Thore Kockerols

indeed, and I solved it in a tailored rewrite of the ImplicitFunctions.jl way of doing it (see lines 3000-3100 [here](https://github.com/thorek1/MacroModelling.jl/blob/autodiff-sylvester/src/MacroModelling.jl)). the problem with ForwarDiff.jl and SparseArrays is that value.(SparseVectorA) and...

I don't think that's the case really. As in ForwardDiff with sparse matrices works but the way they are handled within ImplicitDifferentiation doesn't seem to work. I would lean towards...

Exactly! The workaround I used to get it to work in the sparse vector + ForwardDiff case was to extract the values and partials in a different way: ```Julia function...

I don't think that's relevant in the ImplicitDiff case. As in when I tested it against FiniteDifferences it worked out fine. But I might be overseeing something here (head is...

kk i see (now with a bit more brainpower :) ). then there might be a separate issue related to partials not being extracted properly in the implicitdiff code

Great stuff that zygote works. That would be the most efficient backend in the sparse part of my code anyway. @forward diff: at least you get a chance to build...

I used it to get AD working but once I needed sparse matrices I abandoned it again and wrote a bespoke version instead. here is an example to pack multiple...

that would be one way. I have no preference in that regard

I am also not sure how to do it. What you are saying about the generalized eigenvalue decomposition sounds like a good idea and I found this post which seems...

I'm using the package in my tests and am getting ready for 1.9... but it fails due to the issues addressed in this PR would be great to have it...