In group decision making (GDM) situations, it is quite natural that the decision makers who may have different background and knowledge will provide their preferences by means of different linguistic term sets. Specifically, multi-granular linguistic term sets that are not uniformly and symmetrically distributed will be employed. To deal with this type of GDM problems, this paper proposes a consensus-based GDM model by using two existing 2-tuple linguistic representation models (i.e., the Herrera and Martínez model and the Wang and Hao model), which we called the GDM model based on multi-granular unbalanced 2-tuple linguistic preference relations. First, the framework of the GDM model with multi-granular unbalanced 2-tuple linguistic preference relations is proposed. Then, the transformation function is obtained to relate multi-granular unbalanced linguistic preference relations with uniform balanced linguistic preference relations. Further, a consensus model is presented to help the decision makers reach a consensus. This consensus model not only provides a new way to simultaneously manage individual consistency and group consensus in a linear programming model, but also minimizes information loss (or consensus cost) when reaching the established consensus level. Finally, an example is given to illustrate the feasibility and validity of the proposed model.