nvalchemi.distributed.spec.SPEC_EWALD_HALO

nvalchemi.distributed.spec.SPEC_EWALD_HALO = MLIPSpec(distribution=DistributionSpec(policy=HaloStoragePolicy(scatter_mode='halo_correction', gather_mode='halo_read'), custom_ops=(), third_party_helpers=(), adapters=(), shard_fields=('positions', 'charges', 'atomic_numbers')), owned_only_outputs=frozenset(), all_reduce_outputs=frozenset(), output_kinds={'energy': <OutputKind.PER_GRAPH: 'per_graph'>, 'forces': <OutputKind.PER_NODE: 'per_node'>, 'stress': <OutputKind.PER_GRAPH: 'per_graph'>, 'atomic_energies': <OutputKind.PER_NODE: 'per_node'>}, system_reductions=True, node_energy_key=None, node_virial_key=None, gp_replicate_geometry=False, outputs=None, compile=None)

halo storage. Real-space pair interactions on halo-padded inputs follow the standard halo path. Reciprocal-space dispatch is declarative via custom_ops: stage 1’s handler does owned-slice + all-reduce so the wrapper’s forward stays distribution-agnostic. Per-atom energy scatter to per-system totals routes through per_system_reduce().

Type:

Ewald summation