Not yet. Currently, the way to do this is to write down your log prob function, call it within a pm.model and wrap the resulting tensor with a potential like this:
@pm.model()
def model():
# something
yield pm.Potential(your_log_prob_function(*args, **kwargs))