Can the variance-covariance matrix for the GLMs be exposed to Python/R? It's being calculated behind the scenes already if compute_p_values is set to True.