xnn.common.models.hub.registry.list_models#

xnn.common.models.hub.registry.list_models(cache_dir=None, *, cached_only=False, format=None, architecture=None, tag=None, details=False)[source]#

List the pre-trained models available to load.

The registry (models.json plus register_pretrained()) is listed together with every complete model directory found in the cache, so models fetched by URL or DOI, or copied in from another machine, show up by their cache key.

Parameters:

  • cache_dir (str or pathlib.Path, optional) – Cache to inspect. Defaults to default_model_cache_dir().

  • cached_only (bool, optional) – Only models that load without a download.

  • format (str, optional) – Only models published in this format ("xnn", "mace-torch").

  • architecture (str, optional) – Only this xnn architecture ("mace", "schnet", …).

  • tag (str, optional) – Only models carrying this tag.

  • details (bool, optional) – Return one dict per model (name, architecture, format, cutoff, number of elements and heads, license, cache status, description) instead of the names.

Returns:

Names (sorted), or rows when details=True.

Return type:

list of str or list of dict

Examples

>>> from xnn.common.models import list_models
>>> list_models(architecture="mace")[:2]
['mace-mh-0', 'mace-mh-1']
On this page
Show Source