Error using Enzyme to autodiff wrt Flux neural net parameters

General Usage
6

If that’s the case, then I really think once should heed the advice of option a then (rewriting the code to not be conditionally active). It turns out that Enzyme’s activity analysis proved that a lot of code didn’t need to be differentiated (and was hindered by seemingly this?

Stacktrace:
  [1] _trainmap
    @ ~/.julia/packages/Optimisers/a4OnF/src/destructure.jl:114
  [2] _Trainable_biwalk
    @ ~/.julia/packages/Optimisers/a4OnF/src/destructure.jl:110
  [3] ExcludeWalk
    @ ~/.julia/packages/Functors/LbNAu/src/walks.jl:126

In particular it believes it is storing a non-differentiable matrix into a larger data structure which it thinks is differentiable (perhaps conservatively).

@mcabbott does this code ring any bells or give hints as to where something is going awry?