search

lucidrains / equiformer-pytorch

Implementation of the Equiformer, SE3/E3 equivariant attention network that reaches new SOTA, and adopted for use by EquiFold for protein folding
MIT License
247 stars 23 forks source link

Enable `input_degrees > 1` #11

Closed anton-bushuiev closed 1 year ago

anton-bushuiev commented 1 year ago

Currently, setting input_degrees > 1 and passing a fiber dict to forward leads to a bug because forward tries to apply token embedding and positional embedding to a dict rather than a tensor with type-0 features. This PR fixes this minor issue.

lucidrains commented 1 year ago

@anton-bushuiev thank you for this!