With an increased flexibility in the production new scheduling techniques are necessary to accommodate this change. Though there have already been published many scheduling algorithms fostering this demand for flexibility, there is no common ground on a benchmark data set to compare these approaches against each other. Therefore, this paper aims at the generation of a benchmark data set for the flexible job shop problem (FJSP) with setup and transportation times on which different scheduling algorithms can be evaluated. The data set is specified by several key parameters from which FJSP are created. The use and advantage of the large-scale online scheduling (LOS) data set is exemplified by its application on a Reinforcement Learning online scheduling algorithm and dispatching rules. Furthermore, backward compatibility is established with the former FJSP notation.