Collected sources and patterns will appear here. Add from search or the patterns library.
(Tensor<Irreps>, Tensor<Irreps>) -> Tensor<Irreps>
Compute tensor products of equivariant representations channel-wise (depthwise) before linear channel mixing.
Problem it solves
Standard Clebsch-Gordan tensor products scale quadratically with the channel dimension, causing computational bottlenecks in 3D transformers.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.