Mixtures of Poisson Generalized Linear Models for high dimensional count data clustering. The (multivariate) responses can be partitioned into set of blocks. Three different parameterizations of the linear predictor are considered. The models are estimated according to the EM algorithm with an efficient initialization scheme