ForwardDiff gradient dispatch issue with Float32 ComponentArrays in DiffEqFlux test #1134
Description
Description
The DiffEqFlux test in the Core test suite is failing on Julia lts (1.10) due to a MethodError in the ForwardDiff gradient dispatch when using Float32 ComponentArrays.
Error
MethodError: no method matching (::OptimizationBase.var"#13#43"{DifferentiationInterfaceForwardDiffExt.ForwardDiffGradientPrep{Tuple{...}, ForwardDiff.GradientConfig{ForwardDiff.Tag{...}, Float32, 12, ComponentArrays.ComponentVector{ForwardDiff.Dual{...}, ...}}, ...}, ...})(...)
Affected Test
test/diffeqfluxtests.jl- Neural ODE optimization with BFGS
Workaround
The test has been temporarily skipped in the CI to unblock other tests.
Related
This may be related to DifferentiationInterface or ForwardDiff handling of ComponentArrays with Float32 element types.