xnn.gnn.models.mace_foundation.foundation_to_xnn#

xnn.gnn.models.mace_foundation.foundation_to_xnn(source, head=None, dtype=None, cache_dir=None)[source]#

Resolve, load and convert a foundation checkpoint (or model).

The one-call form behind MACE.from_foundation(). A registered alias, URL or Zenodo DOI goes through the model hub (xnn.common.models.from_pretrained()), which converts the checkpoint once and caches the converted model; a local file or an already-loaded module is converted directly.

Parameters:

source (str, Path or torch.nn.Module) – Alias / URL / DOI / local path of a checkpoint, or an already-loaded mace-torch model.
head (str, optional) – Head to keep for multi-head checkpoints.
dtype (torch.dtype or str, optional) – Final dtype; None keeps the checkpoint’s.
cache_dir (str or Path, optional) – Model hub cache directory (see default_model_cache_dir()).

Returns:

The converted model.

Return type:

MACE

Raises:

TypeError – If source names a model that is not a MACE.