In many areas of science, complex phenomena are modeled by stochastic\nparametric simulators, often featuring high-dimensional parameter spaces and\nintractable likelihoods. In this context, performing Bayesian inference can be\nchallenging. In this work, we present a novel method that enables amortized\ninference over arbitrary subsets of the parameters, without resorting to\nnumerical integration, which makes interpretation of the posterior more\nconvenient. Our method is efficient and can be implemented with arbitrary\nneural network architectures. We demonstrate the applicability of the method on\nparameter inference of binary black hole systems from gravitational waves\nobservations.\n