Sai Krishna Kishore Nori

Thank you for your comment explaining the reasons for the failure @mcabbott. Can `functors` help in this case to remove the shadow on the parameters, so that their effects can...

It would be great to also if there are fixes for the following error in the above MWE. I had this coming up in other places as well even when...

Thank you for the insights! I ll try to dig deeper with ReverseDiff over ForwardDiff The `Zygote.adjoint` construction for StaticArrays, tried out here (https://github.com/FluxML/Zygote.jl/issues/570#issuecomment-606903246) works. Now the error is gone...

Great, this would fix the problem of name clash when using `BenchmarkTools.jl`

Not a solution, but an other alternative fix for this is ```julia u(x) = prod(sinpi, x.data) ``` like you mentioned it might be best and general to specialize `TensorValue` family...

Just to update, the following variation MWE where we loop over all the keys, is a work around. (So the problem is with the unavailability of rules and methods for...

After some trial and error, I have a generic form of the above work around, for which `Zygote.gradient` works, ```julia function mwe_generic(x::Vector) y = x.^2 collection = Dict(:a => x,...

The above workaround unfortunately doesn't work for `IdDict`, seems like it is hitting a `ccall` which Zygote doesn't propagate through, see the following: ```julia function mfe_IdDict(x::Vector) y = x.^2 collection...

Hi @carlodev, thank you for the possible solution. I'll work it out on paper and get back. At first sight it seems to be alright to me, although I am...

> > Hav't included a test, as I ll then have to include the .msh file as well, or may be I can add a test to GridapGmsh.jl pakcage, which...