G-and-k distributions with skewness and kurtosis

You could try a Normal with SinhArcsinh transformation. Again that isn’t available in PyMC3 so would require coding yourself.