Adaptivity of Stochastic Gradient Methods for Nonconvex Optimization