Speeding up bayesian multiple regression

You should use the pairplot to first check where in the parameter space is problematic. My guess is that your intercept is overparameterized - usually we just do intercept = pm.Normal('intercept', 0., 100.).