Closed frostedoyster closed 3 weeks ago
Difficult to say. The way we plan to use last-layer features with multi-headed models, the features are unitless and the units should belong to the weights of the last layer.
[With one-headed models one could see these features as having units and the weights of the last layer being unitless]
The last-layer features have units of
""
, which however has the meaning of unknown units inmetatensor
and triggers some warnings upon wrapping of models inMetatensorAtomisticModel
. This PR changes the units of the last-layer features to"unitless"
.Contributor (creator of pull-request) checklist
📚 Documentation preview 📚: https://metatrain--255.org.readthedocs.build/en/255/