Open vmoens opened 4 years ago
This should be working in v0.4.9, added in #531.
I'm working on a new BatchedArray implementation BTW. The rules are defined inside too, you could just use this package: https://github.com/Roger-luo/BatchedArrays.jl (it's not as mature as NNlib) but could be faster.
Would be useful to have the adjoint for batched matrix multiplication (which causes foreigncall error now). e.g.