Skip to content

Migrate NequIP integration to nequip >=0.7 (NumPy 2 / Python 3.13 support) #82

Description

@k-yoshimi

Background

abICS currently pins nequip >=0.5.6, <0.7 (pyproject.toml) because the abICS
nequip solver/trainer (abics/applications/latgas_abinitio_interface/nequip.py,
nequip_trainer.py) target the pre-0.7 nequip API
(nequip-train / nequip-deploy, nequip.data.AtomicData, nequip.utils.Config,
torch.jit.load("deployed.pth")).

This series uses numpy.in1d, removed in NumPy 2, so it requires numpy<2.
Because Python 3.13 has no numpy 1.x wheels, the NequIP path cannot run on
Python 3.13, and the ActiveLearnNequip CI job is therefore skipped on 3.13
(see #81).

Goal

Migrate the abICS NequIP integration to a numpy-2 / Python-3.13 compatible
nequip (>=0.7, currently up to 0.18.0; requires-python >=3.10, numpy
unconstrained). nequip 0.7.0 is "a major backwards-incompatible update with
breaking changes throughout the code", so this is a real migration, not a
version bump.

Scope (likely)

  • Rewrite nequip.py (solver) and nequip_trainer.py for the new nequip API
    (new CLI nequip-compile / nequip-package instead of nequip-deploy,
    new config format, new data structures).
  • Update nequip-allegro (the pinned 0.3.0 is tied to old nequip; newer
    Allegro has a different package/API).
  • Update tests/integration/active_learn_nequip inputs and install_nequip.sh,
    and re-enable the ActiveLearnNequip CI job on Python 3.13.
  • Update the JA/EN docs and lift the <0.7 cap in pyproject.toml.

References

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions