@Jan_Strube since we currently have a deeponet in ci already, this is a bit surprising (Reactant.jl/benchmark/nn/neural_operators.jl at 592c165632e9298bbf7fc5b04659c8f0ce7fac5d · EnzymeAD/Reactant.jl · GitHub)?
Maybe you can make a PR including your test case and it can help us see what’s up (And ensure no future regressions)