Closed devmotion closed 1 year ago
Honestly, we could also just completely remove the AD-related stuff in Bijectors.jl. It's only relevant for the ADBijector
or whatever it's called, which I don't think is actually used by, well, anyone :shrug:
EDIT: We would still want to support Enzyme, so not arguing against this PR. But just saying the setadbackend
should IMO just be thrown away at some point.
Closed in favour of #242
I assume there will be quite a few test failures.
(Ideally, we would use some interface package for the AD packages instead of defining them in every package but that's the easiest way to move forward with Enzyme support in Turing...)