Description
As mentioned in #1428 (comment), there are some issues with the current setup of the documentation. It is decoupled from the repos where it actually belongs to, which leads to redundant, contradicting and outdated documentation (as e.g. in the case of AbstractMCMC - the documentation on turing.ml is outdated and the interface is only described correctly in the README in the AbstractMCMC repo). Additionally, users of DynamicPPL or AbstractMCMC do not know about the documentation and developers of these packages can't properly update the documentation when adding features to their repos apart from changing the README if Turing still uses the previous version. I assume it also leads to the impression that all these packages are designed specifically for Turing which is (not always) the case.
One suggestion would be to move documentation to the repos to which it belongs (similar to what is done in JuliaGaussianProcesses) and only link to these documentations from the Turing webpage. One could link to the documentation of the version that is used by Turing.jl, but it would still be possible for developers to update it when they add new features/change the interface/etc. and users of these packages that do not use Turing could still read the latest version of the docs.