We study generalized partial functional linear models and aim at conducting inference for these models. A Bahadur representation for both functional and scalar estimates is developed, and we find the joint distribution of the scalar estimates and the functional part to be asymptotically independent. A penalized likelihood ratio test has been proposed to do hypothesis testing relating to the parameters, whose null limit distribution is verified as a normal distribution. Simulation studies provide numerical support for the asymptotic properties. Data of air pollution are used to illustrate our methodology.