This article studies the leaderless consensus problem of heterogeneous multiple networked Euler-Lagrange systems subject to persistent disturbances with unknown constant biases, amplitudes, initial phases, and frequencies. The main characteristic of this study is that none of the agents has information of a common reference model or of a common reference trajectory. Therefore, the agents must simultaneously and in a distributed way: achieve consensus to a common reference model (group model); achieve consensus to a common reference trajectory; and reject the unknown disturbances. We show that this is possible via a suitable combination of techniques of distributed 'observers,' internal model principle and adaptive regulation. The proposed design generalizes recent results on group model learning, which have been studied for linear agents over undirected networks. In this article, group model learning is achieved for Euler-Lagrange dynamics over directed networks in the presence of persistent unknown disturbances.