Φυtorch is an effort to collect high-performance utilities generally useful to physicists who deal with computations (simulation, data analysis, etc.). It is based on PyTorch and thus offers support for massive parallelism on graphics processing units (GPUs) and seamless automatic differentiation. This chapter briefly introduces its main functionalities before presenting an application to efficient gradient-based variational Bayesian inference from \(\mathscr{O}\) (106) mock SNæ Ia, based on a new analytic formulation of the cosmographic distance integrals for a general Λ-cold dark matter (ΛCDM) model with non-zero curvature and radiation.