Oh that’s interesting. I didn’t know about this issue. This is indicating that QuadratureAdjoint is only compatible with f(du,u,p,t) forms and not f(u,p,t). That seems to be a missing spot in the testing matrix and would be worth making an MWE and issue to fix.
That said,
If you were using an implicit method with a Newton solver, like TRBDF2, yes it would give a major speedup. With VCABM, no it’s not going to give one in general. So InterpolatingAdjoint with ZygoteVJP is a good choice for that kind of setup + model.
And this complexity is why I am trying to push more and more into just building better default handling because the choice of adjoints is far more complex than most people should have to deal with 