops by @njzjz in #132
- chore(deps): bump actions/checkout from 6 to 7 by @dependabot[bot] in #133
- feat: enable MACE pt_expt torch.compile training by @njzjz in #134
- feat: add optional MACE cuEquivariance training by @njzjz in #129
- docs: add README highlights by @njzjz-bot in #135
- [pre-commit.ci] pre-commit autoupdate by @pre-commit-ci[bot] in #109
- refactor: Refactor MACE wrapper helper code by @njzjz in #136
Full Changelog: v0.1.4...v0.2.0