We include the code for the applications of the results in "Geometry of Sensitivity: Twice Sampling and Hybrid Clipping in Differential Privacy with Optimal Gaussian Noise and Application to Deep Learning" to DP-SGD for privately training nueral networks.