Not important for InferenceData, since potentials don't produce any values that remain in the trace (unless a user wraps them in a Deterministic, which already accept dims), but it would be neat for graphviz.
It's a simple as forwarding dims here:
|
model.add_named_variable(var) |