Bayesian regression in PyMC3

I am relatively new to PyMC3 and using it to simply fit some observed data to a linear model and estimate the coefficients. However, the sd values that i get using pm.summary(trace) and the ones that i get in the diagonal of the covariance matrix pm.trace_cov(trace) after sd = np.sqrt(np.diag(cov)) are not same. What am i missing here?
Thanks a lot for the help.

Could you post some code to illustrate the issue (with toy data if necessary)?