Accuracy of the distribution obtained by Vi in cmdStanPy

I tried using Vi in cmStanPy but the resulting distribution obtained is quite inaccurate as compared to the one obtained with the default method, is there some way to enhance the accuracy of the results obtained from it?

what do you mean by “enhance the accuracy”?

you might take a look at these papers:
[2108.03782] Pathfinder: Parallel quasi-Newton variational inference

also this:
[1802.02538] Yes, but Did It Work?: Evaluating Variational Inference