GPflow seems to only support multi-output for SVGP. Is it possible to use this multi-output support for other models (e.g. SGPR)? For example:
kernel = mk.SharedIndependentMok(gpf.kernels.RBF(D), P)
feature = features.InducingPoints(X[:M,...].copy())
m = gpf.models.SGPR(X, Y, kernel, feature)
This is an identified issue (https://github.com/GPflow/GPflow/issues/1209), that currently has comparatively low priority for the GPflow core developers - but we'd be very happy for you to join us and contribute features! There is now a public GPflow slack for ease of discussion.